Studying Wind Energy/Bird Interactions: A Guidance Document · This document is organized in two parts. Part I (chapters 1-2) presents the general reader with a framework for considering

Prepared for the Avian Subcommittee and NWCCDecember 1999

STUDYINGWIND ENERGY/BIRD INTERACTIONS:

A GUIDANCE DOCUMENTMETRICS AND METHODS FOR DETERMINING OR MONITORING POTENTIAL IMPACTS

ON BIRDS AT EXISTING AND PROPOSED WIND ENERGY SITES

AcknowledgmentsPrincipal Authors:

Richard Anderson, California Energy CommissionMichael Morrison, California State University, Sacramento

Karin Sinclair, U.S. Department of Energy/National Renewable Energy LaboratoryDale Strickland, WEST, Inc., Cheyenne, Wyoming

Contributing Authors:

Holly Davis, U.S. Department of Energy/National Renewable Energy LaboratoryWilliam Kendall, USGS Patuxent Wildlife Research Center

(formerly with U.S. Fish and Wildlife Service) Avian Subcommittee of the NWCC

With the full support of the NWCC Avian Subcommittee, Robert Thresher and Holly Davis of NREL developed the strategy for making this document a reality. They also selected a writingteam and a review team. Abby Arnold and Janet Stone of RESOLVE, Inc. of Washington DCfacilitated this work and obtained commitments from the writing and review team members.

Review Team:

Donald Bain, Oregon Department of Natural ResourcesKeith Bildstein, Hawk Mountain SanctuarySidney Gauthreaux, Clemson University

Walter George, Bureau of Land ManagementThomas Gray, American Wind Energy Association

Lawrence Mayer, Good Samaritan Regional Medical CenterKenneth Pollock, North Carolina State University

Steven Ugoretz, Wisconsin Department of Natural ResourcesW. John Richardson, LGL Ltd., King City, Ontario, Canada

Handbook compilation, editing and review facilitation providedby Susan Savitt Schwartz, Editor

NWCC Logo and Document DesignLeslie Dunlap, LB Stevens Advertising and Design

Document Layout and ProductionSusan Sczepanski, National Renewable Energy Laboratory

Cover photo courtesy ofNational Renewable Energy Laboratory

Printed on recycled paper

STUDYINGWIND ENERGY/BIRD INTERACTIONS:

A GUIDANCE DOCUMENTMETRICS AND METHODS FOR DETERMINING OR MONITORING POTENTIAL IMPACTS

ON BIRDS AT EXISTING AND PROPOSED WIND ENERGY SITES

Prepared for the Avian Subcommittee and NWCC

National Wind Coordinating Committeec/o RESOLVE1255 23rd Street, Suite 275Washington, DC 20037www.nationalwind.org

December 1999

PrefaceThis guidance document is a product of the Avian Subcommittee of the National Wind CoordinatingCommittee (NWCC). The NWCC was formed in 1994 as a collaborative endeavor composed of representativesfrom diverse sectors including electric utilities and their support organizations, state utility commissions, statelegislatures, consumer advocates, wind equipment suppliers and developers, green power marketers, environ-mental organizations, and state and federal agencies. The NWCC identifies issues that affect the use of windpower, establishes dialogue among key stakeholders, and catalyzes appropriate activities to support the devel-opment of an environmentally, economically and politically sustainable commercial market for wind energy.

The NWCC Avian Subcommittee was formed to better understand and promote responsible, credible, and com-parable avian/wind energy interaction studies. In addition to this document, the National Wind CoordinatingCommittee will be placing wind energy-related materials on its web site: www.nationalwind.org

For comments on this guidance document or questions on wind energy permitting, contact the National WindCoordinating Committee Senior Outreach Coordinator c/o RESOLVE, 1255 23rd Street NW, Suite 275,Washington, DC 20037; phone (888) 764-WIND, (202) 944-2300; fax (202) 338-1264; [email protected].

ii Studying Wind Energy/Bird Interactions: A Guidance Document

Table of Contents

EXECUTIVE SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1Site Evaluation Biology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2Basic Experimental Design and Level 1 Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2Advanced Experimental Design and Level 2 Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3Risk Reduction Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4

PART I

CHAPTER 1: INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6Purpose and Scope of this Document . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7Metrics, Methods and Study Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9Organization of this Document . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9

CHAPTER 2: SITE EVALUATION BIOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11Site Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12Gathering Additional Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16Establishing More Rigorous Research Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16

PART II

CHAPTER 3: BASIC EXPERIMENTAL DESIGN AND LEVEL 1 STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29The Philosophy of Study Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23Design/Data-Based Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25Improving the Reliability of Study Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36Case Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41Summary and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .47

CHAPTER 4: ADVANCED EXPERIMENTAL DESIGN AND LEVEL 2 STUDIES . . . . . . . . . . . . . . . . . . . . . . . .50Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51Manipulative Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51Conceptual Framework for Population Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .53Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55Survivorship and Population Projections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .60Determining Cumulative Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .63Recommendations for Study Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .63

CHAPTER 5: RISK REDUCTION STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65Assessing Avian Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65Review of Wind Energy Plant Hazards to Birds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71Risk Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .72Study Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .73

APPENDIX A: LITERATURE CITED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77

APPENDIX B: INDEX OF KEY TERMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .85

APPENDIX C: NATIONAL WIND COORDINATING COMMITTEE MEMBERS AND KEY PARTICIPANTS . .87

iii

List of FiguresFigure 3-1. Idealized sketch of an impact indicator in a Before-After Design with five time periods (T) of interest where an abrupt change coincides with an impact and is followed by a return to baseline conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28

Figure 3-2. Sketch of point estimates of an impact indicator in an idealized BACI design over four time periods with slight indication of recovery after the incident. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .29

Figure 3-3. Sketch of point estimates of an impact indicator in an idealized BACI design where interaction with time indicates recovery from impact by the third time period following the incident. . . . . .29

Figure 3-4. Idealized sketch of results from a Reference-Impact Design where a large initial difference in the impact is followed by a shift to parallel response curves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30

Figure 4-1. Interaction between the number of fatalities per bird passes within 50 meters of treated and untreated turbines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52

List of TablesTable 1-1. Examples of Avian Research at U.S. Wind Energy Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10

Table 3-1(a-c). Recommended decision matrix for design and conduct of Level 1 studies. . . . . . . . . . . . . . .49

iv Studying Wind Energy/Bird Interactions: A Guidance Document

INTRODUCTIONIn the 1980s little was known about the potentialenvironmental effects associated with large scalewind energy development. Although wind turbineshave been used in farming and remote locationapplications throughout this country for centuries,impacts on birds resulting from these dispersed tur-bines had not been reported. Thus early wind energydevelopments were planned, permitted, constructed,and operated with little consideration for the poten-tial effects on birds.

In the ensuing years wind plant impacts on birdsbecame a source of concern among a number ofstakeholder groups. Based on the studies that havebeen done to date, significant levels of bird fatalitieshave been identified at only one major commercialwind energy development in the United States.Research on wind energy/bird interactions hasspanned such a wide variety of protocols and vastlydifferent levels of study effort that it is difficult tomake comparisons among study findings. As a resultthere continues to be interest, confusion, and con-cern over wind energy development’s potentialimpacts on birds. Some hypothesize that technologychanges, such as less dense wind farms with larger,slower-moving turbines, will decrease the number ofbird fatalities from wind turbines. Others hypothe-size that, because the tip speed may be the same orfaster, new turbines will not result in decreased birdfatalities but may actually increase bird impacts.Statistically significant data sets from scientificallyrigorous studies will be required before eitherhypothesis can be tested.

Purpose and Scope of This DocumentBird mortality is a concern and wind power is apotential clean and green source of electricity, mak-ing study of wind energy/bird interactions essential.An important first step in understanding these inter-actions and assessing potential effects is to use thesame terminology and conduct research that willproduce credible and comparable results. This guid-ance document seeks to:

1. Provide a reference document for use by allstakeholders that will, if followed, produce abody of information adequate to:

• assess the suitability of a proposed wind plantsite with regard to birds of concern

• assess the potential effects of a wind plant onbirds of concern

• evaluate the potential effects of wind energytechnology on birds

2. Provide sufficiently detailed and clearly under-standable methods, metrics, and definitions foruse in the study of wind energy/bird interactions

3. Promote efficient, cost-effective study designs,methods, and metrics that will produce compa-rable data and reduce the overall need for somefuture studies

4. Provide study designs and methods for the col-lection of information useful in reducing risk tobirds in existing and future wind plants

There is no “cookbook” approach to research. Notall jurisdictions will require information on birds orbird research in conjunction with permitting a windenergy development. Many situations will requiresite-specific knowledge and expert recommenda-tions on how to proceed with study design andmethods. This document provides an overview forregulators and stakeholders concerned with windenergy/bird interactions, as well as a more technicaldiscussion of the basic concepts and tools for study-ing such interactions.

Organization of the DocumentThis document is organized in two parts.

Part I (chapters 1-2) presents the general reader witha framework for considering wind energy/bird inter-actions, which typically are studied within the con-text of wind energy development site screening,selection, permitting, and project operation.

Executive Summary 1

Executive Summary

• Chapter 1 - Introduction

• Chapter 2 - Site Evaluation Biology

Part II (chapters 3-5) provides detailed discussion ofmetrics, methods, and study design issues for basicand advanced wind energy/bird interaction studies.Geared toward the more technical reader, thesechapters are intended to provide regulatory staff andtechnical advisors to the various stakeholders with a common understanding of what constitutes scien-tifically rigorous research methodology and its applications.

• Chapter 3 - Basic Experimental Design and Level1 Studies

• Chapter 4 - Advanced Experimental Design andLevel 2 Studies

• Chapter 5 - Risk Reduction Studies

Additional sections at the end of the documentinclude a list of Literature Cited, and an Index of KeyTerms.

SITE EVALUATION BIOLOGYGiving adequate consideration to bird resourcesearly in the site evaluation process can reduceexpense, project delays, and stakeholder frustration,and help in complying with permitting and legalrequirements. Local expertise and advice may provequite valuable in determining what information isrequired by regulatory and permitting agencies. Abrief written assessment for each site being evaluat-ed should include information obtained from:

1. sources of existing information, including localexpertise, literature searches, and naturalresource database searches for sensitive speciesor for areas used by a large number of birds

2. reconnaissance studies

3. vegetation mapping, habitat evaluation and the use of information about wildlife habitatrelationships.

In many cases, existing information is adequate todetermine whether a site is biologically suitable orunsuitable for wind energy development. In somecases, on suitable sites, the existing information willbe adequate and defensible for regulatory and envi-ronmental law purposes. If not, the developer andpermitting agency may want to discuss additionalinformation needs and specify objectives. On-site

surveys and monitoring using approprate sampledesign, metrics, and methods can supply short- or long-term information needs effectively and efficiently. Additional on-site information-gatheringmay focus on:

• species of special concern

• breeding bird species

• migrating birds

• wintering birds

• nocturnal vs. diurnal bird activity

• species known to be susceptible to collision

• special situations.

Again, bird biological information must be clearlydocumented and sufficient for making reasonableestimates of bird impacts.

BASIC EXPERIMENTAL DESIGN ANDLEVEL 1 STUDIESLevel I studies should detect major impacts on birdsand assist in the design of wind energy projects toreduce these impacts where necessary. Constructionof a wind plant is not a random occurrence.Potential wind plant sites are relatively unique, creating the potential for study design problems.Moreover, many of the issues related to wind plantimpacts on birds are based on relatively rare events.Determination and analysis of impacts thus will sel-dom be based on clear-cut statistical tests, but ratheron the weight of evidence developed from the studyof numerous impact indicators, over numerous timeperiods, at numerous wind plants.

Protocols for bird studies will, by necessity, be site-and species-specific. They will be influenced by thestatus of the wind energy project, the area of interest,the issues and species of concern, cooperation oflandowners, and also by budget considerations andavailable time.

Summary of Recommendations forDesigning Level 1 Studies:

1. Clearly define: study objectives (questions to beanswered), the area, the species, and the timeperiod of interest; the area of inference, theexperimental unit (and sample size), and thesampling unit (and subsample size); and theparameters to measure.

2 Studying Wind Energy/Bird Interactions: A Guidance Document

2. Select relatively uncorrelated impact indica-tors, measure as many relevant covariates asfeasible, and identify obvious biases.

3. The Before-After Control Impact (BACI) designis preferred. Collect data for two or more timeperiods before and again after construction onthe assessment area (wind plant) and multiplereference areas. Consider matching pairs of sam-pling units (data collection sites) within eachstudy area based on criteria which are relativelypermanent features.

4. Use a probability sampling plan; stratify on rela-tively permanent features and only for short-term studies. Use a systematic sampling plan forlong-term studies; spread sampling effortthroughout area and time periods of interest,and maximize the number of experimental units(sample size).

5. Develop detailed standard operating proce-dures (SOPs) prior to the initiation of field work,and select methods that minimize bias.

6. Make maximum use of existing data and consid-er some preliminary data collection where littledata exists.

7. When data are unavailable before constructionthen combine multiple reference areas withother study designs, such as the gradient-response design.

8. Maximize sample size within budgetary constraints.

9. Univariate analysis is preferred, especiallywhen relying on weight of evidence.

10. Have the plan peer reviewed with an emphasison developing comparable and credible information.

Usually, Level I studies will serve to focus futureresearch on areas if significant biological impactsappear likely.

ADVANCED EXPERIMENTAL DESIGNAND LEVEL 2 STUDIESTesting hypotheses generated by the results ofLevel 1 studies requires more in-depth (Level 2) studies, including both manipulative experimentsand modeling techniques.

Manipulative experiments. Observational studiescan be used to evaluate risk reduction managementoptions for existing and new wind plants. However,by allowing control of such factors as natural envi-ronmental variation which tend to confound obser-vational studies, manipulative experiments couldsignificantly improve the understanding of how these factors relate to the risk of bird collisions withturbines.

Conceptual framework for population modeling.A population is quantified in terms of birth rate,death rate, sex ratio, and age structure. The spatialstructure of a population has an important role ingenetics, and ultimately, survival. Because most“populations” actually are metapopulations com-posed of many subparts, even impacts occurring in a small geographic area can disrupt immigration andemigration between local subpopulations, resultingin a much wider effect on the population than isimmediately evident. Moreover, small impacts canhave serious consequences for the persistence ofsmall populations.

Survivorship and population projections. Wildlifepopulation projections can be made using variousmodels which provide a numerical tool for deter-mining growth rate and age structure of populations,facilitating growth projections.

A review of major wildlife and ornithological studiespublished during the past 20 years suggests that onlyvery broad generalizations can be drawn regarding“normal” survival rates of bird populations. Becauseinteryear variability in survivorship is large even inhealthy populations, the value of short-term (1-2year) evaluations of a population of concern is ques-tionable. The literature indicates that even a relative-ly minor change in survivorship can have substantialpopulation impacts, and that in most cases adult sur-vivorship is critical to maintaining a viable popula-tion. These studies indicate the importance ofdetermining survivorship in evaluating the effects ofwind plants on birds, and suggest the value of mod-eling structures in guiding this determination.

Determining cumulative effects. The cumulativeeffects of a wind plant on a population over timecould apply to the birds in and immediately aroundthe wind plant, or could manifest itself in popula-tions or subpopulations some distance away throughchanges in immigration and emigration. The cumu-lative effects resulting from the expansion of anexisting wind plant also are extremely difficult to quantify in the field without a tremendous

Executive Summary 3

expenditure of time and funds. Establishing a rigor-ous and focused modeling framework becomesessential for hypothesizing the potential impactsgiven a variety of scenarios. In this way, inferencecan be drawn from data collected over the shortterm as it applies to likely longer-term impacts usingprojections of various population models.

Recommendations for Level 2 Study Design1. Develop a sound modeling framework initially

to prevent the pursuit of ad hoc, unfocusedresearch studies.

2. In many situations, quantification of adult sur-vivorship is an essential step in determining thestatus of the population of interest. Data on survival published in the literature is adequate to allow broad generalizations to be maderegarding “adequate” survival for populationmaintenance.

3. Determine the spatial structure of a populationto place the status of various life history parame-ters into context.

4. Quantify reproductive output and breedingdensity. In combination with knowledge of thepopulation’s spatial structure, this can provide agood idea of the status of the population—espe-cially important when adult survivorship cannoteasily be determined.

5. Habitat loss usually is a factor causing thedecline of a species.

RISK REDUCTION STUDIESMethods of assessing avian risk. In assessing avianrisk with the purpose of eliminating or reducing thatrisk, it is essential to quantify both the use of a siteand the deaths associated with that use. The ratio ofdeath to use (risk) becomes a measure, expressed asmortality, or the rate of death (or injury) associatedwith bird utilization of the wind energy site.Following the epidemiological approach, mortality is the outcome variable—the variable that theresearcher considers most likely to shed light on thehypothesis about the mechanism of injury or death.Determining the mechanism of injury or deathallows the development of appropriate methods toreduce the risk to a bird of being in a wind plant.

In testing modifications to turbines or wind plants, itis important to separate bird mortality from bird uti-lization. Only by separating utilization from riskdoes it become possible to know if a modification

that reduces utilization of a wind plant has a positiveor negative effect on the population.

Methods of study design. There are four logical andsequential tasks that the investigator must accom-plish when designing a study of wind energy/birdinteractions.

1. Isolate the hypothesis of mechanism that isbeing tested.

2. Choose a measure of injury-death frequency thatbest isolates the hypothesis being tested.

3. Choose a measure of effect that uses the meas-ure of injury-death frequency and isolates thehypothesis being tested.

4. Design a study that insures maximum statisticaleffectiveness within budgetary and physical constraints.

If risk is defined as the ratio of dead or injured birdsto some measure of utilization, then the choice ofthe use factor, or denominator, is critical. The idealdenominator is the unit that represents a constantrisk to the bird. Great care must be taken in identify-ing the factor measuring bird use of a wind energydevelopment (e.g., bird abundance, passes near aturbine, nesting success). Indirect factors, such aschanges in habitat, prey quality and quantity, andnesting sites, can affect bird use of a wind plant andmust be considered in study design.


PART I

5

INTRODUCTIONWind energy first was used to generate electricity inthe United States nearly 100 years ago. Commercialwind energy developments now operate in 15 states.Additional projects are planned for completion in1999 and 2000 in at least 16 states, including sixstates where commercial wind energy is not current-ly being used. The use of wind energy also is grow-ing rapidly in many other countries. While windenergy, like other renewable energy resources, offersthe prospect of significant environmental benefits,the effects of wind energy developments on birdshave raised important legal, ecological, and oftenemotional issues in the permitting and operation ofwind plants.

When California’s first large wind plants were beingdeveloped in the late 1970s and early 1980s, stateagencies raised some concerns about the potentialimpacts of these facilities on birds. However, theenvironmental effects associated with wind energydevelopment were poorly understood in the 1980s.

In the ensuing years wind plant impacts on birdsbecame a source of concern among a number of thestakeholder groups. Between 100 to 300 raptorsincluding 40 golden eagles (Aquila chrysaetos) wereestimated to die annually in the Altamont Pass WindResource Area (WRA), California (Orloff & Flannery1992). This attracted the attention and scrutiny ofsome groups and has directly affected the permittingof some wind plants. Studies conducted at sites otherthan Altamont Pass WRA indicated few birds werebeing killed (Anderson et al. 1996b, Strickland et al.1996). Thus, we know that wind energy can bedeveloped in a way that minimizes the potential riskto birds, either by design or as a function of theabundance and type of bird species within the gen-eral area of the WRA. However, because studieshave been conducted using a variety of protocols

and with vastly different levels of effort, it was, andcontinues to be, very difficult to make comparisonsof study findings. As a result, there continues to beinterest, confusion, and concern over potentialimpacts on birds from wind energy development.

PURPOSE AND SCOPE OF THISDOCUMENTThis document is intended as a guide to personsinvolved in designing, conducting, or requiring windenergy/bird interaction studies. It is hoped that byoffering guidance with respect to language, methods,metrics (units of measurement), and study designconcepts, this document will lead to future studiesusing methods and metrics that are, as much as ispractical, consistent with accepted scientific prac-tices. Specifically, our aims are to:

1. Provide a reference document for use byresearchers, biologists, and regulatory andwildlife agencies that will, if followed, producea body of information adequate to:

• assess the suitability of a proposed wind plantsite with regard to birds of concern

• assess the potential effects of a wind plantproject on birds of concern

• evaluate the potential effects of the imple-mentation of wind energy technology onbirds.

2. Provide sufficiently detailed and clearly under-standable methods, metrics, study designs anddefinitions for use in the study of windenergy/bird interactions.

3. Promote efficient, cost-effective study designs,methods, and metrics that will produce


CHAPTER 1:Introduction

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6

PURPOSE AND SCOPE OF THIS DOCUMENT . . . . . . . . . . . . . . . . . . . . . .6

HISTORICAL PERSPECTIVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7

METRICS, METHODS AND STUDY DESIGN . . . . . . . . . . . . . . . . . . . . . . .9

ORGANIZATION OF THIS DOCUMENT . . . . . . . . . . . . . . . . . . . . . . . . . .9

comparable data which could reduce the overallneed for some future studies.

4. Provide study designs and methods for the col-lection of information useful in reducing poten-tial risk to birds in existing and future windplants.

Using generally agreed upon methods and metricsshould help to enhance both the credibility and thecomparability of study results, including the resultsof studies conducted at different sites with differentstudy objectives.

The benefits of achieving these objectives are mani-fold. If study methods and metrics are generallyagreed-upon, stakeholders can focus on the implica-tions of study results rather than on debating thevalidity of the data and how it was obtained. If dif-ferent studies generate comparable results, the totalset of wind energy/bird interaction data will beincreased. This in turn should help in understandingthe differences and similarities between wind energydevelopments, in anticipating potential avian issuesat yet-to-be-developed wind energy sites, and ingenerating a body of knowledge about how windenergy development and operation affects birds thatcan be disseminated to the public. It also shouldlead to a more efficient use of research and monitor-ing budgets.

It is neither possible nor appropriate to provide adetailed “cookbook” approach to every site-specificsituation. Not all jurisdictions will require informa-tion on birds or bird research in conjunction withpermitting a wind energy development. In jurisdic-tions requiring bird research, the information in thisdocument can be used to develop standard operat-ing procedures (SOPs) and/or study designs.However, many situations will require site-specificknowledge and expert recommendations as to whichstudy design and methods are most appropriate.

This document provides an overview as well as amore technical discussion of the basic concepts andtools for studying wind energy/bird interactions.Establishing standard metrics, methods, and studydesigns does not reduce the potential for adverseimpacts, mitigate impacts, or guarantee a siting per-mit. It can help ensure that credible, acceptable, scientifically rigorous bird information is gatheredwherever such information is required for wind energy site development.

While the list of metrics provided in this document is not exhaustive, the technical and biological

information needs and approaches presented in thisdocument can support informed decisions regardlessof the size of the wind energy development projector the number of birds potentially affected. The metrics, methods, and study designs described inthis document can be used for a range of projects,and some attention will be given to the differencebetween large and small projects.

For each of the metrics we describe, we will attemptto point out their relative advantages, disadvantages,and underlying assumptions. Project-specific proto-cols should be developed to accomplish specificstudy objectives. The optimal protocol will varydepending on the study objective.

The basic concepts presented here apply to all birdspecies; however, the appropriate study methodsimplemented will vary depending on whether theprimary species of interest is large (e.g., raptors) orsmall (e.g., passerines), nocturnal or diurnal, migra-tory or resident, and so on. Methods also will varydepending on the objectives of the study. Studyobjectives must be clearly defined in order to deter-mine the appropriate study design. The intent of thisdocument is not to advise regulators on what theobjectives of a study of avian impact should be, butrather to give guidance on how to conduct a scientif-ically defensible study that achieves specified objec-tives, using methods and metrics that can bemeaningfully compared against an agreed-uponbenchmark.

HISTORICAL PERSPECTIVEStudies have established that wind energy generationsystems can sometimes kill birds. Depending uponthe situation, this may be viewed as a serious prob-lem. Although many bird species have been affect-ed, raptors have received the most attention in theU.S. (Anderson and Estep 1988, Estep 1989, Howelland Noone 1992, Orloff and Flannery 1992, Hunt1995, Luke and Watts 1994, Martí 1994, Howell1995).

The detection of dead raptors at the Altamont PassWRA (Anderson and Estep 1988, Estep 1989) trig-gered concern on the part of regulatory agencies,environmental and conservation groups, resourceagencies, and wind and electric utility industries.This led the California Energy Commission and theplanning departments of Alameda, Contra Costa,and Solano counties to commission the first exten-sive study of bird fatality at the Altamont Pass WRA(Orloff and Flannery 1992).

Introduction 7

Other North American and European research ofwind energy/bird interactions have documenteddeaths of songbirds (Orloff and Flannery 1992,Pearson 1992, Winkelmann 1994, Higgins et al.1995, Anderson et al. 1996b) and waterbirds(Pearson 1992, Winkelmann 1994). Research atTarifa, Spain identified a high griffon vulture (Gypsfulvus) fatality rate (Martí 1994). Bats also have beenkilled at wind energy facilities (Higgins et al. 1995).Of the numerous commercial sites in operation inthe United States today, bird fatalities of any signifi-cant level have been identified only at Altamont PassWRA, the largest U.S. commercial wind energy sitewith over 6,000 turbines. Two other large Californiasites, in the Tehachapi and San Gorgonio WRAs, do not appear to have the same problem with birdfatalities.

In 1992, the California Energy Commission andPacific Gas and Electric Company sponsored a windenergy/bird interaction workshop focusing on windenergy effects on birds. This workshop brought manyinterested parties together to discuss the issue and itsevaluation, thus taking an initial step toward thedevelopment of a nationwide approach. A researchprogram directed by Kenetech Windpower, Inc.focused on the sensory and behavioral aspects ofwind energy/bird interactions and representedanother significant early effort to address the avianfatality issue. At the same time, the U.S. Departmentof Energy/National Renewable Energy Laboratory(NREL) initiated a program to identify and prioritizeresearch needs, provide technical advice, and fundor cost-share numerous research projects.

In July 1994, a national workshop was held inDenver, Colorado. Sponsored by NREL, the U.S.Department of Energy, the American Wind EnergyAssociation, the National Audubon Society, theElectric Power Research Institute, and the Union ofConcerned Scientists, that workshop attempted tobring together existing information and concernabout wind energy/bird interactions. One majorfocus was on systematizing the search for the factorsresponsible for avian deaths from wind energy facili-ties, and on placing efforts to reduce avian fatalityon a firm, scientific basis (Proceedings, NationalAvian Windpower Planning Meeting [NAWPMProceedings] 1995).

Shortly afterward, the National Wind CoordinatingCommittee (NWCC) formed an Avian Subcommitteeto carry forward the work, begun at the NREL work-shop, of identifying and setting priorities for windenergy/bird interaction studies. The Subcommitteehas provided advice to funding agencies, promoted

communication among participants in wind energydevelopments regarding approaches to resolvingwind energy/bird conflicts, and facilitated the devel-opment of standard protocols for conducting windenergy/bird interaction studies.

The NWCC felt that the interested parties needed abetter understanding of the effect of wind energydevelopment on birds and that they needed tounderstand whether fatality levels and risk vary fromone WRA to another around the nation. Yet defini-tive research results on this complex questionrequire numerous studies over a period of severalyears — studies that often are field-intensive, time-consuming, and costly.

In September, 1995, the Avian Subcommittee spon-sored a second national workshop in Palm Springs,California, to facilitate communication among avianresearchers, regulators, and groups needing goodscientific information to review wind energy devel-opment proposals. An outcome of this meeting wasthe recommendation that a group of ornithologists,statisticians, and environmental risk specialistsdevelop a set of study protocols and measures ofwind energy/bird interactions that could be adoptedby the NWCC. Studying Wind Energy/birdInteractions: A Guidance Document is the result ofthat effort. It is hoped that this document will facili-tate the comparison of results from wind energy/birdstudies in different areas, and that it will lead toimproved understanding of potential causal factorsin wind energy/bird interactions.

Produced by the Avian Subcommittee, this docu-ment has been reviewed by a wide range of stake-holders and has been endorsed by the NWCC as avaluable reference that could be used throughout thenation. A separate NWCC document, Permitting ofWind Energy Facilities Handbook (NWCC 1998),was developed “to help stakeholders make permit-ting decisions in a manner which assures necessaryenvironmental protection and responds to publicneeds.” The Handbook provides an overview of thebasic features of a wind project and discusses thepermitting process. It also describes many of theissues that may arise in the permitting process andprovides tradeoff considerations and strategies fordealing with the issues. The potential impact of winddevelopment on bird resources of concern is one ofthese issues. Permitting of Wind Energy Facilitiesalso provides information on the steps and partici-pants involved in the permitting process of a windplant project.


Results of the early research at the Altamont Passincreased scrutiny and caution during the permittingof new wind plant developments, often resulting incostly delays. Until recently, there were no researchresults from other U.S. avian studies to conclusivelyprovide support for the belief by some that not allwind developments would result in the same level of bird fatalities as was happening at the AltamontWRA. However, recent research at Tehachapi,California, has indicated that the Tehachapi PassWRA (Anderson et al. 1996b) and the Altamont PassWRA may differ — most importantly, that raptor usemay be much lower in the Tehachapi Pass WRA.Yet, this comparison suffers from the fact that a com-mon set of metrics were not used in these studies.More recently, early results from avian research atother wind sites where many of these metrics arecomparable suggest that wind turbines can be sitedin a manner that reduces the potential for impact onbird resources of concern (see Table 1-1).

METRICS, METHODS AND STUDYDESIGNThe information contained in this document pro-vides guidelines to conduct most required windenergy/bird interaction studies. In addition, one ofthe goals of this document is to provide common ter-minology for those involved in conducting windenergy/bird interaction studies. Three commonlyused terms in this document are metrics, methods,and study design. Metrics are measurements, con-cepts, and relationships, such as miles per hour or,in the case of wind energy/bird interactions: bird uti-lization rate, mortality (a rate of fatality), risk, and soon. Methods refer to observational or manipulativestudy techniques used to document bird location,numbers, use, behavior, and other associated param-eters. Study design, which is part of methods, setsforth how, what, when, and where samples will beselected. The study design will need to be tailored tothe specific project, whereas the metrics and othermethods may not require modification from study tostudy.

For research to be found defensible, the metrics andmethods should be scientifically credible and com-ply with the needs of legal and regulatory processes.

ORGANIZATION OF THIS DOCUMENTThis document is organized in two parts. Part I(chapters 1-2) presents the general reader with aframework for considering wind energy/bird inter-actions. Following this chapter (Introduction),Chapter 2 - Site Evaluation Biology discusses infor-mation needs and sources as well as the on-site

inventory and monitoring work that may be con-ducted at a wind energy site.

Part II of the document (chapters 3-5) providesdetailed discussion of metrics, methods, and studydesign issues for basic and advanced windenergy/bird interaction studies. Geared toward themore technical reader, these chapters are intendedto provide regulatory staff and technical advisors tothe various interested parties with a common under-standing of what constitutes scientifically rigorousresearch and its applications.

Chapter 3 - Basic Experimental Design and Level 1Studies discusses basic monitoring and research thatmay be conducted prior to construction and opera-tion or during operation when potential impacts arepoorly understood. Level 1 research typicallyincludes observational studies designed to detectpotential effects of large magnitude. Such studieslook at risk and cumulative effects.

Chapter 4 - Advanced Experimental Design andLevel 2 Studies discusses more specific (Level 2)studies focusing on manipulative experiments andbird population effects, including field and/ormodel-based studies. This type of research is con-ducted if a regulator is sufficiently concerned that apopulation problem exists or could result from awind resource area. This concern might arise fromthe results of a Level 1 study. This research can usevarious forms of population modeling.

Chapter 5 - Risk Reduction Studies discusses Level 2studies that focus on ways to reduce bird risk. This is research in applied problem solving; it addressesan acknowledged problem and normally involvesdesigns of manipulative studies including treatments.

Additional sections at the end of the documentinclude a list of Literature Cited, and an Index of KeyTerms defined or explained in this document.

Introduction 9


Site Name Purpose of Study Status Results* Size** Length and Location of Study

Altamont Evaluate impacts of perching On-going Many bird species killed 5400 turbines ~2.5 yearsWRA, CA and mortality on a wide variety annually; particular concern at site/940

of turbine types. for raptor fatalities turbines observed;

Conduct behavioral observations 800 turbinesof all bird species on a wide serached forvariety of turbine types. fatalities

Tehachapi Compare bird use, behavior, and Field work A variety of bird species are 5000 turbines ~4 yearsWRA, CA mortality at small and large completed killed annually, but not at at site/ 2700

turbines; compare bird use, levels that are deemed turbines behavior, and mortality at tubular biologically significant included in and lattice towers; compare bird researchuse, behavior, and mortality near and away from turbines

San Gorgonio Compare bird use, behavior, and Field work A variety of bird species are 3500 turbines ~2 yearsWRA, CA mortality at small and large completed killed annually, but not at at site/2700

turbines; compare bird use, levels that are deemed turbines behavior, and mortality at tubular biologically significant included in and lattice towers; compare bird researchuse, behavior, and mortality near and away from turbines

Ponnequin, To document: avian use of a 1st year Little avian activity noted; Phase I – Up to CO relative abundance on the project field work post-construction monitoring 21 turbines 3 years

area pre- and post- construction; completed; will seek to assess impact installedto dateuse of existing power line poles post- of turbinesand fence posts; raptor nesting construction Phase II – up to populations; burrowing activities monitoring 27 turbines/none of ground squirrels; any avian on-going yet installedfatalities. Research will be con-ducted pre- and post-construction.

Searsburg, VT To assess impacts of wind Study com- No major changes in species 11 turbines at ~3 yearsdevelopment on breeding and pleted; final composition were found, but site/11 turbinesmigrating birds. going the numbers of several interior included in

report is forest breeding birds were researchBefore/after study design was used. through lower after construction than

NREL peer- before and several edge review species were more numerous process after construction.

Further study was recommended.

Buffalo Ridge, Uses control/impact on Phase I 3 years Use by some avian groups is Phase I – ~3 yearsMN and before/after and control/impact field work lower than expected within 73 turbines

(BACI) on Phase II to evaluate completed; wind plants, likely due to installed/impacts on wildlife from each 2 years pre behavioral avoidance and sampledphase of development and the and 1 year reduced habitat effectiveness. cumulative impact to wildlife post- Mortality low in comparison Phase II – from all wind energy development. construction to other wind plants in the 143 turbinesTo provide information that can be monitoring U.S. Relatively high incidence installed/used to reduce impacts to wildlife for Phase II of bat mortality appears sampledof subsequent developments unique among wind plants.

Foote Creek Uses BACI to evaluate impacts on 2 years field Turbine strings placed back Phase I – 3-4 yearsRim, WY wildlife from each phase of work com- from rim edges to minimize 69 turbines development and the cumulative pleted: post- impact with raptors; post- installed/all

impact to wildlife from all wind construction construction monitoring surveyedenergy development. To provide monitoring continuesinformation that can be used to and fatality Phase III – up toreduce impacts to wildlife of searches 33 turbines/nonesubsequent developments beginning yet installed

* Results are based on draft documents that will soon be published.** Total number of turbines at site/number of turbines in study

Table 1-1. Examples of Recent Avian Research at U.S. Wind Energy Sites

INTRODUCTIONThe risk to birds is among the many considerationsweighed during site evaluation of a wind plantdevelopment project. This chapter discusses types ofavian biology information that may be helpful inestimating risk for birds associated with proposedwind plant site development. The information pro-vided herein includes sources of information andfield survey techniques for gathering sufficient birdinformation for many site evaluation decisions.

Site evaluation may start with one or more sitesbeing considered and compared for acceptability asa wind energy development site. The site evaluationmay be complete after gathering simple and straight-forward existing and on-site information, or it mayprogress to levels of more detailed information if thedeveloper or permitting agency deems it warranted.Site evaluation normally will utilize only a portion ofthe information sources and research/informationgathering methods discussed in this document. Thedecision as to what constitutes adequate avian biolo-gy information may need to be reconsidered at morethan one stage of the site evaluation. Specific infor-mation on birds may be needed to obtain compli-ance with local permitting requirements, ifapplicable, as well as state and federal environmen-tal laws. Although this discussion focuses on birds,bat species may also be considered.

For a discussion of the various aspects of the permit-ting process regarding wind energy facilities, seePermitting of Wind Energy Facilities: A Handbook(NWCC 1998). Chapter 4 of the Handbook providesan overview of where, why, when, and how biologi-cal resources and bird and bat resources may beconsidered during the permitting process.

