Interpretation of landscape pattern and habitat change for ... et al 2003 CJFR.pdf · et al. 2003), American martens (Martes americana) (Chapin et al. 1998; Hargis et al. 1999), and

Interpretation of landscape pattern and habitatchange for local indicator species using satelliteimagery and geographic information system datain New Brunswick, Canada

Matthew G. Betts, Steven E. Franklin, and Ron G. Taylor

Abstract: We measured the extent and rate of habitat change and interpreted landscape metrics for fragmentation inthe Fundy Model Forest, New Brunswick, from 1993 to 1999 using geographical information system baseline data up-dated with landscape changes detected on Landsat satellite imagery. We report on three categories of landscape metrics(habitat cover, patch size, and nearest neighbour), which we interpret as applicable to potentially fragmentation-sensitive local indicator species in specific habitat types. Between 1993 and 1999, 5.6% of forest land in the FundyModel Forest was estimated by satellite image analysis as having >30% of canopy cover removed, primarily as a resultof forest harvesting treatments. In four of five habitat types, the rate of habitat loss from harvesting outpaced habitatreplacement due to forest growth. Changes in landscape pattern metrics indicate that fragmentation has occurred ineach of the five indicator species habitat types over the available time period; furthermore, the rate of fragmentationexceeded the rate of habitat loss. Declines in the number and area of mixedwood patches dominated the fragmentationof the landscape in this region. More attention to the spatial distribution of harvesting activities may be necessary tochange this trend in landscape pattern in the future.

Résumé : Nous avons mesuré l’étendue et le taux de changement des habitats et interprété diverses mesures de frag-mentation à l’échelle du paysage à la Forêt modèle de Fundy, au Nouveau-Brunswick, de 1993 à 1999. L’étude a étéréalisée à l’aide d’un système d’information géographique dont les données des conditions de bases ont été mises àjour selon les changements détectés dans le paysage grâce à des images satellitaires Landsat. Nous présentons trois ca-tégories de mesures du paysage (superficie des habitats, taille des îlots et plus proches voisins), que nous jugeons perti-nentes pour des espèces indicatrices locales potentiellement sensibles à la fragmentation. De 1993 à 1999, nousestimons que 5,6 % des terrains forestiers à la Forêt modèle de Fundy ont perdu >30 % de leur couvert arborescent,principalement en raison de l’exploitation forestière. Dans quatre des cinq types d’habitats, le taux de perte des habitatsdû à l’exploitation excédait le taux de remplacement dû à l’accroissement de la forêt. L’évolution temporelle des mesu-res du paysage indique que la fragmentation s’est produite dans chacun des cinq types d’habitats durant la périodeconsidérée; par ailleurs, le taux de fragmentation dépassait le taux de perte des habitats. La réduction du nombre et dela superficie des îlots de forêt mixte domine la fragmentation du paysage dans cette région. À l’avenir, il faudra possi-blement accorder plus d’attention à la répartition spatiale des activités de récolte pour modifier cette tendance dansl’évolution de la structure du paysage.

[Traduit par la Rédaction] Betts et al. 1831

Introduction

Measurement and interpretation of landscape pattern orstructure and change over large areas are a research chal-lenge that has emerged in recent decades with the accumula-

tion of scientific literature supporting the hypothesis thatlandscape-scale biodiversity is negatively influenced byanthropogenic changes to landscape pattern (Andren 1994;Paton 1994; Beier and Noss 1998; Bender et al. 1998;Mazerolle and Villard 1999). Quantifying landscape patternand composition with landscape metrics has several advan-tages, including the quantitative documentation of trends inlandscape development and interpretation of landscape frag-mentation (Zheng et al. 1997; Kitzberger and Veblen 1999).A wide range of landscape metrics has been used to explainthe distribution and abundance of species (Mazerolle andVillard 1999; Saveraid et al. 2001), and specific questionsand hypotheses have been tested about the population healthof potentially fragmentation-sensitive species (Andren 1994;Miller et al. 1997; Dijak and Thompson 2000). For example,studies of timber wolves (Canis lupus lycaon) (Mladenoff etal. 1995), grizzly bears (Ursus arcturus horribilus) (Nielsen

Can. J. For. Res. 33: 1821–1831 (2003) doi: 10.1139/X03-104 © 2003 NRC Canada

1821

Received 20 January 2003. Accepted 11 April 2003.Published on the NRC Research Press Web site athttp://cjfr.nrc.ca on 29 September 2003.

M.G. Betts.1 Greater Fundy Ecosystem Research Group, P.O.Box 45111, University of New Brunswick, Fredericton,NB E3C 2G6, Canada.S.E. Franklin. Department of Geography, University ofCalgary, Calgary, AB T2N 1N4, Canada.R.G. Taylor. Nova Forest Alliance, Box 208, George Street,Stewiake, NS B0N 2S0, Canada.

1Corresponding author (e-mail: [email protected])

I:\cjfr\cjfr3310\X03-104.vpSeptember 19, 2003 5:07:15 PM

Color profile: Generic CMYK printer profileComposite Default screen

et al. 2003), American martens (Martes americana) (Chapinet al. 1998; Hargis et al. 1999), and songbirds (Lichstein etal. 2002) have adopted multivariate techniques, such as re-gression, to develop predictive models of the influence oflandscape structure on wildlife populations.

A number of issues must be addressed when using land-scape metrics in quantifying landscape structure and docu-menting change over time for use in wildlife models ormanagement plans. Critical determinants of the quality andeffectiveness of landscape quantification are the scale, ori-gin, and reliability (or accuracy) of the input data (Wickhamet al. 1997; Franklin et al. 2000a; Shao et al. 2001). Manygeographical information system (GIS) databases have onlypartial coverage of multijurisdictional areas; such data areoften highly dependent on manual stratification or interpre-tation of mapping units on aerial photography, which is ex-pensive to produce frequently. GIS databases from differenttime periods may contain inconsistent forest cover attributesmaking cross-year comparisons untenable. Some digital re-mote sensing image sources can be unreliable because ofcloud coverage, seasonal differences, or sensor incompatibil-ities (Woodcock et al. 2001). However, despite these short-comings, if used in combination, existing GIS and remotesensing data may represent the best available data for com-parison of change over large areas and long time periods(Franklin et al. 2002c).

