20 Species List

7/31/2019 20 Species List

1/21

749

The Auk 119(3):749769, 2002

ESTIMATING SPECIES RICH N ESS OF TROPICAL BIRD

COMMUNITIES FROM RAPID ASSESSMENT DATA

SEBASTIAN K. H ERZOG,1,4 MICHAEL KESSLER,2 AN D THOMAS M. CAHILL3

1In st itut fu r Vogelforschun g Vogelwart e Helgoland, A n der Vogelwarte 21, 26386 W ilhelm shaven, Germany ;2A lbrecht-von-Haller In st it ut fu r Pfl anz enwis senschaften, A bteilu ng Sy st ematische Botan ik, Unt ere Karsp ule 2,

37073 Got ti ngen, Germany; and3Canadian Environmental Modelling Centre, Trent University, 1600 West Bank Drive, Peterborough,

Ont ario K9J 7B8, Canada

ABSTRACT.Rapid assessment su rveys of tropical bird commun ities are increasingly used

to estimate species richness a nd to d etermine conservation p riorities, but resu lts of different

stud ies are often n ot comparable du e to the lack of standar dization. On the basis of computer

simulations and six years of field testing, w e evaluated the r ecently p roposed 20-species-

list survey method and statistical estimators for assessing species richness of tropical bird

commu nities. This meth od generates a sp ecies-accumu lation cur ve by subd ividing consec-

utive observations of birds into lists of 20 species, thus relating cumu lative species richnessto the nu mber of observations rather than time or space and th ereby accoun ting for mod erate

differences in observer qualification and field conditions. Species accumulation curves from

computer-simulated communities and two empirical data sets from Bolivia were analyzed

with nine species richness estimators to evaluate estimator accuracy with respect to varia-

tions in species-list size, samp le size, species-pool size, and commun ity structu re. For em -

pirical and most simulated data sets, the MMMEAN estimator p erformed best, but i t was

more sensitive to differences in commu nity structur e than most other estimators. The CHAO

2 estimator, which was recommended by previous studies, performed reasonably well but

was considerably more sensitive to sample size than MMMEAN. The bootstrap and first-

and second -order jackknife estimators p erformed poorly. We recommend using MMMEAN

or, when standa rd deviations of richness estimates are ind ispensable, CHAO 2 with 10-spe-

cies lists for estimating species richness of tropical bird communities and propose a set of

standard survey rules. Careful examination of estimator accumulation curves is required,

however, and a technique based on the ratio between estimator and species accumulation

curve is suggested to control for the confounding effects of sampling effort. Overall, the

species-list method combined with statistical richness estimation is d oubtlessly mu ch m ore

standardized and valuable than simple comparisons of one-dimensional locality lists and

represents a promising tool for conservation assessment and the stud y of avian diversity

patterns in the tropics. Received 29 June 2001, accepted 15 April 2002.

RESUMEN .Cada vez se u san con m ayor frecuencia evalua ciones rap idas de comu nida des

de aves tropicales para estimar la riqueza d e especies y para d eterminar p rioridad es de con-

servacion , pero los resultados de d iferentes estud ios a menu do no son comparables debido

a la falta d e estand arizacion . Basad os en simu laciones realizad as en compu tad oras y en seisan os d e evaluaciones d e cam po, evaluam os el m etod o d e mu estreo lista d e 20 especies

recientemente p ropuesto y los estimadores estadsticos p ara determinar la riqueza d e es-

pecies de comunidad es d e aves trop icales. Este m etod o g enera cu rvas d e acum ulacion de

especies su bdividiendo observaciones consecutivas d e aves en listas d e 20 especies. As, re-

laciona la riqueza acum ulada de especies con el n u mero de observaciones y no con tiemp o

o espacio, incorpora ndo de este m odo d iferencias mod eradas en la h abilidad del observador

y en las condiciones del tiempo. Analizamos curvas acumuladas de especies originadas a

partir de comunidades simuladas y de d os juegos de datos emp ricos de Bolivia empleando

nueve estimadores d e la riqueza de esp ecies par a evaluar la exactitud de los estimadores con

relacio n a l tam an o d e las listas d e esp ecies, tam an o d e la mu estr a, tam an o de l set de esp ecies

y estructura d e la comunidad. Para juegos de datos emp ricos y para la mayor a de los si-

mu lados, el estimad or MMMEAN fue el mejor, pero fue m as sensible que otros estimad ores