Permitting processes often have a defined time line,usually beginning with the formal filing of a permitapplication. Avian information will normally be col-lected during the pre-application period and may besimple and straightforward or more complicateddepending on the avian resources and specific situa-tion. In cases where more detailed information isneeded, timing becomes important. Certain types ofbiological resource information are seasonal and canonly be gathered in the field at certain times of year.Bird activity may need to be looked at during morethan one season. Depending on the species presentand the information requirements, more than oneyear of data-gathering may be required. All site-spe-cific information should be well documented andmore detailed information, if needed, should beobtained using standard, credible techniques, met-rics, methods, and study design. (See chapters 3, 4,and 5 of this document for specific guidance.)

This chapter focuses on where and how to obtainbiological resource information, mostly about birds.How much information is desired and over what

Site Evaluation Biology 11

CHAPTER 2: Site Evaluation Biology

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11

SITE EVALUATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12

Sources of Existing Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12

On-Site Visit and Habitat Relationships . . . . . . . . . . . . . . . . . . . . . .13

Is the Existing Information Adequate and Defensible? . . . . . . . . . . . .14

GATHERING ADDITIONAL INFORMATION . . . . . . . . . . . . . . . . . . . . . .14

Planning to Gather Additional On-site Information . . . . . . . . . . . . . .14

Short-term On-site Surveys and Monitoring . . . . . . . . . . . . . . . . . . .14

DOCUMENTATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16

Is the Biological Information Adequate? . . . . . . . . . . . . . . . . . . . . . .16

ESTABLISHING MORE RIGOROUS RESEARCH DESIGNS . . . . . . . . . . . .16

time period it will need to be gathered is a decisionmade by the project proponent or the permittingauthority, possibly with other stakeholder input. (See the “Biological Resources Tips” section on page41 of the Permitting of Wind Energy Facilities: AHandbook [NWCC 1998].) It is valuable to under-stand the bird resource-related laws, standards, regu-lations, and ordinances of the project site areas. It isalso useful to clarify early in the wind plant site eval-uation process any project-specific and jurisdiction-specific legal and biological information that may beneeded (NWCC 1998).

SITE EVALUATIONWhen one or more sites are under consideration,some quick and easy methods can be utilized todetermine the type of bird resources on and near thesite. Information-gathering at this stage can covermany variables and is intended to eliminate prob-lematic surprises late in the permitting process andduring operation. By conducting an appropriateassessment, the wind plant proponent and permittingauthority will be able to estimate potential bird risk.A written assessment of each site considered shouldinclude:

1. Information from existing sources, including:

• local expertise

• literature searches

• natural resource database searches for sensi-tive species, for bird species known to be sus-ceptible to collision events, and for areasused by large numbers of birds

2. Reconnaissance surveys

3. Vegetation mapping, habitat evaluation, andwildlife habitat relationships

4. Consideration as to whether the existing infor-mation and site visit information will allow com-pliance with, and be defensible for, regulatoryand environmental law purposes.

Sources of Existing InformationLocal ExpertiseSeeking out one or more local experts familiar withthe site(s) being considered can save time as well asprovide valuable information. Local experts canquickly identify potential bird and other biologicalconcerns or issues at the site(s) under consideration.They may have an established working relationship

with or knowledge of other persons or resources thatcan be utilized to provide valuable biological, regu-latory, and legal information. Interviews should bedocumented in a written report. Local expertise caninclude the following:

• state fish and game agents/biologists

• federal wildlife agents/biologists (e.g., U.S. Fish & Wildlife Service, Bureau of LandManagement, U.S. Forest Service, U.S.Geological Survey)

• university professors/graduate students

• Partners in Flight representatives

• National Audubon Society representatives

• Hawk Migration Association of North Americarepresentatives

• bird observatory representatives

• other knowledgeable parties.

In pre-permit evaluation of the Columbia Hills windpower site, the proponent for the site discovered thatthe State of Washington's wildlife agency had histor-ical records of several bald eagle day roost sites near the site. A reconnaissance level survey of thesite discovered a night roost used by a small numberof eagles. This information was used in the finaldesign of the wind plant and, had the project pro-ceeded, would have resulted in the company elimi-nating at least one string of turbines potentiallyplacing birds using the roost at risk.

Literature SearchA literature search can provide valuable informationabout bird resources using the area. Environmentaldocuments previously prepared for the site or sitearea may be useful, as may other types of reports.Research results from other wind energy facilitiescan be used in site evaluation to identify trends orsimilarities that may either point to few problemsanticipated, or raise concerns. As more wind ener-gy/bird interactions research results become avail-able, these results may be helpful for creatingscreening matrices or screening calculations. Manysources of literature will be gray literature, i.e.,reports or studies published incidentally by anagency or stakeholder group as distinguished fromarticles published in independent, regularly pub-lished, peer-reviewed journals. (Gray literature may


or may not be peer-reviewed.) Gray literature canprovide good, useful information; however, thevalue of a specific piece of gray literature should bedetermined by an experienced biologist with knowl-edge of the species of special interest in the area.

Natural Resource Database SearchMost federal, state, and local agency offices andmany conservation organizations (e.g., The NatureConservancy, California Native Plant Society) main-tain databases of sensitive resources in the area oftheir jurisdiction or focus. These databases can bevaluable for determining whether sensitive birdspecies and other sensitive resources are known touse the potential site or vicinity. This informationusually consists of known bird locations, so bear inmind that the site in question may never have beeninventoried for bird resources. Therefore, a databasesearch may come up negative because no one haslooked, or because sensitive bird resources using thesite have not yet been detected. It is important tounderstand how the databases are constructed inorder to interpret and use the results appropriately.

Sensitive species (including species of special con-cern) are those species that are protected by federalor state law, or are listed or monitored by govern-ments, agencies, or environmental groups for variousreasons. Legal protection for these species can varyfrom federal and/or state endangered species laws tolocal agency policies. Having knowledge of the lawsand policies of the project site area is very impor-tant. It is the responsibility of the developer to under-stand these laws and policies and their ramificationsfor the project. Some sources for information regard-ing sensitive species and other special situations (e.g.concentration areas, flyways, etc.) include:

• National Audubon Society Christmas bird counts

• breeding bird surveys/census/maps sites - available from various sources

• state heritage documents and maps

• state bird atlases/bird books

• state and federal endangered and threatenedspecies lists and occurrence information

• federal, state, and local resource agency offices

• state wildlife habitat relationship programs

• Ducks Unlimited

• National Audubon Society state and federalwatch lists

• other sources.

On-Site Information GatheringReconnaissance StudiesReconnaissance studies are on-site surveys used by abiologist to get a general feel for the site, topogra-phy, habitat, bird use and potential use, and forspecies that may use the site. This type of survey canprovide valuable information for an environmentalassessment. Depending on the site and speciesknown to occur there, reconnaissance studies com-bined with other easily-gathered information (suchas local expertise and literature) can provide ade-quate information to estimate potential impacts. Inother situations, it may provide information that canhelp focus more detailed studies of bird resources.

Vegetation Mapping, Habitat Evaluation andWildlife Habitat Relationships Each site should be visited by a trained and experi-enced biologist with specific knowledge of andexperience detecting the bird species and other nat-ural resources of the project site and vicinity. (Thisvisit may coincide with the reconnaissance survey.)Plant and animal species observed on the projectsite and vicinity should be documented. The vegeta-tion series should be identified and mapped at anappropriate scale (e.g. 1” = 500'). Wildlife habitatrelationships are complex, but there may be infor-mation available that will assist with determiningbird species’ use of a habitat. Many states have a“Wildlife Habitat Relationships Program.” Thesedatabases and the vegetation maps can be evaluatedto develop lists of species that may utilize the site. Insome situations the probability of species use/occur-rence may be discussed. Specific habitat attributesand elements may also provide clues to species use.See Morrison et al. (1998) for a review of the litera-ture on wildlife-habitat relationships.

Uses of the area may include such activities asbreeding/nesting, migrating through, wintering,migratory stopover, and foraging. Signs of significantbird use, sensitive species use, or of use by collision-susceptible species are early warnings that may leadto additional investigations or to site abandonment.

Sensitive species use (or likely use) is one determi-nant of a project’s potential for significant impact. If the value of the site for sensitive species is wellknown, more detailed studies may be needed (seechapter 3). If a potential site has a high likelihood for


biological conflicts, it may not be worth the timeand cost of detailed site evaluation work (NWCC1998). If the potential for bird risk is certain to below, then very little additional information may beneeded.

Is the Existing Information Adequate andDefensible?At this stage of the site evaluation process, thedeveloper may want to consider whether existinginformation is adequate and defensible for the per-mit application. Is the biological information ade-quate to make a determination of the likelihood ofcompliance with any applicable regulatory and envi-ronmental laws? Are there potentially any sensitivespecies that may be significantly affected? Whatadditional information may be needed?

GATHERING ADDITIONALINFORMATIONPlanning to Gather Additional On-siteInformationWhen planning to gather additional bird informationfor the permitting process, the permitting agency anddeveloper may want to discuss future informationand monitoring needs. Because cost and time arevaluable considerations, it is important first to speci-fy information needs and establish clear researchobjective(s), and then to use appropriate sampledesign, metrics, and methods. Monitoring, usingoptions discussed in chapter 3, can supply short-term or long-term information needs in an effectiveand efficient manner. Obtaining this more detailedavian resource information requires more scientifi-cally rigorous methods. Chapters 3-5 provide infor-mation and options for moderate to higher levels ofresearch needs.

Short-term On-site Surveys and MonitoringShort-term on-site surveys/monitoring refers to multi-ple visits to a site to document bird use or someother needed information of value. When simplermethods have been exhausted, but sufficient con-cern persists regarding the presence and use of thesite by sensitive species or the numbers and types of species using the site, monitoring may be neededto learn more about the site’s avian resources andprovide information needed to make permittingdecisions with reasonable confidence.

Depending on the site, short-term on-sitesurveys/monitoring may focus on one or more of thefollowing:

• species of special concern

• breeding bird species

• migrating birds

• wintering birds

• nocturnal bird activity

• species known to be susceptible to collision

• special situations.

Not all of these variables and efforts may be neededon a given project site. The effort expended in gath-ering the following types of information will dependon the knowledge of the site and perceived risk tobirds. It may be useful to discuss which types ofinformation are warranted for each project with thepermitting agency, and possibly with others, early inthe pre-permit application process so that the issuedoes not arise late in the process and cause costlydelays. See chapter 3 of Permitting Wind EnergyFacilities: A Handbook for a discussion of the per-mitting process.

When needed, this information should be gatheredusing the appropriate research options discussed inchapters 3-5.

Species of Special Concern A comprehensive literature and on-site search maybe conducted to predict the likelihood the site isused (or not used) by species of special concern.Species of special concern are species listed by thestate or federal governments as threatened or endan-gered, and those species that are afforded other legalprotections. Other species may also be considered inthis category. If species of special concern areknown to use the site, then additional inventoryefforts may be required to better understand their useof the site, time of use, and to estimate potentialadverse affects. Established species-specific invento-ry protocols are sometimes required by regulatory orresource agencies to ensure that adequate inventorymethods and techniques are employed. These proto-cols can be obtained from the permitting authority.In situations where the species may use the site butthis has not been determined with certainty, it isimportant for the developer to work with the permit-ting authority to determine the next steps needed.


Breeding Bird Species Part of inventorying a site for bird resources includesthe identification of bird species of concern thatbreed and nest on or near the site. Birds nestingnearby may include the site in their home range. Thevalue of the area may be for nesting, foraging (feed-ing), or other common or irregular uses related tonesting. An example would be golden eagles nestingnear but off the site and regularly foraging on-site.

Bird species come in different sizes, and exhibit dif-ferent behaviors. Large species such as golden eaglesmay range up to 15 kilometers (km) from the nestarea (Hunt 1995); smaller species may have a muchsmaller activity area (e.g. < 1km). It is valuable tounderstand the species under consideration in orderto assess potential effects. Identification of thesebreeding species is season-constrained, so it isimportant to identify the need for this type of infor-mation early in the pre-permit application period inorder to conduct a site evaluation at the appropriatetime of year.

Migrating Birds Many birds are migratory. Migrating species’ use of asite depends on many variables. Some birds maypass through the project area only during the fall andspring migratory periods. During the time they are inthe project area they may or may not exhibit behav-ior that puts them at risk. Some birds (includingmany passerines, or song birds) mostly migrate atnight, while other birds (such as raptors and vultures)mostly migrate during daylight hours. Waterfowl andshorebirds migrate during day and night. Mostmigrating birds fly at higher altitudes above theground than wind turbines are tall. However, migrat-ing birds fly at other than normal altitudes for differ-ent reasons. Examples include birds flying closer tothe ground surface during bad weather conditions ordue to topographical features such as ridge tops.This type of information may be used to assist in esti-mating possible risk to birds. If on-site migratory birdinformation is gathered, sampling and methodoptions from chapter 3 (Level 1 sampling tech-niques) should be considered (e.g. random samplesites using bird utilization counts at some regular fre-quency during the migratory months).

Wintering Birds Wintering birds are those that spend their winters inthe project area. These birds leave for nestinggrounds in the early spring. Wintering birds (andbats) normally are leaving a colder place with limit-ed food for a warmer locale with an adequate foodsupply. Many birds, such as raptors, become more

social or at least more tolerant of other raptors dur-ing the winter. During nonbreeding periods, birdsoften flock or form groups. They forage and/or movetogether in ways they would not during breedingseason. Concentrations of wintering birds and theirspecific behavior may or may not put them at risk.Surveys for wintering birds may need to be conduct-ed during the winter. It is valuable to choose sampledesigns and methods that will provide adequate anddefensible information (see chapter 3).

Nocturnal Bird ActivityMany birds are active mostly during daylight, butsome birds are more active during the low light peri-ods at sunset and sunrise. Owls, a few other birdspecies and bats, are normally active at night. Birdsactive during low light and at night may be resident,breeding, migrating, or wintering bird species.Activity during low light and night-time periods canresult in collision with wind turbines (Anderson et al.1996b). There have been no known incidents oflarge numbers of bird kills in wind plants during thenocturnal period. Currently, the reports in wind plantdevelopments have been of infrequent kills. Currenttechnology for nocturnal bird utilization monitoringcan be costly and is not as well established as day-time observational methods. We discuss nocturnaltechniques for the reader’s information in the case itmay be applicable.

Information about nocturnal bird activity may beobtained using remote sensing methods such asradar (Gauthreaux 1996a), ceilometers, acousticmonitoring (Evans and Mellinger 1999) or othernight-time inventory techniques. Except in specificsituations, existing techniques for nocturnal surveysmay not be adequately refined nor validated to pro-vide needed information at the level of confidencerequired. For example, marine radar can be used toget numbers, altitudes, flight speeds and directionsof birds at higher altitudes above the ground surface,but has difficulty at lower altitudes above the groundsurface where the information may be important inassessing the potential for collision risk with windturbines. Species identification cannot be made withmarine radar. Acoustic monitoring techniques holdpromise, but are still under development at this writing.

Species Known to be Susceptible to Collision Some species or species groups, such as raptors,have shown a greater tendency to collide with windturbines than have other species, such as ravens(Anderson et al.1996b). As the results of moreresearch and monitoring studies become available,


additional species may be identified as being moreor less at risk of collision. Such information shouldbe employed to evaluate the acceptability of sites.

Special Situations Birds may utilize specific areas more than otherareas on the proposed wind plant site.Understanding those activity areas and modifyingthe project commensurately can be very valuable.Avoiding high use areas or areas used by species ofspecial concern can be effective in minimizingimpacts. Another example of a special situation is aspecies of special concern utilizing the potentialwind energy development site during the time ofyear that coincides with a low wind period (e.g.Altamont Pass gets less than 2% of its wind inDecember and January). Sensitive species utilizingthe area during a low wind period may be presentedwith fewer turning turbine blade collision possibili-ties than when turbines are operating a greater per-centage of the time. This type of information can beuseful in specific siting decisions and site suitabilitydecisions. Gathering this type of information willnormally require scientifically rigorous methods (seechapter 3) in order to obtain results that provide theconfidence needed for these type of decisions.

DOCUMENTATIONIt is important to document in writing how, what,when, and where all biological information wasobtained throughout the site evaluation process.Written documentation ensures that credibility canbe determined for both the bird biological informa-tion and how it was gathered. The integration of thesite evaluation information into a written report thatdescribes the resources and estimates potentialimpacts is valuable and often required.

Is the Biological Information Adequate?Has adequate biological information been gathered?Adequate information is the amount and type ofinformation needed to be in compliance with regula-tory and environmental laws, ordinances, regula-tions, and standards of the jurisdiction(s) involved.Meeting the test of adequacy requires that the bio-logical information (written report) is both sufficientand sufficiently clear to allow for reasonable esti-mates of bird impacts. The types of information dis-cussed in this chapter should be adequate to assistwith making many project decisions.

ESTABLISHING MORE RIGOROUSRESEARCH DESIGNSPart II of this guidance document addresses the sci-entific methods useful when looking at general and

specific issues relating to wind energy/bird interac-tions. These scientific methods can be used in stud-ies that may be short-term or long-term, and thatmay be small or large in magnitude, time, and cost.It is important to match study designs/methods withneeds regarding project size and perceived impacts.In particular, chapter 3 offers the reader several alter-native designs for studies to provide permitting infor-mation as well as data needed to satisfy possibleoperational monitoring requirements. Using theinformation in this document will help to ensure thatpre-permit application, and operational monitoringresearch, if needed, uses credible study design,methods, and metrics.

Chapters 4 and 5 present concepts and discussissues related to advanced studies and modelingtechniques that may be needed in some situations toassess population effects and risk reduction strate-gies. The site evaluation process should alert thedeveloper to problematic sites. However, if problemsarise later, chapters 3-5 contain information that willassist with their resolution.


PART II

17


CHAPTER 3: Basic Experimental Design and

Level 1 StudiesINTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19

The Traditional Experimental Design Paradigm . . . . . . . . . . . . . . . . .19

THE PHILOSOPHY OF STUDY DESIGN . . . . . . . . . . . . . . . . . . . . . . . . . .23

Design/Data-Based Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23

Model-Based Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23

Mixtures of Design/Data-Based and Model-Based Analyses . . . . . . .23

Levels of Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24

DESIGN/DATA-BASED STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25

Designs with Control (Reference) Areas . . . . . . . . . . . . . . . . . . . . . .25

Impact-Reference Design (After Treatment) . . . . . . . . . . . . . . . . . . . .26

Designs Without Reference Areas . . . . . . . . . . . . . . . . . . . . . . . . . . .26

IMPROVING THE RELIABILITY OF STUDY DESIGNS . . . . . . . . . . . . . . . .28

Use of More than One Reference Area . . . . . . . . . . . . . . . . . . . . . . .28

Collection of Data Over Several Time Periods . . . . . . . . . . . . . . . . .28

Interpretation of Area-by-Time Interactions . . . . . . . . . . . . . . . . . . . .29

Model-Based Analysis and Use of Concomitant Site-Specific Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30

Sampling The Area Of Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30

Sampling Plans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .31

DATA ANALYSIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36

Univariate Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36

Multivariate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37

Meta-Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38

Habitat Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .38

Cumulative Effects Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39

Statistical Power and the Weight of Evidence . . . . . . . . . . . . . . . . . .39

CASE STUDIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41

Proposed Plant - Buffalo Ridge, Minnesota . . . . . . . . . . . . . . . . . . . .41

Existing Plant - Tehachapi and San Gorgonio Pass, California . . . . . .43

SUMMARY AND CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .47

INTRODUCTIONPublic interest in the impact of the wind industry on birds has led some state and federal agenciesresponsible for permitting wind plants or protectingpotentially affected avian species to require studiesto:

• predict the potential effect of proposed windplants on birds

• evaluate the actual effects on birds of windplants in operation

• determine the causes of bird fatalities

• evaluate methods for reducing risk of bird fatality.

This chapter provides guidance to regulators, indus-try developers, scientists, and interested members ofthe public as to how such studies may be conductedin a manner that will withstand scientific, legal, andpublic scrutiny. While wind power presents some-what unique environmental perturbations, the prin-ciples involved in designing studies of its effect onbirds are the same as for other environmental perturbations.

This chapter does not provide detailed design proto-cols and standard operating procedures for quantifi-cation of impacts in all wind plant situations.Instead, it discusses how the quantification of effectsfits into the various philosophies of design, conduct,and analysis of field studies. This chapter drawsheavily from guidelines for design and statisticalanalysis of impact quantification studies of oil spillsby McDonald (1995). Examples of two existing pro-tocols measuring wind power effects on bird speciesare discussed.

In a perfect world, impacts would be measured with-out error. For example, bird fatalities on the site of awind plant could simply be counted. However,when a complete count or census is impossible thenimpacts must be detected by the use of scientificstudy and statistics. The ultimate objective of statis-tics is to make inferences about a population (groupof units) from information contained in a sample(Scheaffer, et al. 1990). Statistical or inductive infer-ences are made properly in reference to:

1. the design and protocol by which the studies areconducted in the specific study areas;

2. the specific time period of the study; and,

3. the standard operating procedures (SOPs) bywhich data are collected and analyzed.

If either the design protocol or the SOP is inade-quately documented, then the study is not replicableand its validity is uncertain. In such a situation, it isimpossible to know the proper extent of statisticalconclusions and there would necessarily be less sci-entific confidence in the statistical inferences. Thiswould result in less confidence in inferences basedon expert opinion, as opinion in this case should fol-low statistical inferences. A common practice in eco-logical studies is the extension of study conclusionsbeyond the specific study areas to unstudied areas.This practice is acceptable and often necessary,albeit risky, as long as the assumptions are specifiedand it is clear that the extrapolation is based onexpert opinion. When the extrapolation is presentedas an extension of statistical conclusions it is animproper form of data analysis. Deductive inferencesthat extend beyond the specific study areas to drawgeneral conclusions about cause-and-effect aspectsof operating a wind plant may be possible if enoughindependent studies of different wind plants identifysimilar effects. However, statistical inferencesbeyond the study areas are not possible; nor shouldthis be the primary objective of quantification ofimpact, given the unique aspects of any develop-ment.

The Traditional Experimental DesignParadigmThe traditional design paradigm for the true experi-ment is defined in terms of the following principles(Fisher 1968; Pollock 1996):

• Control. The scientist tries to control (standard-ize) as many variables as possible except forthose associated with the different treatmentconditions that are to be compared.

• Randomization. The scientist randomly allocatestreatments to experimental units so that the val-ues of variables not controlled are allocatedequally over units (at least on average).

• Replication. Each treatment is allocated to multi-ple independent experimental units so thatunexplained or inherent variation can be quanti-fied. Information about the amount of inherentvariability is needed for valid statistical testing.

Two additional methods are useful for increasing theprecision of studies when the number of replicatescannot be increased:

Basic Experimental Design and Level 1 Studies 19

1. Group randomly allocated treatments withinhomogeneous groups of experimental units(blocking).

2. Use analysis of covariance when analyzing theresponse to a treatment to consider the addedinfluence of variables having a measurable influence on the dependent variable.

The study of wind energy development impacts ismade difficult by the relatively large area potentiallyaffected, the relative scarcity of many of the speciesof primary concern, and the relative scarcity of theevents being measured (e.g. mortality, use of a par-ticular turbine by a particular species).Quantification of the magnitude and duration ofimpacts due to a wind plant necessarily requires anobservational design, because the area to receive thewind plant and the areas designated as the refer-ences (controls) are not selected by a random proce-dure. Observational studies also are referred to as“sample surveys” (Kempthorne 1966), “planned sur-veys” (Cox 1958), and “unplanned experiments /observational studies” (National Research Council1985). See Manly (1992) and McKinlay (1975) for adiscussion of the design and analysis of observation-al studies. Impact studies typically are large fieldstudies, as opposed to manipulative experiments orobservational studies in subjectively selected smallhomogenous areas. Data are collected by visualdetection of an event in time and space.

Finally, in studies of wind plant effects on birds, aswith many environmental impact studies, conclu-sions concerning cause-and-effect of wind plants arelimited. Practically speaking, identical “control”areas seldom exist; similar “reference” areas must beused instead. Moreover, there is no random assign-ment of treatment, and replication is usually impos-sible. Wind plant sites are selected because they arevery windy, there exists a market for the power pro-duced, and there is an existing infrastructure (e.g. apower grid). These sites tend to be relatively uniquetopographically, geographically, and biologically,and are difficult to duplicate, at least in a relativelysmall area. Even if all the potential wind sites areknown in an area, the decision regarding where tolocate the plant is never a random process. Finally,the expense of a wind plant makes replicationimpractical. Thus, one does not have a true experi-ment.

In all studies of impact, including wind plantimpacts, it is essential that a few basic study princi-ples be followed. The following is a brief discussionof some of the more important principles. For a

detailed discussion of these principles see Green(1979) and Skalski and Robson (1992).

Know the QuestionIt is essential that the question being addressed bythe research be clearly understood. Research ques-tions form the basis for developing researchhypotheses, and help to define the parameters forcomparing hypothesized outcomes with actualresearch results. (See section on Data Analysis for amore direct discussion of hypothesis testing.) Thedesign of the study protocol depends on the questionbeing addressed. The protocol that addresses thequestion of wind plant risk to individual birds is sub-stantially different from a protocol addressing therisk to a population of birds. A clear understandingof the question increases the efficiency of theresearch. It is a waste of time and money to collectvast quantities of data with the idea that their mean-ing will become obvious after the data are analyzed.The outcome of a study is more likely to be useful ifan appropriate study design is followed and all inter-ested parties have a clear understanding of theresearch question. Studies of wind plant impacts onbirds should allow the research question to beaddressed through inductive (statistical) inferences aswell as deductive inferences (expert opinion). Theseinferences should help provide a sound scientificbasis for development of protocols for quantificationof wind power impact.

ReplicateReplication means repetition of the basic experiment(Krebs 1989) within each time and location of inter-est, producing multiple sets of independent data.Essential for statistical inference, replication allowsthe estimation of variance inherent in natural sys-tems and reduces the likelihood that chance eventswill heavily influence the outcome of studies. Properstatistical inference must also keep the proper exper-imental unit in mind. In studies of wind power theexperimental unit may be a turbine, a string of tur-bines, or the entire wind plant. Using the wrongexperimental unit can lead to errors in the identifica-tion of the proper sample size and estimates of sam-ple variance. Confidence in the results of studiesimproves with increased replication; generallyspeaking, the more replication in field studies thebetter.

The concept of replication often is confused in theconduct of environmental studies; what constitutesreplication of the basic experiment depends on theobjective of the study. For example, if the objectiveis to compare bird use of a wind plant to bird use in


a similar area without the wind plant, replicationmay be achieved by collecting numerous independ-ent samples of bird use throughout the areas andseasons of interest. In this case the sample size forstatistical comparison is the number of samples ofbird use by area and season. However, if the objec-tive is to estimate the effect of another wind plant —or wind plants in general — on bird use, then theabove wind plant constitutes a sample size of one,from which no statistical comparisons to other sitesare possible. The statistical extrapolation of datafrom one study site to the universe of wind plants isone of the more egregious examples of pseudorepli-cation as defined by Hurlbert (1984) and Stewart-Oaten et al. (1986).

When estimating the appropriate sample size in anexperiment, a good rule to follow is that the analysisshould be based on only one value from each sam-ple unit. If five sample plots are randomly located ina study area, then statistical inferences to the areashould be based on five values — regardless of thenumber of birds which may be present and meas-ured or counted in each plot. If five animals are cap-tured and radio-tagged, then statistical inferences tothe population of animals should be based on a sam-ple of five values, regardless of the number of timeseach animal is relocated. Repeated observations ofbirds within a plot or repeated locations of the sameradio-tagged animal are said to be dependent forpurpose of extrapolation to the entire study area.Incorrect identification of data from sampling units isa common form of pseudoreplication that can giverise to incorrect statistical precision of estimatedimpact. It becomes obvious that replication is diffi-cult and costly in environmental studies, particularlywhen the treatment is something as unique as awind plant.

RandomizeLike replication, an unbiased set of independent datais essential for estimating the error variance and formost statistical tests of treatment effects. Althoughtruly unbiased data are unlikely, particularly in envi-ronmental studies, a randomized sampling methodcan help reduce bias and dependence of data andtheir effects on the accuracy of estimates of parame-ters. A systematic sample with a random start is onetype of randomization (Krebs 1989).

Collecting data from “representative locations” or “typical settings” is not random sampling. Iflandowner attitudes preclude collecting samplesfrom private land within a study area, then sampl-ing is not random for the entire area. In studies

conducted on representative study areas, statisticalinference is limited to the protocol by which theareas are selected. If private lands cannot be sam-pled and public lands are sampled by some unbi-ased protocol, statistical inference is limited topublic lands. The selection of a proper samplingplan is a critical step in the design of a project andmay be the most significant decision affecting theutility of the data when the project is completed. Ifthe objective of the study is statistical inference tothe entire area, yet the sampling is restricted to asubjectively selected portion of the area, then thereis no way to meet the objective with the studydesign. The inference to the entire area is reducedfrom a statistical basis to expert opinion.

Control and Reduce ErrorsThe precision of an experiment (density of repeatedmeasures of the same variable) can be increasedthrough replication, but this is expensive. As dis-cussed by Cochran (1977) and Cox (1958), the pre-cision of an experiment can also be increasedthrough:

1. use of experimental controls

2. refinement of the experimental techniquesincluding greater sampling precision withinexperimental units

3. improved experimental designs including strati-fication and measurements of non-treatment factors (covariates) potentially influencing theexperiment.

Use of experimental controls. Good experimentaldesign should strive to improve the precision of con-clusions from experiments through the control (stan-dardization) of related variables (Krebs 1989). In theevaluation of the effect of some treatment (e.g., ananti-perching device) on the frequency of perchingon wind turbines, it would be most efficient to studythe devices on the same model turbine, controllingfor turbine type. One could evaluate the effect ofwind turbines on bird use by making comparisonswithin vegetation types and thus control for theeffect of vegetation. However, standardization ofrelated variables is often difficult in field studies.

An alternative to standardizing variables is to useinformation that can be measured on related vari-ables in an analysis of covariance (Green 1979). For example, understanding differences in raptor use between areas is improved when considered in


conjunction with factors influencing use, such as therelative abundance of prey in the areas.

Precision can also be improved by stratification, orassigning treatments (or sampling effort) to homoge-nous strata, or blocks, of experimental units.Stratification can occur in space (e.g., units ofhomogenous vegetation), and in time (e.g., samplingby season). Strata should be small enough to maxi-mize homogeneity, keeping in mind that smallerblocks may increase sample size requirements. Forexample, if vegetation is used to stratify an area,then the stratum should be small enough to ensure arelatively consistent vegetation pattern within strata.However, stratification requires some minimum sam-ple size necessary to make estimates of treatmenteffects within strata. It becomes clear that stratifica-tion for a variable (say vegetation type) at a finer andfiner level of detail will increase the minimum sam-ple size requirement for the area of interest. If addi-tional related variables are controlled for (e.g.,treatment effects by season), then sample sizerequirements can increase rapidly. Stratification alsoassumes the strata will remain relatively consistentthroughout the life of the study, an assumption oftendifficult to meet in long-term field studies.

Minimizing bias. Sampling (study) methods shouldbe selected to minimize bias in the outcome of thestudy. Green (1979) provides several examples ofbias introduced by study methods. In field studies itis probable that study methods will always introducesome bias. This bias can be tolerated if it is relativelysmall, measurable, or consistent among study areas.For example, the estimation of bird use within windplants and reference areas may be accomplished byvisual observation. The presence of the observer nodoubt influences bird use to some extent. However,if the observations are made the same way in bothareas then the bias introduced by the study methodshould have little influence on the measured differ-ence in use between the two areas, which is theparameter of interest. Methods introducing severebias should be avoided.

Size and distribution of study plots. The size anddistribution of study plots also is an important com-ponent of the study method. Skalski et al. (1984)illustrated how field designs that promote similarcapture (selection) probabilities in the different pop-ulations being compared result in comparisons withsmaller sampling error. Green (1979) points out thatplot size makes little difference if organisms are dis-tributed at random throughout the study area, butthat use of a larger number of smaller plots increases

precision with aggregated distributions. Since aggre-gated distributions are the norm in nature, it general-ly is better to use a larger number of smaller plotswell distributed throughout the study area or stra-tum.

Cost, logistics, the behavior of the organism beingstudied, and the distribution of the organism willdetermine plot size. Use of larger plots usuallyallows the researcher to cover more area at a lowerunit cost (e.g. cost/hectare sampled). Also, plots canbe so small that measurement error increases dra-matically (e.g. is the study subject in or out of theplot) or the variance of the sample increases becausethe detection of the organism is rare, resulting in adata set with a lot of zeros. As a rule, the smallestplot size practical should be selected.

Shape of study plots. The shape of study plots is animportant consideration. For example, fixed plot andline-intercept sampling work well with commonplant and animal species. In fixed plot samplingthere is an attempt at complete census of some char-acteristic within selected units. Assuming some formof unbiased sampling is conducted, fixed plot sam-pling should result in equal probability of selectionof each plot. Point-to-item and line transect sam-pling are more effective when sampling less com-mon items. However, line-intercept, somepoint-to-item methods (e.g., plotless estimates ofbasal area), and some applications of line transectmethods (e.g. when larger objects are more easilyseen) are biased in that larger individuals are morelikely to be included in the sample. The selection ofthe appropriate size and shape of study plot must bemade on a case by case basis and is an importantcomponent of the study protocol.

Pilot StudiesA few data can do wonders for the design of envi-ronmental studies. Environmental studies shouldmake maximum use of existing data. When little orno data exist, a pilot study can provide preliminarydata useful in evaluating estimates of needed samplesize, optimum sampling designs, data collectionmethods, the presence of environmental patternsand other factors which can affect the success of thestudy. Pilot studies can vary from reconnaissancesurveys to the implementation of a draft protocol ina portion of the study area for a relatively short peri-od of time. It may be false economy to try to savemoney by avoiding some preliminary data collectionthat could dramatically improve the quality of astudy. In the absence of data on the study area, thefirst time period of study often becomes the pilot


study. If the first period of study suggests majorchanges in the protocol, then the value of the firstdata set may be relatively low in the ultimate analy-sis of impacts, an important consideration fordesigns dependent on pre-impact data. While pilotstudies are not absolutely necessary they are recom-mended when the lack of data and/or delay due tostudy requirements are major concerns.

THE PHILOSOPHY OF STUDY DESIGN Statistical conclusions are made under two broadand differing philosophies for making scientific inferences: design/data-based and model-based.Widespread confusion surrounds these philosophies,both of which rely on current data to some degreeand aim to provide “statistical inferences.” There is acontinuum from strict design/data-based analysis(e.g., finite sampling theory [Cochran 1977] and ran-domization testing [Manly 1991]) to pure model-based analysis (e.g., habitat suitability indices/habitatevaluation procedures [HSI/HEP (U.S. Fish andWildlife Service 1980)] using only historic data[USDI 1987]). A combination of these two types ofanalyses is often employed, resulting in inferencesbased on a number of interrelated arguments.

Design/Data-Based AnalysisIn strict design/data-based analysis, basic statisticalinferences concerning the study areas are justified bythe design of the study and data collected (Cochran1977, Scheaffer et al. 1990). Computer intensive sta-tistical methods (e.g., randomization, permutationtesting, etc.) are available without requiring addi-tional assumptions beyond the basic design protocol(e.g., Manly 1991). Design/data-based statisticalconclusions stand on their own merits for theagreed-upon:

• impact indicators

• procedures to measure the indicators

• design protocol.

Re-analysis of the data at a later time cannot declarethese basic statistical inferences incorrect. The datacan be re-analyzed with different model-basedmethods or different parametric statistical methods;however, the original analysis concerning the studyareas will stand and possess scientific confidence ifconsensus is maintained on the conditions of thestudy (bulleted items above).

Model-Based AnalysisPredictive methods estimate risk and impact throughthe use of models. In the extreme case of model-based analysis where no new data are available, allinferences are justified by assumption, are deduc-tive, and are subject to counter-arguments. The morecommon model-based approach involves the combi-nation of new data with parameters from the litera-ture or data from similar studies by way of atheoretical mathematical/statistical model. An exam-ple of this approach in the evaluation of wind plantimpacts on bird species is the demographic model-ing of a bird population combined with use of radio-telemetry data to estimate the influence of the windplant on critical parameters in the model. Thisapproach is illustrated by the telemetry studies inAltamont Pass, California, as described by Shenk etal. (1996).

Mixtures of Design/Data-Based and Model-Based AnalysesOften inferences from study designs and data requiremixtures of the strict design/data-based and puremodel-based analyses. Mixtures of study designswould include those analyses where:

1. design/data-based studies are conducted on afew important bird species

2. manipulative tests are conducted using surrogatespecies to estimate the effect of exposure to wind turbines on species of concern (Cade1994)

3. deductive professional judgment and model-based analyses are used to quantify impacts oncertain components of the habitat in the affectedarea.