Equally important are issues associated with the specificecological implications and interpretation of landscape met-rics and the change in metrics over time (Krummel et al.1987; Baskent and Jordan 1995; Hulshoff 1995; Sachs et al.1998; Baskent 1999; Wang and Moskovits 2001). Some met-rics are not necessarily relevant to species or ecological pro-cesses (Cale and Hobbs 1994). Expert opinions on “optimalranges” of landscape pattern appear to vary broadly(Lundquist et al. 2001). These realities leave forest managerswith little guidance when interpreting ecological implica-tions of landscape metrics and landscape change research.

In this study in the Fundy Model Forest (FMF), NewBrunswick, the extent and rate of habitat loss were measuredand landscape fragmentation interpreted from 1993 to 1999using available GIS data and Landsat satellite imagery.Landscape metrics were computed for two time periods froman existing forest inventory GIS database updated withchanges detected in the imagery. The methods for thischange detection approach are described in the remote sens-ing literature by Collins and Woodcock (1996), Cohen et al.(1998), Cohen and Fiorella (1999), and Franklin et al. (2001,2002a, 2002b, 2002c) and are briefly summarized here. Ourobjective was to examine the metrics and forest changeswith reference to potentially fragmentation-sensitive localindicator species. By “grounding” measurement of landscapechange in hypothesized stand-level and landscape-levelspecies–habitat relationships, we have attempted to highlightthe ecological relevancy of the change observed in a selec-tion of landscape metrics.

Study area

The area of the FMF (4500 km2) extends north of the Bayof Fundy in New Brunswick, Canada (66.08°W–64.96°W,46.08°N–45.47°N). Landownership in the FMF is 63% small

private woodlots, 17% large private holdings (J.D. Irvingfreehold land), 15% provincial Crown (public) land, and 5%national park. All of the FMF lies within the Acadian ForestRegion (Rowe 1972). The FMF area is characterized by89% forest cover, a maritime climate, and rolling topogra-phy (Woodley 1998). The forest cover is primarily intoleranthardwood (white birch (Betula papyrifera Marsh.), bigtoothaspen (Populus grandidentata Michx.), trembling aspen(Populus tremuloides Michx.), and red maple (Acerrubrum L.)), tolerant hardwood (American beech (Fagusgrandifolia Ehrh.), sugar maple (Acer saccharum Marsh.),and yellow birch (Betula alleghaniensis Britt.)), andmixedwood communities (either tolerant hardwoods or intol-erant hardwoods combined with red spruce (Picea rubensSarg.) and balsam fir (Abies balsamea (L.) Mill.)). However,pure softwood communities (red spruce, balsam fir, whitespruce (Picea glauca (Moench) Voss), or black spruce(Picea mariana (Mill.) BSP)) exist in low-lying areas andalong the Bay of Fundy coast. Pure pine stands (easternwhite pine (Pinus strobus L.) or jack pine (Pinus banksianaLamb.)) are uncommon but exist primarily in postburn areasof the FMF. Intensive forestry activities are common in allareas of the FMF except for Fundy National Park.

Methods

Habitat mappingFive major habitat types for New Brunswick were identi-

fied for this study based on relationships with local indicatorspecies defined by the New Brunswick Department of Natu-ral Resources and Energy (NBDNRE) (Table 1): old toleranthardwood habitat (OTHH), old hardwood habitat (OHWH)(made up of intolerant hardwood tree species listed above),old mixedwood habitat (OMWH), old spruce–fir habitat(OSFH), and old pine habitat (OPIH) (Table 2). To representthe five habitat conditions, different habitat maps were de-veloped using data obtained from the forest inventory GISdatabase for 1993. This spatial forest inventory comprisesdigitized interpreted 1 : 12 000 aerial photographs. The as-sumption underlying the maps is that habitat-related standstuctures (e.g., number of cavity trees, number and size ofsnags, presence of shrub layer), which are not directly mea-sured in the available GIS data or by satellite image analy-sis, can be inferred from the image data or polygonattributes such as age-class and species composition. Thisassumption appears to hold for coarsely defined habitattypes (Dussault et al. 2001).

First, to represent the baseline year of 1993, we extractedpolygon attributes from the existing GIS forest inventory da-tabase. We selected 1993 as the baseline year in this proce-dure because 1993 was the most recent year for whichcomplete and comprehensive land cover and associated for-est structure data were available. To categorize detailedstand conditions contained in the GIS into the five generalhabitat categories, we applied a “forest community group”algorithm (Table 1). Simple logical decision rules based onthe structural conditions, percent composition of the mostcommon tree species in forest stands (NBDNRE 2000), andmaturity classes were used to aggregate polygonal attributesinto the generalized habitat classes. For example, if the pri-mary and secondary species in a stand were hardwood spe-

© 2003 NRC Canada

1822 Can. J. For. Res. Vol. 33, 2003



cies and softwood species, respectively, the stand would beassigned to the “mixedwood” habitat category.

Second, August cloud-free Landsat-5 TM or Landsat-7ETM+ satellite images were obtained for 1992, 1995, 1997,1998, and 1999. These images were processed digitally inthe following sequence to produce a change detection maplayer (Franklin et al. 2001, 2002b).

Geometric correctionEach image was registered to the N.B. Stereographic Pro-

jection employed in the GIS inventory data. More than 40ground control points located by GPS survey on road inter-sections and distinct natural features were used to develop athird-order polynomial transformation with less than 0.5pixel root mean square error.

Atmospheric correctionTo correct measured radiances for illumination differences

and atmospheric conditions on each image date, a model-based atmospheric algorithm was applied (Richter 1990).First, a standard atmosphere was estimated based on visibil-ity conditions in each image. Second, a set of pseudo-invariant features was selected; radiances were then matchedto actual reflectance characteristics under each image solarillumination conditions. Finally, the image radiance valueswere converted to estimated surface reflectances based onthese training data and selected best-fit coefficients.

Tasseled Cap transformationEach Landsat image was transformed into brightness–

greenness–wetness indices using the Tasseled Cap transfor-mation procedure for reflectance data described by Crist(1985). This procedure was suggested by Collins and Wood-cock (1996) as an ideal approach when mapping distinct for-est canopy changes or mortality with Landsat imagery. Thedifference between the multitemporal wetness indices (alsoknown as the enhanced wetness difference index) has beenshown in earlier work in New Brunswick to discriminatesilvicultural and partial harvest canopy disturbances with aper-pixel accuracy of approximately 71% (Franklin et al.2000b); subsequent polygon-based accuracy assessment hasindicated no error of omission (Franklin et al. 2002a).