4 E-mail: skh erzog @compu serve.com


2/21

750 [Auk , Vol. 119H ERZOG , KESSLER, AN D CAHILL

a diferencias en la estructura de la comunidad. El estimador CHAO 2, recomendado por

estu dios pr evios, funciono razon ablement e bien, per o fue consid erablemen te m as sen sible

al tam ano de la mu estra qu e MMMEAN. El bootstrap y los estimad ores de primer y se-

gun do orden de jackknife fun cionaron mal. Recomend amos u sar MMMEAN o, cuan do se

requ ieren estim aciones de la d esviacion estan da r d e las estim aciones de riqu eza, CH AO 2

con listas de 10 especies para estimaciones de riqueza de especies de comunidades de aves

tropicales, y p roponem os u na serie de reglas estand ares d e mu estreo. Sin em bargo, es ne-

cesario examinar cuidad osamente los estimadores de curvas acumu lativas, y sugerimos una

tecnica basad a en el cociente entre el estimad or y la curva acum ulada de esp ecies para con-

trolar la distorsion creada p or efectos de esfuerzo d e mu estreo. En resum en, el me todo de

la lista d e especies combinad o con la estim acion estad stica de la r iqueza es sin dud a mucho

ma s es tan da r y valido que comp araciones sim ples d e listas d e esp ecies unid imen sionales, y

representa un a herr amienta prom etedora p ara evaluaciones de conservacion y el estud io de

los patr ones d e d iversidad de aves en el trop ico.

Q UANT I FYI NG T HE SPE CIE S richness of bird

commu nities has gained increasing imp ortance

i n e nv ir o n m en t a l i m p act a ss es sm e n ts (e.g .Fjeldsa 1999), conservation pla nn ing (Bibby et

al. 1992, Stotz et al. 1996), an d ecological re-

search (Huston 1994, Rosenzweig 1995). In

Holarctic regions, where species richness is low

and avian commu nities are well characterized,

s t an d a r d i ze d co u n t a n d ce n su s m e t ho d s a r e

available (Holmes et al. 1986, Bibby et al. 2000).

In the tropics, however, sp ecies-rich commu -

nities inhabit highly complex, heterogeneous

environments and those method s are often d if-

ficult to apply (Terborgh et al. 1990, Remsen

1994, Poulsen et al. 1997a). Detailed quantita-

tive studies of tropical bird communities are

highly labor intensive (e.g. 12 person-months

by Terborgh et al. 1990) and require a variety of

methods such as mist-netting, spot-mapping,

point counts, and observation of mixed species

flocks to achieve a high d egree of comp leteness

(Remsen and Parker 1983, Terborgh et al. 1990,

Poulsen 1994, Remsen 1994, Gram and Faaborg

1997). They are therefore limited to few sites in

selected habitats (e.g. Blake et al. 1990, Ter-

borgh et al. 1990, Cohn-Haft et al. 1997, Rob-inson et al. 2000).

In the light of increasing forest destruction

(Dale et al. 1994) and wide gaps in the under-

standing of tropical bird commu nities, several

researchers have recently applied a rapid as-

sessment a pp roach to maximize d ata collection

w ith limited fu nd s, time, and person nel (Parker

and Bailey 1991, Parker et al. 1993, Poulsen et

al. 1997a, Poulsen and Krabbe 1998). To find

general patterns in responses of avian species

richness to distur bance, many spatial and tem-poral replicates will be necessary, instead of

condu cting exact coun ts at a few sites. This is a

principal reason for developing rapid assess-

men t metho d s. Unfortu nately, lack of stand ard -

ization w ith respect to su rvey method, observ-er qu alification, area, time, weather, and season

have largely preclud ed quantitative comp ari-

sons of rapid assessment stud ies.