Strict adherence to design/data-based analysis inquantifying injuries often may be impossible, but it isrecommended that the design/data-based analysis beadhered to as closely as possible. The value of indis-putable design/data-based statistical inferences on atleast a few impact indicators cannot be overempha-sized in establishing confidence in the overallassessment of impact due to wind plants. However,in some circumstances model-based methods pro-vide a suitable alternative to design/data-basedmethods. The advantages, limitations, and appropri-ate applications of model-based methods are dis-cussed further in chapter 4 and in Gilbert (1987),Johnson et al. (1989), and Gilbert and Simpson(1992).


Levels of StudiesTo simplify discussion of studies of wind plant effectson birds, we have divided the discussion of specificstudy protocols into discussions of broad, less inten-sive (Level 1) studies and more detailed (Level 2)studies.

Level 1 studies include pre-permitting baseline stud-ies, risk assessment studies, and monitoring studiesdesigned to detect the relatively large effects of oper-ating wind plants. Studies to determine the relativerisk of wind plants to species and communities, aswell as monitoring studies, normally would beLevel 1 studies. Level 2 studies involve detailed stud-ies of one or more bird populations and manipula-tive studies designed to determine the mechanismsof fatality or risk. Basic research on fatality path-ways, the quantification of risk to populations, andthe evaluation of risk reduction management prac-tices normally involve Level 2 studies. This chapterprovides an introduction to the basics of studydesign and discusses Level 1 studies in some detail.For the remainder of this chapter, the wind plant isconsidered to be a treatment in a scientific study andthe area affected by the wind plant to be the assess-ment area.

Level 1 studies generally will be useful in the follow-ing situations:

• A wind plant site is selected but the distributionof turbines and turbine strings has not beendetermined and turbine siting could be influ-enced by information on potential risk to birdspecies.

• The decision to construct a wind plant has beenmade, but development will proceed in phasesbased on assessment of impacts of constructionand operation of the initial phase.

• A wind plant exists but the extent and impor-tance of bird impacts is unknown.

• Other studies or credible information on birduse and habitat suggest impacts are likely.

Once the decision is made to conduct Level 1 stud-ies, the following potential issues must be identifiedand considered.

1. The area of interest (area to which statisticaland deductive inferences will be made). Optionsinclude the plant site, the entire WRA, the localarea used by birds of concern, or the bird

population potentially affected (in this case pop-ulation refers to the group of birds interbreedingand sharing common demographics).

2. Time period of interest. The period of interestmay be (for example) diurnal, nocturnal, season-al, or annual.

3. Species of interest. The species of interest maybe based on behavior, fatalities in existing windplants, abundance, or legal/social mandate.

4. Potentially confounding variables. These mayinclude landscape issues (e.g. large scale habitatvariables), biological issues (e.g. variable preyspecies abundance), land use issues (e.g. rapidlychanging crops and pest control), weather, studyarea access, etc.

5. Time available to conduct studies. The timeavailable to conduct studies given the projectdevelopment schedule will often determine howstudies are conducted and how much data canbe collected.

6. Budget. Budget is always a consideration forpotentially expensive studies. Budget should notdetermine what questions to ask but will influ-ence how they are answered. It will largelydetermine the sample size, and thus the degreeof confidence one will be able to place in theresults of the studies.

7. Project magnitude. Project magnitude will oftendetermine the level of concern and the requiredprecision.

Level 1 studies to quantify risk and impacts of windplants typically will use an observational design withstudy areas not selected by random procedure.Observational studies also are referred to as “samplesurveys” (Kempthorne 1966), “planned surveys”(Cox 1958), and “unplanned experiments/observa-tional studies” (National Research Council 1985).The objective of observational studies is usually anestimate of parameters necessary to describe the sta-tistical population, such as density, survival rates,natality, and habitat use (Skalski and Robson 1992).In this case, the statistical population is defined asthe group of animals or other objects of study. SeeManly (1992) and McKinlay (1975) for excellent dis-cussions of the design and analysis of observationalstudies.


An observational study of the impacts of a windplant on bird species is not a true experimentbecause selection of the area to receive the windplant and selection of the areas to be the referencesare not by a random procedure. The wind resourceassessment area may consist of several disjoint sub-regions affected by wind turbines. These disjoint seg-ments of the wind plant may be further stratified intomajor vegetation types. A potential undeveloped ref-erence site may have areas within its boundary thatappear similar to the wind plant and may also bestratified by the same major vegetation types. Eventhough the logic used in the study of these areas isthat both the assessment area and the reference areaare stratified into vegetation types, and study sitesare randomly selected from within strata, these sub-regions are not independent replicates of the windplant. Random selection of study sites/organismsfrom assessment and reference areas is known assubsampling. In the end, in an Impact-Referencestudy design, only one wind plant in one area isavailable for comparison to one or more subjectivelyselected reference areas.

DESIGN/DATA-BASED STUDIESBoth design/data-based and model-based methodsmay be used in Level 1 studies. Both methods bene-fit from historic and current data collected accordingto repeatable and reliable field studies. This sectioncontains designs that are most appropriate for Level1 studies, but can be used in Level 2 studies. Studiesfollowing the recommended designs are repeatable.Statistical results from repeated sampling followingthe same design would apply to the same universe ofstudy; whether the universe of study is an assessmentarea, an assessment population, or a time period ofinterest.

There are several alternative methods of study whenestimating impact. The following designs arearranged approximately in order of reliability, forsustaining confidence in the scientific conclusions. Itmust be understood that no one method is alwaysbest; the method selected for a particular study willdepend on a number of issues, as discussed below.

We also discuss designs for studies that make com-parisons between assessment areas and areas withsimilar physical and biological characteristics. Theseareas often are termed control areas but are not truecontrols in the experimental sense (i.e., a near per-fect match to the assessment area). Since good con-trol areas seldom exist in field studies, we will usethe term reference area instead. The term is definedin the same way as Stewart-Oaten (1986) and others

have used the term control area: an area representa-tive of the assessment area. The term “referencearea” appropriately illustrates that, in observationalstudies, the differences between an assessment areaand an area to which it is compared must be consid-ered in light of the high degree of natural variabilityamong any two sites.

Designs with Control (Reference) AreasThe Before-After/Control Impact Design (BACI)The Before-After/Control (reference)-Impact (BACI)design is common in the literature (e.g., Stewart-Oaten 1986), and has been called the “optimalimpact study design” by Green (1979). It is equiva-lent to the paired control-treatment design proposedby Skalski and Robson (1992). The term BACI is socommon in the literature that the letter C must beretained in its name, even though we use the term“reference area” rather than “control area.”

The BACI design is very desirable for impact deter-mination because it addresses two major impactstudy design problems.

1. Impact indicators, such as the abundance oforganisms, vary naturally through time, so anychange observed in an assessment area betweenthe pre- and post-impact periods could conceiv-ably be unrelated to the treatment (e.g., the con-struction and operation of a wind plant). Largenatural changes are expected during an extend-ed study period.

2. There always are differences in the indicatorsbetween any two areas (again, consider birdabundance). Observing a difference betweenassessment and reference areas following thetreatment does not necessarily mean that thewind plant was the cause of the difference. Thedifference may have been present prior to con-struction. Conversely, one would miss a windplant impact if the abundance of the indicatoron the reference area were reduced by someother perturbation concurrent with constructionof the wind plant.

The BACI design helps with these difficulties. By col-lecting data at both reference and assessment areasusing exactly the same protocol during both pre-impact and post-impact periods one can ask thequestion: Did the average difference in abundancebetween the reference area(s) and the wind plantarea change after the construction and operation?


The BACI design is not always practical or possible.Adequate reference areas often are difficult to locate,and while preliminary analysis may satisfy the per-mitting agency that a project may proceed, the plan-ning of wind plant projects does not always allowenough time for a full-scale pre-impact study period.The multiple time periods necessary for this designusually increase the cost of study. Additionally, alter-ations in land use or disturbance occurring overthese time periods and reference areas complicatethe analysis of study results. Caution should be usedwhen employing this method in areas where poten-tial reference areas are likely to incur relatively largealterations or changes that impact the species beingstudied. In the case of small homogeneous areas ofpotential impact and where a linear response isexpected the impact gradient design may be a moresuitable design. If advanced knowledge of a windplant location exists, the area of impact is somewhatvaried, and species potentially impacted are wideranging, the BACI design is preferred for observa-tional studies of impact.

Matched Pairs in the BACI DesignMatched pairs of sites from assessment and referenceareas often are subjectively selected to reduce thenatural variation in impact indicators (Skalski andRobson 1992). Eberhardt (1976) labeled designsusing this matching “pseudo-experiments” becauseof the lack of randomization and true replication oftreatments and control conditions. Statistical analysisof these pseudo-experiments is dependent on thesampling procedures used for selection of sites andthe amount of information collected on concomitantsite-specific variables. For example, sites may berandomly selected from the assessment area andeach subjectively matched with a site from a refer-ence area. In this case the area of inference is to theassessment area, and the reference pairs simply actas an indicator of baseline conditions.

When applied to a wind plant or other non-randomperturbations (treatments), the extent of statisticalinferences when matched pairs are used in the BACIdesign is limited to the assessment area. The infer-ences also are limited to the protocol by which thematched pairs are selected. If the protocol for selec-tion of matched pairs is unbiased, then statisticalinferences comparing the assessment and referenceareas are valid and repeatable. The selection ofmatched pairs for extended study contains similarrisks associated with stratification. The presumptionis that, with the exception of the treatment, the pairsremain very similar — a risky proposition in long-term studies.

Primary references for design and analysis areSkalski and Robson’s (1992: chapter 6) Control-Treatment Paired (CTP) design and Stewart-Oaten’s(1986) Before/After-Control/Impact-Pairs (BACIP)design. If there are modifications of the basic struc-ture of the design, then statistical analysis of theresulting data will not follow standard textbookexamples.

Impact-Reference Design (After Treatment) The Impact-Reference Design is for quantification ofimpact in studies where the impact indicators meas-ured on the assessment area are compared to meas-urements from one or more reference areas. In theImpact-Reference Design, data collected after thewind plant is operational are contrasted between theassessment and reference areas. The Impact-Reference design is considered because proposedand existing wind plants often lack “before construc-tion” baseline data from the assessment area and/ora reference area. In these cases, the BACI design isnot applicable and an alternative must be found.Assessment and reference areas are censused or ran-domly subsampled by an appropriate observationaldesign. Design and analysis of wind plant impacts inthe absence of pre-impact data follow Skalski andRobson’s (1992: chapter 6) recommendations foraccident assessment studies.

Differences between assessment and reference areasmeasured only after the impact might be unrelatedto the impact, because site-specific factors differ. Forthis reason, differences in natural factors betweenassessment and reference areas should be avoided asmuch as possible. However, differences usually willexist. Reliable quantification of impact must includeas much temporal and spatial replication as possible.Additional study components, such as the measure-ment of other environmental factors that might influ-ence impact indicators, may also be needed to limitor explain variation and the confounding effects ofthese differences. Environmental indicators often aretermed covariates because analysis of covariancemay be used to adjust the analysis of a random vari-able to allow for the effect of another variable.

Designs Without Reference AreasImpact-Gradient DesignsThe Impact-Gradient Design is for quantification ofimpact in relatively small assessment areas onhomogeneous environments. If potentially impactedspecies have relatively small home ranges (e.g.passerines) and a gradient of response is anticipated,this design is a preferred approach to impact studies.When this design is appropriate, associated costs


should be less than for those designs requiring base-line data and/or reference areas and impacts can beestimated with more confidence.

Analysis of the Impact-Gradient Design is based onan analysis of the relationship between the impactindicator and distance from the hypothesized impactsource — in this case, wind turbines. In effect, theassessment area includes the reference area on itsperimeter. This design does not require that theperimeter of the assessment area be free of impact,only that the level of impact be different. If a gradi-ent of biological response(s) or distance is identified,the magnitude of differences can be translated intowhat can be presumed to be at least a minimum esti-mate of the amount of impact. This Impact-GradientDesign would be analogous to a laboratory toxicitytest conducted along a gradient of toxicant concen-trations. An example might be an increasing rate offledgling success in active raptor nests or a decreasein passerine mortality as a function of distance fromthe wind plant.

In a field study, there likely will be naturally varyingfactors whose effects on the impact indicators areconfounded with the effects of the impact. Thus it isimportant to have supporting measurements ofcovariates to help interpret the gradient of responseobserved in the field study. In the example ofdecreased mortality in passerines, an obvious covari-ate to consider would be vegetation type.

Data collected from these studies may also be ana-lyzed from the philosophy of the designs with refer-ence areas if one discovers that a gradient ofresponse is absent but a portion of the study areameets the requirements of a reference area. Theimpact gradient design can be used in conjunctionwith BACI, Impact Reference and Before-Afterdesigns.

Before-After DesignsThe Before-After Design is for the quantification ofimpact when measurements on the assessment areabefore the impact are compared to measurements onthe same area following the impact. This design isconsidered because it is possible that large-scalemonitoring of birds within an area might be under-taken if enough concern exists for their securitywithin a potential WRA. Government agencies orprivate industry may monitor impact indicators overlong periods of time, and reliable baseline data mayexist. If so, measurements can be made after theincident using exactly the same protocol and SOPs.

However, observed differences might be unrelated tothe incident, because confounding factors alsochange with time (see the above discussion of theBACI design). Reliable quantification of impact usu-ally will include additional study components tolimit variation and the confounding effects of naturalfactors that may change with time.

Because of the difficulty in relating post-impact dif-ferences to treatment effects in the absence of datafrom reference areas, injury indicators can be partic-ularly useful in detecting impacts using Before-AfterDesign. The correlation of exposure to toxic sub-stances and a physiological response in wildlife hasbeen documented well enough for some substancesto allow the use of the physiological response as abiomarker for evidence of impact. Examples of bio-markers used in impact studies include the use ofblood plasma dehydratase in the study of lead expo-sure, acetylcholinesterase levels in blood plasma inthe study of organophosphates, and the effect ofmany organic compounds on the microsomalmixed-function oxidase system in liver (Peterle1991). The number of dead birds in some definedarea determined by necropsy to be caused by a windplant could be used as such an indicator. It is possi-ble that existing biomarkers (e.g., biomarkers indi-cating stress) might also have some application toestimating wind plant impacts on birds.

Costs associated with conducting the Before-AfterDesign should be less than that required for designsrequiring reference areas. Statistical analysis proce-dures include the time-series method of interventionanalysis (Box and Tiao 1975). An abrupt change inthe impact indicator at the time of the impact mayindicate the response is due to the perturbation (saya wind plant). Scientific confidence is gained thatthe abrupt change is due to the wind plant if theimpact indicator returns to baseline conditionsthrough time after making adjustments of factors inthe wind plant apparently related to observedimpacts (Figure 3-1).

If the impact indicator returns to baseline conditionsduring the operation of the wind plant, impactswould be considered short-term and the absence oflong-term impacts would be suggested. However,interpretation of this type of response without refer-ence areas or multiple treatments is difficult andsomewhat subjective. This type of design is mostappropriate for short-term impacts, rather than forlong-term projects such as a wind plant.


IMPROVING THE RELIABILITY OF STUDYDESIGNSUse of More than One Reference AreaUse of two or more reference areas increases thereliability of conclusions concerning quantificationof impact (Underwood 1994). Reliability and validityof a scientific study for quantification of impact oftenwill be questioned on the basis that “the referencearea is not appropriate for the assessment area.”Consistent relationships between the assessmentarea and each of two (or more) reference areas willgenerate far more scientific confidence in the resultsthan if a single reference area is used. This scientificconfidence will likely be increased more than wouldbe expected given the increase in number of refer-ence areas. This is true whether the wind plant isconcluded to have “an important impact” or “noimportant impact.” The use of multiple referenceareas has the disadvantage of increased cost.

With two or more reference areas, one will be ableto compare the impact indicators between differentreference areas during the assessment period.Multiple reference areas also will allow a compari-son of impact indicators from the assessment areawith the mean of impact indicators from two or

more reference areas. For example, consider a windplant and two reference areas outside the influenceof but in the same general area as the wind plant. Ifapproximately the same differences exist betweenthe impact indicators on the wind plant and each ofthe reference areas before construction, then this“replication in space” usually gives scientists moreconfidence when making deductive professionaljudgments regarding post construction impacts.

In practice, impact indicators for the three areas willbe plotted and examined for relative changes beforeand after construction of the wind plant. Assumingall three areas have similar trends in impact indica-tors before impact and reference areas have similartrends after impact, tests for differences will bebetween the mean of the impact indicators for multi-ple reference areas and the value of the impact indi-cator for the wind plant. By studying the effect of afew important covariates on the impact indicator onthe wind plant and reference areas, it may be possi-ble to adjust raw data before comparisons of meanvalues are made. For example, if nestling survival ishighly correlated with prey abundance it might bepossible to adjust survival rates for differences inprey on reference and assessment areas before test-ing for wind plant effects.

Collection of Data Over Several TimePeriodsCollection of data on the study areas for several timeperiods before and/or after the impact also willenhance reliability of results. Confidence in the rela-tionship of assessment and reference areas isimproved. Figure 3-2 illustrates results from a BACIdesign with two periods for data collection beforethe wind plant impact and two periods of data col-lection following the wind plant development. Inthis sketch there is only a slight indication of recov-ery after the construction of the wind plant.Statistical tests or other analyses (e.g., confidenceintervals) unique to the subsampling plan used indata collection will be required for judging whetherstatistically significant differences exist between thepoint estimates.

For example, assume data on a response variable —say the number of fledglings per active nest — existfor two years before construction and two years afterconstruction of a wind plant with one referencearea. Assume also that the data meet the assump-tions necessary for use of analysis of variance(ANOVA). ANOVA would be used to test for inter-action among study sites and years, the primary


Period: 1 2 3 4 5

X

BEFORE AFTER

Figure 3-1. Idealized sketch of an impact indicator in aBefore-After Design with five time periods (T) of interestwhere an abrupt change coincides with an impact and isfollowed by a return to baseline conditions.

indicator of an effect due to the development. Asignificant interaction effect may indicate that a pre-treatment difference between a developmentarea and reference areas is not equal to the post-treatment difference. Additional comparisons couldbe made, such as the comparison of the meanresponse pre-treatment with the response each yearpost-treatment or with the mean over all years post-treatment. Results would be presented graphically to illustrate point estimates and precision (confi-dence intervals or standard errors). The statisticalinference would be limited to the two areas and the four years.

The specific test used depends on the response variable of interest (count data, percentage data,continuous data, categorical data, etc.) and the subsampling plan used (point counts, transectscounts, vegetation collection methods, GIS dataavailable, radio-tracking data, capture-recapturedata, etc.). Often, classic ANOVA procedures will be inappropriate and nonparametric, or other com-puter-intensive methods will be required.

Interpretation of Area-by-Time InteractionsNon-parallel responses for impact indicators plottedover time on assessment and reference areas are saidto exhibit area-by-time interaction (Figure 3-3).

If abrupt changes in the relationship of assessmentand reference areas occur following the impact andare followed by a return to baseline conditions, thenscientific confidence is gained for the conclusionthat the abrupt changes were due to the impact. Thisinteraction is illustrated in Figure 3-4, where the dif-ference between the impact indicator on the refer-ence and assessment areas represents the magnitudeof an impact. Also, a return to a relationship similarto baseline conditions provides additional scientificconfidence that comparison of assessment area andthe subjectively selected reference areas is appropri-ate for estimating impact (Skalski and Robson 1992).In the case of a wind plant, recovery suggests achange in bird behavior reducing risk, a temporaryimpact due to construction, or a change in the windplant (e.g., safer turbines).

Evidence of significant area-by-time interaction isespecially important in an Impact-Reference Design,because this may be the only factor which aids inestimating the difference, if any, between the refer-ence areas and assessment area in the absence of the


Period: 1 2 3 4 BEFORE AFTER

REFERENCE AREA

WIND PLANT

Figure 3-2. Sketch of point estimates of an impactindicator in an idealized BACI design over four timeperiods with slight indication of recovery after theincident.

Period:1 2 3 4 5

BEFORE AFTER

REFERENCE AREA

WIND PLANT

Figure 3-3. Sketch of point estimates of an impactindicator in an idealized BACI design where interactionwith time indicates recovery from impact by the third timeperiod following the incident.

impact. This situation is illustrated in figure 3-4 withan idealized presentation of a large differencebetween the assessment and reference area follow-ing the impact, which is followed by a return toapproximately parallel responses of data plotted overtime. (Note that this figure, like the others in thischapter, is an idealized hypothetical presentation.Real data points would necessarily include errorbars.) This interaction could indicate that impactswere temporary or that a significant change has beenmade in the operation of the wind plant (say installa-tion of safer turbines or removal of turbines responsi-ble for the impact).

Model-Based Analysis and Use ofConcomitant Site-Specific Variables Pure design/data-based analysis often is not possiblein impact studies. For example, bird abundance inan area might be estimated on matched pairs ofimpacted and reference study sites. However care-fully the matching is conducted, uncontrolled factorsalways remain that may introduce too much varia-tion in the system to allow one to statistically detectimportant differences between the assessment andreference areas. In a field study, there likely will benaturally varying factors whose effects on the impactindicators are confounded with the effects of theincident. Data for easily obtainable random vari-ables that are correlated with the impact indicators(covariates) will help interpret the gradient of

response observed in the field study. These variablesordinarily will not satisfy the criteria for determina-tion of impact, but can be used in model-basedanalyses for refinement of the quantification ofimpact (Page et al. 1993, Smith 1979). For example,in the study of bird use on the Wyoming wind plantsite, WEST Inc. (1995) developed indices to preyabundance (e.g. prairie dogs, ground squirrels, andrabbits). These ancillary variables are used in model-based analyses to refine comparisons of avian preda-tor use in assessment and reference areas. Land usealso is an obvious covariate that could provideimportant information when evaluating differencesin bird use among assessment and reference areasand time periods.

Indicators of degree of exposure to the impact-pro-ducing factor also should be measured on samplingunits. As in the Impact-Gradient Design, a clearimpact-response relationship between impact indi-cators and degree of exposure will provide corrobo-rating evidence of impact. These indicators also canbe used with other concomitant variables in model-based analyses to help explain the “noise” in datafrom natural systems. For example, the size of tur-bines, the speed of the turbine blades, the type ofturbine towers, etc. can possibly be considered indicators of the degree of exposure.

Sampling The Area Of InterestIn this section, the word sample means either theprocess by which units of observation in a specificarea are selected, or the actual collection of unitsselected for study. The study area consists of either a finite or an infinite universe of sampling units. Forexample, a small site might be divided into a finiteset of 1-m x 1-m plots, each having an opportunityto be selected in the sample. A sample of plots isselected from the area and measurements are madeof indicators such as the number and biomass ofplants or animals on each plot. In this case, the wordsample refers more to the location of the units thanto the specimen (plant, animal, sediment, etc.) col-lected from the unit.

If one is interested in the set of animals or plants liv-ing on (or influenced by) the assessment or referencestudy sites, then a second universe exists: namely,the population of animals or plants. The word popu-lation in this case refers to the group of organismsunder study (the statistical population) and not nec-essarily to the biological population. This seconduniverse also can be sampled and used to make sta-tistical inferences to the group of organisms living in or influenced by the study area. For example, the


Period: 1 2 3 4 BEFORE AFTER

REFERENCE AREA

WIND PLANT

WIND PLANT

Figure 3-4. Idealized sketch of results from a Reference-Impact Design where a large initial difference in theimpact is followed by a shift to parallel response curves.

impact of a wind plant on breeding pairs of raptorsmay extend >20 km from the turbines (determinedby the range of the birds) and a capture-recapturemodel-based study may be undertaken of the breed-ing pairs within the WRA and a 20 km radius. In thiscase the marked animal is treated as the sample unit.All of the techniques for study of animal or plantpopulations in field ecology (plotless methods fromforestry, capture-recapture methods from wildlife science, etc.) become candidates for study of theimpacts of a wind plant. Chapter 4 provides moredetail on the detailed study of wind plant impacts on mobile wildlife.

Two Levels of SamplingFor a smaller wind plant with a less extensive assess-ment area, the entire area may be the study site,resulting in only one level of sampling. However,wind plants may affect relatively large areas, inwhich case the WRA as a whole is “sampled” forstudy sites, and each of these sites is then sampled,resulting in two levels of sampling. (In a study of rap-tor use on the 60,619 acres of the Wyoming WRA,18 study sites were selected for a second level ofsampling.) In addition, present technology does notallow direct measurement of some environmentalindicators (e.g., the number of passerine nests byspecies) on even moderately large areas. Destructivesampling, which permanently changes the sampledpoint or line (e.g., by removing vegetation), also maybe required; and only a small part of the site can bedestructively sampled without changing the verynature of the site.

If this second level of sampling (subsampling) is conducted according to an acceptable randomizeddesign, then statistical inferences can be made to thestudy sites. But inferences beyond the study sites tothe assessment and reference areas will be deductiveunless the first level of sampling (study-site selection)also was conducted according to an acceptable ran-domized procedure.

It should not be surprising that two different studieson the same wind plant may yield conclusions thatdiffer, given that:

1. study sites within the wind plant may be select-ed using different criteria

2. subsampling protocols and SOPs for measure-ment (or estimation) of indicators at a site maydiffer between the two studies.

This again emphasizes the importance of rigorousselection and documentation of sampling protocolsand SOPs so that the conclusions drawn from astudy can be defended. It also illustrates the impor-tance of having similar protocols for the study ofimpacts on birds by a new technology (wind tur-bines) in widely separated areas. However, even ifidentical areas, designs, and SOPs are used, resultsof studies based on independent sample units willfluctuate because of natural variation within the areaand variation in the application of methods.Resolution of such apparently conflicting results mayrequire intensive investigation of sampling designs,sampling protocols, sample processing, and dataanalysis by experts in the specific biological areasand experts in study design and statistical analysis.

Sampling PlansStatistical inferences can be made only with refer-ence to the protocol by which study sites and/orstudy specimens are selected from the assessmentand reference areas. Statistical inferences also arereferenced to the protocol used for subsampling (orcensus) of units from sites and to the SOPs for meas-urement of impact indicators on subsampled units.Sampling plans can be arranged in four basic cate-gories (Gilbert 1987):

1. haphazard sampling

2. judgment sampling

3. search sampling

4. probability sampling

Sampling plans that are most likely to be used duringimpact quantification associated with wind plantdevelopment are discussed below. (See Gilbert1987:19-23, Gilbert and Simpson 1992 and Johnsonet al. 1989 for other common variations of probabili-ty sampling.)

Haphazard SamplingGilbert (1987:19) noted that:

“Haphazard sampling embodies thephilosophy of ‘any sampling location will do.’This attitude encourages taking samples atconvenient locations (say near the road) ortimes, which can lead to biased estimates ofmeans and other population characteristics.Haphazard sampling is appropriate if thetarget population is completely


homogeneous… This assumption is highlysuspect in most environmental studies.”

Haphazard sampling has little role to play in provid-ing data for statistical inferences, because results arenot repeatable. Information from haphazard sam-pling may be appropriate for preliminary reconnais-sance of an area, but the information can be usedonly in making deductive arguments based on pro-fessional judgment.

Judgment Sampling“Judgment sampling means subjective selection ofpopulation units by an individual [the researcher]”Gilbert (1987:19). Gilbert is not much more enthusi-astic about judgment sampling than haphazard sam-pling:

“If the [researcher] is sufficientlyknowledgeable, judgment can result inaccurate estimates of population parameterssuch as means and totals even if allpopulation units cannot be visually assessed.But it is difficult to measure the accuracy ofthe estimated parameters. Thus, subjectivesampling can be accurate, but the degree ofaccuracy is difficult to quantify” (1987:19).

As in haphazard sampling, judgment sampling maybe appropriate for preliminary reconnaissance of anarea, but has little role to play in providing data forstatistical inferences, because results are not repeat-able. Judgment sampling has a role to play in under-standing and explaining the magnitude and durationof an impact, but inferences are deductive anddepend on professional judgment.

Search SamplingSearch sampling requires historical knowledge ordata indicating where the resources of interest exist.For example, a study of factors causing bird fatalitiesmight be limited to the portion of the wind plantwhere bird use is common. Searching for “hotspots,” which is discussed more fully under the “costcutting procedures” section below, is a form ofsearch sampling. The validity of this proceduredepends on the accuracy of the information guidingwhere and when to search. The procedure alsoplaces a great deal of emphasis on the collection ofaccurate data over time and space to guide thesearch.

Probability SamplingProbability sampling refers to the use of a specificmethod of random sampling of sites from the

assessment and reference areas, or sampling of units from assessment and reference sites (Gilbert1987:20). Randomization is necessary to make prob-ability or confidence statements concerning themagnitude and/or duration of impact (Johnson et al.1989). Examples of random sampling plans includesimple random sampling (random sampling), strati-fied random sampling (stratified sampling), randomstart systematic sampling (systematic sampling), andsequential random sampling (sequential sampling).

These sampling plans (and others, especially formobile animals) can be combined or extended togive an amazing array of possibilities. Johnson et al.(1989: 4-2) recommend: “If other more complicatedsample designs are necessary, it is recommendedthat a statistician be consulted on the best design,and on the appropriate analysis method for thatdesign.” For example, after stratification of theimpact or reference area, one might use systematicsampling within strata for location of points to mark,release, and recapture mobile animals.

Random sampling. Random sampling requires thatthe location of each sample site (unit) be selectedindependently of all other sites (units). Such sam-pling plans have “nice” mathematical properties, butrandom locations are usually more clumped andpatchy than expected. Entire regions of special inter-est may be under- or over-represented. Some scien-tists mistakenly believe that random sampling isalways the best procedure. Random sampling shouldbe used in assessment or reference areas (sites) onlyif the area is very homogeneous with respect to theimpact indicators and covariates. Because this is sel-dom if ever the case, researchers should try to avoidrelying solely on random sampling.

Stratified sampling. Stratified sampling is a proce-dure designed to guarantee that the sampling effortwill be spread out over important subregions calledstrata which are identified in advance. Importantstrata are identified, and sites within strata are select-ed for study. Similarly, sites might also be stratifiedfor subsampling. Unless otherwise indicated, it isimplicit that locations for sample sites within strata(units within sites) are randomly or systematicallylocated.

Stratification is the division of an area into relativelyhomogeneous components. Strata may be subareason a map of the known range of the species of inter-est. Stratification may also be by reference to someknown characteristic of the species of interest (e.g.,areas of high and low numerical density) or by some environmental variable (e.g., vegetation type)


potentially influencing the species’ response to aperturbation.

Strata must not overlap, and all impact/referenceareas of interest must be included. Study sites (sam-pling units) must not belong to more than one stra-tum. Also, statistical inferences cannot be drawntoward differences in impact indicators for any por-tion of strata unavailable for sampling. It may bepossible to make professional judgments concerningthe magnitude and duration of impact on thoseareas, but conclusions will be made without the aidof inductive statistical results. As an example, in thestudies of golden eagles in Altamont (Hunt et al.1995) some private lands were not accessible fortrapping eagles. The resulting relocation data mustbe analyzed with the knowledge that the radio-tagged sample is not a random sample of the population.

Stratification often will be used in impact studies forquantification of impact within strata and for con-trasting the impacts of the incident between strata.For example, it may be of interest to investigate theimpacts of a wind plant in different vegetation types(a potential stratification) where the objective is tomake statistical inference to each vegetation typewithin the wind plant. This type of analysis isreferred to as using “Strata as domains of study... inwhich the primary purpose is to make comparisonsbetween different strata...” (Cochran 1977: 140). Inthis situation, the formulas for analysis and for allo-cation of sampling effort (Cochran 1977: 140-141),are quite different from formulas appearing in intro-ductory texts such as Scheaffer et al. (1990). Thestandard objective considered in textbooks is to min-imize the variance of summary statistics for all stratacombined (e.g., the entire wind plant).

It is usually stated in textbook examples that a pri-mary objective of stratification is improved precisionbased on optimal allocation of sampling effort intomore homogeneous strata. This sounds nice, butthere is a problem with this objective. It may be pos-sible to create homogeneous strata with respect toone primary indicator (or a few indicators), but thereare often many indicators measured, and it is notlikely that the units within strata will be homoge-neous for all of them. For example, one could stratifya study area based on vegetation and find that thestratification works well for indicators of impactassociated with overstory vegetation. But because ofmanagement (e.g., grazing), understory vegetationmight be completely different and make the stratifi-cation unsatisfactory for indicators of impact meas-ured in the understory. Further, anticipated reduction

in variance for the primary indicators may not occuror may be in the range of 5% to 10% and thus notsubstantially better than random sampling.Systematic sampling with post-classification intodomains of interest (subpopulations) in the spirit ofthe U.S. EPA Environmental Monitoring andAssessment Program (Overton et al. 1991) may per-form better than stratified random sampling. (See thediscussion on systematic sampling below.)

Factors on which to stratify in quantification ofimpact associated with wind plant developmentcould include physiography/topography, vegetation,land use, turbine type, etc. Strata should be relative-ly easy to identify by the methods that will be usedto select strata and study sites within strata, and ofobvious biological significance for the indicators ofimpact. Spatial stratification is a major help whenstudy is of relatively short duration and very few sites(units) are misclassified. The reality is that the bestlaid plans go astray. Some potential study sites willbe misclassified in the original classification (e.g., apond on the aerial photo was actually a parking lot).The short-term study may turn into a long-term studyin which interests migrate toward complicatedanalysis of subpopulations (Cochran 1977: 142-144)which cross strata boundaries, and strata maychange (e.g. the corn field has become a grassland).In long-term studies investigators are likely to behappiest with the stratification procedure at thebeginning of the study. Benefits of stratification oncharacteristics such as vegetative cover type, densityof prey items, land use, etc. diminish quickly asthese phenomena change with time.

A fundamental problem is that strata normally are ofunequal sizes and, thus, units from different stratahave different weights (importance values) in anyoverall analysis to be conducted. Consider the rela-tively complex formulas for computing an overallmean and its standard error based on stratified sam-pling (Cochran 1977: 87-95). In the analysis of sub-populations (subunits of a study area) which belongto more than one stratum (Cochran 1977: 142-144),formulas are even more complex for basic statisticssuch as means and totals. The influence of theseunequal weights in subpopulations is unknown formany analyses such as ordination or multidimen-sional scaling. Many analyses of studies ignore theseunequal weights and assume the units from differentstrata are selected with equal probability.

Stratification is often based on maps, but studies usu-ally suffer from problems caused by inaccurate mapsor data concerning impact sites, reference sites, and


vegetation types at the time study sites are randomlyselected. The basic problems are two:

1. misclassified sites have no chance of selectionin the field SOPs used by investigations

2. unequal probability of site selection is intro-duced within strata.

It may be necessary to stratify with little prior knowl-edge of the study area; but if possible, stratificationshould be limited to geographic stratification withexcellent maps, and the minimum number of stratashould be used (preferably no more than three orfour). Covariates that are potentially correlated withthe magnitude and duration of impact should bemeasured on the study sites (or on subsampling unitswithin sites). Some analyses such as ordination andmultidimensional scaling may require additionaloriginal mathematical research for justification oftheir use.

Systematic sampling. Systematic sampling distributesthe locations of sites (units) uniformly over the area(site) with a random starting rule. Mathematicalproperties are not as “nice” as for random sampling,but the statistical precision generally is better(Scheaffer et al. 1990). Systematic sampling has beencriticized for two basic reasons. First, the arrange-ment of points may fall in step with some unknowncyclic pattern in the response of impact indicators.This problem is addressed a great deal in theory, butis seldom a problem in practice. If there are knowncyclic patterns in the area, one should use them toone’s advantage to design a better systematic sam-pling plan.

Second, in classical finite sampling theory (Cochran1977), variation is assessed in terms of how muchthe result might change if time could be backed upand a different random starting point could beselected for the uniform pattern. For a single uniformgrid of sampling points or plots (or a single set ofparallel lines) this is impossible, and thus variationcannot be estimated in the classical sense. Variousmodel-based approximations have been proposedfor the elusive measure of variation in systematicsampling (Wolter 1984).

Aside from the criticisms, systematic sampling worksvery well in the following situations:

1. design/data-based analyses conducted as if ran-dom sampling had been conducted (effectivelyignoring the potential correlation between

neighboring locations in the uniform pattern of asystematic sample [Gilbert and Simpson 1992])

2. encounter sampling with unequal probability(Otis et al. 1993, Overton et al. 1991)

3. the model-based analysis commonly known as“spatial statistics,” wherein models are proposedto estimate impact using the correlation betweenneighboring units in the systematic grid (see, forexample, Kriging [Johnson et al. 1989: chapter10]).