Final change layerThe final satellite image map was overlayed with clearcuts

on both Crown land and J.D. Irving freehold land mappedwith the use of global positioning systems. Clearcuts onother landownerships were detected on each of the imagedates with the enhanced wetness difference index procedure;all such changes were then compiled in a single map layerrepresenting all forest cover changes that occurred between1993 and 1999. We masked agricultural and other nonforestchanges with simple GIS-based rules.

We accounted for forest growth and the consequent devel-opment of new habitat in 1999 by including as habitat allstands that were defined as “young” in the 1993 New Bruns-wick forest development survey (NBDNRE 1986). This is anoverestimate of the amount of “new” habitat because theyoung age-class extends for 10–20 years, whereas ourchange detection only extended 7 years; therefore, it is prob-able that the majority of young stands remained in this age

© 2003 NRC Canada

Betts et al. 1823

Hab

itat

type

Str

uctu

ral

feat

ures

defi

ned

byN

BD

NR

E*

For

est

com

mun

ity

grou

pM

atur

ity

clas

ses

Mai

ncr

iter

ia

Old

hard

woo

dC

row

ncl

osur

e20

–80%

,sn

ags

>10

cm,

shru

bla

yer,

mid

stor

yla

yer

Har

dwoo

dO

verm

atur

e,m

atur

e,im

mat

ure

Prim

ary

and

seco

ndar

ysp

e-ci

esar

eha

rdw

ood

Old

spru

ce–f

irC

row

ncl

osur

e>

40%

,ca

vity

tree

s>

40cm

,sn

ags

>30

cm,

live

tree

s>

30cm

,sh

rub

laye

r,w

oody

debr

isB

alsa

mfi

r,bl

ack

spru

ce,

othe

rso

ft-

woo

d,pu

resp

ruce

,sp

ruce

–fir

,sp

ruce

Ove

rmat

ure,

mat

ure,

imm

atur

ePr

imar

yan

dse

cond

ary

spec

ies

are

soft

woo

dO

ldm

ixed

woo

dC

avit

ytr

ees

>30

cm,

crow

ncl

osur

e>

40%

Bal

sam

fir,

blac

ksp

ruce

,ot

her

soft

woo

d,pu

resp

ruce

,sp

ruce

–fir

,†sp

ruce

,†

hard

woo

d,†

tole

rant

hard

woo

d†

Ove

rmat

ure,

mat

ure,

imm

atur

eP

rim

ary

spec

ies

isha

rd-

woo

dan

dse

cond

ary

spec

ies

isso

ftw

ood

Old

pine

Cro

wn

clos

ure

>40

%Ja

ckpi

ne,

pine

Ove

rmat

ure,

mat

ure,

imm

atur

ePr

imar

ysp

ecie

sis

pine

Old

tole

rant

hard

woo

dC

avity

tree

s>

45cm

,cr

own

clos

ure

>40

%,

snag

s>

10cm

Tol

eran

tha

rdw

ood

pure

Ove

rmat

ure,

mat

ure,

imm

atur

ePr

imar

yan

dse

cond

ary

spe-

cies

are

tole

rant

hard

woo

d

*New

Bru

nsw

ick

Dep

artm

ent

ofN

atur

alR

esou

rces

and

Ene

rgy.

† Prim

ary

spec

ies

isso

ftw

ood

and

seco

ndar

ysp

ecie

sis

hard

woo

d.

Tab

le1.

GIS

fore

stco

mm

unit

ygr

oup

deci

sion

rule

sus

edto

deve

lop

the

base

line

habi

tat

map

for

1993

.



© 2003 NRC Canada

1824 Can. J. For. Res. Vol. 33, 2003

category until 1999. However, in the absence of spatially ex-plicit forest growth models for all landownerships, thismethod provided a coarse estimate of habitat replacementover this period. Because habitat change estimates based onthis approach are thought to be conservative, we have re-ported both (i) habitat changes incorporating potential forestgrowth (reported as 1999a) and (ii) habitat changes with noinclusion of young age-classes (reported as 1999). Thesedata can be considered as “confidence intervals” withinwhich actual habitat change and forest fragmentation levelsactually lie.

Landscape metricsNo single metric is capable of reflecting the diversity of

landscape composition and pattern (Davidson 1998). Whilemany metrics have been used to interpret landscape frag-mentation, we sought to analyze those most likely to be rele-vant to native biodiversity. Total proportion of suitablehabitat, edge effects, and patch size are some of the mostfrequently cited as ecologically important landscape metrics(Paton 1994; Bender et al. 1998). Configuration metrics(e.g., connectivity, isolation, and contagion) have also beenreported as significant (Beier and Noss 1998). Riitters et al.(1995) used factor analysis on 26 metrics computed on 86land cover maps and found that 87% of the metric variationwas explained by the first six factors (interpreted as compos-ites of landscape area, patch shape and distribution, and mapclass complexity). In another study, Popplewell et al. (2003)statistically analyzed a suite of metrics and reported thatfour (edge density, mean patch size, mean nearest neighbour,patch size covariance) were most strongly related to relativegrizzly bear population densities; they suggested that the se-lection of landscape metrics be guided by the need to cap-ture total landscape area, diversity of patch types, andvarious aspects of the patches themselves (such as their size,shape, variability in size and shape, distance, and isolation).Based on these suggestions, we selected three categories ofmetrics for analysis of landscape change in the FMF:(i) habitat area is the amount of habitat within a landscapethat meets the requirements of particular indicator species,(ii) patch size metrics relate the maximum, minimum, mean,and frequency distribution of contiguous habitat patcheswithin a landscape, and (iii) landscape configuration is rep-resented by measures of the proximity of habitat patches toone another, such as the mean proximity and frequency dis-tribution of nearest forest patches for the five habitat types.

We used the published spatial habitat criteria of indicatorspecies for patch size and interpatch distance in the interpre-

tation of the metrics (Table 2) (NBDNRE 2000). For exam-ple, minimum interpatch distance criteria were based onlikely dispersal distances for the indicator species repre-sented by each habitat type in Table 1. Minimum patch sizecriteria were based on (i) home range size and (ii) assumedpatch size for maintenance of viable populations (Beaudette2000). These critieria were developed as part of a forest andwildlife management indicator species approach that as-sumed that if the habitat requirements of a set of mature for-est specialists were met, populations of other more generalistspecies would also be maintained. To generate landscapemetrics, we used Patch Analyst 2.0 (Rempel 1999; Elkie etal. 1999), an extension of ArcView spatial analyst (ESRI,Redlands, Calif.). Nearest neighbour distances were calcu-lated as the distance between each patch and the nearestpatch of the same habitat type. All distances reported are thedistance from edge to edge of habitat patches.