Un t il r e ce n tly, t h e m o s t f re qu e n t ly u s e d

method for inventories of tropical bird com-

mun ities was mist-netting, wh ich was often

combined w ith u nstandardized visual obser-

vations and tape record ings (Karr 1981, 1990;

Poulsen 1994; Schmitt et al. 1997). Mist-netting

und oubtedly h as an advantage in reducing bi-

ases introduced by varying observer experi-

ence and qu alification, but it is noneth eless su b-

ject to a va riety of other bi as es , su ch as net

avoidance, weather, habitat structure, and be-

havioral differences between species and indi-

viduals of the same species (Karr 1981, 1990;

Jenni et al. 1996; Remsen and Good 1996). Oth-

er major disadvan tages of mist-nettin g are high

labor intensity, low time efficiency, the com-

paratively small proportion of the total com-

mu nity sampled, and a strong bias toward s un-

d e r st or y s p e cie s (G r am a n d Fa a bo r g 19 97,

Whitman et al. 1997).The recent increase in knowledge of vocali-

zations (Parker 1991, Budney and Grotke 1997),

comm ercially available reference record ings

(e.g. Mayer 2000), and high -quality field gu ides

has resulted in an increased use of acoustical

and visual observations in rapid assessments.

However, even highly experienced observers

are subject to a variety of biases, mainly relat-

ing to varying detectabilities between species

or between seasons for any particular species

(e.g. Karr 1981, 1990; Oniki and Willis 1982a, b,1983; Verner 1985; Verner and Milne 1990),

which are compounded by differences in ob-


3/21

July 2002] 751Est im atin g Tropical Species Richn ess

server skill and observation techn ique (Sauer et

al. 1994). Thus, p ur ely observational d ata mu st

always be considered with caution (Cohn-Haft

et al. 1997). On the oth er han d, aud itoryvisual

method s are m uch more time efficient, report a

considerably larger p ortion of local avifaun asthan mist netting (Whitman et al. 1997), and

the presence of most species at any given lo-

cality can be d ocumented by tape recording

(Parker 1991).

Some aud itoryvisual rap id assessments em-

p l oy ed s t an d a r d s u r vey t ech n i qu e s s u ch a s

po int coun ts (Pou lsen and Krabbe 1998) or line

transects (Karr 1971). Although those methods

generate quantifiable data with well-estab-

lished sampling protocols, they have several

disadvantages for rapid assessments: they (1)are often d ifficult or impossible to ap ply un der

tropical field conditions (e.g. in steep, inacces-

sible montane forests), (2) require the exclusion

of observations between sample intervals from

the analysis, which further redu ces the already

limited amoun t of obtainable d ata, (3) require

h i gh o b se rv er q u a li fica tio n (P ou l s en e t a l.

1997a), and (4) tend to underestimate richness

and abundance of some bird groups (e.g. noc-

turnal species; Bibby et al. 2000). Thus, an ideal

survey or analysis method should lack the lim-itations of rigid standardization while ensuring

comp arability between stud ies.

MacKinnon and Phillips (1993) suggested a

quantitative approach to analyzing aud itory

visual survey data that accounts, at least to

some deg ree, for d ifferences in effort, observer

q u a li fica tio n , a n d w e at h er (P ou l se n e t a l.

1997a ). In t h at m e th o d , o bs er va tio n s a r e

grou p ed in to consecutive lists of 20 species and

a sp ecies accum u lation cur ve is generated from

adding those species not recorded on any pre-vious list to the total sp ecies numb er, wh ich is

then plotted as a function of list number. It is

crucial to includ e even observations that cann ot

be positively identified at first (Poulsen et al.

1997a). Hence, observers can devote the neces-

sary time to become completely familiar with

the avifauna an d are forced to tap e record an d

track dow n u niden tified vocalizations. Because

the method relates species richness to the num-

ber of observations rather than time or area,

t h is m e t ho d a llo w s co m p a r is on o f d a t a o b -tained by d ifferent observers or u nder varying

field conditions. Of course, no method will ac-

count for strong deficiencies in observer quali-

fication or extreme w eather.

Although attractive du e to its simplicity and

the possibility for comparing different studies,

the 20-species-list method (Poulsen et al.