The design and analysis in Case (1), above, is oftenused in evaluation of indicators in relatively small,homogeneous study areas or small study areaswhere a gradient is expected in measured values ofthe indicator across the area. Ignoring the potentialcorrelation and continuing the analysis as if it is jus-tified by random sampling can be defended, espe-cially in impact assessment, primarily because froma statistical perspective the analysis is conservative.Estimates of variance treating the systematic sampleas a random sample will tend to overestimate thetrue variance of the systematic sample (Hurlbert1984, Scheaffer et al. 1990, Thompson 1992). Thebottom line is that systematic sampling in relativelysmall impact assessment study areas followingGilbert and Simpson’s (1992) formulas for analysis isa good plan. This applies whether systematic sam-pling is applied to compare two areas (assessmentand reference), the same area before and followingthe incident, or between strata of a stratified sample.

One of the primary reasons given for preference ofstratified sampling (see above) over systematic sam-pling is that distinct rare units may not be encoun-tered by a uniform grid of points or parallel lines.Hence, scientists perceive the need to stratify, suchthat all units of each distinct type are joined togetherinto strata and simple random samples are drawnfrom each stratum. As noted above, stratified ran-dom sampling works best if the study is short term,no units are misclassified, and no units change strataduring the study. Systematic sampling has been pro-posed to counter these problems in U.S. EPA’s longterm Environmental Monitoring and AssessmentProgram (EMAP), as described by Overton et al.(1991). Unequal probability sampling is almostinescapable, but the problems associated with mis-classified units and units that change strata over timecan largely be avoided. For long-term impact assess-ment, monitoring, or when problems with misclassi-fication and changes in land use are anticipated, one should consider systematic sampling strategies


similar to those proposed for EMAP (Overton et al.1991).

As an example, in the case of the Buffalo Ridge Sitein Minnesota (Strickland et al. 1996), a potentialstrategy was to stratify the WRA and reference areaby vegetation type. However, the two major vegeta-tion types present on the study area were fallowlands, reserved from tillage under the ConservationReserve Program (CRP), and lands being activelyfarmed. These vegetation types are both very influ-ential on bird use. Nevertheless, both vegetationtypes are likely to change within the next year or so.Thus, stratification on vegetation might be attractivein year 1 of the study, only to be completely inap-propriate in year 2 or 3. To overcome the problem, asystematic grid of points with a random startingpoint was established covering each study area. Fora given year, bird use within vegetation types on theassessment and reference areas are compared statis-tically as if the points were randomly located, even ifvegetation types change from year to year.

Cost-Cutting Sampling ProceduresOne of the biggest problems with large-scale fieldstudies is that they are very expensive. Estimating thenumber of birds of a large number of species usingan area is a prime example. Some of the standardsampling procedures that may reduce costs of field-work are presented below. These techniques shouldbe considered in design of all field studies.

Double sampling and Smith’s two-stage samplingprocedure. The basic idea of double sampling is thateasy-to-measure/economical indicators are meas-ured on a relatively large subset or census of sam-pling units in the assessment and reference areas. Inaddition, the expensive/time-consuming indicatorsare measured on a subset of the sampling units fromeach area. As always, easily obtainable ancillarydata should be collected. Analysis formulas areavailable in Cochran (1977). The ideas for doublesampling are simple to state and the method is easyto implement.

Smith’s (1979) two-stage sampling procedure is avariation of the general double sampling method.Basically, Smith’s suggestion is to over-sample in aninitial survey when knowledge concerning impactsis most limited and record economical easy-to-meas-ure indicators. For example, bird use (an index toabundance sampled according to a probability sam-ple) might be taken during a pilot study, allowingone to identify species most likely affected. In thesecond stage and with pilot information gained, the

more expensive and time-consuming indicators, e.g.,the actual number of individuals, might be measuredon a subset of the units. If the correlation betweenthe indicators measured on the double-sampledunits is sufficiently high, precision of statisticalanalyses of the expensive/time-consuming indicatoris improved.

Ranked set sampling. Ranked set sampling is a tech-nique originally developed in estimation of biomassof vegetation during study of terrestrial vegetation;however, the procedure deserves much broaderapplication (Muttlak and McDonald 1992, Stokes1986). The technique is best explained by a simpleillustration. Assume 60 uniformly spaced samplingunits are arranged in a rectangular grid in a WRA.Measure a quick, economical indicator of bird risk(say bird use) on each of the first three units, rank-order the three units according to this indicator andmeasure an expensive indicator (say bird fatalities)on the highest ranked unit. Continue by measuringbird use on the next three units (numbers 4, 5, and6), rank order them, and measure fatalities on thesecond-ranked unit. Finally, rank order units 7, 8,and 9 by bird use and measure fatalities on the low-est ranked unit; then start the process over on thenext nine units. After, completion of all 60 units, a“ranked set sample” of 20 units will be available onthe fatalities. This sample is not as good as a sampleof size 60 for estimating the number of bird fatalities,but should have considerably better precision than astandard sample of size 20.

Ranked set sampling is most advantageous when thequick, economical indicator is highly correlated withthe expensive indicator. These relationships need tobe confirmed through additional research. Also, themethodology for estimation of standard errors andallocation of sampling effort is not straightforward.

Sequential sampling. In sequential sampling, a statis-tical test is used to evaluate data after the impactindicator is measured on a subset of units or batch ofunits selected for sampling (Johnson et al. 1989:chapter 8, Mukhopadhyay et al. 1992). The results ofeach sequential test determine whether another sub-set of sampling units or batch of units will be collect-ed and analyzed. The procedure has obviousadvantages in certain situations where a large num-ber of samples are collected for laboratory analysis.In field studies, the estimate of certain biases, suchas the estimate of scavenger removal of carcasses bymonitoring carcasses placed in the field, might bene-fit from sequential sampling.


Johnson et al. (1989) gives an excellent presentationof the basic formulas for sequential analysis usingsimple random sampling. However, any variation inthe simple random sampling protocol (or simple sys-tematic sampling protocol) results in computationalrequirements not described in standard textbooks.Unexpected complexities are introduced into statistical procedures, because the “sample size” is a random variable (i.e., one cannot determine inadvance the number of sampling units which will be analyzed).

Adaptive sampling. In adaptive sampling the proce-dure for selecting sites or units to be included in thesample may depend on values of the variable ofinterest observed during the survey (Thomson andSeber 1996). Adaptive sampling takes advantage ofthe tendency of plants and animals to aggregate anduses information on these aggregations to directfuture sampling. Adaptive sampling could be consid-ered a method for systematically directing searchsampling.

As an example of adaptive sampling, say the windplant is divided into a relatively large number ofstudy units. A survey for bird carcasses is conductedin a simple random sample of the units. Each studyunit and all adjacent units are considered a “neigh-borhood” of units. With the adaptive design addi-tional searches are conducted in those units in thesame neighborhood of a unit containing a carcass inthe first survey. Additional searches are conducteduntil no further carcasses are discovered. As withsequential sampling, computational complexities areadded because of the uncertainty of the sample sizeand the unequal probability associated with theselection of units.

Sampling intensity. Usually the largest source ofvariation in impact indicators is due to natural varia-tion among sampling units across study areas andtime, not measurement and subsampling error (e.g.,determining the cause of death through blindnecropsy). Precision of statistical procedures andpower to detect important changes in impact indica-tors usually will be most influenced by an increasein the number of independent sampling units in theassessment and reference areas. A rule of thumb forimproving statistical precision is to increase thenumber of independent field sampling units. If pre-liminary or pilot data are available, optimal alloca-tion of financial resources to increase precision instatistical procedures (i.e., stratification) should beconsidered.

Searching for hot spots. Methods of searching forhot spots (i.e., areas within the assessment areawhich have high values of the impact indicator) maybe valuable under certain conditions — includingthe evaluation of whether impacts are significant andcontinuing. Johnson et al. (1989: chapter 9) model ahot spot as a localized elliptical area with values ofthe impact indicator above a certain standard. If asampling study does not find hot spots, then confi-dence is gained in the conclusion that the area is notimpacted above the standard or that impacts are notcontinuing above the standard. Techniques involvesystematic sampling from a grid of points arranged ina certain pattern and judgment that there are no hotspots of impact if none of the points yield valuesabove a given standard. This technique will be mostapplicable in wind plant monitoring studies whereregulatory standards for mortality exist, the study isof limited duration, and no reference areas are available.

Johnson et al. (1989: chapter 9) provide an excellentintroduction to the technique and give the analysesfor two basic approaches. If hot spots are detectedthen a decision must be made whether it is neces-sary to fully quantify the impact over the assessmentarea or just within the hot spots. For wind plants,monitoring for mortality might consider thisapproach if more extensive sampling suggests hotspots (e.g., end row turbines, turbines near wetlands,etc.). However, if bird use of the wind plant changes,more extensive monitoring may be required to iden-tify hot spots.

DATA ANALYSISUnivariate Analyses It is assumed that quantification of impact will bebased on measurements for indicators that satisfy thecriteria for determination of impact. For these indica-tors, conducting a series of independent univariateanalyses is recommended. For example, the numberof dead birds found per square kilometer (km2) ofwind plant surveyed following a year of operationmight be estimated and compared to the number ofdead individuals found per km2 on a reference area.During the same year of the same study, the numberof fledglings produced per nest might be estimatedand compared among the study areas.

It is recommended that impact and recovery of abiological community be defined in terms of individ-ual impact indicators. Examples of impact indicatorsinclude the number of individuals of a particularspecies, biomass of a particular species, and number of species present. Recovery is considered


incomplete and an impact exists in the biologicalcommunity as long as any differences (positive ornegative) in indicators can be detected betweenassessment and reference areas within the particularstudy design used (Page et al. 1993, Stekoll et al.1993). It is also recommended that:

• the biological community be characterized interms of relatively uncorrelated indicators thatare impact indicators; and that

• individual tests of direct and more understand-able measures of community response be usedrather than the multivariate indices mentionedbelow.

As an example, several comparisons of impact indi-cators — e.g., the numbers of several species andthe biomass of those same species — are madebetween a wind plant and reference areas. Thespecies selected should be relatively unrelated eco-logically (e.g. golden eagles and several species ofpasserines and shore birds). In the analysis of impactthe percentage of biological indicators that are sig-nificantly different (positive or negative) when testedat a given level of significance (Page et al. 1993,Stekoll et al. 1993) is used to determine the directionand magnitude of the impact. This use of a relativelylarge number of individual comparisons is related tothe vote-counting method of meta-analysis (Hedgesand Olkin 1985, Hedges 1986).

In spite of the recommendation above that indicatorsbe uncorrelated, the indicators (e.g. number of indi-viduals of a species) will always be correlated to acertain extent. Thus, the votes used in determiningimpact (i.e., the P-values from the indicators) are notindependent. Admittedly the procedure is ad hoc ifapplied only once after the impact, because theexpected percentage of significant differences isunknown (under the hypothesis that assessment andreference areas have the same distributions for indi-cators). However, impact to the community can beinferred if, for example:

• In a BACI design (with data collected before andfollowing the impact) there is an abrupt increasein the percentage of significant differences fol-lowing the incident (the inference will be morereliable if the abrupt increase is followed by areturn to baseline levels, i.e., recovery); or,

• In an Impact-Reference design (with several timeperiods of data collected following the impact)there is a large percentage of significant differ-ences relative to the size of the test (e.g., α =

0.05) immediately following the impact which isfollowed by a reduction in the percentage (theinference will be more reliable if the percentagedecreases to about 5%).

This form of data analysis increases the likelihood ofType I errors and makes the interpretation of resultsin studies with a large number of impact indicatorsdifficult. The assessment of the statistical significanceof differences is also more subjective than with mul-tivariate tests, placing a greater burden on theresearcher in evaluating the results. However, uni-variate tests help interpret results in terms of biologi-cal significance. As mentioned above, somecorrelation among impact indicators usually willexist and univariate analyses will help with the inter-pretation of the significance of this correlation in thedetermination of impact. In the univariate analysisthe detection of obvious impacts and their cause willbe more straightforward and more easily defendedwhen compared to multivariate indices of impact.

Multivariate AnalysisThere is a great deal of interest in simultaneousanalysis of multiple indicators (multivariate analysis)to explain complex relationships among many differ-ent kinds of indicators over space and time. This isparticularly important in studying the impact of aperturbation on the species composition and com-munity structure of flora and fauna (Page et al. 1993,Stekoll et al. 1993). These multivariate techniques(Gordon 1981, Green 1984, James and McCulloch1990, Ludwig and Reynolds 1988, Manly 1986,Pielou 1984, Seber 1984) include multidimensionalscaling and ordination analysis by methods such asprincipal component analysis and detrended canoni-cal correspondence analysis (Page et al. 1993). Ifsampling units are selected with equal probability bysimple random sampling or by systematic samplingfrom the assessment and reference areas, and nopseudo-experimental design is involved (e.g., nopairing), then the multivariate procedures are applicable.

It is unlikely that multivariate techniques will direct-ly yield impact indicators (i.e., combinations of theoriginal indicators) which meet the criteria for deter-mination of impact. The techniques certainly canhelp explain and corroborate impact if analyzedproperly within the study design. However, datafrom many recommended study designs are not easi-ly analyzed by those multivariate indices, because,for example:


• in stratified random sampling, units from differ-ent strata are selected with unequal weights(unequal probability)

• in matched pair designs, the inherent precisioncreated by the pairing is lost if that pair bond isbroken.

Meta-AnalysisMeta-analysis is a relatively new approach asapplied to the analysis of ecological field studies. Itinvolves the combination of statistical results fromseveral independent studies that all deal with thesame issue (Hedges and Olkin 1985, Hedges 1986).While many biologists and statisticians are unfamil-iar with its application, meta-analysis has been wellknown and widely used in some fields (e.g. psychol-ogy, medical research) for quite some time. It may beextremely important for use of historical and base-line data in impact assessment. The simplest form ofmeta-analysis is easy to understand. If several inde-pendent statistical comparisons are made on thesame impact indicator but with relatively low sam-pling intensity, then it is possible that none are sig-nificant at the traditional level of P ≤ 0.05. However,all or most significance levels may be “small” (e.g.,all Ps are ≤ 0.15) and suggestive of the same type ofimpact. The probability that, for example, three ormore independent tests would, by chance, indicatethe same adverse impact if there were no actualimpact from the perturbation, is itself an unlikelyevent. The combined results may establish impactdue to the incident with overall significance levelP ≤ 0.05.

For a second illustration, historic scientific studies ina given assessment area may have addressed thesame basic objective, but were conducted by differ-ent protocols with varying degrees of precision. It isdifficult to combine original data from such studies,but it may be possible to combine results of statisti-cal tests using meta-analysis to establish a reliablemeasure of baseline conditions.

For a third illustration of potential use of meta-analy-sis, consider stratified random sampling, where sam-pling intensity within a given stratum (e.g.,vegetation type) is not sufficient to reject the classi-cal null hypothesis of “no impact.” If the point esti-mates of effect are in the same direction and indicateimpact, then the statistical results might be com-bined across strata (e.g., vegetation type) by meta-analysis to establish the overall conclusion of impactat an acceptable level of precision.

Discussion of all aspects of the emerging field ofmeta-analysis is beyond the scope of this document(see, for example, Burnham 1995, Draper et al.1992, Durlak and Lipsey 1991, Hedges and Olkin1985, Hedges 1986, and Hunter and Schmidt 1990).Meta-analysis should be considered if several his-toric or baseline studies have been conducted. Itmay also be of value if several independent studiespoint in the same direction of impact, but individual-ly lack the usual scientific requirements for statisticalinferences that the impacts are “real.”

Habitat SelectionHabitat selection by birds may be of interest in eval-uating potential risk associated with existing or newwind plants. Manly et al. (1993) provide a unifiedstatistical theory for the analysis of selection studies,and a thorough review of this resource is recom-mended for anyone considering this type of study. Inresource selection studies, the availability of aresource is the quantity accessible to the animal (orpopulation of animals) and the use of a resource isthat quantity utilized during the time period of inter-est (Manly et al. 1993). When use of a resource isdisproportionate to availability then the use is selec-tive (i.e. the animal is showing a preference for theresource).

Scientists often identify resources used by animals(e.g. vegetation type, food, etc.) and document theiravailability (usually expressed as abundance or pres-ence/absence). These studies are usually carried outto identify the long-term requirements for the man-agement or conservation of an animal population.The differential selection of resources provides infor-mation about the ecology of birds that should alsoimprove the assessment of risk posed by potentialwind plants. Resource selection could also be usedin Level 2 model-based analyses of such things asthe difference in mortality associated with turbinedesign. Using most of the designs previously dis-cussed, resource selection models can be used toevaluate mortality and other metrics indicating riskto birds as a function of distance to various turbinetypes.

In most Level 1 studies it will be impossible to iden-tify unique animals. However, using observations ofanimals seen from randomly or systematically cho-sen points it would be possible to use resource vari-ables with known availability (e.g., vegetation) aspredictor variables. For example, if it appears that acertain vegetation type is preferentially selected forhunting by red-tailed hawks within 0.5 km of a nest,then one could predict that the risk of impact would


increase if turbines were constructed on preferredhunting habitat <0.5 km from a nest. Alternatively,the study area could be classified into available unitscharacterized on the basis of a set of predictor vari-ables such as vegetation type, distance to water, dis-tance to a nest, and distance to a turbine. Thepresence or absence of use of a sample of unitscould then be used to assess the effect of the predic-tor variables on bird use. In the case where studyplots are searched for the presence or absence ofdead birds, resource selection could be used to evaluate the effect of a set of predictor variables on mortality.

Cumulative Effects Analysis Cumulative effects are a hot and difficult topic in theevaluation of environmental impacts. Neufeldt andGuralnik (1988) define cumulative as “increasing ineffect, size, quantity, etc. by successive additions.”As is often the case, a relatively simple term takes ona very complicated meaning when applied to naturalresources and their response to perturbations. Tocomplicate matters, the term is defined differently byfederal law such as the National EnvironmentalPolicy Act and its implementing regulations. Suter etal. (1993) classifies cumulative effects into the fol-lowing categories:

• Nibbling - the cumulative effects of a number ofactions which have similar small incrementaleffects. For example, the additions of individualturbines to a wind plant.

• Time-Crowded Perturbations - the cumulativeeffects that occur when actions are so close intime that the system has not recovered from theeffects of one before the next one occurs. Forexample, if impacts from wind turbines areinfluenced by birds’ experience with the struc-tures, one could anticipate some learnedresponse to the turbines over time, possiblyreducing risk. One could hypothesize that rapiddevelopment of a wind plant might have agreater impact on birds than phased develop-ment of the same facility.

• Space-Crowded Perturbations - the cumulativeeffects that occur where actions are so close inspace that the areas within which they caninduce effects overlap. For example, bird riskmay be influenced by turbine and turbine stringspacing.

• Indirect Effects - The cumulative effects thatoccur when the direct effects of actions are not

space- or time-crowded, but their indirect effectsare. For example, the change in land use result-ing from a wind plant may not affect bird use orcause increased mortality but may affect habitatquality, either positively or negatively.

Cumulative effects analysis involves the study of theinteraction of wind plant structures, other land uses,and the ecology of birds. Effects of wind plants onbirds may be additive, increasing mortality beyondwhat might occur without the plant; or effects maybe compensatory, simply replacing other sources ofmortality. Effects of wind plants may be synergistic;that is, a wind plant in combination with anotherland use may result in an increased rate of bird mor-tality greater than the sum of increased mortalitieswhich might occur due to each individual develop-ment. Or, effects may be antagonistic, in which caseassociation with some other variable would reduceimpacts from the wind plant. Finally, impacts of awind plant may increase to a limit or threshold ofeffect. As with testing hypotheses of first order directeffects, the key to a successful analysis is the proto-col by which the data are collected.

For example, if one wishes to evaluate the cumula-tive effects of individual turbines then the protocolshould be designed appropriately. Study designsshould have individual turbines as the basic sam-pling units and impact indicators (say mortality)should be attributable to individual turbines.Covariates should relate to the basic sampling unit(e.g. end-row turbine versus within-row turbine). Itshould be obvious that cumulative effects analysisrequires much more extensive study than simplymeasuring impact indicators associated with singleturbines or even entire wind plants. Cumulativeeffects analysis associated with population effects(Level 2 Studies) would be even more complicatedand expensive.

Statistical Power and the Weight ofEvidenceTraditionally in scientific research, a null hypothesis —that there is no difference in the value of an indica-tor between reference areas and assessment areas orthat there is a zero correlation between two indica-tors along their gradients — is adopted as the “strawman” that must be rejected in order to infer that anindicator has changed or that a cause-and-effectrelationship exists. Although this approach has per-vaded the scientific method and discipline of statis-tics for nearly a century, it usually places the burdenof scientific proof of impact on regulators. The clas-sical use of a null hypothesis protects only against


the probability of a Type I Error (concluding thatimpact exists when it really does not, i.e., a falsepositive). Often the significance level is required tobe below α = 0.05 before the conclusion of impactis considered to be valid. The probability of a Type IIError (concluding no impact when in fact impactdoes exist, i.e., a false negative) is commonlyignored and is often much larger than 0.05. The riskof a Type II error can be decreased by conductinglarger, more expensive studies or, in some situations,through use of better experimental design and/ormore powerful types of analysis. In general, thepower of a statistical test of some hypothesis is theprobability that it rejects the null hypothesis when itis false. An experiment is said to be very powerful ifthe probability of a Type II Error is very small.

The traditional statistical paradigm is geared to pro-tect against a “false positive,” but the interest of theregulator is protection against a “false negative.” Amore fair statistical method is needed to balanceprotection against the two possible errors. The stan-dard paradigm is clumsy at best and is not easilyunderstood by many segments of society. For a dis-cussion of an alternative paradigm see McDonald(1995), McDonald and Erickson (1994), andErickson and McDonald (1995).

Scientists often are concerned with the statisticalpower of an experiment, that is, the probability ofrejecting a null hypothesis when it is false. In thecase of wind plant monitoring, the null hypothesiswill usually be that there is no impact to birds.Accepting a “no impact” result when an experimenthas low statistical power may give regulators and thepublic a false sense of security. The power of the testto detect an effect is a function of the sample size,the chosen α value, estimates of variance, and themagnitude of the effect. The α level of the experi-ment is usually set by convention, if not by regula-tion, and the magnitude of the effect in anobservational study is certainly not controllable.Thus, sample size and estimates of variance usuallydetermine the power of observational studies. Manyof the methods discussed in this chapter are directedtoward reducing variance in observational studies.When observational studies are designed properly,the ultimate determination of statistical power issample size.

The lack of sufficient sample size necessary to havereasonable power to detect differences betweentreatment and reference areas is a common problemin field studies described in this chapter. Estimates ofdirect mortality can be made in a given year through

carcass searches, but tests of other parameters forany given year (e.g., avoidance of wind plant by birdspecies) may have relatively little power to detect aneffect of wind power on the species of concern. Thelack of power is a concern and should be addressedby increasing sample size, through the use of othermethods of efficient study design described above,and by minimizing measurement error (e.g., the useof the proper study methods, properly trained per-sonnel, etc.). However, most field studies will resultin data that must be analyzed with an emphasis ondetection of biological significance when statisticalsignificance is marginal. For a more complete studyof statistical power see Cohen (1973), Dallal (1992),Fairweather (1991), and Peterman (1990). Computer-intensive methods allow estimates of variance andstandard error when complicated designs make stan-dard estimates of variance problematic (Manly1991). Such methods can be useful in calculatingconfidence intervals and in tests of hypotheses usingdata with non-standard distributions. Computer-intensive methods also can be used with pilot datato predict necessary sample sizes to meet objectivesfor precision.

The trend of differences between reference andimpact areas for several important variables maydetect impacts, even when tests of statistical signifi-cance on individual variables have marginal confi-dence. This deductive, model-based approach isillustrated by the following discussion. The evalua-tion of effects from wind energy developmentincludes effects on individual birds (e.g., reductionor increase in use of the area occupied by the tur-bines) and population effects such as mortality (e.g.,death due to collision with a turbine). Several out-comes are possible from the bird studies. For exam-ple, a decline in bird use on a new wind plantwithout a similar decline on the reference area(s)may be interpreted as evidence of an effect of windenergy development on individual birds. The pres-ence of a greater number of carcasses of the samespecies near turbines than in the reference plotsincreases the weight of evidence that an effect canbe attributed to the wind plant. However, a declinein use of both the reference and development area(i.e., an area with wind turbines) in the absence oflarge numbers of carcasses may be interpreted as aresponse unrelated to the wind plant. Data oncovariates (e.g., prey) for the assessment and refer-ence area(s) could be used to further clarify thisinterpretation.

The level at which fatalities are considered signifi-cant is subjective and will depend on the species


involved. Even a small number of carcasses of a rarespecies associated with turbine strings may be con-sidered significant, particularly during the breedingseason. A substantial number of carcasses associatedwith a decline in use relative to the reference area,particularly late in the breeding season during thedispersal of young, may be interpreted as a possiblepopulation effect. The suggestion of a populationeffect may lead to Level 2 studies.

CASE STUDIES Proposed Plant - Buffalo Ridge, Minnesota The following describes a monitoring protocol forevaluating the cumulative effects on passerines andshore birds of proposed wind energy development inthe Buffalo Ridge area of southwestern Minnesota(Strickland et al. 1996). The initial implementation of the protocol monitors the effects of the existing25 MW Phase I development and the proposed100 MW Phase II wind plant. Phase I, constructedby Kenetech Wind power, Inc., consists of 73 Model33 M-VS turbines and related facilities. Phase I islocated in the approximate center of the windresource area (WRA).

Phase II consists of 143 turbines constructed in 1997and 1998 by Enron Wind Corporation. Phase II islocated in the northwestern portion of the WRA.Facilities capable of generating an additional100 MW are planned for the WRA by early 1999.Plans call for additional phases of development andthe eventual production of 425 MW of electricitywithin the WRA. The results of the first year’s studyare presented in Johnson et al. (1997).

The primary goals of monitoring are to evaluate therisk to bird species from each phase of developmentand the cumulative risk to bird species from all windpower development in the WRA. The secondarygoal of monitoring is to provide information whichcan be used to reduce the risk to bird species fromsubsequent developments. While the elements of theplan are somewhat technical, it was developed incooperation with and received review from individu-als and groups with an interest in the wind plant andits potential effect on birds (stakeholders). The planalso was peer-reviewed before implementation.

The sampling design is a combination of the impact-reference area and the BACI design. Bird use andmortality are measured on plots located at varyingdistances from turbines following the samplingProtocol A proposed by Manly et al. (1993). Thesampling plan allows for mortality estimation for the

entire wind plant and reference areas and an estima-tion of use standardized by unit area and unit effort.

The BACI design combines collection of data beforeand after Phase II and III of wind power develop-ment with collection of data from multiple referenceareas. The impact-reference design involves collec-tion of data from the existing development and mul-tiple control areas. Reference areas are as similar aspossible to the wind plant development areas, bothphysically and biologically. Four areas are studied:the existing wind plant (Phase I) (denoted EW), thenorthwest development area (Phase II) (denotedNW), the southeast development area (Phase III)(denoted SE), and a permanent reference area(denoted REF). The southeast site serves as a refer-ence area prior to its development; data collected forthe site also provide pre-construction data for itsfuture development. Future development sites willbe identified as soon as possible and pre-construc-tion data collection will begin at these sites. As withthe SE site, new sites monitored prior to develop-ment will act as references to developed sites. Bysampling both the reference and the impact areasbefore and after wind power development, bothtemporal and spatial controls are utilized, optimiz-ing the impact design (Green 1979). Adding futuresites within the Buffalo Ridge area into the monitor-ing effort will address cumulative impacts of development.

Relative use by bird species is measured throughpoint count surveys conducted during daylighthours. Two types of point count surveys are conduct-ed. The first method involves making point counts ofbirds for a short duration (five minutes) at a largenumber of relatively small plots across the studyareas. Passerines, shore birds, and other smallerbirds (PSB) are the targets of these surveys, sincethey have smaller home ranges and cannot bedetected at large distances. The second methodinvolves making point counts of birds for a longerduration (30-60 minutes) from a systematicallyselected subset of the points with large viewing areasin each study area. Raptors and other larger birds(RLB) are the target of these surveys, since they canbe detected from larger distances and appear to befairly rare within the study area. In the PSB surveys,raptors and other larger birds are recorded whendetected, but observers concentrate on a muchsmaller viewing area to minimize missing smallbirds. In the RLB surveys, observers concentrate ondetecting all raptors and other large birds, but alsorecord unusual or rare observations of small birds. Inboth surveys, distance to the observation is recorded


so data can be standardized by both area and effort(time). Mortality is measured through carcass search-es at turbines and within a systematically locatedsubset of reference plots where use is estimated.

Data AnalysisThe project is in its second year. The followinganalyses are planned at the end of each year.Species lists are generated by study period and studyunit (EW, NW, SE, and REF). The number of birdsseen during each point count survey is standardizedto a unit area and unit time surveyed (observed density). For example, if four horned larks are seenduring a five-minute interval at a station with a stan-dardized viewing area of 0.031 km2, these data arestandardized to 4/0.031=~129 horned larks/km2.Relative density corrected for visibility bias(Buckland et al. 1993) is estimated by species, if data are sufficient, using the program DISTANCE(Laake et al. 1993).

Data are tabulated and plotted to illustrate differ-ences in bird use between: seasons, times of day,stations, turbine sites and non-turbine sites, vegeta-tion type, flight height, and study areas. As data areaccumulated over time, ANOVA techniques will beused to test for differences in bird use and possibleinteractions between: seasons, times of day, studyareas, turbine sites and non-turbine sites, and pre-versus post-development. Non-parametric methodssuch as randomization tests (Manly 1991) will beused when assumptions for the parametric methodscannot be met (e.g., equality of variances). Statisticalanalysis of bird use and flight height as a function ofdistance from turbines, vegetation type, and otherphysical characteristics will be conducted usingresource selection techniques (Manly et al. 1993,McDonald et al. 1995, Pereira and Itami 1991).Other standard statistical procedures will be used asappropriate. For important tests of hypotheses, thepower (i.e., probability of rejecting the hypothesis ofno difference in means) is calculated for variouseffect sizes based on baseline studies and initial datacollected during monitoring, as soon as enough dataexists for estimates of variance.

Resource/Habitat Selection AnalysesStatistical tools used in habitat selection studies areapplied to the bird use data for investigation of habi-tat selection as well as the effects of the turbines onthe avian resource. Data collected prior to develop-ment of the wind plant (Phase II and Phase III) canbe used to determine what important factors appearto be related to presence/absence of a bird species

or the magnitude of use by bird species. For exam-ple, through multiple regression techniques it maybe shown that use by a species of bird is related tothe amount (percentage of area) of land protectedunder the Conservation Reserve Program (CRP) with-in the vicinity of the point count or to distance fromthe nearest wetland. Using presence/absence data atthe point count location, logistic regression (Hosmerand Lemeshow 1989) can be used to estimate therelative probability that an area will be used as afunction of the characteristics of the area. For exam-ple, it may be shown that distance to the nearestwetland is related to the probability of use for aspecies, and that areas at (for example) 300 metersare twice as likely to have bird use by this species as areas at 500 meters. These functions may be use-ful in developing a data layer in a GIS system indi-cating those regions within a development whichhave the highest probability of use by the givenspecies. This information may be useful in siting turbines in future phases.

Using bird use data collected within areas with tur-bines (existing wind plant and northwest site afterconstruction), these same resource selection tech-niques can be used to evaluate effects of the turbineson use by bird species. For example, logistic regres-sion models may show that a bird species has ahigher probability of using an area that is far fromturbines (i.e., possible avoidance of turbines).Multiple regression models may be used to deter-mine if distance to turbines is negatively related tothe magnitude of bird use.

Data collected at the point (bird use,presence/absence, habitat) are used in the logisticand multiple regression analyses. Because repeatedcorrelated measures are made of these variables atthe point, bootstrapping techniques (Manly 1991,Ward et al. 1996) will be used to estimate the preci-sion and confidence in the coefficients of the regres-sion analyses and to avoid pseudoreplication.

Analysis of Flight Height Flight height categories will be recorded for everybird observed corresponding to flights below, within,and above the rotor swept area. More than one cate-gory may be recorded for each bird observed. Forexample, the proportion of individual birds of agiven species observed to be within the rotor sweptarea at least once during the observation period maybe analyzed. Also, the proportion of observations byspecies/groups within each flight height category(i.e., below, within, or above rotor swept area) can


be calculated and plotted. In a similar fashion to thehabitat selection and regression methods describedpreviously, logistic and ordinary regression can beused to predict the probability a bird will use a loca-tion as a function of the characteristics of the loca-tion. In this analysis, observations used in theanalysis are individual birds with potentially corre-lated observations within study plots, so bootstrap-ping techniques will again be used to incorporatepoint to point and survey to survey variability and toavoid pseudoreplication in reporting precision of sta-tistical results.

This type of analysis can be used to predict the prob-ability a bird (by species) will fly within areas of riskassociated with a wind plant, say the rotor sweptarea of turbines, based on the characteristics of thearea surrounding turbines. Use within an area of riskis considered an index of risk of exposure. True riskmust consider other factors such as species ability toavoid collisions while flying close to turbines andother wind plant features.

SummaryThis protocol embodies many of the sampling andanalysis principles described throughout this chap-ter. While the overall design philosophy is a BACIdesign, the control/impact design is required to eval-uate the effects of Phase I because of the lack of pre-construction data. In addition, there will be only twoyears of pre-construction data for Phase II, reducingthe value of the BACI design for evaluating theeffects of this phase. It is obvious that the assessmentof impacts will require long-term monitoring withthe strongest analysis of effect occurring after thedevelopment of Phase III. This delays the strongestdetection of effect until much of the wind plant isconstructed. A stronger study design would havebeen possible if the BACI design had been imple-mented before construction of Phase I. The protocolalso does not include nocturnal use studies.However, the carcass searches should detect thedeath of nocturnal and migrant species.

Although multiple reference areas are used, the useof one reference area that is likely to be developedin the near future is less desirable than having bothreference areas outside the WRA. The protocol illus-trates the necessity of balancing the desire for astrong study design with budgetary constraints.Finally, the species most likely to be affected by thewind plant may be the red-tailed hawk, the only rap-tor species breeding in significant numbers in thearea, or it may be a water bird. If Level 1 studies verify that a single or small group of species is at

greatest risk, Level 2 studies may be required toquantify population effects.

Many of the weaknesses in this study relate to timingand budgetary constraints. As a general rule, the ear-lier research begins in the development of a windplant the more likely the research will provide dataimportant to decisions related to the future of thedevelopment.

Existing Plant - Tehachapi and SanGorgonio Pass, CaliforniaA number of studies have been conducted on exist-ing plants in the United States and Europe. Colson(1995), PNAWPPM (1995), and Orloff and Flannery(1996) have summarized the results of these studies.Most of these studies were observational, involving avariety of protocols, and were conducted at what wehave defined as Level 1 and Level 2. The mostrecent attempt at preparing a detailed protocol forthe conduct of bird studies at an existing wind plantis described by Anderson et al. (1996a) and by mod-ifications to this protocol (Anderson, personal com-munication). The following is a description of theirprotocol and modifications. Like the Minnesota pro-tocol, the Anderson protocol was developed incooperation with and was reviewed by individualsand groups with an interest in wind power and itspotential effect on birds (stakeholders). The protocolwas also peer-reviewed before implementation.

The protocol is designed to determine if an operatingWRA results in an increased risk of bird mortality.The protocol was implemented at wind plants insouthern California in the Tehachapi Pass WRA (PilotStudy). A modification of this protocol is beingimplemented in studies at Tehachapi and the SanGorgonio Pass WRA in southern California (Phase II).

Pilot StudyThe purpose of the pilot study was measuring birduse and mortality in wind plants and in surroundingundeveloped areas and for developing several indi-cators of bird risk. The protocol utilizes standardpoint count methodology to determine the relativeabundance and utilization rates of all bird speciesusing study areas within the WRAs. The protocolalso includes estimating mortality at each pointwhere use is estimated. The protocol proposes theuse of the ratio of mortality to use, as a function ofdistance from turbines, to estimate the amount ofrisk of bird fatality attributable to the development ofwind energy. The protocol also proposes the use ofthis ratio, in combination with the rotor size and


hours of operation, to make comparisons of risk ofbird fatality among different sizes of turbines.

Results from the first year of study on the TehachapiPass WRA are contained in Anderson et al. (1996b).The WRA is heavily developed with numerous windenergy companies operating wind plants with a vari-ety of turbine types. Initial studies have focused onfour companies making up approximately 80 per-cent of the development within the WRA. The WRAis divided (stratified) into three primary study areascorresponding to three distinct geographical loca-tions with operating wind plants.

1. The West Ridge, owned and/or operated byEnron, is at the highest elevation and is dominat-ed by grasslands and Sierra Nevada foothillsvegetation (scattered woodlands, subshrubs, andoak chaparral).