Results

Between 1993 and 1999, 20 450 ha of forest land was de-termined by satellite image analysis as representing changedforest conditions (Fig. 1). This area represents 4.5% of thetotal land area and 5.6% of the forested land in the FMF. In1993, 7423 ha (2.0%) of forest was classified as “young”, aproportion of which will have aged into old forest habitatcategories. Of the five habitat types examined, mixedwoodand spruce–fir habitat types appear to be under the most se-rious harvesting pressure (Table 3). These two habitat typeshave been reduced by nearly 1% per year. Further, the ma-jority (63%) of new mixedwood habitat contains shade-intolerant species. Mixedwood habitat made up of shade-tolerant tree species declined by 9.6% over 7 years. In allcases, as of 1999, the total amount of habitat in each cate-gory constituted less than 20% of the FMF landscape.

Mean patch size declined for all habitat types except pine(Table 4). Hardwood habitat mean patch sizes decreased themost markedly. The number of “large” habitat patches (ac-cording to NBDNRE patch size criteria) also decreased forhardwood, tolerant hardwood, and mixedwood, indicatingthat the decline in mean patch size was likely not simply dueto the splitting of small patches (Table 5; Fig. 2). Mixed-wood habitat patches have been the most heavily influencedby the changes observed. Over the 7-year period, nine of121 patches of mixedwood greater than 60 ha were removedor reduced in size (a reduction of 11.6% in total largemixedwood patch area).

Habitat type Indicator species Spatial requirements

Old tolerant hardwood Barred owl (Strix varia) Patch >20 ha, interpatch distance >3 kmWhite-breasted nuthatch (Sitta carolinensis) Patch >40 ha, interpatch distance <1 km

Old hardwood Northern goshawk (Accipiter gentiles) Patch >20 ha, interpatch distance >3 kmHairy woodpecker (Picoides villosus) Patch >30 ha, interpatch distance <1 km

Old spruce–fir American marten (Martes americana) Patch >375 ha, width >1000 mOld pine Pine warbler (Dendroica pinus) Patch >15 ha, interpatch distance <1 kmOld mixedwood Northern flying squirrel (Glaucomys sabrinus) Patch >60 ha, interpatch distance <1 km

Table 2. Spatial requirements and indicator species of five New Brunswick Department of Natural Resources and Energyhabitat types used in fragmentation analysis.



The patch size distribution for all mature habitat revealsthat large patches have declined in number (Fig. 3), and, inmost cases, area (Fig. 4) since 1993. Small patches (<5 ha)increased in both number and area. This may be largely a re-sult of the splitting of large patches by new roads and cuts.While a small percentage of the total number of patches arelarge, they still constitute a substantial part of the FMF landbase. As of 1999, 42% of mature forest habitat existed inpatches that were large enough to meet the requirements ofindicator species. However, this is a decrease from the 1993level of 44%.

© 2003 NRC Canada

Betts et al. 1825

Fig. 1. Spatial distribution of landscape change in the Fundy Model Forest as detected in satellite image analysis.

Habitat type 1993 1999 1999a% of landscape,1999a

% change,1999 (1999a)

Old hardwood 76 217 71 019 73 448 15.8 –6.8 (–3.6)Old mixedwood 60 201 55 478 57 278 11.7 –7.9 (–4.9)Old pine 3 917 3 539 4 437 1.3 –9.6 (11.7)Old spruce–fir 49 927 45 825 47 104 10.2 –8.21 (–5.6)Old tolerant hardwood 29 466 27 950 28 756 6.2 –5.14 (–2.4)Total 219 728 203 811 211 023 45.3 –7.24 (–3.9)

Note: 1999a data incorporate potential forest growth.

Table 3. Change in habitat area (ha), 1993–1999.

Habitat type 1993 1999 1999a% change, 1999(1999a)

Old hardwood 22.6 15.8 16.3 –30 (–28.2)Old mixedwood 13.5 9.5 9.6 –29.6 (–28.7)Old pine 6.5 6.5 7 0 (7.1)Old spruce–fir 13.0 8.7 8.71 –33.3 (–33.0)Old tolerant hardwood 23.3 18.1 18.4 –22.3 (–21.0)


Table 4. Change in mean patch size (ha), 1993–1999.



Results of mean nearest neighbour statistics indicate thatthe distance between patches decreased from 1993 (191.3 m)to 1999 (187.3 m). Histograms of nearest neighbour dis-tances for patches in each habitat category reveal that themajority of patches fell into the most proximal category (0–500 m) (Fig. 5). This is well within the 1000-m nearestneighbour distance identified for all NBDNRE spatial indi-cator species. However, since 1993 the number of patches inthe 0–500 m category declined for all habitat types except

spruce–fir and pine habitat. In mixedwood, hardwood, toler-ant hardwood, and pine habitats, large percentages of habitatwere beyond NBDNRE’s suggested minimum neighbourdistances for indicator species (OMWH: 32.9%, OPIH:47.9%, OTHH: 18.6%). Hardwood (OHWH) can be inter-preted as the least fragmented of the habitat types with only10.5% in nearest neighbour categories greater than 1000 m.While no substantial differences existed in amount of land-scape change on land managed by three major landowner

© 2003 NRC Canada

1826 Can. J. For. Res. Vol. 33, 2003

Habitat type

NBDNRE spatialobjective for patchsize (ha)

No. ofpatches,1993

Area, 1993(ha)

No. ofpatches, 1999(1999a)

Area, 1999(1999a) (ha)

% change inarea, 1999(1999a)

Old hardwood 30 509 52 242 477 (491) 46 069 (48 384) –11.8 (–7.3)Old mixedwood 60 121 14 858 104 (112) 12 281 (13 130) –17.3 (–11.6)Old pine 15 51 1 895 46 (64) 1 669 (2 209) –11.9 (14.2)Old spruce–fir 375 6 10 876 6 (6) 10 122 (10 122) –6.9 (–6.9)Old tolerant hardwood 40 169 17 108 163 (165) 15 723 (16 385) –8.1 (–4.2)


Table 5. Change in number and area of habitat patches, 1993–1999, according to New Brunswick Department of Natural Re-sources and Energy (NBDNRE) spatial criteria.

Fig. 2. Spatial distribution of patches meeting size criteria for five New Brunswick Department of Natural Resources and Energy(NBDNRE) habitat types in the Fundy Model Forest (1999).



groups (Crown, industrial freehold, private woodlots), pri-vate woodlots were the most heavily cut over the 1993–1999period (Fig. 6).

According to the landscape metrics most closely associ-ated with an interpretation of landscape fragmentation, therate of fragmentation in the FMF occurred more rapidly thanhabitat loss (Fig. 7). Only in the case of OTHH and OSFHnearest neighbour analysis did habitat loss occur at a greaterrate than fragmentation.

Discussion

Landscape analysisSoftwood and mixedwood habiats contained the largest

area of changes in the 1993–1999 period in the FMF; toler-ant hardwood forest habitat had the lowest amount of changeas detected by the satellite image analysis. This low ratecould reflect a shift in management practices towardsmaller-scale patch cuts and other partial cuts in this habitattype on Crown land (NBDNRE 2000). Decreases inmixedwood (4.9%) and tolerant hardwood habitat (2.4%)over the 1993–1999 period could be of some concern due tothe poor capacity of tree species associated with these foresttypes for regenerating from stand-replacing disturbances,such as clear-cutting (Archambault et al. 1998). With forestgrowth taken into account using our simple age-class model(see Methods), pine habitat appeared to increase over the1993–1999 period. However, this result should be inter-preted with caution because a portion of “new” pine habitatis made up of old jack pine plantations, and it is question-able whether plantations will have the same structural attrib-utes as unplanted old pine forest (Freedman et al. 1994).

No habitat types were found to cover more than 20% ofthe total FMF landscape. In a modeling exercise, Fahrig(1998) predicted that fragmentation effects should not affectpopulation survival until breeding habitat for an organismcovers less than 20% of the landscape. Many species proba-bly use several of the habitat types identified by NBDNREand thus may be unaffected by fragmentation because totalavailable habitat is still greater than 20%. However, indicatorspecies that, by definition, should “specialize” on a single

habitat type are the most likely to be sensitive to the effectsof fragmentation over and above the effect of habitat lossalone. While the interpretation of landscape metrics for frag-mentation is variable, it is clear that the process of fragmen-tation is continuing.

Our results indicate that the relationship between habitatloss and fragmentation is not proportionate in the FMF. Therate of habitat loss is generally exceeded by the rate of frag-mentation. Decreases in the number and area of largemixedwood patches are one of the most significant trends re-vealed in this study. Tolerant mixedwood and hardwoodwere the most predominant forest types in presettlementsouthern New Brunswick (Zelazny et al. 1997). In pre-settlement times, mature mixedwood and tolerant hardwoodwould probably have existed in the largest patches (Zelaznyet al. 1997; Lorimer 2001). Species adapted to mixedwoodat the stand scale could also be adapted to the landscapeconfiguration of this forest type (Walters 1998). For exam-ple, mixedwood species such as northern flying squirrel(Vernes 2001) and American marten (Payer and Harrison1999) are reported to be sensitive to landscape fragmentationin northeastern New England and the Maritimes. There isalso evidence that survival rates of white-breasted nuthatch(which may be dependent on tolerant hardwood forest) arenegatively correlated with habitat fragmentation (Dohertyand Grubb 2002). Conversely, pine stands probably existedin a naturally patchy distribution across the landscape. It isless likely that species adapted to this forest type are as sen-sitive to habitat isolation. Such hypotheses need to be testedby determining the historical pattern of the landscape withthe use of modeling and ecological land classification (e.g.,Zelazny et al. 1997; Gustafson et al. 2000). Continuation ofresearch on spatial species–habitat relationships should pro-vide information on species-specific sensitivities to fragmen-tation.

Declining mean nearest neighbour values for mature for-est as a whole in the FMF is the combined result of (i) thesplitting of single large patches into two or more adjacentsmaller patches that have low nearest neighbour distances,(ii) the emergence of new habitat in close proximity to exist-

© 2003 NRC Canada

Betts et al. 1827

Fig. 3. Patch size distribution by patch frequency for 1993 and1999a landscapes.

Fig. 4. Patch size distribution by area for 1993 and 1999a land-scapes.



ing habitat patches, and (iii) the attrition of distant patchesthat may have previously boosted mean nearest neighbourvalues. This ambiguous result indicates the potentially mis-leading nature of “mean” statistics in landscape analysis(Gustafson et al. 2000). A more fruitful approach to nearestneighbour analysis is to report the full range of nearestneighbour values (the distance from each patch’s edge iscalculated to its nearest neighbour’s edge). Further, ratherthan lumping all mature forest into a single habitat categoryfor nearest neighbour computation (which does not reflectecological reality for many indicator species identified inNew Brunswick), we reported these values in terms of spe-cific NBDNRE habitat types. In this analysis, distance be-tween large patches of most habitat types increased between1993 and 1999. About 30% of the mixedwood habitat is innearest neighbour categories greater than the minimum 1 kmrecommended by NBDNRE (2000). Species with lowvagility, such as the northern flying squirrel, would likelyexperience difficulty colonizing this proportion of habitatpatches (Reunanen et al. 2000; Vernes 2001). However, it isimportant to note that this minimum distance does not in-clude the smaller habitat patches that fall between thepatches that meet patch size criteria. It is possible that taxacould be using such smaller patches as “stepping stones” formovement between larger patches (Forman 1998). Minimum

distance between patches may be a function of the matrixtype. Patches that are separated by an agricultural field aremore likely to be more isolated for some species thanpatches separated by a 10-year-old clearcut (Verboom et al.1991; Schmiegelow et al. 1997).

Management implicationsAll habitat types, except perhaps pine, are being reduced

in area at a rate greater than they are regenerating as a resultof forest growth in the FMF. Further, it cannot be assumedin the longer term that recent cutovers will regenerate tohave the same tree species composition and structure thatthey did preharvest. This is particularly true for toleranthardwood and mixedwood habitats; the majority of newmixedwood habitat is often made up of shade-intolerant treespecies. Accounting for forest growth, tolerant mixedwoodhabitat declined at a rate well below replacement (9.6%from 1993 to 1999). This may have serious implicationsover the long term if species exist that specialize on thishabitat type. One possible implication is that forest harvest-ing patterns should be developed for these stand types sothat tree species composition is maintained (Woodley andForbes 1997; J.D. Irving Sussex District 2001).