2. The Middle Area, owned and/or operated byCannon and FloWind Corporation, contains amix of grasslands, mountain foothills, and smallpatches of Joshua trees and desert scrub-shrub.

3. The East Slope, operated by SeaWest,Tehachapi, Inc., is dominated by vegetation ofthe Mojave Desert (junipers, Joshua trees, andcreosote bush).

In spite of the differences in vegetation, all threeareas are structurally similar and are dominated bygrassland with patches of shrubs and sub-shrubs.

Each of three primary study areas is defined as adeveloped subregion of the WRA and a buffer is des-ignated as a non-developed comparison area. Thestudy areas are further stratified by natural plantcommunities. Bird use is determined by recordingobservations of birds at points established alongtransects within the study area using the point countmethod. Random starting points are selected withinthe study area with the starting points stratified bynatural communities. From these starting points, ran-dom angles are used to determine each transectdirection. Points for making bird counts are estab-lished every 300 meters along each transect where5- and 10-minute counts are made of any bird seen,regardless of distance from the point. Transects maycross plant community boundaries. Unlike standardpoint counts, birdcalls are not used in detection ofbirds in an effort to avoid the potential bias againstdetecting birdcalls within the wind plant wherebackground noise is stronger.

Point counts are made throughout daylight hoursand throughout each of four seasons. A minimum of250 points are sampled each season and a minimumof 1,000 points are sampled each year. All birds seenfrom a point are recorded and the horizontal dis-tance to the bird and its height above ground areestimated. Bird activities are recorded, including fly-ing, perching, soaring, hunting, foraging, and prox-imity to wind plant structures.

A circular area within a 50-meter radius of eachpoint is thoroughly searched for dead birds or birdparts. Dead birds or bird parts of any age are count-ed and specimens are left in the field, potentially tobe counted again if they fall within a subsequentcarcass search area. Additional data (covariates) arecollected at the site including estimated time sincedeath, cause of death, type of injury, distance tonearest turbine and distance to nearest structure.

Transect length varies depending on a variety of fac-tors (e.g., change in land ownership affecting access,fatigue, etc.). Bird utilization counts and carcasssearches are made during standardized hours of theday. The proposed duration of the study is one totwo years. Anderson et al. (1996a) indicate thatfuture phases of this study and potentially other stud-ies will be expanded to include the San GorgonioPass WRA in southwestern California.

Data AnalysisWith the exception of some details of the measure-ment methodology, the measure of the above met-rics, the analysis of the data, and the testing ofhypotheses is similar to the Buffalo Ridge protocol.What sets this protocol apart from Buffalo Ridge isthe use of the area surrounding active portions of thewind plant as a control area. The design assumesthat the surrounding area is an adequate control area(impact-control design with no pre-constructiondata). It is also an example of the protocol for con-taminant-gradient designs discussed earlier.

Measuring use and mortality at the same points forthe WRA and a control is a strong point of this proto-col. Providing an index to mortality as a function ofuse for a treatment and reference area is essential tothe interpretation of the index. Attributable risk isalso an important issue. In the absence of an esti-mate of attributable risk, all mortalities found withinthe WRA will be attributed to the wind plant.Plotless methods allow recording of all birds seen(i.e., observers record what they see without regardfor distance), not just those within a fixed-size


sample plot. However, since utilization rate may bemost appropriate when considered as use per-unit-area per-unit-time, the data can be trimmed to fitwithin a fixed area if necessary by measuring the distance to each bird observed.

The protocol results in data that will allow the devel-opment of several indices of risk including bird uti-lization rate and bird mortality. The protocol alsocalculates: bird risk, defined as the ratio of bird mor-tality to bird utilization rate; attributable risk, definedas the difference between background mortality andmortality caused by the wind plant; and rotor swepthour, defined as the product of area swept by a tur-bine rotor and the hours of turbine operation. Rotorswept hour is used to estimate a rotor swept hourrisk, defined as one (1) divided by the product ofrotor swept hour and bird risk. This value is pro-posed as a method for comparing risk associatedwith different rotor swept areas of turbine sizes inrelation to the time they operate when data on birdmortality and use exist.

Following this protocol a determination of the effecton bird risk of operation of a wind plants requiresone of two assumptions:

1. that the surrounding area is an adequate control area (impact-control design with no pre-construction data); or

2. that relative risk is a function of distance fromthe wind plant and wind plant features, whichalso assumes that the surrounding area is anadequate control.

The empirical use and mortality data and indicescan be analyzed using the same basic methods asproposed for the Minnesota project for comparingtwo or more wind plants. In the Anderson protocol,the comparisons are between the buffer (reference)area and the wind plant. The random angle intro-duces unequal and unknown probability of selectionof sample points within the wind plants and bufferand may introduce some bias. However, if the proto-col is followed from area to area and year to year, anindex can be developed which will allow thedesired comparisons. As with all indices, the result-ing impact indicator is totally dependent on the pro-tocol. If technology changes or methods change thenthe index will change in some unknown way.Absolute estimates of parameters, such as dead birdsper unit area, cannot be computed because of theunequal and unknown probability of sampling. Thegradient-response allows the use of logistic and

ordinary regression models to evaluate indices ofrisk as a function of the distance to turbines by typeand different portions of the wind plant.

Phase I ProtocolThe Phase I studies are currently being conducted in the Tehachapi and San Gorgonio WRAs. The fol-lowing discussion focuses on Tehachapi, Phase I.The objective of Phase I studies is to use the indica-tors of bird risk developed in the pilot study to evaluate differences in risk among six turbine types.Measurement methods are the same as used in pilotstudy. However the sampling plan is substantiallydifferent.

The Phase I sampling plan follows the stratified ran-dom sampling procedure. The WRA is again strati-fied into the three regions described for the pilotstudy. As described earlier, each region can also beconsidered a subdivision of the WRA by company(see above). Each region is further divided into threeapproximately equal sized subregions, along anapproximately north-south orientation. For the pur-pose of this discussion the subregions will be termedthe north, middle, and south subregions. All turbineswithin each region and subregion are classified intoone of the following types:

• large turbine (rotor diameter > 26m) with a lattice support structure

• small turbine (rotor diameter < 22m) with a lattice support structure

• large turbine with a tubular support structure

• small turbine with a tubular support structure

• small turbine with a lattice support structuredeployed as a “windwall”

• standard vertical axis turbine with two blades.

A total of 160 plots, each centered on a turbine, areselected for study. (Note that each plot includes mul-tiple turbines.) Each plot is given a unique numberand a random sample is drawn from each subregionin proportion to availability. The subregions are usedto insure that the sample of turbines is well distrib-uted (as much as possible given their patchy distri-bution) across the region.

Bird utilization is estimated for each plot four timeseach quarter-year for Phase I studies, and fatalitiesare estimated for each plot once each quarter-year


using the methods described. Plots are visited in ran-dom order for utilization surveys until all turbinesare visited and then the process is repeated for fourvisits. Plots are visited in random order for carcasssearches and are continued until all turbines are vis-ited. Utilization surveys are conducted during morn-ing hours and carcass searches are conducted duringthe remainder of the day. Carcass searches takeapproximately four times as long as utilization sur-veys, explaining the increased number of utilizationsurveys during the three-month survey period.

Data AnalysisPhase I studies had not evolved to the point of dataanalysis at the time this chapter was written andAnderson et al. (1996b) does not include a proposedanalysis. Two possible analyses of the Phase I studyare outlined below. These two possibilities may notbe the most desirable and are certainly not offered asa substitute for the analysis that may be used whenthe project is completed.

In most scientific studies, there exist several compet-ing analyses. Which analysis to use in any one studydepends in part on study objectives, responses ofinterest, and available covariate information. In gen-eral, it is best to analyze available data in a mannerconsistent with the way they were collected. Forexample, if “treatments” are randomly applied towind turbines or wind resource areas, the datashould probably be analyzed using acceptedANOVA techniques (Neter et al. 1985). If the experi-mental design of such a study is appropriate,ANOVA will provide a powerful, defensible, andeasily replicated analysis.

Analysis 1: analysis of variance (incomplete block).The first analysis views the data as coming from a“blocked” experiment where there are six treat-ments, three manufacturer blocks, and three geo-graphic blocks. The six treatments are: 1)large-lattice, 2) small-lattice, 3) large-tubular, 4)small-tubular, 5) windwall, and, 6) vertical axis. Thethree manufacturer blocks are: 1) Enron, 2) Cannonand FloWind, and, 3) SeaWest. The three geographicblocks are: 1) North, 2) Middle, and 3) South.

In order to analyze all data in one large analysis (i.e.,using all the structure outlined above), there need tobe certain combinations of treatments (turbinetypes/sizes) present in each combination of geo-graphic block and manufacturer block. That is, if alltreatments appear in all combinations of geographicblock and manufacturer block (i.e., all six treatmentsappear in the Enron-North block, all six treatments

appear in the Enron-Middle block, all six in Enron-South, etc.) then the proper combinations exist forone large analysis.

The Tehachapi Pass study was not a designed experi-ment and study managers did not have completecontrol over which turbine types and sizes appearedin blocks. For example, it may be unreasonable toexpect SeaWest to remove some tubular turbinesfrom their management area and construct Enron lat-tice types in their place. This lack of design controlis a big issue for the analysis of Phase I data. Withoutthe above conditions met, we must search for small-er pieces of the large analysis to analyze separately.

One potential analysis might be to combine treat-ments 1, 2, 3, and 4 and ignore the SeaWest block toproduce an analysis of the “treatments”: rotor sizeversus windwall versus vertical axis. This examplewould reduce to an incomplete block design withthree treatments, two blocks, and subsampling of therotor size treatment in both blocks.

Alternatively, one could analyze each manufacturerblock separately by viewing each manufacturer’swind resource area as a separate experiment. Forexample, if we focus only on Enron’s block, it maybe that enough combinations of turbine type/sizeoccur within each geographic block (North, Middle,and South) to achieve design requirements andallow an estimate of error. If so, the analysis wouldbe an incomplete block design with three blocks (thegeographic blocks) and three (or possibly two) treat-ments (large-lattice, small-lattice, and windwall). Asimilar approach can be taken in the other manufac-turer blocks.

Analysis 2: regression (analysis of covariance). Thisanalysis is both an alternative and a complement tothe ANOVA technique outlined above. It should benoted that regression (Neter et al. 1985) and ANOVAare not disjoint analysis techniques, but rather arethe same analysis technique applied to differenttypes of covariates (or explanatory variables).

There may be many reasons to consider regressionmethods, two of which are described below, fol-lowed by examples.

1. When the design does not lend itself to a strictANOVA approach, it is possible that a regressionapproach, which considers other (and often con-tinuous) covariates, will yield an analysis whichis essentially the same as ANOVA.


2. If the blocking structure of the study does not doan adequate job of controlling excess variation,it is possible that a regression approach will.

Suppose that even by focusing on theCannon/FloWind block, the regular ANOVAapproach does not yield an analysis because thereare not enough turbines of each type in each of thenorth, middle, and south geographic blocks. Aregression approach would allow one to disregardthe geographic blocks and instead record the lati-tude of each turbine under study. In this approach,the latitude of each turbine becomes a surrogate forthe geographic blocking and would potentially con-trol for the same extraneous factors as the geograph-ic block would have had they produced an analysis.Furthermore, latitude represents a continuous covari-ate. To test for differences between turbine types,researchers must be willing to assume that there isno latitude by turbine type interaction, or at leastassume some simple form for the interaction.

As a second example, suppose that prevailing windspeed affects bird mortality at all types of turbines(suppose higher wind means higher mortality irre-spective of turbine type). Furthermore, assume thatthe existing blocking factors (manufacturer and geo-graphic) do not adequately control for wind speedbecause the blocks are too big to accurately repre-sent wind speed in a local area around each turbine.A regression approach would solve this problem byestimating average wind speed at each turbine andby using average wind speed as a covariate insteadof the blocking factors.

SUMMARY AND CONCLUSIONProtocols for bird studies will, by necessity, be site-and species-specific. However, all protocols shouldfollow good scientific methods. Many of the issuesrelated to avian impacts of wind power are con-tentious, and settling these issues will be assisted bygood scientific studies. However, many of the issuesrelated to wind power impacts on birds are based onrelatively rare events. First, producing electricitycommercially with wind power is a relatively recentdevelopment. With the possible exception of fatali-ties at the Altamont WRA, bird fatalities in existingwind plants are a relatively infrequently documentedevent. Many of the bird species of major concern arealso rare. Second, as pointed out in this chapter, theconstruction of a wind plant is not a random occur-rence and potential wind plant sites are relativelyunique, making selection of reference areas difficult.In spite of these difficulties, bird mortality is a signifi-cant concern and wind power is a potential clean

source of electricity, making study of these issuesessential.

Because impact indicators normally are estimates ofrelatively rare events, analysis of impacts must relyon an accumulation of information. A determinationof impact will seldom be based on clear-cut statisti-cal tests, but usually will be based on the weight ofevidence developed from the study of numerousimpact indicators, over numerous years, at numer-ous wind plants. The selection of the appropriateprotocol must be site- and species-specific. Protocolselection will be influenced by the status of the windpower project (existing or proposed), the area ofinterest, the issues and species of concern, coopera-tion of landowners, and so on. Decisions aboutmethods, designs, and sample sizes will always beinfluenced by budget considerations.

The following is a summary of important considera-tions when designing Level I observational studies:

1. Clearly define the objectives of the study includ-ing the questions to be answered, as well as thearea, the species, and the time period of interest.

2. Clearly define the area of inference, the experi-mental unit (and sample size), and the samplingunit (and subsample size).

3. Clearly define the parameters to measure, selectimpact indicators which are relatively uncorre-lated to each other, measure as many relevantcovariates as possible, and identify obvious bias-es. Impact indicators should allow for the deter-mination of impact following generally acceptedscientific principles and as defined by the stan-dards agreed to by stakeholders.

4. The BACI design is the most reliable design forsustaining confidence in scientific conclusions.Data should be collected for two or more timeperiods before and again two or more time peri-ods after construction of the wind plant on boththe assessment area (wind plant) and multiplereference areas. Consider matching pairs of sam-pling units (data collection sites) within eachstudy area based on matching criteria which arerelatively permanent features (e.g. topography,geology). If the BACI design can not be imple-mented then other appropriate designs shouldbe used.

5. Use a probability sampling plan, stratify on rela-tively permanent features, such as topography,


and only for short-term studies; use a systematicsampling plan for long-term studies, spread sam-pling effort throughout area and time periods ofinterest, and maximize sample size.

6. Develop detailed SOPs prior to the initiation offieldwork and select methods that minimizebias.

7. Make maximum use of existing data and consid-er some preliminary data collection where littleinformation exists.

8. When pre-construction data are unavailablethen combine data collection on multiple refer-ence areas with other study designs such as thegradient-response design.

9. Maximize sample size within budgetary con-straints.

10. Univariate analysis is preferred, especiallywhen determining impacts by a weight of evi-dence approach.

11. Have the plan peer-reviewed.

Each wind energy project will be unique and deci-sions regarding the study design, sampling plan, andparameters to measure will require considerableexpertise. There is no single combination of studycomponents appropriate for all situations. However,at the risk of oversimplification, Table 3-1 contains asimple decision matrix to assist in the design of windenergy/bird interaction studies.

Level I studies should detect major sources of impacton species of interest and assist in the design of windenergy projects to reduce impacts on birds. Whenuncertainty on avian risk exists Level I studies shouldalso identify sites where there is a low probability ofrisk to these species. More often than not, the prod-uct of Level I studies will be to focus future researchon areas where significant biological impacts appearlikely or identify that no further research is needed.



(a) Design Options Study Conditions Recommended Design Potential Design Modification

• Pre-impact Data Possible BACI Matching of study sites Matched Pair • Reference Area Indicated on assessment and Design With BACI

reference areas possible

• Pre-impact Data Not Impact-Reference Matching of study sites Matched Pair Possible on assessment and Design With • Reference Area Indicated reference areas possible Impact-Reference

• Pre-impact Data Possible Before-After• Reference Area Not Indicated

• Small Homogenous Area Impact-Gradient1 of Potential Impact 1Impact-Gradient design can be used in conjunction with BACI, Impact Reference, and Before-After designs.

Table 3-1(a-c). Recommended decision matrix for the design and conduct of Level 1 studies.

(b) Sampling Plan Options

Sampling Plan Recommended Use

Haphazard/Judgment Sampling Preliminary Reconnaissance

Probability Based Sampling:

• Simple Random Sampling Homogenous area with respect to impact indicators and covariates

• Stratified Random Sampling Strata well defined and relatively permanent, and study of short duration

• Systematic Sampling Heterogeneous area with respect to impact indicators and covariates, and study of long duration

(c) Parameters To Measure Parameter Empirical Description

Abundance/Relative Use Use per unit area and/or per unit time as an index2

Mortality Carcasses per unit area and/or per unit time

Reproduction Young per breeding pair of adults

Habitat Use Use as a function of availability

Covariates Vegetation, topography, structure, distance, species, weather, season, etc.

2Can be summarized by activity/behavior for evaluation of risk.

CHAPTER 4:Advanced Experimental Design and

Level 2 StudiesINTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51

Manipulative Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51

Population Effects and Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . .51

MANIPULATIVE EXPERIMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51

CONCEPTUAL FRAMEWORK FOR POPULATION MODELING . . . . . . . .53

Demography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .53

Genetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54

Environmental Stochasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54

Life History Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54

Ecological Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55

Additive vs. Compensatory Mortality . . . . . . . . . . . . . . . . . . . . . . . .55

MODELING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55

Uses of Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55

Types of Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .56

Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57

Effective Population Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .58

Model Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59

SURVIVORSHIP AND POPULATION PROJECTIONS . . . . . . . . . . . . . . . .60

Population Viability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61

DETERMINING CUMULATIVE EFFECTS . . . . . . . . . . . . . . . . . . . . . . . . . .63

RECOMMENDATIONS FOR STUDY DESIGN . . . . . . . . . . . . . . . . . . . . . .63


INTRODUCTIONLevel 1 observational field studies can be importantin making statistical inferences regarding the magni-tude and extent of impacts of proposed and existingdevelopments. Level 2 studies employ controlledmanipulative experiments to determine cause-and-effect relationships or make use of mathematicalpopulation modeling to improve the basis for deduc-tive professional judgments regarding the impact ofwind energy developments on birds. This chapterdiscusses both manipulative experiments and popu-lation models.

Manipulative ExperimentsIn the case of wind energy, manipulative experi-ments (also known as “comparative experiments”[Cox 1958, Kempthorne 1966] and “randomizedexperiments” [National Research Council 1985])usually will be conducted to evaluate risk reductionmanagement options for existing and new windplants. For example, turbine characteristics such assupport structure type, rotor swept area, and turbinecolor have been suggested as factors affecting birdrisk in wind plants. Observational studies, such asAnderson et al. (1996b) can be used to evaluatesome of these risk factors. However, manipulativeexperiments could significantly improve the under-standing of how these factors relate to the risk of birdcollisions with turbines. Manipulative experimentshelp determine treatment effects by allowing controlof such factors as natural environmental variation,which tend to confound observational studies.

Population Effects and ModelingAlthough methods are available for making empiri-cally-based estimations of potential impacts on pop-ulations, population models represent an alternativeand sometimes complementary approach. Protocolsusing population models have the advantage of pro-viding results of impact assessment with relativelylimited empirical data and allow the evaluation ofdata needs in a stepwise fashion, and should becapable of providing at least a preliminary indicationof potential responses of birds to wind energy devel-opments. But as discussed below, population modelsdo have many limitations. Further, it is doubtful ifpopulation modeling will be indicated in most winddevelopments.

The main goal of this chapter is to develop a frame-work for Level 2 studies that can be generalized tomost bird species for evaluation of potential windplant impacts. This is accomplished by:

• developing a conceptual framework based onthe major factors that can influence the persist-ence of a wild population

• briefly reviewing the basic approach to manipu-lative experiments as well as the various modelsthat can aid in estimating population status andtrend, including methods of evaluating modelstructure and performance

• reviewing survivorship and population projections

• developing a framework for determining thecumulative effects of wind energy developmenton birds.

This chapter does not argue against rigorous design-based (field) studies. Rather, it describes how analternative, model-based approach can assist withevaluation of wind energy/bird interaction issues.Before proceeding to a detailed discussion of popu-lation effects and modeling, a brief discussion ofmanipulative experiments is offered.

MANIPULATIVE EXPERIMENTSManipulative experiments may be useful in windenergy/bird interaction studies. They satisfy two cri-teria:

1. Two or more “treatments” (one of which usuallyis a control, or reference treatment) are to becompared for study of cause-and-effect relation-ships on impact indicators.

2. Treatments are randomly assigned to experimen-tal units (Hurlbert 1984).

If treatments are not randomly assigned to experi-mental units, the experimental design becomesobservational (as in Chapter 3), and the informationgained on cause-and-effect relationships is muchreduced (Cox 1958, Kempthorne 1966). Designs forstudying impacts of a wind plant can never be trulymanipulative, because the area/population to beimpacted by the plant and the reference areas/popu-lations are not randomly selected by the researcher.

In manipulative experiments the statistical inferenceis still the protocol by which the study is conducted,the criteria by which study sites are selected, thesource of the treatment materials, and the amount ofreplication in time and space. For example, if twowind plants are selected for the study of some treat-ment and the treatment and references are randomly

Advanced Experimental Design and Level 2 Studies 51

assigned within the two plants, there exist two inde-pendent studies. Statistical inference is limited to theeffect of the selected treatment as applied in thestudy on the wind plant where it is applied for thetime period of application. The results of the twoindependent studies can be used in the subjectiveassessment of the potential effect of the treatment onother wind plants.

Any design used in laboratory experiments ormanipulative field experiments are of use in Level 2studies of wind energy/bird interactions, and a com-plete discussion of these options is beyond the scopeof this document. For details on study design see ref-erences such as Cox (1958), Box et al. (1978), Green(1979), and Hurlbert (1984). All of the design princi-ples and the basic sampling designs contained inChapter 3 are appropriate for manipulative experi-ments. However, it is worth repeating Krebs (1989)that “every manipulative ecological field experimentmust have a contemporaneous control..., randomizewhere possible..., and, because of the need for repli-cation, utilize at least two controls and two experi-mental areas or units.”

The following example illustrates the use of commondesign principles (described in Chapter 3) in theevaluation of a hypothetical risk reduction treatmentincluded in the design of a newly constructed windplant. This is just one example from among analmost infinite number of potential designs. Supposea new wind plant is constructed consisting of 120turbines distributed in 12 turbine strings, each with10 turbines. Also suppose a two-year study is con-ducted to evaluate a treatment applied to some ofthe turbines hypothesized to reduce the risk of birdcollisions with turbines. Finally, assume that risk ismeasured by the relative amount of bird use andbird carcasses located within study plots centeredaround treated and untreated turbines.

The first year of the study estimates the avian behav-ior, use, and mortality at the newly constructed windplant without installation of a treatment. In year two,the treatment is applied and the reduction in risk dueto the treatment is evaluated through the measure-ment of avian behavior, use, and mortality at tur-bines with and without the treatment.

In year one of the study, avian use and mortality aremeasured on plots containing turbines without treat-ment; in year two, use and mortality are measuredon plots containing turbines both with and withoutthe selected treatment. All twelve turbine strings aresurveyed for avian use, behavior, and mortality, so acensus in space within the wind plant is achieved. It

is assumed that if a bird comes into the defined criti-cal zone surrounding the turbines (some distancefrom turbines), then the bird is potentially at risk ofinjury. If the bird does not enter the critical zone, itis assumed that the bird is not at risk of injury. Risk isthus defined as use within a certain distance of a tur-bine. Fatalities are measured and an estimate ismade of mortality per unit of use. Risk may also bedefined as a change in mortality per unit of use.

There are two basic paradigms regarding the analysisof these data. One paradigm is that the samplingdesign is a matched pairs design (randomized blockwith two treatment levels). The second paradigm isthat this is a manipulative study embedded in a largeobservational study using a BACI design. In the firstparadigm, the effectiveness of the treatment is evalu-ated by testing the interaction between year andtreatment. A two-factor repeated measures analysisof variance is conducted using the mortality rate(number of carcasses per search divided by bird useper visit per observation point) as the dependentvariable. Figure 4-1 illustrates the mean mortality perunit of bird use near turbines by year and treatment.There appears to be an interaction between year andtreatment; the mean is relatively stable for the non-treated turbines, whereas the mean for the treatedturbines decreased in year 2. Given that a statisticaltest for interaction corroborates our interpretation ofthe graph, statistical tests of treatment effects shouldbe conducted within each year. Bird fatality neartreated turbines is significantly less than near the


1 2 Year

UntreatedTurbine Plots

TreatedTurbine Plots

Interaction Plot for Avian Use

0.20

0.19

0.18

0.17

0.16

0.15# fa

talit

ies

per/

pass

es w

/i 50

m o

f tur

bine

s

Figure 4-1. Interaction between the number of fatalitiesper bird passes within 50 meters of treated and untreatedturbines.

non-treated turbines in year 2, indicating that thetreatment does appear to reduce the risk to birds.

The second paradigm recognizes that the turbines(and turbine strings) are not random effects becausethe wind plant, turbine strings, and turbines are notrandomly located. According to this paradigm, this isa pseudo-experiment with an unreplicated observa-tional study over time and space. The analysis wouldfollow statistical analyses for BACI designs (Skalskiand Robson 1992) as described in Chapter 3.

In the above example, the design could be modifiedby applying the treatment and reference in year oneto the selected subset of turbines and switching thetreatment and reference turbines the second year.While this design slightly strengthens the study, itwould be practical only if the risk reduction treat-ment were relatively easy and inexpensive to apply.

Manipulative studies can be very complex.However, because of the cost of treating wind tur-bines, most studies will by necessity be limited tosimple designs evaluating a small number of treat-ments. Manipulative studies will be most valuableinitially in evaluating treatments on individual windplants. As data accumulate, subjective inference ona more global scale will be possible. However, caremust be taken to avoid extrapolating the effective-ness of a treatment at one or a few wind plants to allwind plants.

CONCEPTUAL FRAMEWORK FORPOPULATION MODELINGA severe and continuing problem in ecology is deter-mination of the proper population of study. A typicaldefinition of population illustrates the vagueness ofthe concept: a population is a group of organisms ofthe same species living in a particular space at a par-ticular time (Krebs 1985). A population is quantifiedin terms of birth rate, death rate, sex ratio, and agestructure. Unfortunately, the absolute values for eachof these parameters is determined in large part bythe geographic boundaries the user draws. Recently,much interest has been shown in population struc-ture, primarily because of advances in understandingthe role that the spatial structure of a population hason genetics, and ultimately, survival. It is now under-stood that most “populations” are actually metapop-ulations composed of many subparts. This conceptrelates directly to determination of impacts on ani-mals. First, even impacts occurring in a small geo-graphic area can disrupt immigration and emigrationbetween local subpopulations, and thus the impactcan have a much wider effect on the population

than immediately evident. Second, small impactscan have serious consequences to the persistence ofsmall populations if population size is allowed todrop below a certain threshold point (e.g., the effec-tive population size). Thus, the “population” or sub-sets thereof must be clearly defined prior to initiationof any study.

The major factors that can influence the persistenceof a population are:

• demography

• genetics

• environmental stochasticity

• life history parameters

• ecological factors

• additive vs. compensatory mortality.

These factors, reviewed below, could be consideredwhen developing a study plan for evaluating a spe-cific impact of development.

DemographyDemographic “accidents” leading to extinction aremost likely to occur in small populations. Becauseindividuals do not survive for the same length oftime, and individuals vary in the number of offspringthey bear, such effects are sensitive to populationsize. These effects can be ignored if the population islarger than about 100 (depending on age structure).This minimum number will be inadequate, however,if the population is effectively divided into manylocal subpopulations, which is likely in mostregions. In such cases, substantial mortality (by anycause) in one subpopulation can negatively impactadjacent subpopulations by changing rates of immi-gration and emigration. Boyce (1992) summarizedthat variation in population parameters attributableto environmental stochasticity will be more impor-tant than demographic uncertainty with regard to theprobability of actual extinction.

This is not to say, however, that population projec-tions are not an important part of evaluating a popu-lation. Leslie matrix and similar stage-structuredmodels (described below) can give insight into theprocesses of population growth. For example, thesensitivity of the population growth rate, lambda, toperturbations in vital rates for a Leslie-type modelcan be solved analytically. Understanding how


growth rate changes in response to perturbations atvarious stages in life table analysis may help directmanagement strategies. For example, adult survivaltends to be a very sensitive demographic parameterin long-lived species, whereas fecundity can bemore important in short-lived species.

At low densities, a positive relationship between percapita population growth rate and population size(known as an Allee effect) can occur. The conse-quences of such an effect for a population areimportant because it can create a threshold or criti-cal population size below which extinction is proba-ble. For example, limitations to juvenile dispersalcan create an extinction threshold in territorialspecies (Lande 1987).

A model partitioned into spatial subunits can be dif-ficult to analyze, although increasing understandingof metapopulation dynamics is forcing such efforts(e.g., Wootton and Bell 1992). Spatial heterogeneityand dispersal can stabilize population fluctuations,whether the fluctuations are caused by natural orhuman-induced factors. Unfortunately, no generalstatements can be made regarding: the influence ofcorridors; minimum distances between habitatpatches; the ability of dispersers to locate suitableareas; the size and shape of suitable habitat patches;the ability of animals to travel across, or survivewithin, marginal habitats; and so forth. These limita-tions must be considered when interpreting results ofany model.

Quantifying adult survivorship tells one a lot aboutpopulation status (Lande 1988). Adult survivorship isusually very high, especially in long-lived species(such as raptors). Also, lambda has been shown tobe most sensitive to changes in adult survivorship. Inaddition, in most monogamous species, it is femalesurvivorship that is usually the most important topopulation persistence (e.g., Wootton and Bell1992). Thus, in cases where information on popula-tion status is needed, quantifying adult survivorshipprovides a preliminary, basic indication of the statusof the population.

GeneticsModels of genetic variation have a central role in theconservation of populations. We are particularlyinterested in how small population size can result ininbreeding depression and a reduction in geneticvariation, both of which can lead to extinction(Boyce 1992). Local populations or subpopulationsmay contain the genetic diversity that is necessary to

ensure survival of the species within a region, oreven throughout its range.

Boyce (1992) concluded that modeling genetics isnot likely to be as important as modeling demo-graphic and ecological processes in evaluating pop-ulation persistence. He based this conclusion, inpart, on the fact that we do not yet understandgenetics sufficiently to use it as a basis for manage-ment. Thus, practical considerations were the over-riding factor in his conclusion. Genetics will be ofpriority in small, isolated populations, but areunlikely to have direct applicability in studies ofwind energy/bird interactions.

Environmental StochasticityRandom environmental events (e.g., catastrophicfires, hurricanes, disease, and so forth) can have pro-nounced effects on small populations. Such factorscan also have pronounced effects on large popula-tions that are spatially divided into subpopulations.Here, factors such as dispersal will determine thefate of a subpopulation driven to very low numbersby a catastrophic event. It is also important to under-stand the variance structure of the population; thatis, how environmental stochasticity affects theorganism. A major problem here, however, is thatsampling variance may overwhelm attempts todecompose variance into individual and environ-mental components. Thus, the relative importance ofenvironmental stochasticity must be based on anunderstanding of the spatial distribution of the popu-lation under study. This conclusion relates directly tothe discussion above on the importance of determin-ing the spatial structure of the population in anyevaluation of potential impacts.

Life History ParametersThe animal characteristics that we collectively calllife-history parameters include longevity, lifetimereproductive output, young produced per breedingattempt, age at dispersal, survivorship, sex ratio, timebetween breeding attempts, and various other char-acteristics. The absolute expression of each of thesecharacteristics is usually determined by age of theindividual; they change during the lifetime of an ani-mal. For example, young and old individuals tend toproduce fewer viable young than do animals in theirprime. In addition, these factors can interact in vari-ous ways that modify the expression of other factors.

Life-history parameters are used in the developmentof some population-projection models. For example,combining various ranges of parameters can yieldsubstantially different rates of adult survival. Such


analyses provide guidance on the rates of mortalitythat can be sustained under varying expressions oflife history traits. Once such relationships are under-stood, researchers have the opportunity to monitorselected life history traits as part of an assessment ofthe status of a population. For example, if previouswork shows that the timing of breeding is correlatedwith reproductive output, and thus with populationsize for the year, monitoring time of breeding canprovide an early warning of potential population-level problems.

Naturally, this is a simplistic view of a complex setof factors that determine population size and trend.However, it illustrates the utility of developing evenrudimentary models as part of the evaluation of thestatus of a population. Much of the data necessaryfor input into such models is available in the litera-ture (see discussion below).

Ecological FactorsTemple (1986) found that for endangered birds, 82%are listed as endangered as a result of habitat loss,44% due to excessive take, 35% due to introduc-tions of exotics, and 12% due to chemical pollutionor the consequences of natural events. Developmentof a wind plant alters the environmental conditionsavailable to birds; thus, habitat (defined as a species-specific entity) is added, removed, and modified dur-ing and after development, depending on the birdspecies being considered. Quantifying changes inhabitat for species of concern thus may give insightinto potential changes (positive or negative) causedby wind energy development.

Temple (1986) listed “excessive take” as the secondmost important factor causing declines in small pop-ulations. Generalizing “take” to include any artifi-cially-induced mortality implies that mortalitycaused by wind turbines could be quantified to giveinsight into the magnitude of the effect. It might beeasier to quantify and model habitat parameters, andtheir influence on some index of population abun-dance and life-history traits, than it is to adequatelymodel demographics. As noted above, delineatingpopulation or subpopulation boundaries is extremely difficult.

Additive vs. Compensatory MortalityMuch interest has been expressed by individuals inthe wind industry regarding the specific type of mor-tality functioning on wind plants. That is, do windplant-related deaths add to the total number ofdeaths experienced by a population; or, are suchdeaths simply part of the total number of birds that

would have died from some other cause? The formersituation, termed additive mortality, is of concernbecause it means wind plants are potentially causingpopulation-related impacts. The latter situation,termed compensatory mortality, would render windplant deaths of relatively less concern because thetotal number of deaths has not been increased by thewind energy development.

The issue of additive versus compensatory mortalityhas been a central topic of debate in the ecologicalliterature for decades. In large part, this is becausehunting advocates have used the possibility of com-pensatory mortality to argue that hunting onlyremoves animals that would have died anyway.Unfortunately, little quality data exists on this sub-ject. In the case of wind plants, no studies have beenconducted that help to elucidate the additive versuscompensatory issue. It is not appropriate to assumethat compensatory mortality takes place, especiallywith regard to the specific age and sex segment ofthe population (i.e., even if overall deaths might bethe same, the segment of the population experienc-ing the deaths would likely be different). Thus, theissue of compensatory versus additive mortalitycould be further investigated, and it is probably notappropriate to assume a specific form of mortality isoccurring at this time in wind plants.

MODELINGUses of ModelingIn many model-based analyses of populations, acentral part of impact assessment is development ofa model predicting the survival rates required tomaintain a population. The strategy is to determinesurvival rates required to sustain populations exhibit-ing various combinations of the other parametersgoverning population size. To be useful in a widerange of environmental situations and useable forpeople with varying expertise, the model must bebased on simple mathematics.

The use of models (of all types) soared beginning in the 1980s. In fact, modeling is now a focus ofmuch interest, research, and management action in wildlife and conservation biology. But as in allaspects of science, models have certain assumptionsand limitations that must be understood beforeresults of the models can be properly used.Modeling per se is neither “good” nor “bad”; it is the use of model outputs that determines the valueof the modeling approach.

The use of population models to make managementdecisions is becoming common. For example, such


models are playing a large role in management plansfor such threatened and endangered species as thespotted owl (Strix occidentalis, all subspecies), deserttortoise (Gopherus agassizii ), Kirtland’s warbler(Dendroica kirklandii ), various kangaroo rats(Dipodomys spp.), and so forth.

Two general uses of models should be distinguished:

1. using models to give insight into how an ecolog-ical system behaves

2. predicting the outcome of a specific situation.

In the first case, the model helps guide decisionswhen used in combination with other reliable data,whereas in the second case model assumptions andresults must be tested in a quantitative manner (i.e.,model validation).

Types of ModelsLife TablesLife tables are one of the oldest means of examiningmortality in animals; simply, they summarize sur-vivorship by age classes in a cohort of animals. Abasic life table requires only that age, the number ofindividuals surviving to the beginning of each ageclassification, and the number of deaths in each ageclass be known; mortality and survival rates can becalculated from these data. There is only one inde-pendent column in a life table; all the others can becalculated from entries in any one column. Thisdependency requires that great care be taken in con-structing the table, and that large sample sizes begathered. Grier (1980) and Buehler et al. (1991) useda deterministic life-table model to calculate survivor-ship and population growth in bald eagles.