Decreases in patch size and nearest neighbour values formost habitat types suggest that more attention be given to

© 2003 NRC Canada

1828 Can. J. For. Res. Vol. 33, 2003

Fig. 5. Nearest neighbour frequency histograms for four New Brunswick Department of Natural Resources and Energy habitat types.Note that no change occurred for old spruce–fir habitat (OSFH) patches (not shown). Analysis was performed only on patch sizesidentified as sufficient to meet spatial requirements of indicator species for each habitat type (Table 2).



the spatial distribution of cuts. Crown land policies, such asmaximum block size (100 ha), potentially preclude themaintenance of large patches and effectively prevent the re-generation of large patches in the future. A “patchwork”landscape pattern is being created that is unlikely to haveexisted in the presettlement era. Of the minimum patch sizeguidelines recommended by NBDNRE, only the 375-hapatch size for OSFH is required by policy. Not surprisingly,this is the only habitat type that did not exhibit a decrease inthe number of habitat patches over the 1993–1999 period.Without actively planning for contiguous patches of otherhabitat types, even within the context of sustainable manage-ment for timber volume (J.D. Irving Sussex District 2001),there will likely be a continued decline in mean patch sizeand the number of patches that meet spatial requirements ofindicator species. It will be important to monitor populationsof indicator species to determine if trends in abundancereflect changes in habitat area and the degree of fragmenta-tion. Because habitat patches span landownership bound-aries, it will also be increasingly important to developtransboundary approaches to habitat planning, particularlyon private woodlots where, in this study, landscape changeoccurred at the highest rate (Betts et al. 2002).

ConclusionWe generated a 1993 baseline habitat map using existing

GIS forest inventory data developed from aerial photo-interpretation, and we updated this map with changes ob-served in Landsat satellite imagery acquired in 1995, 1997,1998, and 1999. The habitat types were related to indicatorspecies and native biodiversity through inferences about spe-cies habitat preferences and forest structure. Landscape met-rics were computed to summarize the landscape structureover the 1993–1999 time period; we interpret several ofthese metrics to analyze the fragmentation process that hasoccurred over the multijurisdictional area of the FMF. Mostforest habitats decreased in area, nearest neighbour dis-tances, and patch sizes. The rate of habitat loss was ex-ceeded by the rate of fragmentation. Decreases in thenumber of mixedwood patches and their sizes appeared to

dominate the fragmentation of the landscape; more attentionto the spatial distribution of harvesting activities may benecessary to change this landscape pattern in the future.Some low-vagility species, such as the northern flying squir-rel (recognized as an indicator species by NBDNRE 2000),may have difficulty maintaining population levels in suchhighly dispersed mixedwood landscapes.

Some important limitations of this approach exist andmust be considered. Interpreting the ecological implicationsof landscape metrics and landscape change remains prob-lematic. Substantial information gaps preclude inferenceabout population-level effects of the habitat composition andconfiguration changes reported in this study. Detailed con-figuration or fragmentation sensitivities of indicator speciesremain largely unknown. It is possible, for example, thatpatch size criteria for indicator species have been underesti-mated. Conversely, species could be engaging in “habitatsupplementation”: maintaining home ranges across a largerarea of fragmented landscape to compensate for lack of con-tiguous habitat (Dunning et al. 1992). Further, it is likelythat many species perceive landscapes as gradients of habitatresulting in “source” and “sink” areas rather than in terms ofcategorical “habitat” versus “nonhabitat” (Donovan et al.1995). More information on the spatial and stand-level sensi-tivities of landscape configuration sensitive taxa needs to becollected if the ecological implications of landscape metricsare to be fully understood. It is unknown how many patchesof a certain size are necessary in the region to maintain a vi-able population. This information could potentially be ob-tained by a combination of detailed species life historyinformation, species–habitat models based on local data, andspatially explicit population viability analysis (Dunning etal. 1995; McCarthy et al. 2001). In the most rapidly chang-ing habitat types, indicator species could be monitored to de-termine whether population trends match measured variationin landscape composition and pattern. And finally, data defi-ciencies can overwhelm even the most dedicated spatialanalysis project. For example, GIS data often are not spa-tially comprehensive and may be of dubious accuracy and

© 2003 NRC Canada

Betts et al. 1829

Fig. 6. Percent mature forest area removed on major FundyModel Forest land tenures during 1993–1999. Total change is thesum of both bars; change incorporating forest growth is shownby the dark gray bar.

Fig. 7. Relationship between the proportion of habitat loss andfragmentation. Each point represents the ratio between percentchange in a fragmentation metric and percent change in habitatfor each habitat type. The diagonal line represents 1:1 (propor-tionate) change.



vintage; remotely sensed data, on the other hand, may be ofinsufficient resolution or quality (e.g., cloud cover) and re-quire substantial image processing resources and skill.

Acknowledgments

Funding was provided by the Natural Sciences and Engi-neering Research Council of Canada, Canadian Forest Ser-vice, and FMF. Thanks to Graham Forbes and Jeff Bowmanand two anonymous reviewers for valuable comments thathelped to greatly improve the manuscript. Thanks are alsodue to Dan Beaudette for information on NBDNRE habitatcriteria.

References

Andren, H. 1994. Effects of habitat fragmentation on birds andmammals in landscapes with different proportions of suitablehabitat: a review. Oikos, 71: 355–366.

Archambault, L., Morissette, J., and Bernier-Cardou, M. 1998. For-est succession over a 20-year period following clearcutting inbalsam fir yellow birch ecosystems of eastern Quebec, Canada.For. Ecol. Manage. 102: 61–74.

Baskent, E.Z. 1999. Controlling spatial structure of forested land-scapes: case study towards landscape management. Landsc.Ecol. 14: 83–97.

Baskent, E.Z., and Jordan, G.A. 1995. Characterizing spatial struc-ture of forest landscape. Can. J. For. Res. 25: 1830–1849.

Beaudette, D. 2000. Habitat definitions for vertebrate forest wild-life. New Brunswick Department of Natural Resources and En-ergy, Fredericton, N.B.

Beier, P., and Noss, R. 1998. Do habitat corridors provide connec-tivity? Conserv. Biol. 12: 1241–1252.

Bender, D., Contreras, T., and Fahrig, L. 1998. Habitat loss andpopulation decline: a meta-analysis of the patch size effect.Ecology, 79: 517–533.

Betts, M., Knox, J., and Forbes, G. 2002. A landscape ecologicalapproach to private woodlot planning in New Brunswick, Can-ada. Nat. Areas J. 22: 311–317.

Cale, P.G., and Hobbs, R.J. 1994. Landscape heterogeneity indices:problems of scale and applicability, with particular references toanimal habitat description. Pac. Conserv. Biol. 1: 183–193.

Chapin, T.G., Harrison, D.J., and Katnik, D.D. 1998. Influence oflandscape pattern on habitat use by American marten in an in-dustrial forest. Conserv. Biol. 12: 1327–1337.

Cohen, W.B., and Fiorella, M. 1999. Comparison of methods offorest conifer change detection using Landsat Thematic Mapperimagery. In Remote sensing change detection: environmentalmonitoring methods and applications. Edited by R. Lunetta andC.D. Elvidge. Taylor & Francis, London, U.K. pp. 89–102.

Cohen, W.B., Fiorella, M., Gray, J., Helmer, E., and Anderson, K.1998. An efficient and accurate method of mapping forestclearcuts in the Pacific Northwest using Landsat imagery.Photogramm. Eng. Remote Sens. 64: 293–300.

Collins, J.B., and Woodcock, C.E. 1996. An assessment of severallinear change detection techniques for mapping forest mortalityusing multitemporal Landsat TM data. Remote Sens. Environ.56: 66–77.

Crist, E.P. 1985. A TM Tasseled Cap equivalent transformation forreflectance factor data. Remote Sens. Environ. 17: 301–306.

Davidson, C. 1998. Issues in measuring landscape fragmentation.Wildl. Soc. Bull. 26: 32–37.

Dijak, W.D., and Thompson, F.R. 2000. Landscape and edge ef-fects on the distribution of mammalian predators in Missouri. J.Wildl. Manage. 64: 209–216.

Doherty, P.F., and Grubb, T.C. 2002. Survivorship of permanent-resident birds in a fragmented forested landscape. Ecology, 83:844–857.

Donovan, T.M., Thompson, F.R., Faaborg, J., and Probst, J.R.1995. Reproductive success of migratory birds in habitat sourcesand sinks. Conserv. Biol. 9: 1380–1395.

Dunning, J., Danielson, B., and Pulliam H. 1992. Ecological pro-cesses that affect populations in complex landscapes. Oikos, 65:169–175.

Dunning, J., Stewart, D., Danielson, B., Noon, B., Root, T.,Lamberson, R., and Stevens, E. 1995. Spatially explicit popula-tions models: current forms and future uses. Ecol. Appl. 5: 3–11.

Dussault, C., Courtois, R., Huot, J., and Ouellet, J.P. 2001. The useof forest maps for the description of wildlife habitats: limits andrecommendations. Can. J. For. Res. 31: 1227–1234.

Elkie, P.C., Rempel, R.S., and Carr, A.P. 1999. Patch Analystuser’s manual: a tool for quantifying landscape structure. On-tario Ministry of Natural Resources, Thunder Bay, Ont. NWSTTech. Man. TM-002.

Fahrig, L. 1998 When does fragmentation of breeding habitat af-fect population survival? Ecol. Model. 105: 273–292.

Forman, R.T. 1998. Land mosaics. The ecology of landscapes andregions. Cambridge University Press, Cambridge, U.K.

Franklin, S.E., Moskal, M., Lavigne, M., and Pugh, K. 2000a. In-terpretation and classification of partially harvested forest standsin the Fundy Model Forest using multitemporal Landsat TMdigital data. Can. J. Remote Sens. 26: 318–333.

Franklin, S.E., Dickson, E.E., Farr, D.M., Hansen, M.J., andMoskal, L.M. 2000b. Quantification of landscape change fromsatellite remote sensing. For. Chron. 76: 877–886.

Franklin, S.E., Lavigne, M., Moskal, M., Wulder, A., andMcCaffrey, T.M. 2001. Interpretation of forest harvest condi-tions in New Brunswick using Landsat TM enhanced wetnessdifference imagery (EWDI). Can. J. Remote Sens. 27: 118–128.

Franklin, S., Lavigne, M., Wulder, M., and McCaffrey, T. 2002a.Large-area forest structure change detection: an example. Can.J. Remote Sens. 28: 588–592.

Franklin, S.E., Lavigne, M.B., Wulder, M.A., and Stenhouse, G.B.2002b. Change detection and landscape structure mapping usingremote sensing. For. Chron. 78: 618–625.

Franklin, S.E., Hansen, M.J., and Stenhouse, G.B. 2002c. Quan-tifying landscape structure with vegetation inventory maps andremote sensing. For. Chron. 78: 866–875.

Freedman, B., Woodley, S., and Loo, J. 1994. Forestry practicesand biodiversity with particular reference to the Maritime prov-inces. Environ. Rev. 2: 33–77.

Gustafson, E., Shifley, S., Mladenoff, D., Nimerfro, K., and Hong,H. 2000. Spatial simulation of forest succession and timber har-vesting using LANDIS. Can. J. For. Res. 30: 32–43.

Hargis, C.D., Bissonette, J.A., and Turner, D.L. 1999. The influ-ence of forest fragmentation and landscape pattern on Americanmartens. J. Appl. Ecol. 36: 157–172.

Hulshoff, R. 1995. Landscape indices describing a Dutch land-scape. Landsc. Ecol. 10: 101–111.

J.D. Irving Sussex District. 2001. Forest management plan forFundy Crown timber licence: Sussex. J.D. Irving Sussex Dis-trict, Saint John, N.B.

Kitzberger, T., and Veblen, T. 1999. Fire-induced changes in north-ern Patagonian landscapes. Landsc. Ecol. 14: 17–33.

© 2003 NRC Canada

1830 Can. J. For. Res. Vol. 33, 2003



Krummel, J.R., Gardner, R.N., Sugihara, G., O’Neill, R.V., andColeman, P.R. 1987. Landscape patterns in a disturbed environ-ment. Oikos, 48: 321–324.

Lichstein, J., Simons, T., and Franzreb, K. 2002. Landscape effectson breeding songbird abundance in managed forests. Ecol. Appl.12: 836–857.

Lorimer, C. 2001. Historical and ecological roles of disturbance ineastern North American forests: 9,000 years of change. Wildl.Soc. Bull. 29: 425–439.