Simple Lotka ModelsThe annual geometric growth rate of a population is represented by lambda, also known as the finiterate of population increase. At time t the populationsize is lambda times its value at time t - 1, Nt =lambda(Nt-1). The population is increasing if lambda > 1, is constant if lambda = 1, and isdecreasing if lambda < 1. For example, if lambda =1.04, then the population was growing at the rate of4% per period during the time sampled. For purpos-es of calculation, this formula is usually presented asNt = N0ert, where e is the base of the natural loga-rithm, and r is the instantaneous rate of populationincrease (Johnson 1994).

Eberhardt (1990) developed a modeling schemebased on approximations of the Lotka equations

using the grizzly bear (Ursus arctos) as an example.Parameters used to develop the model included littersize, proportion of female cubs, breeding interval inyears, and reproductive rate. The utility of thisapproach was that estimates of these parameterswere available in the literature. Eberhardt used varia-tions among these estimates (e.g., litter size rangedfrom 1.65 to 2.36) to calculate ranges of female sur-vival rates to provide information about the scope ofsuch rates needed to sustain populations. Each userof the Eberhardt scheme could select the particularcombination of demographic parameters thought tobe most appropriate for a particular situation.

Leslie Matrix ModelsMatrix models subsume classical life table analysisas a special case but have capabilities that go farbeyond that analysis. As summarized by McDonaldand Caswell (1993), they:

• are not limited to classifying individuals by age

• lead easily to sensitivity analysis

• can be constructed using the life cycle graph, anintuitively appealing graphical description of thelife cycle

• can be extended to include stochastic variationand density-dependent nonlinearities.

McDonald and Caswell present a detailed descrip-tion of the formulation and application of matrixmodels to avian demographic studies.

The numbers in the body of the matrix are transitionprobabilities for survival and progression into otherstages, while the numbers on the top row of thematrix represent stage-specific fecundity values. The term in any particular row and column can bethought of as the contribution of an individual in theage class represented by that column in year t to theage class represented by that row in year t + 1. Thepopulation can be projected from one year to thenext by repeating the process into the future. Thus,we term this matrix the population projection matrix,or more popularly, the Leslie matrix after its develop-er (Leslie 1945).

A Leslie matrix can be built from estimates of fecun-dity and survival probabilities, and populationgrowth may be projected for any number of timeperiods by pre-multiplying the age distribution ateach time period by the Leslie matrix to get the newage distribution for the next time period. Population


projections using Leslie matrices is a usefulapproach to the analysis of demography (Jenkins1988). They provide a numerical tool for determin-ing growth rate and age structure of populations. TheLeslie matrix also is useful for illustrating and study-ing the transient properties of populations as theyconverge to the stable state.

Stage-based matrices, analogous to the age-basedLeslie, can be used to analyze population growth forspecies in which it is difficult to age individuals, orwhere it is more appropriate to classify them into lifestages or size classes rather than by age; these mod-els are generally referred to as Lefkovitch (1965)stage-based models. It is extremely difficult to deter-mine the specific age of most birds and mammalsafter they reach adulthood. In the case of raptors, thefocus of concern in many wind energy develop-ments, young (juveniles) and subadults can usuallybe aged up until adulthood (through differences inplumage and soft tissues, and sometimes eye color).Further, adult raptors can often be placed into cate-gories based on breeding status.

Lefkovitch models assume a single, well-mixed pop-ulation with no spatial structure and no densitydependence in the variables. Thus, they assumehomogeneous probabilities of survivorship andbreeding success within each stage, independent ofother factors such as age. The models can be modi-fied to incorporate spatial population structure andanalyze it in the context of different managementoptions for a population (e.g., see Wootton and Bell1992 for development and review; and Ruckelshauset al. 1997 for problems with using developing mod-els). However, such spatially-explicit models arebeyond the scope of this document, and are moredetailed than necessary for most wind energy applications.

Case StudyThe impact of the Altamont Wind Resource Area onthe golden eagle population resident there is beingevaluated with the aid of a Lefkovitch stage-basedmodel (see Shenk et al. 1996 for complete modeldevelopment). The model is being developed by ateam of scientists assembled by NREL; field data arebeing collected by a team of biologists headed byDr. Grainger Hunt. The overall goal of the project is to determine the finite rate of population growth(lambda) based on birth and death rates for thedefined golden eagle population around the WRA. If the estimate of lambda is >1, then the populationwill be assumed to be stable or increasing. If lambdais <1, then no definite conclusion regarding the

impact of the WRA on the population can be made.This is because there could be many reasons for adeclining growth rate. The model will, however, provide a quantitative approximation of the currentstatus of the eagle population. Further, parameterestimates of survival and fecundity will assist in evaluating the status of the population through com-parisons with the same information for other popula-tions. For example, if the survival of some segmentof the population is relatively low, and that segmenthas been shown to sustain turbine-related deaths,then further study would be indicated. The value ofthe model-based approach is that it provides a spe-cific structure for the field studies to follow, includ-ing understanding of the sample sizes necessary toreach desired estimates of the parameters.

The group developing the Altamont model wasfaced with time (three years for field study) and mon-etary constraints. Because of these constraints, timeeffects would have to be ignored (i.e., interyear variability could not be determined), and samplingwould have to focus only on those elements essen-tial to model development. Regardless of the lengthof study chosen (i.e., relatively longer- or shorter-term), modeling helps to focus the field sampling,thus making for an efficient and justifiable expendi-ture of funds.

The modeling group next evaluated the sample sizesnecessary to estimate survival with a minimum pre-cision of 10%. The eagle population could be bro-ken into four general categories: adult breeders(territory holders; >4 years old), adult floaters (non-territory holders; >4 years old, subadults (1-4 yearsold), and juveniles (<1 year old). Eight total cate-gories resulted when sex was considered.Preliminary analyses indicated that at least 25 indi-viduals would be necessary for each of the eightclasses. Based on time and funding constraints, itwas determined to be infeasible to gather this manysamples. Further discussion indicated that all adultfloaters and subadults could be combined and con-sidered “nonterritory holders.” Most demographicmodels only consider the demographics of femalesbecause of their central role in the production ofyoung. (Wootton and Bell 1992). However, becausemale eagles can have a key role in determining nest-ing success (G. Hunt, Univ. California, Santa Cruz,pers. comm., various dates 1996-97), male eagleswere also marked. Immigration and emigration wereassumed to have no influence on lambda because of the difficulties in estimating these parameters. Ofcourse, the potential influence that such assumptionsmay have on model results must be consideredwhen drawing conclusions from the study. The


assumptions and constraints necessary in theAltamont eagle study are typical of real-world modeling situations.

Effective Population SizeAs discussed above, small populations are suscepti-ble to extinction because of random loss of geneticvariation and random demographic events. In the“ideal” theoretical population, the rate of loss ofgenetic variation is inversely proportional to thepopulation size. Of course, the reproductive behav-ior of natural populations is far from ideal. To try tolink natural and idealized populations, Wright(1931) defined the “effective population size” (Ne) asthe size of an ideal population whose genetic com-position is influenced by random processes in thesame way as the natural population.

When Ne is small, the population can rapidly losegenetic variation. However, Ne has no set relation-ship to actual population size, and its precise estima-tion is complex. Two approaches have been used toestimate Ne: genetic and ecological. The geneticmethods directly quantify the effects on genetics of aparticular effective population size, whereas the eco-logical methods are indirect and depend upon themeasurement of ecological parameters that arethought to influence a particular effective populationsize.

Although direct measurement of the effective popu-lation size by a genetic method is the most appropri-ate, there are several problems associated with itsdetermination. First, the method requires the gather-ing of a large amount of genetic information.Although new technologies are reducing this prob-lem, it is still beyond the capabilities of mostresearchers. Second, the confounding factors ofimmigration and population subdivision, and thepossibility that even relatively low levels of sometypes of evolutionary selection are having a largeinfluence on the estimated Ne.

Ecological methods depend on theory linking partic-ular ecological parameters, usually based on demog-raphy or behavior, to changes in Ne. Wright (1938)established the relationship linking the effective pop-ulation size to the population sex ratio and to thevariance in reproductive success among individuals.For example, variation in family size inflates thevariance in reproductive success and thus reducesthe effective population size.

Various formulas have been developed to estimatethe effective population size. Harris and Allendorf

(1989) evaluated several of these methods. Hill’s(1972) original equation and its derivatives wereconsistently the most accurate. Nunney and Elam(1994) developed a related approach that required aminimum amount of information while still giving agood estimate of Ne. Termed the “minimal” method,it requires the estimation of six parameters: (1) meanmaturation time to adulthood for both males andfemales; (2) mean adult life span for each sex, andoverall; (3) estimation of generation time; estimationof variation in (4) male and (5) female reproductivesuccess per breeding season; and (6) estimation ofthe adult sex ratio. This method is designed to pro-vide an estimate of the effective population size inlong-lived species using the minimum of data possi-ble derived from the literature and short-term study.The method is most effective if survivorship is age-independent, which is common in many natural,long-lived populations (not including juveniles).

Nunney and Elam (1994) argued that the minimalmethod (and ecological methods in general) providedata that can be used to predict changes in effectivepopulation size as the conditions confronting thepopulation change. Thus, it functions well in moni-toring populations over time. Genetic methodsdetermine what the effective population size hasbeen over the last or several generations, but theyprovide no insight into why this has been the pre-vailing value. Therefore, they recommend ecologicalmethods when it is practical, so that the effect of dif-ferent management options on the effective popula-tion size can be estimated. They note, however, thatthe demographic information needed to provide areliable estimate of Ne can often be difficult toobtain. As noted above, it is unlikely that this levelof data collection will be indicated in most windenergy applications.

There has been continuing debate over the mini-mum size a population must maintain to ensurelong-term persistence (perhaps 100 generations).During the 1980s and into the 1990s, geneticistsestimated that the minimum effective populationsize was 500 or fewer breeding individuals. Newgenetic evidence suggests, however, that this formerestimate is far too low, and could easily rangebetween 1000 and 10,000 individuals. This newestimate is based on consideration of the effect thatmutations have on the fitness of the organism at lowpopulation sizes (Lande 1995, Lynch et al. 1995).

It is difficult to make broad generalizations on theeffective population size of organisms. For example,small populations (<100 adults) have been shown to persist for extended periods of time because of


adaptations to local environmental conditions (e.g.,Reed et al. 1986, Grant and Grant 1992, Nunney1992). Evaluation of effective population size maybe appropriate in preliminary analyses of a popula-tion. Such evaluations can help prioritize species tostudy and help determine the level of concern thatshould be placed on deaths in a population beforeinitiating a full-scale population study.

Model EvaluationBart (1995) provided an excellent review of the stepsnecessary in evaluating the appropriate uses of apopulation model. The following outline is summa-rized from his paper. There are three major compo-nents of model evaluation that should be included inall studies: model objectives, model description, andanalysis of model reliability. The latter component isfurther divided into four important criteria.

Model ObjectivesAs noted above, all studies should list the specificobjectives for which model outputs will be used,and the reliability needed for those outputs. Will theoutput be used only as part of a much larger set ofinformation — or will management decisions bebased on model results? The precision needed in allcases should be specified; there are no pre-estab-lished standards.

Model DescriptionThe general structure and organization of the modelshould be detailed. This description should includethe basis for classifying the environment (e.g., vege-tation types used for analysis), the number of sex andage classes, the behavior of the animals (e.g., breed-ing times, dispersal), and so on. For example, ifsexes or age classes are lumped (because of sample-size considerations), then the behavior of the sexesand age classes is assumed to be equal. Likewise, ifdata on any aspect of the model are lumped acrossyears, then time is held constant and assumed tohave no overriding impact on the model. Most deci-sions reduce the complexity of the model, which inturn reduces its reality. Careful consideration andjustification of any such decisions must be includedin model description.

Analysis of Model ReliabilityThere are four major types of model reliability toevaluate: structure, parameter values, secondary pre-dictions, and primary predictions. Each type shouldreceive attention, with emphasis on the particulartype that the management will focus on.

Model structure. The realism of each assumptionabout the model should be fully assessed using anyinformation available. Naturally, the first source ofinformation here is the scientific literature about ani-mal behavior, habitat relationships, population struc-ture, and demographics. If little information isavailable on the species of interest, then data onrelated species should be consulted. The impact thateach assumption should have on model resultsshould be clearly discussed. Some assumptions like-ly will have minimal impact, while others may havepotentially severe influence on the model. In somecases the decision will have to be made that insuffi-cient information is available on this or closely relat-ed species for any meaningful evaluation of themodel to be made. In such cases, the model — ifdeveloped — is of the purely descriptive form andshould only function in identifying likely areas uponwhich field research (to fill the data gaps) shouldfocus. However, information is usually availablewith which at least a preliminary model structurecan be based.

Parameter values. The most reasonable estimate ofmean values and ranges for each parameter shouldbe developed. Again, the literature should first beconsulted. However, field studies may have to beconducted to provide reasonable estimates of certainparameter values. Unfortunately, the wildlife litera-ture provides little in the way of strong data on sur-vivorship of animals, especially where data onspecific sex-age classes are needed. The reality ofthe situation usually demands that a short-term (1-3year) study be initiated to provide the missing data.Because these studies usually focus on either rarespecies or isolated populations, it may be necessaryto ignore yearly variations and lump across time toachieve an adequate sample size. As discussedabove, the ramifications of this type of simplificationmust be carefully evaluated. It also is almost alwaysthe case that certain age classes (e.g., nonbreedingadults in raptors) will have to be combined; addi-tionally, in most animals age cannot be readilydetermined after adulthood is reached.

Secondary predictions of the model. Secondarypredictions are intermediate outputs of the modelthat can be used to better understand the populationand help evaluate the reliability of the final model.Each of these outputs is a function of two or moreinput variables. Comparing them to empirical data,to data for similar species, or just plain ecologicalcommon sense helps identify how reliable the modelwill be (and where weaknesses exist). Examples ofsecondary outputs include the distribution of age


classes at first breeding, territory occupation, and soon.

Primary predictions of the model. Primary predic-tions are the outputs of primary interest; this is theinformation used to determine project impacts andmake management decisions. Predicted modelresults should be compared to reality either by com-paring them with empirical data, or by running sim-ulations that can be compared with known (past)population values. That is, if the model fits past(known) trends, then it is more likely to be properlyforecasting future values. Unfortunately, little dataare usually available because few animals have beenadequately studied. Evaluations of models, however,are not truly independent if available empirical dataare used to develop the model in the first place; test-ing the model predictions with the same data resultsin a biased validation.

SynthesisThe goal should be to present a realistic and unbi-ased evaluation of the model. It is preferable to pres-ent both a best and worst case scenario for modeloutputs, so that the range of values attainable by themodel can be evaluated. For example, with a basicLeslie Matrix Model of population growth, knowingwhether the confidence interval for the predicted(mean) value for lambda (rate of population growth)includes a negative value provides insight into thereliability of the predicted direction of populationgrowth.

The process of model development and evaluationmay show that the predictions of the model are sufficiently robust to existing uncertainties about theanimal’s behavior and demography that high confi-dence can be placed in the model’s predictions. Apoor model does not mean that modeling is inappro-priate for the situation under study. Rather, even a poor model (i.e., a model that does not meet studyobjectives) will provide insight into how a popula-tion reacts to certain environmental situations, andthus provide guidelines as to how empirical datashould be collected so that the model can beimproved. Modeling is usually a stepwise process.Confidence intervals can be calculated to quantifythe amount of variability associated with model outputs (Bender et al. 1996).

SURVIVORSHIP AND POPULATIONPROJECTIONSMajor wildlife and ornithological journals (Journal of Wildlife Management, Condor, Auk, Journal ofRaptor Research) published during the past 20 years

were reviewed for this chapter to determine if anycommonality existed among species with regard toannual survivorship. Most data in the articles exam-ined were based on either short-term (usually 1-3years) telemetry studies, or long-term analyses ofband returns. Most of the band return data wereobtained from waterfowl harvested by hunters.

In summary, only very broad generalizations can bedrawn regarding “normal” survival rates of avianpopulations. Further, interyear variability in survivor-ship is large even in healthy populations, whichmakes short-term (1-2 years) evaluations of a popu-lation of concern suspect. Bellrose (1980) summa-rized survival rates for waterfowl, concluding thatimmature ducks show 60% to 70% first year mortali-ty, but that subsequent (adult) yearly loss is only35% to 40% (or survival of about 60% to 65%).More recent studies confirm these general values ofBellrose. For example, Smith and Reynolds (1992)found that survivorship in mallards ranged fromabout 60% to 70%, and that the population shouldnot decline in abundance. Unfortunately, most stud-ies that present survivorship data provide no infor-mation on population trends or projected populationpersistence; most showed survivorship values similarto those summarized by Bellrose (1980; e.g., seeConroy et al. 1989, Chu et al. 1995, Reynolds et al.1995). Haramis et al. (1993) and Hohman et al.(1993) found what they called “high” survivorshiprates of over 90% in canvasbacks. Foster et al.(1992) examined survival of northern spotted owls in four study areas across 2-4 years by radioing 213owls. Annual survivorship ranged between 67% and100% with most between 80% to 94%; no informa-tion on population persistence was provided.

In glaucous-winged gulls, Reid (1988) found 85%annual survival in adults, 80% in second year, and61% in first year birds. Using these survival valuesand other population parameters to construct aLeslie matrix model, he calculated a lambda of 1.05.This rate of population growth compared favorablyto the observed rate of growth.

For bald eagles in Maryland, Bowman et al. (1995)provided survivorship data for 1-6 year old ageclasses (their table 1), and using a deterministic life-table model, predicted a finite population growthrate of 5.8% per year. They found, however, that asimulated 12% decrease in minimum adult survival(from 83% to 73%) eliminated population growth.Their review of the literature showed that their estimated survival rate exceeded those previouslypublished.


Likewise, Bowman et al. (1995) found that after firstyear, survivorship for bald eagles in Alaska wasabout 90%; survivorship within the first year was71%. They too used a deterministic life table to cal-culate a lambda of 1.02 (2% annual populationgrowth). They illustrated through sensitivity analysisthat their model was robust to changes in reproduc-tive rates and annual survival rates for first-yeareagles, but sensitive to changes in survival rates forafter first-year eagles (see their fig. 2). Sensitivityanalysis suggested that lambda turned <1 when afterfirst-year survival dropped to only 88%; lambdadropped to about 93% when survival was loweredto 82% (other parameters also were modified).

Conway et al. (1995) conducted an experimentalevaluation of the effects of removal of nestlingprairie falcons on the breeding population in anattempt to simulate the impacts of falconry. Theyremoved 138 of 451 nestlings (31% of natality) from20 territories during 1982-89, along with a referencearea. They found no overall difference in nestingsuccess and productivity between treatments andreferences, although treatments were lower than ref-erences in two years of study. Their results suggestedthat intensive harvest of nestling prairie falcons mayadversely affect some local population parameters,but harvests were sustainable and probably did notaffect local population size. Because only about0.2% of all prairie falcon natality is harvested annu-ally in the United States, such a loss has no impacton population numbers or persistence (relative to themuch higher level of harvest they simulated). Thisstudy is an excellent example of experimental evalu-ation of the impacts of loss of young, and indicatesthe resilience of raptor populations to loss of young.

These studies are important because they indicatethat even a relatively minor change in survivorshipcan have substantial population impacts. They alsoindicate the importance of determining survivorship,as guided by a modeling structure, in evaluations ofeffects of wind energy developments on birds. Theliterature clearly indicates that, in most cases, adultsurvivorship is critical to maintaining a viable population.

Population Viability AnalysisA population viability analysis (PVA) is a complexprocess that considers all factors that affect theprocesses of a species’ population dynamics. Suchfactors can include demographic, genetic, and environmental stochasticity, plus life history andhabitat-use parameters; dispersal, competition, andpredation also need to be considered. By formally

trying to understand these processes and how theymight influence the species, our general knowledgeof how an environmental impact might affect thespecies is formed. The determination of biologicaleffect made by the Forest Service in their biologicalevaluation procedure is usually based on some typeof PVA, and PVAs are now being included by theFish and Wildlife Service in some recovery plans forendangered species.

Thus, a PVA concerns birth, death, immigration, andemigration rates and how environmental factors(including a treatment) affect these rates over time.Models are used within a PVA process, ranging fromsimple verbal to complex mathematical versions (asreviewed above). A PVA should thus be considered a “process” rather than a specific model in and ofitself. It entails evaluation of available data and mod-els for a population to anticipate the likelihood ofpopulation persistence over some period of time.The “minimum viable population” (MVP) modelingscheme, where an estimate of the minimum numberthat constitutes a viable population is estimated, isclosely related. PVA embraces MVP, but withoutseeking to arrive at an absolute population mini-mum. An MVP can be considered in overall determination of the PVA.

There are few published or peer reviewed PVAs, andmost that are available provide only a vague outlineof model structure or use general “rules of thumb”that are burdened with severe assumptions. Thereare no specific guides for completing a valid PVA.This is understandable, however, because each situ-ation is unique due to differing environmental condi-tions and differences in the proposed impact(s). Theadvantage of the PVA over an MVP is found in thefact that sufficient data to derive reliable estimatesfor all parameters to develop an MVP is not practicalin most cases (for the reasons developed above).

Hindering the fuller use of PVAs in the decision-making process is a general misunderstanding oftheir strengths and limitations. Lindenmayer et al.(1993) and Boyce (1992) carefully reviewed thistopic, with the former outlining the followingstrengths of a PVA.

1. It produces an explicit statement of the ecology of the species and identifies missing data.

2. It synthesizes interacting factors and identifies trends in population behavior.

3. It identifies processes threatening to the species.


4. It can be used in defining minimum critical areas and designing reserves.

5. It can enhance on-ground management and decision making.

6. It has applications in species recovery, reintroduction, and captive breeding programs.

It is clear that PVAs have applicability in a widerange of management scenarios, which increases theneed for managers from all disciplines to understandtheir functioning.

Of particular interest in application of PVAs to windenergy development are Lindenmayer (et al.)’s points1, 2, 3, and 5. By describing the general ecology ofthe species (point 1), users from all educationalbackgrounds are able to better understand the prob-lems confronting the scientist in making predictionsregarding likely environmental impacts of develop-ment. For example, the needs of a raptor for certaintypes and sizes of prey when feeding young, or theinfluence of a skewed sex ratio on territory occupan-cy during breeding, are complicated issues that mustbe described. These factors might interact (point 2)because only a certain sex is able to efficientlyexploit the prey available in the project area; devel-opment might change this prey availability becauseof disturbance to the ground and changes in plant-species composition (point 3). Development ofpoints 1-3 naturally leads to fulfillment of their point5, namely enhancement of sound decision-makingregarding both permitting of the project, and in thecase where permitting is allowed, modification ofthe project to avoid potential impacts on the speciesof concern (in the above example, avoiding unwant-ed changes in prey availability through habitat man-agement). The development of a complete PVA willseldom be necessary in permitting or evaluation ofwind-energy developments. Great care must be usedin developing a PVA because of the difficultyinvolved in gathering data and producing reliableresults.

From this review (above) of factors known to influ-ence population persistence, it appears that weshould:

• worry about adult survival in long-lived species(including most raptors), but fecundity in short-lived species

• not try to quantify genetics, but rather examineestimates of effective population size

• worry about the spatial structure of populationsof concern (which by implication includesimmigration and emigration)

• carefully evaluate how life-history parameterscould interact to influence population persist-ence (e.g., sex ratio and productivity).

As noted by Lindenmayer et al., however, PVAs —as with many models — are only as strong as thedata available for use in their development. Andbecause all models are simplifications of ecologicalinteractions, PVAs by their sheer complexity tend toconflate errors. Further, because of substantial differ-ences in life-history parameters among species, nogeneric PVA model is available. This greatly compli-cates the use of a PVA, because one must be familiarboth with the ecology of the species as well as witha complex set of mathematical formulations(Lindenmayer et al. list many of the models avail-able). As such, most models for PVA analysis mustbe modified to meet the particular requirements of a given project.

The use of PVAs is thus hindered not by somethinginherently wrong with the concept per se, but ratherby the inherent complexity of biological systems.PVAs are simply an attempt to formalize the com-plexity of nature. As such, PVAs may enhance deci-sion-making by formally identifying the processunder study, thus providing a list of the data avail-able and the data still needed to make a rationaldecision regarding project impacts. In fact,researchers and managers alike are increasing theiruse of PVAs (Mace and Lande 1991).

Boyce (1992) and Lindenmayer et al. (1993)reviewed many of the PVAs that have been con-structed. The summary presented by Lindenmayer etal. (1993: Table 1) is especially useful in highlightingthe fact that each of the PVAs they reviewed wereable to identify the primary cause of risk to the pop-ulation. As they noted, habitat loss was the primaryrisk factor in most of the situations evaluated. Incases where habitat loss was not the primary factor,a species- and site-specific factor was identified as ofprimary concern. For example, hurricanes — whichfall into the general category of environmental sto-chasticity — were the risk factor for several small,geographically isolated populations (Puerto Ricanparrot [Amazona vittata], key deer [Odocoileus vir-ginianus clavium]). Other site-specific impacts werefound to be the primary risk agent in other isolatedpopulations, including ski resort development, hunt-ing, and logging.


These results from the PVA are mostly intuitive: onewould expect that a small, isolated populationwould be negatively impacted by factors heavilyimpacting the location where members of the popu-lation remain. The ability of a population to adjust tochanges in its habitat can be predicted through care-ful study of the behavior of individuals in the popu-lation; that is, through determination of theclassification of individuals as either “specialists” or“generalists”. By definition, habitat is a species- orpopulation-specific phenomenon (Morrison et al.1998). As such, changes to its habitat must havesome impact on individuals in the area, given that itshabitat has been properly characterized by theresearcher.

The question arises, then, if results of PVAs are basi-cally intuitive, what value does actual creation of aPVA offer? The answer is found in the fact that byconstructing a PVA, the researcher is able to show ina systematic and analytical fashion that his/her intu-ition was indeed logical. Perhaps a proper test of aPVA would involve evaluation of a hypothesis basedon researcher knowledge and intuition. Further, aPVA allows knowledge to be gained on the interac-tions of various life-history parameters, and theirimpact on population numbers.

DETERMINING CUMULATIVE EFFECTSThere are two major aspects to cumulative effectanalysis that are directly related to wind energydevelopment. The first concerns cumulative effectson a population over time. That is, are effects (posi-tive or negative) caused by the wind plant relativelysubtle over a short period of time, so that only alonger-term study will reveal the trend of impact?This impact could apply to the birds in and immedi-ately around the wind plant, or could manifest itselfin populations or subpopulations some distanceaway through changes in immigration and emigra-tion. This type of influence is extremely difficult toquantify in the field without a tremendous expendi-ture of time and funds. Here, it becomes essentialthat a rigorous and focused modeling framework beestablished so that the potential impacts can behypothesized given a variety of scenarios (e.g., levelsof death). In this way, inference can be drawn fromdata collected over the short term as it applies tolikely longer-term impacts using projections of vari-ous population models.

The second issue with regard to cumulative effectsconcerns the expansion of an existing wind plant.The comments in the preceding paragraph stillapply, but the issue is complicated by the continuing

development of the wind plant. No information isavailable on how bird populations respond to windplant expansion. In particular, we do not know if therelationship between number of turbines and num-ber of deaths is linear, or if it plateaus at some point.Further, we do not know if the potential benefits of awind plant to certain bird species (e.g., potentialincrease in prey for raptors) reaches some optimallevel given a certain size of the wind plant. Hereagain, the most efficient approach would be tomodel the likely responses of a population to simu-lated changes in prey abundance and deaths, andthen compare the resulting population projects withwhat is found initially in the field. These results willindicate the level of concern that should be appliedto bird deaths.

As detailed in chapter 3, proper experimentaldesigns must be implemented for analysis of theresponse of birds to wind energy development. It isbeyond the scope of this chapter to describe all ofthe various designs and analyses possible. The stan-dard call for adequate treatments and references,including pre-treatment data, apply here as well. Theadvantage of designing a study of cumulative effectsas a wind plant expands is that good referencespotentially exist in the areas that are scheduled fordevelopment at some point in the future. The onlyweakness here is that, if the wind plant is fully devel-oped, the references will eventually disappear;allowances must be designed for this eventuality(e.g., locating areas that could be suitable for windenergy development, but are unlikely to be so developed).

Land uses unrelated to wind development also couldimpact bird populations inhabiting a wind plant. Forexample, residential housing, commercial develop-ment, roads, and agriculture could influence birdson or near a wind plant. It is not the purpose of thisdocument, however, to discuss the myriad non-windfactors that could be part of a complete analysis ofthe cumulative effects of human activities on birdpopulations. Such an endeavor would involve athorough environmental impact assessment.

RECOMMENDATIONS FOR STUDYDESIGNBelow is a summary of the primary points discussedin this chapter.

1. Manipulative studies can be an effective meansof determining the response of birds to treat-ments or experiments designed to test behavior,


such as procedures designed to identify methodsfor reducing the risk of bird deaths.

2. Developing a sound modeling framework mayhelp identify the critical aspects of the popula-tion that should be studied, even if a formalmodel is not calculated.

3. In many situations, quantification of adult sur-vivorship is an essential step in determining thestatus of the population of interest. Data on survival published in the literature is adequate to allow broad generalizations to be maderegarding “adequate” survival for populationmaintenance.

4. Determining the spatial structure of a population —whether it is divided into subpopulations — isimportant in that it places the status of variouslife history parameters into context.

5. Quantifying reproductive output and breedingdensity, when combined with knowledge of thepopulation’s spatial structure, provides a goodidea of the status of the population. This will beespecially important when adult survivorshipcannot easily be determined.

6. Habitat loss is usually a factor causing thedecline of a species. Quantification of habitatuse, including factors such as food abundance,can be an important part of evaluation of a population’s status.

7. Compensatory mortality should not be assumedto be operating with regard to wind plant relatedmortalities.

8. It is likely that Leslie matrix models will be mostuseful when predicting the response of locallyabundant subpopulations. Here, enough individ-uals are present for a population trend to be estimated.

9. Determination of the effective population size(Ne) likely will be useful in evaluating the statusof rare subpopulations. A rapid determination of the likely lower critical threshold for the sub-population is necessary.


INTRODUCTIONThis chapter outlines guidelines for studying the riskof death to birds in wind energy developments, andfor studying the effect of treatments to reduce thatrisk. It is important to bear in mind that there are twocomponents to this issue:

• The effect of a wind energy facility on bird utilization of a site

• The risk of death to birds using the site (birdmortality)

Thus, we are interested in quantifying both the useof a site and the deaths associated with that use.Great care must be taken in identifying an appropri-ate measure of bird use of a wind energy develop-ment (e.g., bird abundance, passes near a turbine,nesting success). Indirect factors, such as changes inhabitat, prey quality and quantity, nesting sites, andmany other factors can affect bird use of a windenergy development and must be considered instudy design.

The goal of this chapter is to develop guidelines forstudying methods of reducing the risk of bird death

in wind energy developments. This goal will beaddressed by:

1. developing methods of assessing avian risk

2. reviewing the hazards to birds in wind energydevelopments

3. describing methods of study design for evaluat-ing bird risk in wind energy developments, andfor assessing the effectiveness of treatments inreducing that risk.

Much of the material reviewed herein is taken fromreports previously developed under NREL auspicesby L. S. Mayer, M. L. Morrison, H. Davis, K. H.Pollock, R. L. Anderson, and S. A. Gauthreaux.

ASSESSING AVIAN RISKIn assessing avian risk with the purpose of eliminat-ing or reducing that risk, it is essential to quantifyboth the use of a site and the deaths associated withthat use. The ratio of death to use becomes a meas-ure, expressed as mortality, or the rate of death (orinjury) associated with bird utilization of the windenergy site. Following the epidemiological

Risk Reduction Studies 65

CHAPTER 5Risk Reduction Studies

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65

ASSESSING AVIAN RISK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65

Attributable vs. Preventable Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . .66

Measuring Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .67

REVIEW OF WIND ENERGY PLANT HAZARDS TO BIRDS . . . . . . . . . . .71

Direct Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71

Indirect Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .71

RISK REDUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .72

Plant Siting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .73

On-site Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .73

STUDY DESIGNS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .73

Basic Experimental Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . .73

Treatments and References (Controls) . . . . . . . . . . . . . . . . . . . . . . . .74

Statistical Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .75

approach, mortality is the outcome variable — thevariable that the researcher considers most likely toshed light on the hypothesis about the mechanism ofinjury or death. Mortality thus depends heavily onthe mechanism hypothesized as the cause of injuryor death. Determining the mechanism of injury ordeath allows the development of appropriate meth-ods to reduce the risk to a bird while in a windplant.

In testing modifications to turbines or wind plants, itis important to separate bird mortality from bird uti-lization in order to determine if decreased deathswere due to decreased utilization, decreased risk, orboth. Separating utilization from risk makes it possi-ble to know if a modification that reduces risk ofinjury or death of birds using a wind plant has a pos-itive or negative effect on the population. For exam-ple, modifications in turbine characteristics might beaccompanied by, say, a 30% reduction in the num-ber of dead birds as compared to the wind plantbefore the modifications. If use has remained con-stant, this could be interpreted as a positive outcomefrom the modifications. If, however, use of the windplant by birds drops by 50% following the modifica-tion, then the benefits of the modification are notclear. If the 50% decline in use results from somefactor unrelated to the wind plant then the 30%decline in the number of deaths might actually beconsidered an increase in mortality. Thus, if weignore bird use, we cannot assume that a modifica-tion that is accompanied by reduced deaths willhave a positive or a negative effect on the bird popu-lation in question.

Attributable vs. Preventable RiskAttributable RiskAttributable risk is defined as the proportionalincrease in the risk of injury or death attributable tothe external factor (turbine or wind plant). It com-bines the relative risk imposed by exposure to theexternal factor with the likelihood that a given indi-vidual is exposed to the external factor. Attributablerisk (AR) is calculated as:

AR = (PD - PDUE)/PD

where PD is the probability of death per individualfor the entire study population, and PDUE the proba-bility of death per individual for the population notexposed to the risk. For example, if the probability ofdeath for a randomly chosen individual in the popu-lation is 0.01, and the probability of death in a refer-ence area for a bird flying through a theoretical rotorplane without the presence of blades is 0.0005, then

AR is (0.01 - 0.0005)/0.01 = 0.95. Thus, about 95%of the risk of dying while crossing the rotor plane isattributable to the presence of blades. As noted byMayer (1996), it is the potentially large attributablerisk that stimulates the concern about the impact ofwind energy development on birds, regardless of theabsolute number of bird deaths. Testing a preventivemeasure in a treatment-reference experiment allowsus to determine the change in risk due to the treatment.

Preventable FractionIf impacts are at unacceptably high levels, and pre-ventive measures are deemed necessary, there isinterest in determining the proportion of deathsremoved by a preventive step. The proportion ofdeaths removed by a preventive step is termed thepreventable fraction and is defined as the proportionof injuries or deaths that would be removed if allbirds were able to take advantage of the preventiveintervention (e.g., perch removal). Preventable frac-tion (PLF) is calculated as:

PLF = (PD - PDI)/PD

where PDI is the probability of injury or death giventhe preventive intervention. For example, if popula-tion mortality in the wind plant is 0.01, and mortali-ty for those using the area with perches removed is0.005, then the preventable fraction is (0.01 -0.005)/0.01 = 0.5. Thus, about 50% of the riskwould be removed if all of the perches wereremoved. Note that the attributable risk and prevent-able fraction would be the same value if the inter-vention removed the risk.

Prevented FractionThe prevented fraction is the actual reduction inmortality that occurred because of the preventiveintervention. Prevented fraction (PFI) is calculatedas:

PFI = (PDAI - PD)/PDAI

where PDAI is the probability of injury or death inthe absence of intervention. For example, supposethat 25% of the perches are removed in a treatment-reference experiment. If field studies determine thatmortality is 0.01 for the population and 0.015 forthose living without the prevention (e.g., perches),then the prevented fraction is (0.015 - 0.01)/0.015 =0.33. Thus, about 33% of the risk has been removedby removing 25% of the perches. This and the fol-lowing analyses can be stratified by ages or sexes ofconcern.


It is important to remember that these three meas-ures of effect remove emphasis from the risk to indi-viduals and place emphasis on the risk to thepopulation.

Measuring riskPotential Observational Data There are a limited number of parameters that canbe measured during Level I studies. These studiesnormally will not use marked animals and the observational methods will not allow estimation ofabsolute abundance. However, observational datacan be used to estimate use, which can be consid-ered an index to abundance, where the parametermeasured is an observation of an individual birdover some specific time period. Individual behaviorsalso can be quantified. Observations of use can beclassified according to activity, and thus used to esti-mate the amount of time a particular species spendsperching, soaring, flapping, etc. If these activitiescan be related to risk, they can be used to testhypotheses regarding the impact of wind plants onbird species. For example, it may be assumed thatthe more time a species spends flying at heightsencompassed by the rotor swept area of turbines, the more risk the species faces in a wind plant.Measures of use should allow a comparison ofpotential wind plant sites for differences in risk tobird species. The season of use can indicate the relative abundance of migrants, wintering birds, and breeding populations.

Because many birds migrate at night, estimates ofnocturnal use are of interest, but suitable methodsare still in the developmental stage. To date, howev-er, there is no strong evidence that large-scale killsare occurring at night. Some concern has beenraised that bats might be killed by colliding with tur-bine blades. Cooper (1996) describes the use ofradar for wind plant-related research. Gauthreaux(1996b) describes advances in the use of radar andinfrared techniques in the study of bird use andmigration. Recent work by Evans and Mellinger(1999) and Evans and Rosenberg (1999) describe theuse of acoustic monitoring to estimate species com-position of nocturnal bird migration. These “remotesensing methods” of bird study may provide tools for use in the estimation of impact in the future.Presently they seem most useful in early screening ofwind resource areas for potential conflicts with birdssimilar to the study described by Hawrot andHanowski (1997).

Mortality is the primary indicator of negative impactto individual animals from a wind plant. Mortality

can be calculated from an estimate of fatalities. Touse carcasses in assessing a wind plant as a cause offatalities, all carcasses located within areas surveyed(regardless of species), should be recorded and acause of death determined, if possible (The U.S. Fishand Wildlife Service may assign a cause of death forlegal purposes). Not all carcasses will be whole ani-mals. The condition of each carcass found should berecorded using condition categories such as:

• Intact - carcass that is completely intact, is notbadly decomposed, and shows no sign of beingfed upon by a predator or scavenger.

• Scavenged - entire carcass that shows signs ofbeing fed upon by a predator or scavenger or aportion(s) of a carcass in one location (e.g.,wings, skeletal remains, legs, pieces of skin,etc.).

• Feather spot or feather tract - 10 or more feath-ers at one location indicating predation or scavenging.

The estimated time of death, season of death, andspecies can be important in interpreting fatalities.There is always the possibility that death was notcaused by striking the turbine, so care should betaken in assigning a cause of death (e.g., shooting,poisoning). In certain situations a blind necropsymay be indicated.

In addition to carcasses, observers may discover livebirds that cannot fly or have other physical abnor-malities due to collisions with turbines or otherinjuries. These birds should be captured and exam-ined to determine the cause of injuries. For injuredbirds that cannot be captured, the species, location,and physical abnormalities observed should bedescribed in the data. Injured birds should be treated in accordance with the appropriate laws and regulations.

Impacts of wind plants on reproduction can bemeasured. In Level I studies the most common measure of reproductive performance will bethrough nest surveys. For example, the number anddistribution of active nests within an area potentiallyimpacted by the placement of wind turbines overtime represents an index to the status of the breedingpopulation of raptors. The area influenced by windturbines will extend varying distances depending onthe size of the area utilized by individuals of thespecies of interest. Passerines may range only a fewhundred meters while raptors can range 20 or more


kilometers. For species with multiple nest structuresthe number of breeding pairs is of more value thanoccupied nests when evaluating breeding populationstatus. Other factors that are changing within andaround the wind development, such as roads, hous-ing, and recreational activities, might also impactbirds and should also be considered in any analysis.

Nesting surveys for smaller species such as passer-ines, some shore birds, and ground nesting birds arebest accomplished on foot using ground surveys(Ralph et al., 1993). Unless the area is completelycovered, previously described sampling protocolsshould be followed. For larger species, such as rap-tors, study areas should be surveyed initially whenpossible by air, preferably by helicopter, during theheight of the nesting period. Aerial surveys shouldbe followed immediately by ground surveys to con-firm the species and status of each observed nest.Ground visits to occupied nests should be contin-ued, to confirm the number of young fledged.Surveys should begin early enough to detect earlynesters, such as eagles, and continue until all speciesof interest have begun nesting activity.

Empirical data on nesting pairs should be collectedfor all species of interest. In addition, the numerousreproductive parameters should be estimated to aug-ment empirical data. The number of occupied nestswithin the defined area can be used to estimate rela-tive abundance of nesting species potentially affect-ed by the wind turbines. The following nest andterritory parameters are suggested:

• Occupancy rate - the number of occupied terri-tories (nests) per number of territories (nests)checked.

• Breeding pair density - the number of breedingpairs per area surveyed.

• Reproductive rate - the number of reproductivepairs per number of occupied territories.

• Fledging success rate - the number of pairsfledging young per number of reproductive pairs.

• Breeding rate - the number of young fledged pernumber of reproductive pairs.

Statistical comparisons of these parameters, if suffi-cient data exist, can be made among assessment andreference areas before and after construction.

Data on the above parameters will contain numer-ous biases, most of these related to the samplingmethod, data collection methods used (e.g., radar,visual, etc.), and observer and detection biases.Biases associated with sampling methods have beendiscussed previously. Biases associated with datacollection methods may be found in numerous refer-ence publications including Bibby et al. (1993),Buckland et al. (1993), Bookhout (1994), Edwards etal. (1981), Gauthreaux (1996a), and Reynolds et al.(1980).

Selection of Impact IndicatorsImpact indicators should allow for the determinationof impact following generally accepted scientificprinciples. Stakeholders should believe that the crite-ria for determination of impact will be satisfied bythe indicators at the end of the assessment period.Of course, other indicators that are believed to pro-vide useful information for analysis or for corrobora-tion of results also should be measured. In the end,studies should be designed to:

• quantify indicators that will allow convincingarguments that impacts did or did not occur

• quantify the magnitude and duration of theimpact with acceptable measures of precisionand accuracy

• allow for standardized comparisons among pop-ulations and with results of other studies.

In an ideal world, a study of birds and wind plantswould involve a direct count of birds using or pass-ing through the wind plant, behaviors putting birdsat risk, and a count of fatalities caused by wind tur-bines and related facilities. To count birds andbehaviors one would need to identify individualbirds. To count fatalities one would need to detectcarcasses before removal by scavengers and be 100percent confident of the cause of death. This level ofeffort is not possible in Level 1 studies.

As an alternative, studies of wind energy/bird inter-actions must rely on estimation of parameters thatallow the test of hypotheses. These parameters areoften expressed as rates, similar to epidemiologicalstudies. Mayer (1996) provides an excellent discus-sion of the use of epidemiological measures to esti-mate the effects of wind plants and related facilitieson bird species. He points out the importance ofselecting the appropriate denominator when devel-oping a rate for use in comparisons of effect. For


example, a comparison of the number of bird fatali-ties per turbine among portions of a wind plant,between two turbine types, or among several windplants, is much more meaningful if an estimate ofbird abundance is added to the denominator.

There are a limited number of parameters that onecan measure in a Level 1 study. The more likelyparameter candidates and some potential riskindices are listed here and described below.

• bird utilization counts

• bird utilization rate

• dead bird search

• bird mortality

• removal rate

• observer bias

• detection bias.

There is little doubt that the presence of a wind plantwill increase the risk of individual bird fatalities. Thismay be of great concern if the individual birds at riskhave some special significance, as in the case of anextremely rare species. Risk of individual fatalitiesmay be of interest when planning the design or loca-tion of a new wind plant, evaluating differencesamong turbine types, or when making modificationsin equipment. However, the risk of individual fatali-ties may not necessarily represent a risk to a popula-tion of birds. Studies of risk to individuals andpopulations require separate study designs.Normally, Level 1 studies will be designed to makedirect statistical and deductive inference to risk toindividuals and indirectly indicate risk to popula-tions. Level 2 studies normally will be needed toestimate risk to populations.

Metrics DefinitionsBird utilization counts. Utilization counts areindices of relative abundance among plots, areas,and seasons. Utilization counts represent observa-tions of individual birds from an observation point ortransect conducted repeatedly over some time peri-od to document behavior and relative abundance ofbirds using the area. The observer counts the lengthof time the bird is within the plot and estimates “birdminutes” of use. The bird utilization counts allowcomparisons among defined time periods (e.g., sea-sons, migration periods, or years), and areas. Bird

activities should include behaviors which could berelated to risk of injury or mortality from wind plantsand might include flying, perching, soaring, hunting,foraging, height above ground, and behavior within50 meters of WRA structures, etc. In situations ofhigh bird density where it is impossible to keep trackof all birds in a plot, use can be estimated for theobservation period by making instantaneous countsrepeatedly during the counting time period.

Bird utilization rate. This term refers to the numberof birds observed or the number of bird minutesrecorded per count period and/or survey plot. Likebird utilization counts, bird utilization rate may beused for comparisons among plots, areas, and sea-sons. One formula for utilization rate is

# birds observed = Bird Utilization Ratetime or time and area

Utilization rates within specified distances of windplant structures (e.g. large and small turbines, differ-ent tower types, etc.), subdivided on the basis of rel-evant environmental covariates (e.g. topographicfeatures, vegetation edge, nesting structures, etc.)can be derived from the bird utilization counts.Rates can be developed for species, taxonomicgroups, all birds observed, natural communities, sea-sons, distance from nearest turbine, turbine type,and other variables. Rates can be calculated for spe-cific behaviors and risk can be evaluated in terms ofthe number of birds observed exhibiting behaviorsthat place them at greater risk. For example, birdsflying at heights within the range of the rotor sweptarea are likely at greater risk than those consistentlyflying at heights above and below the rotor sweptarea. Evaluation of risk based on behavioral data canbe used in a variety of studies of wind energy includ-ing relative comparisons of areas, turbines, andspecies. The choice of a utilization rate is critical;see discussion below.

Dead bird search. Searches are conducted in adefined area with complete coverage to detect birdfatalities. The number of dead birds found at eachsearch site (e.g., a 50-meter diameter circle centeredon the bird utilization count site) is documented.Information is collected which will aid in analysislater in the study. This may include bird species, sex,age, estimated time since death, cause of death, typeof injury, distance and direction to nearest turbine,and distance and direction to nearest structure.

Bird mortality. The number of dead birds document-ed per search site may be termed “bird mortality.”


This is the rate of fatalities. Examples of indices forbird mortality are:

# dead birds , and dead birds turbine unit rotor swept area

Removal rate. This is the rate at which bird carcass-es are removed by scavengers or by other means(e.g., human removal), resulting in their loss todetection by the dead bird search. Information aboutremoval rates is necessary when estimating the totalnumber of dead birds in a given area. The results areused to adjust the number of dead birds detected.This rate may be determined by placing a knownnumber of bird carcasses at randomly chosen loca-tions and monitoring them for removal. Removalrates can be calculated as a rate or rate/area. Thisallows for comparison of removal levels betweendifferent locations or subareas within the WRA. Ifnot detected, significant removal rate differenceswould result in misleading bird risk rates. If removalrates in different areas within the same WRA orbetween WRAs are equal, they will have no effectwhen computing and comparing mortality rates, birdrisk rates, and attributable risk rates.

Observer bias. Observer bias is a quantification ofthe observer’s ability to find dead birds or detect livebirds. One study might quantify the observer’s abilityto find dead birds when a known number of birdsare placed in the search area. Another study mightcompare the field crew’s live bird observations inorder to determine inter-observer differences.

Detection bias. Detection bias is a measure of thedifferences in detection probability due to topogra-phy and vegetative structure. Detection bias may bedetermined through a designed study which includesplacing a known number of dead birds in a varietyof locations with differing topography and vegetativestructure. The detection success can be quantifiedand the probability of detection determined.

Defining UtilizationIf risk is defined as the ratio of dead or injured birdsto some measure of utilization, then the choice ofthe use factor, or denominator, is more importantthan the numerator (number of dead or injuredbirds). In fact, the treatment effect is usually smallrelative to the variability that would arise fromallowing alternative measures of risk. The choicearises from the preliminary understanding of theprocess of injury or death. For example, should thedenominator be bird abundance, bird flight time inthe plant, bird passes through the rotor plane, or

some other measure of use? Unless these measuresare highly correlated with death — which may beunlikely — then the measure selected will result inquite different measures of mortality. Further, thechoice of denominator should express the mecha-nism causing the injury or mortality. If it does not,then it cannot be used to accurately measure theeffectiveness of a risk reduction treatment. There is,however, much uncertainty in the mechanism(s)leading to bird fatalities in wind plants.

Choice of utilization factor. Suppose that bird useor abundance is selected as the denominator, withbird deaths as the numerator, and painted blades asthe treatment. A treatment-reference study deter-mines that death decreases from 10 to 7 followingthe treatment, but use also decreases from 100 to 70(arbitrary units). It thus appears that the treatmenthad no effect because both ratios are 0.1 (10/100and 7/70). There are numerous reasons why bird useof a wind plant could change (up or down) that areindependent of the blade treatment; for example,changes in prey availability, deaths on winteringgrounds, environmental contaminants, change ofland use, and so on. Thus, unless it can be estab-lished that there is a direct link between the numberof birds using the area and flights near a turbine, thisstudy may be seriously flawed. Recording bird flightsthrough the rotor plane of painted blades wouldhave yielded a more correct measure of effect. Inaddition, the use of selected covariates can helpfocus the analysis on the treatment effects. Naturally,the hypothetical study noted above should be ade-quately replicated if implemented. (See chapter 3 forrecommendations on study design.)

Surrogate utilization variables. Utilization is anindicator of the level of at-risk behavior. Thus,adopting a measure of utilization requires theassumption that the higher the utilization, the higherthe fatalities. It is, of course, prohibitive from a prac-tical standpoint to record every passage of a birdthrough a zone of risk (be it a rotor plane or theoverall wind plant). Further, it is usually prohibitiveto accurately census the population and tally alldeaths. Researchers must usually rely on surrogatevariables to use as indices of population size anddeath. A surrogate variable is one that replaces theoutcome variable without significant loss in thevalidity or power of the study. For example,researchers might use the number of birds observedduring 10-minute point counts (i.e., the number ofbirds counted during a 10-minute observation peri-od) as a measure of utilization (for either a treatmentor reference case).


Once a measure of mortality is chosen, a measure ofeffect must be selected. This measure could be therisk ratio, defined as the ratio of mortality in onearea (e.g., wind plant) to that in another area (e.g.,reference). Thus, if mortality in the wind plant is0.01 and that in the reference area is 0.001, the riskratio is 10; the relative (potential) risk of death is 10times greater for a randomly chosen bird in the siteversus one in the reference area. Ideally, such astudy should be adequately replicated, because ref-erences are not perfect matches to their associatedtreated sites. An alternative is to use one of themeasures of attributable risk described above. Thesemeasures have the advantage of combining relativerisk with the likelihood that a given individual isexposed to the external factor. This results in the pro-portional change in the risk of injury or death attrib-utable to the external factor. Whereas the risk ratioignores the absolute size of the risk, the use of attrib-utable risk implies that the importance of the risk isgoing to be weighed by the absolute size of the risk.

REVIEW OF WIND ENERGY PLANTHAZARDS TO BIRDSDirect InteractionsCollision with TurbineThe specific factors causing bird deaths in winddevelopments are not well understood. It has beenproposed that birds die when trying to pass throughthe rotor plane because they cannot see the blades,turbulence, or because they are fixated on a perch orprey item beyond the blades. Birds might also bekilled by striking turbine support structures (wires),or by striking part of a tower, or through electrocu-tion by a turbine-related power line.

Layout of Wind PlantsTo maximize operation time, turbines often areplaced on ridges and upwind slopes. Such locationsplace turbines near updrafts that are commonly usedby soaring birds, including but not restricted to rap-tors. It has also been suggested that turbines placednear valleys and end-row turbines might result in rel-atively higher risk to individual birds. The spacingand height of turbines also could interact to changethe relative risk to individual birds. Thus, themicrositing of a turbine could be influencing avianrisk, and doing so in a complicated manner thatincludes turbine height and spacing, location alonga ridge, and the relationship to other turbines.

Plant OperationThe presence of the turbine, even with stationaryblades, could increase risk to individual birds, espe-cially in periods of poor visibility (fog, rain, night,

dusk or dawn). Obviously, if birds have difficultyseeing blades, then operating a turbine during poorvisibility would likely increase the risk of death forindividual birds. In addition, operating during peakperiods of migration, such as during spring and fall,could increase the absolute number of bird deathssimply because of the large number of individualspassing through the area. To date, however, fewsmall, migratory birds have been shown to be killedin wind plants. We emphasize that these factors areuntested hypotheses and should not be taken to rep-resent management recommendations.

Indirect InteractionsChanges in Species HabitatCentral to understanding how a development suchas a wind plant affects animal habitat is a properunderstanding of the term itself. Unfortunately,“habitat” is a commonly misused term in ecology.First, habitat is a species-specific concept. That is, an area is neither “good” nor “poor” habitat per seexcept with reference to a specific species. Part ofthe misunderstanding comes from the term “habitattype” — originally developed for use in vegetationecology to describe the general type of vegetation inan area — being misapplied to describe an animal’shabitat. The term “vegetation type” should be usedto describe the vegetative community in an area;“habitat” should refer to the environmental charac-teristics used by a specific animal or group ofspecies (see below). To avoid confusion betweenanimal and plant ecology, the term “habitat type”simply should not be used in any context.

Habitat is a multifaceted concept encompassing:

• both the structural and floristic composition ofvegetation

• any number of environmental factors that influ-ence animals, including water, soil properties,salinity, temperature, and so forth

• competitors and predators

• quantity and quality of food

• various other factors.

The amount, or presence or absence of any one ofthese factors can render an area unsuitable, regard-less of the status of the other necessary factors. Forexample, a location that seems otherwise ideal for aspecies might, in fact, be unsuitable because of the


presence (perhaps unseen to the observer) of a predator.

Thus, it is critical that the observer has a firm under-standing of the habitat requirements of a speciesbefore making recommendations regarding the less-ening of a real or potential environmental impact,such as development of a wind plant. As brieflyreviewed below, a host of factors can affect the num-ber and behavior of animals in an area, both directlyand indirectly.

Changes in Prey Quantity and QualityThe substantial changes to an area that accompanydevelopment often cause substantial changes in theabundance and distribution of potential food for ani-mals. Developing a wind plant requires that a net-work of roads be constructed, that pads be formedfor siting the turbines and support buildings, thatpowerlines be buried, and so forth. All of these oper-ations disturb the soil, many of them in permanentways. These activities thus alter the potential habitatfor many species, likely lessening the amount andquality for some species, while increasing it for otherspecies. A notable example of this in wind energydevelopment is the probable enhancement of habitatfor ground burrowing animals (e.g., squirrels,gophers) due primarily to the aforementioned soildisturbance. In many areas throughout the west,such disturbance has resulted in substantial increas-es in ground squirrel abundance (e.g., Salmon 1981,Smallwood et al. 1998). Any activity that also resultsin low grass height also enhances ground squirrelhabitat, including fire, grazing, and mowing. Manyother species of small mammal, including those noc-turnal species seldom seen by people, also react(positively or negatively) to these changes in soil andvegetation. Some of these changes could beephemeral, or can be eliminated through habitatrestoration.

Soil disturbance, along with changes in vegetativestructure, can also change the habitats available fora host of other species. For example, species ofgrassland birds react differently to different densitiesand heights of grasses and shrubs. Predators can beattracted to a disturbed area because of an increasein prey density, or might avoid the area becausetheir prey has decreased, or because of the increasedactivity of people.

Thus, the response of animal species can usually bepredicted within at least wide bounds depending onthe specific conditions present following develop-ment. However, it must be remembered that the

responses are species-specific, and gross generaliza-tions regarding the impact of development on theanimal community cannot be made. For example,increasing the abundance of raptors because preyare more available is not “good” if other specieshave declined or disappeared, and the goal was tomaintain a predevelopment animal community.

Changes in Perches, Nest Sites, and Related ItemsWind plants may increase the number of perchespotentially available to birds for perching and nest-ing. The increasingly common use of tubular towers(rather than lattice towers) is reducing the perchingopportunities on generating equipment in windplants. Although perching by raptors is most obviousbecause of their size, most birds perch above theground, including even grassland ground-foragingspecies such as sparrows, meadowlarks, and larks.Birds perch for a variety of reasons, including to rest,to scan for predators, or to scan for prey. Thus, useof turbines as perches could enhance the habitat ofan area for a species by increasing its hunting suc-cess and/or lowering its susceptibility to predation.As noted above, there is likely some cost/benefit tothe bird that is species-specific, and also probablylocation (wind plant)-specific. That is, decreasingstarvation by increasing hunting success, while alsoincreasing deaths due to striking blades, could resultin an overall reduction in population mortality.Evaluating attributable risk is an important compo-nent in examining the cost/benefit ratio.

RISK REDUCTIONIndividuals from the wind industry and the scientificcommunity as well as individual environmentalistsand regulators have postulated that bird deaths canbe reduced by modifying towers to reduce perching,painting disruptive patterns on turbine blades, modi-fying turbine spacing, and so on. Some have suggest-ed, however, that statistically valid analyses of suchtreatments are not feasible because bird deathappears to be such a rare event. While it may beargued that simply reducing bird use on and aroundtowers is sufficient to conclude that treatments havebeen effective, the weakness of this argument is thatchanges in behavior could also cause increases indeath even if the use around turbines has declined.(For example, a perch guard might successfully pre-vent birds from perching on the tower, but mightalso have the effect of causing a frightened bird to flyinto the blades, indirectly resulting in the very deathit was designed to prevent.) Further, without quantifi-cation of dead birds, no statements can be maderegarding the influence of turbines on the abun-dance and dynamics of bird populations — unless


the turbines displaced the population (see chapter4). If the risk to an individual per visit to a turbinestays the same, then mortality (rate of bird death) hasnot been reduced even if fewer birds visit. Thus, theparameter used to quantify “visit” is an absolutelycritical part of impact assessment.

Plant SitingIt seems intuitive that avoiding areas of concentratedbird activity would eliminate many potential bird-turbine problems. Surveys can be used to determineif a proposed plant site is located in areas of highnesting or seasonal density, or in the range of athreatened or endangered species. Using such datain the site selection and evaluation process canreduce the absolute number of bird deaths. Manyexisting wind plants have avoided areas of concen-trated bird use.

On-site ReductionThere are two major possibilities for reducing therisk to birds on a developed site. First, the site itselfcan be made unsuitable for use by birds or a specificbird species, either directly through micrositing ofthe turbines, or indirectly through changes in habitatparameters (e.g., changing prey type or abundance).Second, the turbines themselves can be madeunsuitable for use by birds (e.g., removing potentialperches on lattice towers).

Micrositing includes the position of turbines relativeto a ridge, spacing between turbines, distance frompotential perch and nest sites, turbine location rela-tive to vegetated gullies or water sources, and soforth. It is important, given the concern over turbine-caused bird deaths, that micrositing include consid-eration of biological factors.

The height of the turbine tower, as well as the lengthof the blade, also could function in bird deaths.Taller towers potentially could expose a narrower orwider range of birds to impacts, although little spe-cific research has been conducted on this factor. Inaddition, the length of the blade changes the rotorswept area, thus potentially changing the opportuni-ty for collisions (Howell 1997).

Birds exhibit variations in activity both within andbetween days. These variations can be quantified bydeveloping activity budgets for species of concern.Based on such data, reliable models could be devel-oped that predict times of maximum risk to birds

Development of a wind plant likely changes preyavailable to birds, both increasing food for certain

species and decreasing it for others. Many of thesechanges can be predicted, and thus the response ofbirds to the changes anticipated. For example, it islikely that numbers of ground squirrels will increasebecause of soil disturbance and decreased grassheight through vegetation management that oftenaccompany wind plant development. Because squirrels are a central part of the diet of many largeraptors, it is likely that an increase in squirrel abun-dance could attract certain raptors to the site. It hasbeen well documented that birds use certain towertypes as perches, potentially increasing the chancesof bird death because of the proximity to the blades(e.g., Orloff and Flannery 1992).

STUDY DESIGNSBasic Experimental Approaches As outlined by Mayer (1996), there are four tasksthat the investigator must accomplish when design-ing a study of wind energy/bird interactions. Thelogic is sequential and nested; each choice dependson the choice made before:

1. Isolate the hypothesis of mechanism that isbeing tested. For example, one might be testingthe hypothesis that birds strike blades whenattempting to perch on a turbine.

2. Choose a measure of injury-death frequencythat best isolates the hypothesis being tested.The two components of this choice are tochoose an injury-death count to use as a numer-ator and a base count (likely utilization) to useas a denominator. It is critical that a relevantmeasure of use be obtained (e.g., passes throughthe rotor plane; occurrence by flight-height categories; use within a certain distance of theturbine).

3. Choose a measure of effect that uses the meas-ure of injury-death frequency and isolates thehypothesis being tested. The key is to decidewhether the relative risk (risk ratio), attributablerisk, or another measure of effect should beused.

4. Design a study that compares two or moregroups (12 is preferable) using the measure ofeffect applied to the measure of injury-deathfrequency chosen. The goals here are to isolatethe effect, control for confounding factors, andallow a test of the hypothesis. Replication isessential.


The ideal denominator in epidemiology is the unitthat represents a constant risk to the bird. The unitmight be miles of flight, hours spent in the site, oryears of life. If the denominator is the total popula-tion number then we are assuming that each birdbears the same risk by being alive. In human epi-demiological studies, the total population size isusually used because we cannot estimate units oftime or units of use. In avian studies, however, actu-al population density is extremely difficult to esti-mate and entire populations are seldom at risk fromthe site. If the risk is caused by being in the area,then deaths per hour in the area is probably the bestepidemiological measure in avian studies. It is thenextrapolated to the population by estimating the uti-lization rate of the area for the entire population.Measuring utilization is difficult, however, and mustbe approached carefully (see chapter 3).

Thus, we have two major alternative ways to calcu-late mortality rate:

1. number of dead birds / number of birds in population

2. number of dead birds / bird use.

The first ratio may be the ideal, but as discussedabove, is usually impractical. The second ratio isfeasible, but will vary widely depending upon themeasure of bird use selected. In addition, if bird useis chosen for the denominator, the background (non-wind site) mortality rate must also be determined forcomparative purposes. Thus, use of ratio (2) shouldbe the focus of further discussion.

Consultations with wind industry personnel haveidentified the need for a measure of operation timethat is easily standardized among wind plants andturbine types, preferably one that considers differ-ences in blade size and operation time. To meetthese requirements, the concept of rotor swept areahas been developed. This is simply the circular areathat a turning blade covers. Rotor swept area is thenconverted to an index that incorporates operationtime as follows:

rotor swept hour = rotor swept area x operation hours

An index of risk is then calculated by using a meas-ure of risk:

rotor swept hour risk = risk measure/rotor swept hour

Here, “risk measure” could be flight passes throughthe rotor plane or any other appropriate measure ofuse (as discussed above). The length of the bladeand speed of rotation are factors that could alsoinfluence risk (e.g., Tucker 1996).

Treatments and References (Controls)Study DesignThere are two main options for experimental units,using either the wind plant or a small plot as thebasis for study.

Wind plant-based study. In this design, a relativelylarge portion of the wind plant serves as an experi-mental unit. For example, a group of 100 turbineswould receive treatment (e.g., perch guards, paintedblades), and a similar group of turbines would serveas a reference. This basic approach could be appliedto both existing and planned wind plants. Unlesspreliminary studies are first conducted, an educatedguess would be necessary to determine how manyturbines to include in an experimental unit. Further,it will usually be difficult to replicate the pairs ofexperimental units if extrapolation to the plant isdesired. With a few pairs (1, 2, or 3), this design ismost comparable to a series of observational studies.Even if treatments are randomly assigned to onemember of the pair, statistical inference is only to thepair and the protocol by which they were selected.With this design, however, extrapolation to the entirewind plant is subjective and possible only if one iswilling to assume that the study sites are representa-tive of the wind plant (i.e., that appropriate criteriawere used to select treatments and references, andthat the study is adequately replicated). Thisapproach can be considered a model-basedapproach.

Small plot-based study. In this design, an individualturbine, or a small group (e.g., a string) of turbinesserves as the experimental unit. For example, pairsof strings of turbines are randomly selected and oneof each pair is randomly selected to receive thetreatment. This design has the advantage of beingcentered on discrete units that can readily beobserved; it is a design-based experiment. Thegreater the area of the plant sampled (replicated), thebetter the basis for extrapolating results to the plant,using deductive inference (professional judgment).

This design is preferred because of the relative easeof gaining an adequate sample size. A relativelylarge number of pairs of units can be analyzed in thesense of a “true” experiment. Extrapolation to the


entire wind plant is possible if treatment and refer-ence units are randomly sampled.

Design ConsiderationsTreatments and references can be reversed after theinitial experimental period. This strengthens the test,and would be useful in the wind plant-based studybecause of the likely small number of replicates possible.

Variable SelectionOne primary variable usually will drive the studydesign, and the initial sample size should be aimedat that variable. It is thus assumed that at least a rea-sonable sample size will be gathered for the other,secondary variables. There can be a sequentialanalysis of sample size as data are collected.

With the small, paired-unit design, one or two pri-mary use variables (e.g., passes through the bladeplane, perch attempts) will likely be adequate. Theminimum number of pairs to be sampled should beas many as can be afforded but no less than 12.However, a greater number of pairs would be desir-able, at least initially. The sampling unit can eitherbe individual turbines, or strings of turbines. It isexpected that string length will range from 5 to 10turbines, depending upon the size and configurationof the wind plant. Portions of longer strings can besubsampled; however, this complicates the analysis.

Designing treatment vs. reference studies for infer-ences on measures of use is feasible. Determinationof mortality is possible using Ratio (2) (see p. 74), butstatistical power to conclude that treatment and ref-erence sites have different mortality rates will below. For example, in a randomized pairs design,most pairs are expected to result in zero mortalities,with tied values and no mortalities on either memberof a pair. The high frequency of zero values effec-tively reduces the sample size for most analyses.

Case Study ApproachCase studies have high utility in evaluating mortality.Here, one collects dead birds inside and outside awind plant, and conducts blind analysis to deter-mine the cause of death. Unfortunately, under mostsituations very few dead birds will be found outsidethe site.

The case study approach suggests that epidemiologi-cal analysis can often be combined with clinicalanalysis to extend the inferential power of a study.Here the clinical analysis would be the necropsies of

the birds. Suppose we are successful at finding deadbirds inside a wind plant. If we look at proportionalmortality — the proportion of the birds killed byblunt trauma, sharp trauma, poisoning, hunting, nat-ural causes, etc. — then the proportions should dif-fer significantly between the plant and the referencearea. The assumption is that the differential bias infinding dead birds within the two areas is uniformacross the causes of mortality and thus the propor-tions should be the same even if the counts differ(i.e., relatively few dead birds found outside the site).

Behavioral and Physiological StudiesObtaining information on the sensory abilities ofbirds should help in designing potential risk-reduc-tion strategies for wind plants and individual tur-bines. Although it may seem intuitive to paint bladesso birds can more readily see them, there are manypossible designs and colors to select from. For exam-ple, what colors can birds see, and how do birdsreact to different patterns? If painting blades causes abird to panic and fly into another turbine, then paint-ing has not achieved its intended goal. Many ofthese questions are best investigated initially in alaboratory setting. Unfortunately, translating lab tofield is an age-old problem in behavioral ecology.Success in the lab using tame and trained birds doesnot necessarily mean success in the field, where amyriad of other factors come into play (wind speedand direction, fog, presence of other birds), and thephysical scales are different. However, initial labstudies can help to narrow the scope of field trials. Asequential process of initial lab testing of treatments,followed by field trials, followed by additional labtrials as indicated, can be implemented.

Researchers under the direction of Drs. HughMcIsaac and Mark Fuller, Boise State University,have been conducting a series of intensive laborato-ry trials to determine the visual acuity of raptors(unpubl. data). Included are trials to determine theability of the birds to differentiate between differentpainted patterns on turbine blades. In addition, theMcIsaac-Fuller research team has initiated field trialsto determine the ability of trained but free-flying rap-tors to avoid painted blades. This research is anexample of how combining laboratory and fieldexperimentation can be conducted to address bird-wind interactions.

Statistical ConceptsExperimental UnitsPairs or groups of turbines with similar environmen-tal conditions and/or breeding (nesting) densities for


the species of interest (and, if possible, similar histo-ry of mortalities) will be the basic sets of experimen-tal units. (Blocks of groups of turbines may improvethe design if there are three or more treatments.) Thenumber of turbines or strings of turbines will bebased on the configuration of the site. Theresearcher should attempt to sample as many as thebudget will allow but a minimum of 12 pairs ofstrings. Treatment should be randomly assigned to amember of each pair of experimental units (treat-ments are randomly assigned to experimental unitsin a block).

Sampling FrequencyThe frequency of sampling (i.e., taking measure-ments at the experimental units) should be based onthe goal and objectives of the project; initial sam-pling can be adjusted after preliminary data are ana-lyzed. Sampling should be stratified by time so thatadequate samples are taken both within andbetween days (temporal replication).

Stratification of sampling by major weather condi-tion (high or low wind; clear or moderate to heavyfog) could be valuable, because weather coulddirectly influence the frequency of bird strikes.However, stratification on weather would be verydifficult, and could be initiated only if funds areavailable for the additional observers necessary totake advantage of such conditions. A more practicaldesign would be to use weather as a covariate. It ishighly recommended that designs be kept simpleand focused, as it is far better to learn a lot aboutone treatment than it is to gain partial information on multiple factors. Sampling can cover a completerange of environmental conditions considered relevant to treatment effects. However, such anapproach often increases variability and may masktreatment effects (unless effects are very large). Seechapter 4 for a more detailed discussion of theseissues.


APPENDIX A:Literature Cited

CHAPTER 1Anderson, R.L. and J.A. Estep. 1988. Wind energy

development in California: impacts, mitiga-tion, monitoring, and planning. CaliforniaEnergy Commission, Sacramento. 12 pp.

Anderson, R.L., J. Tom, N. Neumann, and J. A.Cleckler. 1996b. Avian monitoring and Riskassessment at Tehachapi Pass WindResource Area, California. Staff Report toCalifornia Energy Commission, Sacramento,CA, November, 1996. 40pp.

Estep, J. 1989. Avian mortality at large wind energyfacilities in California: identification of aproblem. Staff report no. P700-89-001.California Energy Commission, Sacramento.

Higgins, K.F., C.D. Dieter, and R.E. Usgaard. 1995.Monitoring of seasonal bird activity andmortality on Unit 2 at the Buffalo RidgeWindplant, Minnesota. Preliminary progressreport for the research period May 1 -December 31, 1994. Prepared by the SouthDakota Cooperative Fish and WildlifeResearch Unit, National Biological Service,South Dakota State University.

Howell, J.A. 1995. Avian mortality at rotor sweptarea equivalents. Altamont Pass andMontezuma Hills, California. Prepared forKenetech Windpower (formerly U.S.Windpower, Inc.), San Francisco, California.

Howell, J.A., and J. Noone. 1992. Examination of avian use and mortality at a U.S.Windpower wind energy development site, Montezuma Hills, Solano County,California. Final report. Prepared for SolanoCounty Department of EnvironmentalManagement, Fairfield, California.

Hunt, G. 1995. A pilot golden eagle populationstudy in the Altamont Pass Wind ResourceArea, California. Prepared by the PredatoryBird Research Group, University ofCalifornia, Santa Cruz, for the NationalRenewable Energy Laboratory, Golden,Colorado. 218 pp. Report No. TP-441-7821.

Luke, A. and A. Watts. 1994. “Bird deaths promptrethinking on wind farming in Spain.”WindPower Monthly, February:14-16.

Martí, R. 1994. “Bird/wind turbine investigations insouthern Spain.” Pages 48-52 inProceedings of the National Avian-WindPower Planning Meeting, Denver, Colorado,20-21 July 1994. Proceedings prepared byLGL Ltd., environmental research associates,King City, Ontario, Canada. Author’saddress: Sociedad Española de Ornitologia,Ctra, de Húmera No. 63-1, 28224 Pozuelo,Madrid, Spain.

National Wind Coordinating Committee (NWCC).1998. Permitting of Wind Energy Facilities:A Handbook. Washington, D.C. 59 pp. +appendices.

Orloff, S. and A. Flannery. 1992. Wind turbineeffects on avian activity, habitat use, andmortality in Altamont Pass and SolanoCounty WRAs. Prepared by BiosystemsAnalysis, Inc., Tiburon, California, forCalifornia Energy Commission, Sacramento,Calif. 163pp + appendices.

Pearson, D. 1992. Unpublished summary ofSouthern California Edison’s 1985 bird mon-itoring studies in the San Gorgonio Pass andCoachella Valley. Presented at a joint PacificGas and Electric Company/California EnergyCommission workshop on wind energy andavian mortality. San Ramon, California.

PNAWPPM. 1995. Proceedings of National Avian-Wind Power Planning Meeting, Denver, CO,20-21 July 1994. Report DE95-004090.RESOLVE Inc., Washington, D.C., and LGLLtd., King City, Ont. 145pp.