Lundquist, J.E., Lindner, L.R., and Popp, J. 2001. Using landscapemetrics to measure suitability of a forested watershed: a casestudy for old growth. Can. J. For. Res. 31: 1786–1792.

Mazerolle, M., and Villard, M.-A. 1999. Patch characteristics andlandscape context as predators of species presence and abun-dance: a review. Écoscience, 6: 117–124.

McCarthy, M., Possingham, H., Day, J., and Tyre, A. 2001 Testingthe accuracy of population viability analysis. Conserv. Biol. 15:1030–1038.

Miller, J.N., Brooks, R.P., and Croonquist, M.J. 1997. Effects oflandscape patterns on biotic communities. Landsc. Ecol. 12:137–153.

Mladenoff, D.J., Sickley, T.A., Haight, R.G., and Wydeven, A.P.1995. A regional landscape analysis and prediction of favorablegray wolf habitat in the northern Great Lakes region. Conserv.Biol. 9: 279–294.

NBDNRE. 1986. Forest development survey. New Brunswick De-partment of Natural Resources and Energy, Fredericton, N.B.

NBDNRE. 2000. A vision for New Brunswick’s forests: goals andobjectives for Crown land management. New Brunswick Depart-ment of Natural Resources and Energy, Fredericton, N.B.

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., and Munro, R.H.M.2003. Development and testing of phenologically driven grizzlybear habitat models. Écoscience, 10: 1–10.

Paton, P. 1994. The effect of edge on avian nest success: howstrong is the evidence? Conserv. Biol. 8: 17–26.

Payer, D., and Harrison, D. 1999. Influence of timber harvestingand trapping on habitat selection and demographic characteris-tics of Marten. Final contract report to Maine Department of In-land Fisheries and Wildlife, Orono, Me.

Popplewell, C., Franklin, S.E., Hall-Beyer, M., and Stenhouse,G.B. 2003. Using landscape structure to classify grizzly beardensity in Alberta Yellowhead Ecosystem Bear ManagementUnits. Ursus, 14: 27–34.

Rempel. 1999. Patch Analyst 2.0. Available from http://www.lakeheadu.ca/~rrempel/patch/index.html [updated 17 July 1999,cited January 2002].

Reunanen, P., Monkkonen, M., and Nikula, A. 2000. Managing bo-real forest landscapes for flying squirrels. Conserv. Biol. 14:218–226.

Richter, R. 1990. A fast atmospheric correction algorithm appliedto Landsat TM images. Int. J. Remote Sens. 11: 159–166.

Riitters, K.H., O’Neill, R.V., Hunsaker, C.T., Wickham, J.D., Yan-kee, D.H., Timmins, S.P., Jones, K.B., and Jackson, B.L. 1995.

A factor analysis of landscape pattern and structure metrics.Landsc. Ecol. 10: 23–39.

Rowe, J.S. 1972. Forest regions of Canada. Department of the En-vironment, Ottawa, Ont. Can. For. Serv. Publ. 1300.

Sachs, D., Sollins, P., and Cohen, W. 1998. Detecting landscapechanges in the interior of British Columbia from 1975 to 1992using satellite imagery. Can. J. For. Res. 28: 23–36.

Saveraid, E.H., Debinski, D., Kindscher, K., and Jakubauskas, M.2001. A comparison of satellite data and landscape variables inpredicting bird species occurrences in the Greater YellowstoneEcosystem, USA. Landsc. Ecol. 16: 71–83.

Schmiegelow, F., Machtans, C., and Hannon, S. 1997. Are borealbirds resilient to forest fragmentation? An experimental study ofshort-term community responses. Ecology, 78: 1914–1932.

Shao, G., Liu, D., and Zhao, G. 2001. Relationships of image clas-sification accuracy and variation of landscape statistics. Can. J.Remote Sens. 27: 33–43.

Verboom, J., Schotman, A., Opdam, P., and Metz, A. 1991. Euro-pean nuthatch metapopulations in a fragmented agriculturallandscape. Oikos, 61: 149–156.

Vernes, K. 2001. Gliding performance of the Northern FlyingSquirrel (Glaucomys Sabrinus) in mature mixed forest of easternCanada. J. Mammal. 82: 1026–1033.

Walters, J. 1998. The ecological basis of avian sensitivity to habitatfragmentation. In Avian conservation: research and manage-ment. Edited by J.M. Marzluff and R. Sallabanks. Island Press,Washington, D.C. pp. 181–192.

Wang, Y., and Moskovits, D. 2001. Tracking fragmentation of nat-ural communities and changes in land cover: applications ofLandsat data for conservation in an urban landscape (ChicagoWilderness). Conserv. Biol. 15: 835–843.

Wickham, J.D., O’Neill, R.V., Riitters, K.H., Wade, T.G., andJones, K.B. 1997. Sensitivity of selected landscape pattern met-rics to land-cover misclassification and differences in land-covercomposition. Photogramm. Eng. Remote Sens. 63: 397–402.

Woodcock, C.E., Macomber, S.A., Pax-Lenney, M., and Cohen,W.B. 2001. Monitoring large areas for forest change usingLandsat: generalization across space, time and Landsat sensors.Remote Sens. Environ. 78: 194–203.

Woodley, S. 1998. State of the Greater Fundy Ecosystem: introduc-tion. In State of the Greater Fundy Ecosystem. Edited by S.Woodley, G. Forbes, and A. Skibicki. Greater Fundy EcosystemResearch Project, Fredericton, N.B.

Woodley, S., and Forbes, G. 1997. Forest management guidelinesto protect native biodiversity in the Fundy Model Forest. NewBrunswick Cooperative Fish and Wildlife Research Unit, Uni-versity of New Brunswick, Fredericton, N.B.

Zelazny, V., Veen, H., and Colpitts, M. 1997. Potential forests ofthe Fundy Model Forest. Department of Natural Resources andEnergy, Fredericton, N.B.

Zheng, D., Wallin, D., and Hao, Z. 1997. Rates and patterns oflandscape change between 1972 and 1988 in the ChangbaiMountain area of China and North Korea. Landsc. Ecol. 12:241–254.

© 2003 NRC Canada

Betts et al. 1831



Documents

Interpretation of landscape pattern and habitat change for ... et al 2003 CJFR.pdf · et al. 2003), American martens (Martes americana) (Chapin et al. 1998; Hargis et al. 1999), and