Strickland, M.D., W.P. Erickson, and L.L. McDonald.1996. Avian monitoring studies - BuffaloRidge Wind Resource Area, Minnesota.Prepared for Northern States Power,Minneapolis, MN. 31pp.

Winkelman, J.E. 1994. “Bird/wind turbine investi-gations in Europe.” Pages 43-48 inProceedings of the National Avian-WindPower Planning Meeting, Denver, Colorado,20-21 July 1994. Proceedings prepared byLGL Ltd., environmental research associates,King City, Ontario. Author’s address:

Literature Cited 77

Birdlife/Vogelbescherming Nederland,Driebergweweg, The Netherlands.

CHAPTER 2Anderson et al., op. cit.

Evans, W.R. and D.K. Mellinger. 1999. “Monitoringgrassland birds in nocturnal migration.”Studies in Avian Biology No. 19:219-229.

Gauthreaux, S. A., Jr. 1996a. “Suggested practices formonitoring bird populations, movements,and mortality in wind resource areas.” pp.88-110 In: Proceedings of National Avian -Wind Power Planning Meeting II, PalmSprings, Calif., 20-22, 1995. Prepared forthe Avian Subcommittee of the NationalWind Coordinating Committee by RESOLVEInc., Washington, D.C., and LGL Ltd., KingCity, Ont. 152 pp.

Hunt, G. 1995, op. cit.

Morrison, M.L., B.G. Marcot, and R.W. Mannan.1998. Wildlife-Habitat Relationships:Concepts and Applications. Second edition.University of Wisconsin Press, Madison.

NWCC 1998, op. cit.

CHAPTER 3Anderson, R.L., J. Tom, N. Neumann, J. Noone, and

D. Maul. 1996a. “Avian risk assessmentmethodology.” Pages 74-85 in Proceedingsof National Avian - Wind Power PlanningMeeting II, Palm Springs, CA, 20-22 Sept.1995. Prepared for the Avian subcommitteeof the National Wind CoordinatingCommittee by RESOLVE Inc., Washington,D.C., and LGL Ltd., King City, Ont. 152pp.

Anderson, et al., 1996b, op. cit.

Box, G.E.P., and B.C. Tiao. 1975. “Interventionanalysis with applications to economic andenvironmental problems.” Journal of theAmerican Statistical Association 70:70-79.

Buckland, S.T., D.R. Anderson, K.P. Burnham, andJ.L. Laake. 1993. Distance sampling:Estimating abundance of biological popula-tions. Chapman & Hall, London, England.

Burnham, K.P. 1995. “Meta-analysis: synthesizingresearch results.” Colorado Cooperative Fishand Wildlife Research Unit, Fort Collins,CO. Page 20 in The Wildlife Society SecondAnnual Conference Abstracts, September12-17, 1995, Portland, Oregon. 153pp.

Cade, T.J. 1994. “Industry Research: Kenetech WindPower.” Pages 36-39 in Proceedings ofNational Avian-Wind Power PlanningMeeting, Denver, CO, 20-21 July 1994.Report DE95-004090. RESOLVE Inc.,Washington, D.C., and LGL Ltd., King City,Ont. 145pp.

Cochran, W.G. 1977. Sampling Techniques, 3rdEdition. John Wiley and Sons, NY.

Cohen, J. 1973. “Statistical power analysis andresearch results.” American EducationalResearch Journal. 10:225-229.

Colson, E. 1995. Avian interactions with wind ener-gy facilities: a summary. Prepared for theAmerican Wind Energy Association.

Cox, D.R. 1958. Planning of Experiments (WileyClassics Library Edition Published 1992).John Wiley and Sons, NY.

Dallal, G.E. 1992. The 17/10 rule for sample-sizedeterminations (letter to the editor). The Am.Stat. 46:70.

Draper, D., D.P. Graver, P.K. Goel, J.B. Greenhouse,L.V. Hedges, C.N. Tucker, J.R. Tucker andC.M. Watenaux. 1992. Combining informa-tion: statistical issues and opportunities forresearch. Washington, DC.: NationalAcademy Press.

Durlak, J.A. and M.W. Lipsey. 1991. “A practitioner’sguide to meta-analysis.” American Journalof Community Psychology, 19(3):291-333.

Eberhardt, L.L. 1976. Quantitative ecology andimpact assessment. Journal of EnvironmentalManagement 4: 27-70.

Erickson, W.P, and L.L. McDonald. 1995. “Tests forbioequivalence of control media and testmedia in studies of toxicity.” EnvironmentalToxicology and Chemistry 14:1247-1256.


Fairweather, P.G. 1991. Statistical power and designrequirements for environmental monitoring.Aust. J. Mar. Freshwater Res. 42:555-567.

Fisher, R. A. 1966. The design of experiments, 8th

Edition. Hafner Publishing Company, NewYork. 248 pp.

Gilbert, R.O. 1987. Statistical Methods forEnvironmental Pollution Monitoring. VanNostrand Reinhold Company, New York,New York, USA.

Gilbert, R.O., and J.C. Simpson. 1992. Statisticalmethods for evaluating the attainment ofcleanup standards. Vol. 3. “Reference-basedstandards for soils and solid media.” PacificNorthwest Laboratory, Battelle MemorialInstitute, Richland, WA. Prepared for theU.S. Environmental Protection Agencyunder a Related Services Agreement withthe U.S. Department of Energy. PNL-7409Vol. 3, Rev. 1/UC-600.

Gordon, A.D. 1981. Classification. Chapman andHall, London.

Green, R.H. 1979. Sampling Design and StatisticalMethods for Environmental Biologists. JohnWiley and Sons, NY.

Green, R.H. 1984. “Some guidelines for the designof biological monitoring programs in themarine environment.” Pages 647-655 in H.H. White (ed.). Concepts in MarinePollution Measurements. University ofMaryland, College Park, MD.

Hedges, L.V. 1986. “Statistical issues in the meta-analysis of environmental studies.”Proceedings of Biopharmaceutical Section,Am. Statist. Assoc., Alexandria, VA.

Hedges, L.V., and I. Olkin. 1985. Statistical Methodsfor Meta-analysis. Academic Press, SanDiego, CA.

Hosmer, D.W. and S. Lemeshow. 1989. AppliedLogistic Regression. John Wiley & Sons.New York.

Hunt, G, R.E. Jackman, T.L. Brown, J.G. Gilardi, D.E.Driscoll, and L. Culp. 1995. A pilot goldeneagle population project in the AltamontPass Wind Resource Area, California. Reportto National Renewable Energy Laboratory,

Golden, CO., Subcontract No. XCG-4-14200 to the Predatory Bird ResearchGroup, University of California, Santa Cruz.212pp.

Hunter, J.E. and F.L. Schmidt. 1990. Methods ofMeta-analysis. Newbury Park: Sage.

Hurlbert, S.H. 1984. “Pseudoreplication and thedesign of ecological field experiments.”Ecol. Monogr. 54:187-211.

James, F.C., and C.E. McCulloch. 1990.“Multivariate analysis in ecology and sys-tematics: Panacea or Pandora’s box?” Ann.Rev. Ecol. Syst. 21:129-166.

Johnson, B., J. Rogers, A. Chu, P. Flyer and R.Dorrier. 1989. Methods for Evaluating theAttainment of Cleanup Standards. Vol. 1.“Soils and solid media.” WESTAT Research,Inc., Rockville, MD. Prepared for U.S.Environmental Protection Agency,Washington, DC. EPA 230/02-89-042.

Johnson, G.D., W.P. Erickson, M.D. Strickland, M.F.Shepherd and D.A. Shepherd. 1997. 1996Avian Monitoring Studies, Buffalo RidgeWind Resource Area, Minnesota. TechnicalReport prepared by WEST, Inc. for NorthernStates Power Co., Minneapolis, MN. 158pp.

Kempthorne, O. 1966. The Design and Analysis ofExperiments, 5th printing. John Wiley andSons, NY.

Krebs, C.J. 1989. Ecological Methodology. Harper &Row, Publishers, NY.

Laake, J.L., S.T. Buckland, D.R. Anderson, and K.P.Burnham. 1993. DISTANCE User’s Guide.Version 2.0. Colorado Cooperative Fish andWildlife Research Unit, Colorado StateUniversity, Fort Collins, CO. 72pp.

Ludwig, J.A., and J.F. Reynolds. 1988. StatisticalEcology: a Primer on Methods andComputing. John Wiley and Sons, NY.

Manly, B.F.J. 1986. Multivariate Statistical Methods:a Primer. Chapman and Hall, London.

Manly, B.F.J. 1991. Randomization and Monte CarloMethods in Biology. Chapman and Hall,London.

Literature Cited 79

Manly, B.F.J. 1992. The design and analysis ofresearch studies. Cambridge UniversityPress, Cambridge, United Kingdom. 353 pp.

Manly, B.F.J., L. McDonald, and D. Thomas. 1993.Resource Selection by Animals - StatisticalDesign and Analysis for Field Studies.Chapman and Hall, London. 177 pp.

McDonald, L.L. 1995. Guidelines for design and statistical analysis of injury quantificationstudies. Prepared for the NationalOceanographic and AtmosphericAdministration, Washington, D.C. 61pp.

McDonald, L.L., and W.P. Erickson. 1994. “Testingfor bioequivalence in field studies: Has adisturbed site been adequately reclaimed?”In Statistics in Ecology and EnvironmentalMonitoring, pp. 183-197 [Eds. D.J. Fletcherand B.F.J. Manly], Otago Conference SeriesNo. 2, University of Otago Press, Dunedin,New Zealand.

McDonald, L.L., W.P. Erickson, B.F.J. Manly, and R.Skinner. 1995. Resource selection functions:guidelines for analysis of multivariate datafrom Geographical Information Systems.Paper presented at TWS Second AnnualConference.

McKinlay, S.M. 1975. “The design and analysis ofthe observational study — A review.”Journal of the American StatisticalAssociation 70:503-518.

Mukhopadhyay, N., R.B. Bendel, N.P. Nikolaidis,and S. Chattopadhyay. 1992. “Efficientsequential sampling strategies for environ-mental monitoring.” Water ResourcesResearch 28:2245-2256.

Muttlak, H.A., and L.L. McDonald. 1992. “Rankedset sampling and the line intercept method:A more efficient procedure.” BiometricalJournal 34:329-346.

National Research Council. 1985. Oil in the Sea:Inputs, Fates, and Effects. National AcademyPress, Washington, D.C. 587 pp.

Neter, J., W. Wasserman, and M.H. Kutner. 1985.Applied linear statistical models - regres-sion, analysis and experimental design.

Richard D. Irwin, Inc. Homewood, Ill.1127pp.

Neufeldt, V. And D.B. Guralnik, Editors. 1988. Thirdcollege edition - Webster’s new world dic-tionary. Webster’s New World, Clevelandand New York. 1574pp.

Orloff and Flannery, op. cit.

Otis, D.L., L.L. McDonald, and M. Evans. 1993.“Parameter estimation in encounter sam-pling surveys.” Journal of WildlifeManagement 57:543-548.

Overton, W.S., D. White, and D.L. Stevens. 1991.Design report for EMAP: Environmentalmonitoring and assessment program.Environmental Research Laboratory, U.S.Environmental Protection Agency, Corvallis,OR. EPA/600/3-91/053.

Page, D.S., E.S. Gilfillan, P.D. Boehm and E.J.Harner. 1993. Shoreline ecology programfor Prince William Sound, Alaska, followingthe Exxon Valdez oil spill: Part 1— Studydesign and methods. DRAFT. ThirdSymposium on Environmental Toxicologyand Risk: Aquatic, Plant, and Terrestrial.American Society for Testing and Materials,Philadelphia, PA.

Pereira, J.M.C and R.M. Itami. 1991. “GIS-basedhabitat modeling using logistic multipleregression: A study of the Mt. Graham RedSquirrel.” Photogrammetric Engineering andRemote Sensing. 57(11): 1475-1486.

Peterle, T.J. 1991. Wildlife Toxicology. Van NostrandReinhold, New York. 322pp.

Peterman, R.M. 1989. “Application of statisticalpower analysis on the Oregon coho salmonproblem.” Canadian Journal of Fish. andAquatic Sci. 46:1183-1187.

Pielou, E.C. 1984. The Interpretation of EcologicalData: a Primer on Classification andOrdination. John Wiley and Sons, NY.

PNAWPPM, op. cit.


Pollock, K.H. 1996. “Assessing avian-wind powerinteractions: Sampling, study design and sta-tistical issues.” Pages 15-24 in Proceedingsof National Avian - Wind Power PlanningMeeting II, Palm Springs, CA, 20-22 Sept.1995. Prepared for the Avian subcommitteeof the National Wind Coordinating Commit-tee by RESOLVE Inc., Washington, D.C.,and LGL Ltd., King City, Ont. 152pp.

Scheaffer, R.L., W. Mendenhall and L. Ott. 1990.Elementary Survey Sampling. PWS-KentPublishing Co., Boston, MA.

Seber, G.A.F. 1984. Multivariate observations. JohnWiley and Sons, NY.

Shenk, T.M., A.B. Franklin, and K.R. Wilson. 1996.“A model to estimate the annual rate ofgolden eagle population change at theAltamont Pass Wind Resource Area.” Pages47-54 in Proceedings of National Avian -Wind Power Planning Meeting II, PalmSprings, CA, 20-22 Sept. 1995. Prepared forthe Avian subcommittee of the NationalWind Coordinating Committee by RESOLVEInc., Washington, D.C., and LGL Ltd., Kingcity, Ont. 152pp.

Skalski, J.R., and D.S. Robson. 1992. Techniques forwildlife investigations: Design and Analysisof capture data. Academic Press, Inc., SanDiego, CA.

Skalski, J.R., M.A. Simmons, and D.S. Robson. 1984.“The use of removal sampling in compara-tive censuses.” Ecology 65:1006-1015.

Smith, W. 1979. “An oil spill sampling strategy.”Pages 355-363 in R.M. Cormack, G.P. Patiland D.S. Robson (Eds.). Sampling BiologicalPopulations. International Co-operativePublishing House, Fairland, MD.

Stekoll, M.S., L. Deysher, R.C. Highsmith, S.M.Saupe, Z. Guo, W.P. Erickson, L. McDonald,and D. Strickland. 1993. “Coastal HabitatInjury Assessment: Intertidal communitiesand the Exxon Valdez oil spill.” Presented atthe Exxon Valdez Oil Spill Symposium,February 2-5, 1993, Anchorage, Alaska.

Stewart-Oaten, A. 1986. The Before-After/Control-Impact-Pairs design for environmentalimpact. Prepared for Marine Review

Committee, Inc., 531 Encinitas Boulevard,Encinitas, CA 92024.

Stewart-Oaten, A., W.W. Murdoch, and K.R. Parker.1986. “Environmental impact assessment:‘Pseudoreplication’ in time?” Ecology67:929-940.

Stokes, S.L. 1986. “Ranked set sampling.” Pages585-588 in S. Kotz and N.L. Johnson (Eds.),C.B. Read (Ex. Ed.). Encyclopedia ofStatistical Sciences Vol. 7. John Wiley andSons, NY.

Strickland et al., op.cit.

Suter, II, G.W., L.W. Barnthouse, S.M. Bartell, T. Mill,D. Mackay, and S. Paterson. 1993.Ecological risk assessment. Lewis Publishers.538pp.

Thompson, S.K. 1992. Sampling. John Wiley andSons, Inc., NY.

Thompson, S.K., and G. A. F. Seber. 1996. Adaptivesampling. John Wiley & Sons, New York.265 pp.

Underwood, A.J. 1994. “On beyond BACI: samplingdesigns that might reliably detect environ-mental disturbances.” EcologicalApplications 4:3-15.

USDI. 1987. Type B technical information docu-ment: Guidance on use of habitat evaluationprocedures and suitability index models forCERCLA application. PB88-100151. U.S.Department of the Interior, CERCLA 301Project, Washington, DC.

U.S. Fish and Wildlife Service. 1980. “HabitatEvaluation Procedures (HEP).” EcologicalServices Manual 102. U.S. Department ofInterior, Fish and Wildlife Service, Divisionof Ecological Services. Government PrintingOffice, Washington, D.C. 84 pp. + appendices.

Ward, D.H., R.A. Stehn, W.P. Erickson, and D.Derksen. 1996. “Response of fall-stagingbrant and other geese to aircraft overflights,at Izembek Lagoon, Alaska.” Submitted toJournal of Wildlife Management, February,1996.

Literature Cited 81

Western EcoSystems Technology (WEST), Inc. 1995.Draft general design, Wyoming windpowermonitoring proposal. Appendix B in DraftKENETECH/ PacifiCorp windpower projectenvironmental impact statement. FES-95-29.Prepared by U.S. Department of the Interior,Bureau of Land Management, Great DivideResource Area, Rawlins, Wyoming andMariah Associates, Inc., Laramie, Wyoming.

Wolter, K.M. 1984. “Investigation of some estimatorsof variance for systematic sampling.” Journalof American Statist. Assoc. 79:781-790.

CHAPTER 4Anderson et al., 1996b, op. cit.

Bart, J. 1995. “Acceptance criteria for using individ-ual-based models to make managementdecisions.” Ecological Applications5:411-420.

Bellrose, F.C. 1980. Ducks, Geese, and Swans ofNorth America. Stackpole Books,Harrisburg, Penn.

Bender, L.C., G.J. Roloff, and J.B. Haufler. 1996.“Evaluating confidence intervals for habitatsuitability models.” Wildlife Society Bulletin24:347-352.

Bowman, T.D., P.F. Schempf, and J.A. Bernatowicz.1995. “Bald eagle survival and populationdynamics in Alaska after the Exxon Valdezoil spill.” Journal of Wildlife Management59:317-324.

Box and Tiao, op. cit.

Boyce, M.S. 1992. “Population viability analysis.”Annual Review of Ecology and Systematics23:481-506.

Buehler, D.A., J.D. Fraser, J.K.D. Seegar, G.D.Therres, and M.A. Byrd. 1991. “Survivalrates and population dynamics of baldeagles on Chesapeake Bay.” Journal ofWildlife Management 55:608-613.

Chu, D.S., J.D. Nichols, J.B. Hestbeck, and J.E.Hines. 1995. “Banding reference areas andsurvival rates of green-winged teal, 1950-89.” Journal of Wildlife Management59:487-498.

Conroy, M.J., G.R. Costanzo, and D.B. Stotts. 1989.“Winter survival of female American blackducks on the Atlantic coast.” Journal ofWildlife Management 53:99-109.

Conway, C.J., S.H. Anderson, D.E Runde, and D.Abbate. 1995. “Effects of experimentalnestling harvest on prairie falcons.” Journalof Wildlife Management 59:311-316.

Cox, op. cit.

Eberhardt, L.L. 1990. “Survival rates required to sus-tain bear populations.” Journal of WildlifeManagement 54:587-590.

Foster, C.C. et al. (7 authors). 1992. “Survival andreproduction of radio-marked adult spottedowls.” Journal of Wildlife Management56:91-95.

Grant, P.R., and B.R. Grant. 1992. “Demography and the genetically effective sizes of twopopulations of Darwin’s finches.” Ecology73:766-784.

Green, op. cit.

Grier, J.W. 1980. “Modeling approaches to baldeagle population dynamics.” WildlifeSociety Bulletin 8:316-322.

Haramis, G.M., D.G. Jorde, and C.M. Bunck. 1993. “Survival of hatching-year female canvasbacks wintering on Chesapeake Bay.” Journal of Wildlife Management57:763-771.

Harris, R.B., and F.W. Allendorf. 1989. “Geneticallyeffective population size of large mammals:an assessment of estimators.” ConservationBiology 3:181-191.

Hill, W.G. 1972. “Effective size of populations withoverlapping generations.” TheoreticalPopulation Biology 3:278-289.

Hohman, W.L., R.D. Pitchert, J.L. Moore, and D.O. Schaeffer. 1993. “Survival of femalecanvasbacks wintering in coastalLouisiana.” Journal of Wildlife Management57:758-762.

Hurlbert, op. cit.


Jenkins, S.H. 1988. “Use and abuse of demographicmodels of population growth.” Bulletin of the Ecological Society of America69:201-207.

Johnson, D.H. 1994. “Population analysis.” Pages419-444 in T. A. Bookhout, ed., Researchand Management Techniques for Wildlifeand Habitats. The Wildlife Society,Bethesda, Maryland.

Kempthorne, op. cit.

Lande, R. 1987. “Extinction thresholds in demo-graphic models of territorial populations.”American Naturalist 130:624-635.

Lande, R. 1988. “Demographic models of the north-ern spotted owl (Strix occidentalis caurina).”Oecologia 75:601-607.

Lande, R. 1995. “Mutations and conservation.”Conservation Biology 9:782-791.

Lefkovitch, L.P. 1965. “The study of populationgrowth in organisms grouped by stages.”Biometrics 21:1-18.

Leslie, P.H. 1945. “On the use of matrices in certainpopulation mathematics.” Biometrika33:183-212.

Lindenmayer, D.B., T.W. Clark, R.C. Lacy, and V.C.Thomas. 1993. “Population viability analy-sis as a tool in wildlife conservation policy:with reference to Australia.” EnvironmentalManagement 17:745-758.

Lynch, M., J. Conery, and R. Burger. 1995. “Mutationaccumulation and the extinction of smallpopulations.” American Naturalist 146:489-518.

Mace, G.M., and R. Lande. 1991. “Assessing extinc-tion threats: toward a reevaluation of IUCNthreatened species categories.”Conservation Biology 5:148-157.

Morrison, et al., op. cit.

National Research Council, op. cit.

Nunney, L. 1992. “Estimating the effective popula-tion size and its importance in conservation

strategies.” Transactions of the WesternSection of The Wildlife Society 28:67-72.

Nunney, L., and D.R. Elam. 1994. “Estimating theeffective population size of conserved popu-lations.” Conservation Biology 8:175-184.

Reed, J.M., P.D. Doerr, and J.R. Walters. 1986.“Determing minimum population sizes forbirds and mammals.” Wildlife SocietyBulletin 14:255-261.

Reid, W.V. 1988. “Population dynamics of the glau-cous-winged gull.” Journal of WildlifeManagement 52:763-770.

Reynolds, R.E., R.J. Blohm, J.D. Nichols, and J.E.Hines. 1995. “Spring-summer survival rates of yearling versus adult mallardfemales.” Journal of Wildlife Management59:691-696.

Shenk, et al., op. cit.

Skalski and Robson, op. cit.

Smith, G.W., and R.E. Reynolds. 1992. “Hunting andmallard survival, 1979-88.” Journal ofWildlife Management. 56:306-316.

Temple, S.A. 1986. “The problem of avian extinc-tions.” Current Ornithology 3:453-485.

Wootton, J.T., and D.A. Bell. 1992. “A metapopula-tion model of the peregrine falcon inCalifornia: viability and management strate-gies.” Ecological Applications 2:307-321.

Wright, S. 1931. “Evolution in Mendelian popula-tions.” Genetics 16:97-159.

Wright, S. 1938. “Size of population and breedingstructure in relation to evolution.” Science87:430-431.

CHAPTER 5Bibby, C.J., N.D. Burgess, and D.A. Hill. 1993. Bird

Census Techniques. Academic Press Inc.257pp.

Bookhout, T.A., Editor. 1994. Research and manage-ment techniques for wildlife and habitats.Fifth ed. The Wildlife Society, Bethesda, Md.740pp.

Literature Cited 83

Buckland et al., op. cit.

Cooper, B.A. 1996. “Use of radar for wind power-related avian research.” Pages 57-72 inProceedings of National Avian - WindPower Planning Meeting II, Palm Springs,CA, 20-22 Sept. 1995. Prepared for theAvian subcommittee of the National WindCoordinating Committee by RESOLVE Inc.,Washington, D.C., and LGL Ltd., King City,Ont. 152pp.

Edwards, D.K., G.L. Dorsey and J.A. Crawford.1981. “A comparison of three avian censusmethods.” Pages 170-176 in C.J. Ralph andJ.M. Scott, Eds. Estimating the numbers ofterrestrial birds. Stud. Avian Biol. 6.

Evans and Mellinger, op. cit.

Evans, W.R. and K.V. Rosenberg. 1999. Acousticmonitoring of night-migrating birds: aprogress report. In Bonney, R., D.N. Pashley,R. Cooper, and L. Niles (eds). “Strategies forBird Conservation: The Partners in FlightPlanning Process.” Cornell Laboratory ofOrnithology. [http://birds.cornell.edu/pifcapemay]

Gauthreaux, op. cit.

Gauthreaux. 1996b. “Bird Migration: methodologiesand major research trajectories (1945-1995).” Condor 98:442-453.

Hawrot, R.Y. and J.M. Hanowski. 1997. Avianassessment document: avian populationanalysis for wind power generation regions--012. NRRI Technical Report No. NRRI/TR-97-23, Center for Water and theEnvironment, Natural Resources ResearchInstitute, Duluth, MN. 14pp.

Howell, J.A. 1997. Avian mortality at rotor sweptarea equivalents, Altamont Pass andMontezuma Hills, California. Transactions ofthe Western Section of The Wildlife Society33: 24-29.

Mayer, L.S. 1996. “The use of epidemiological meas-ures to estimate the effects of adverse factorsand preventive interventions.” Pages 26-39in Proceedings of National Avian - WindPower Planning Meeting II, Palm Springs,CA, 20-22 Sept. 1995. Prepared for the

Avian subcommittee of the National WindCoordinating Committee by RESOLVE Inc.,Washington, D.C., and LGL Ltd., King City,Ont. 152pp.

McDonald, D.B., and H. Caswell. 1993. “Matrixmethods for avian demography.” CurrentOrnithology 10:139-185.

Orloff and Flannery, op. cit.

Ralph, C.J., G.R. Geupel, P. Pyle, T.E. Martin, andD.F. DeSante. 1993. Handbook of fieldmethods for monitoring landbirds. PacificSouthwest Res. Station, Albany, Calif. Gen.Tech. Report PSW-GTR-144. 41pp.

Reynolds, R.T., J.M. Scott, R.A. Nussbaum. 1980. Avariable circular-plot method for estimatingbird numbers. Condor 82(3):309-313.

Salmon, T.P. 1981. Controlling ground squirrelsaround structures, gardens, and small farms.University of California Division ofAgricultural Science Leaflet 21179. 11pp.

Smallwood, K.S., J. Beyea, and M.L. Morrison. 1998.“Animal burrowing attributes affecting haz-ardous waste management.” EnvironmentalManagement 22: 831-847.

Tucker, V.A. 1996. Using a collision model to designsafer wind turbine rotors for birds. Journal ofSolar Energy Engineering 118:263-269.


additive, 39, 53, 55

Altamont Pass WRA, 6, 7, 16, 23

analysis of variance (ANOVA), 28, 29, 42, 46,

area-by-time interaction, 29

assessment area, ix, 24, 25-31, 36, 38, 39, 47

Before-After Design, 27, 28, 49

Before-After/Control (reference)-Impact (BACI), 3,10, 25 - 29, 37, 41, 43, 47, 49, 52

biomarker, 27

case study approach, 75

collision, 2, 3,12, 13, 15, 16, 40, 43, 51, 52, 67,73

compensatory, 39, 53, 55, 64

control (standardization) of related variables, 3, 19,21, 25, 26, 41, 44-47, 51, 73, 78

cumulative effects, 3, 9, 10, 39, 41, 51, 63

dead bird searches (or counts), 27, 36, 39, 44, 45,66 69, 70, 72, 74, 75

demography, 53, 57, 58

design/data-based analysis, 23, 25, 30, 34

detrended canonical correspondence analysis, 37

effective population size, 53, 58, 59, 62, 64, 82, 83

environmental stochasticity, 53, 54, 61, 62

epidemiology, 74

experimental units, 3, 19-27, 51, 74, 75, 76

genetics, 3, 53, 54, 58, 62, 83

gray literature, 13

habitat loss, 4, 55, 62, 64

haphazard, 31, 32

Impact-Gradient Design, 26, 27, 30, 49

Impact-Reference Design, 26, 29, 37, 41

independent data, 20, 21

judgment, 23, 28, 31, 32, 49, 51, 74

lambda, 53, 54, 56, 57, 60, 61

Leslie matrix models, 56, 64

Level 1 studies, 2, 3, 9, 18, 24, 25, 38, 43, 49, 68,69,

Level 2 studies, 2, 3, 9, 24, 25, 39, 41, 43, 50, 51,52, 69

life history parameters, 4, 53, 54, 64

life tables, 56

line transect, 22

long-term studies, 3, 26, 33, 48

Lotka models, 56

measuring, 4, 19, 35, 39, 43-45, 67, 74

metapopulation, 3, 53, 54, 83

methods, 1-4, 6, 7, 9, 11, 12, 14-16, 19, 22, 23, 25,29, 31, 33, 36, 37, 40, 42, 43, 45-48, 51, 58, 64-68,79, 80, 84

metrics, 1, 2, 6, 7, 9, 11, 14, 16, 38, 44, 69, 83

micrositing, 71, 73

minimum viable population, 61

model-based, 9, 23, 25, 30, 31, 34, 38, 40, 51, 55,57, 74

Index of key Terms 85

APPENDIX B:Index of Key Terms

mortality, 1, 4, 9, 10, 20, 27, 36, 38-45, 47, 49, 52,53, 55, 56, 60, 64, 65-67, 69-75, 77, 78, 84

multivariate analysis, 37, 79

nest sites, 72, 73

outcome variable, 4, 66, 70

perches, 66, 72, 73

perching, 10, 21, 44, 67, 69, 72

plant operation, 71

plant siting, 73

point-to-item, 22

population, 3, 4, 9, 10, 16, 19, 20-25, 30-33, 38-41,43, 51, 53-64, 66-69, 72-74, 77, 78, 81-83

population viability analysis (PVA), 61, 82

preventable fraction, 66

prevented fraction, 66

prey, 4, 22, 24, 28, 30, 33, 40, 62, 63, 65, 70-73

principal component analysis, 37

probability, 2, 13, 22, 31-39, 49, 53, 66, 70

proportional mortality, 75

pseudoreplication, 21, 42, 43, 79, 81

random, 2, 15, 19-26, 30, 32-36, 38, 45, 46, 49, 51,53, 54, 58

Reconnaissance studies, 2, 13

reference area, 3, 22, 25-33, 35-37, 39-41, 43, 47,49, 51, 61, 66, 68, 71, 75, 82

relationships, 2, 9, 12, 13, 28, 35, 37, 51, 55, 59, 78

replication, 19-21, 26, 28, 42, 51, 52, 73, 76, 79, 81

risk - attributable, 39, 43, 44, 53, 66, 70-73,

risk ratio, 71, 73

risk reduction, 2-4, 9, 16, 24, 51-53, 70, 72

rotor swept area, 42, 43, 45, 51, 67, 69, 73, 74, 77,84

rotor swept hour, 45, 74

rotor swept hour risk, 45, 74

sampling, 2, 3, 15, 21-23, 26, 28-37, 41, 43, 45, 47-49, 52, 54, 57, 68, 75, 76, 78-82

sampling frequency, 76

stage-based models, 57

stratification, 21, 22, 26, 32-36, 76

stratified, 25, 32-34, 38, 44, 49, 76

study design, 1-4, 6, 7, 9-11, 16, 20, 21, 23-25, 28,31, 37, 39, 40, 43, 48, 52, 63, 65, 69, 70, 73-75,80, 81

surrogate variable, 70

survivorship, 3, 4, 51, 54, 56-61, 64

systematic, 3, 21, 32, 34, 36, 37, 48, 49, 63, 82

univariate analysis, 3, 36, 48

utilization, 4, 9, 15, 43-46, 65, 66, 69, 70, 73, 74

weight of evidence, 2, 3, 39, 40, 47, 48


Leif AndersenNEG Micon USA, Inc.

Dick Anderson*California Energy Commission

Larry Bean*Iowa Department of NaturalResources

Eric Blank Land & Water Fund of the Rockies

R.T. "Hap" BoydEnron Wind Corporation

Matthew Brown*National Conference of StateLegislatures

Jack CadoganUS Department of Energy WindEnergy Program

Kim ChristiansonNorth Dakota Division ofCommunity Services

Steven L. Clemmer*Union of Concerned Scientists

Steve CorneliMinnesota Attorney General's Office

Lisa DanielsWindustry Project

Ed DeMeo*Renewable Energy ConsultingServices, Inc.

Stephen DrydenFPL Energy, Inc.

Steve Ellenbecker*Wyoming Public ServiceCommission

Sam Enfield FPL Energy, Inc.

Chris Flavin*Worldwatch Institute

Larry FlowersNational Renewable Energy Lab

Larry Frimmerman*Ohio Consumer Counsel

Peter Goldman*US Department of Energy Wind Energy Program

Dean GosselinFPL Energy, Inc.

Bob GoughInter-Tribal Council on Utility Policy

Thomas GrayAmerican Wind Energy Association

Rick HaletNorthern States Power

Roger HamiltonOregon Public Utilities Commission

Ed HoltEd Holt & Associates

Karen LaneUtility Wind Interest Group

Ron Lehr*National Association of RegulatoryUtility Commissioners

Chuck Linderman*Edison Electric Institute

Annunciato MarinoPennsylvania Public UtilitiesCommission

Ward MarshallCentral & South West Services

Garry MauroTexas General Land Office

Rudd MayerLand & Water Fund of the Rockies

Chuck McGowinElectric Power Research Institute

Mark P. McGree*Northern States Power Company

Robert MillerFarmer

Pam NelsonSouth Dakota Public UtilitiesCommission

Jim NicholsLincoln County EconomicDevelopment Department, MN

John F. Nunley IIIWyoming Business Council, EnergyOffice

Brian ParsonsNational Renewable Energy Lab

Steve PonderFPL Energy, Inc.

Kevin PorterNational Renewable Energy Lab

Jane PulaskiTexas State Energy ConservationOffice

Thomas H. RawlsGreen Mountain Energy Resources

Leda RoselleTexas General Land Office

Richard SedanoVermont Department of PublicService

Tom SloanKansas State Representative

Charlie SmithElectrotek Concepts Inc.

Pat SpearsInter-Tribal Council On Utility Policy

Randy Swisher*American Wind Energy Association

Michael Tennis*ReGen Technologies/All Energy

Steve WegmanSouth Dakota Govenor's Office

Carol WernerEnvironmental and Energy StudyInstitute

Steve Wiese Planergy, Inc.

Jean WilsonPacifiCorp

National Wind Coordinating Committee Members and Alternates 87

APPENDIX C:National Wind Coordinating Committee Members and

Key Participants*= Steering Committee Members

The NWCC is a collaborative endeavor formed in 1994 that includes representatives from electric utilities and their support organizations, state legislatures, state utility commissions,consumer advocacy offices, wind manufacturers and developers, power marketers, environ-mental organizations, and local, regional, state, tribal and federal agencies. The National WindCoordinating Committee identifies issues that affect the use of wind power, establishes dialogueamong key stakeholders, and catalyzes appropriate activities to support the development of anenvironmentally, economically, and politically sustainable market for wind power.

For additional information or to schedule a wind-avian or siting workshop, please contact:

Senior Outreach Coordinator Phone: 202-965-6398 or 888-764-WINDNational Wind Coordinating Committee Fax: 202-338-1264c/o RESOLVE E-Mail: [email protected] 23rd Street NW, Suite 275Washington, DC 20037

This complete document is available on NWCC’s website: http://www.nationalwind.org

NWCC members include representatives from:

American Wind Energy Association

California Energy Commission

Central and South West Services

Lincoln County Economic Development

Ed Holt & Associates

Edison Electric Institute

Electric Power Research Institute

FPL Energy, Inc.

Greenmountain.com

Hawkeye Power Partners

Inter-Tribal Council on Utility Policy

Iowa Department of Natural Resources

Iowa State Legislature

Kansas State Legislature

Land & Water Fund of the Rockies

Minnesota Attorney General’s Office

National Association of Regulatory Utility Commissioners

National Association of State Energy Officials

National Conference of State Legislatures

NEG Micon USA, Inc.

North Dakota Division of Community Services,Energy Program

Northern States Power Company

Ohio Consumer Counsel

Oregon Public Utilities Commission

PacifiCorp

Pennsylvania Public Utilities Commission

Planergy

ReGen Technologies/All Energy

South Dakota Public Utilities Commission

Windustry Project

Texas General Land Office

Texas State Energy Conservation Office

Union of Concerned Scientists

U.S. Department of Energy, Wind Program

Utility Wind Interest Group

Vermont Department of Public Service

Worldwatch Institute

Wyoming Business Council, Energy Office

Wyoming Public Service Commission

Documents

Studying Wind Energy/Bird Interactions: A Guidance Document · This document is organized in two parts. Part I (chapters 1-2) presents the general reader with a framework for considering