1997a) remains largely untested. Poulsen et al.(1997a) concluded that 20-species lists are bi-

ased like any other bird count, but not more

than other methods and that the method is

suitable for judging (a) when a site is ade-

quately su rveyed, (b) the magn itude of the spe-

cies richness, (c) the relative abundance of each

species and (d) an -index of diversity. These

conclusions, how ever, are based on a variety of

untested assump tions about frequency distri-

butions. In addition, neither MacKinnon and

Phillips (1993) nor Poulsen et al. (1997a) sug-gested standardized survey rules or assessed

the po ssibilities of statistical data a nalysis. In a

later note, Poulsen et al. (1997b) considered po-

tential biases and called for more rigorou s test-

ing before the method can be recommended.

Quantitative comparisons of species accu-

mulation curves have been widely explored

(e.g. Palmer 1990, Baltana s 1992, Bun ge and

Fi tz p a tr ick 1993, C ol w el l a n d C od d i n gt on

1994, Walther and Morand 1998, Gotelli and

Colwell 2001) but non etheless are seldom used

in biodiversity studies (Boulinier et al. 1998).

To estim ate sp ecies richn ess from accum u lation

curves, three types of analyses have been em-

ployed: (1) extrapolation, (2) fitting species-

abundance distributions, and (3) nonparamet-

ric estimators (Soberon and Llorente 1993,

Colwell and Codd ington 1994, Walther et al.

1995, Chazdon et al. 1998, Walther and Morand

1998). However, the performa nce of estimators

and their sensitivity to variations in sampling

protocol, sample size, species r ichness, and

other variables remains largely un known.Based on computer simulations and six years

of field testing, this article presents a quanti-

tative assessment of the species-list method

and recommendations for a standardization of

surveys to allow quantitative compar isons be-

tween studies. We further explore possibilities

of estimating species richness from data gath-

ered with the species-list method by compar-

ing the performance of nine statistical richness

e st im a t or s i nclu d e d i n t h e p r o g ra m ESTI-

MATES 5.0.1 (Colw ell 1997). Sp ecifically, w e as-sessed the following parameters for each esti-

mator: (1) accuracy, (2) sensitivity to sample


4/21


size (i.e. number of species lists), (3) sensitivity

to tr ue sp ecies richn ess, and (4) the influence of

un derlying species-abund ance or species-de-

tectability distributions. An ideal estimator

should have constant, high accuracy irrespec-

tive of chang es in any of the other param eters.

METHODS

COMPUTER SIMULATIONS FOR ESTIMATION TESTS

To evaluate performance and biases of statistical

richness estimators, we used a computer model that

created lists of consecutive observations drawn at

random from species pools of different sizes with a

species-abun dan ce or species-detectability d istribu-

tion according to the exponential decline model:

rxN N ex 0 (1)

where N is the num ber of individuals in species x, N0the num ber of individu als in the most common spe-

cies, and r the exponential rate constant determined

by the number of individuals in the most common

and least common species for a given species-pool

size. Several other models have been proposed to de-

scribe sp ecies-abund ance d istributions in natur e (see

G o te ll i a n d G r av es 19 96 fo r a r e ce n t s u m m a r y ).

However, surveys of tropical bird communities are

sub ject to a variety of biases, such as the considerable

differences in detectability b etween species (Karr

1971, 1981, 1990; Boulinier et al. 1998), and it is thusunlikely that the species-abundance distribution of

rapid assessment data accurately reflects the samp le

commu nitys true structure in most cases. We ex-

amined species-abundance distributions of 15 em-

pirical data sets collected with the species-list meth-

o d (s ee b elow ) u s in g ch i-s qu a r e t es ts . Th r ee

distributions were significantly different from the

exponential decline and four were significantly dif-

ferent from the log normal m odel. Exponential de-

cline was a better fit than log normal in nine cases

and an equal fit in t hree cases. The log norm al model

was a better fit in only three cases. We thu s used the

exponential decline model as the underlying distri-

bution in our simulations unless otherwise stated.

Species richness of model commu nities and the

number of individuals in the most common and rar-

est species were user defined (and thereby the total

num ber of individuals). Abundan ce of the latter was

always set at 1 and that of the former was set to ob-

tain a total of about 7,640 individuals, which is 4

the nu mber of individuals found by Terborgh et al.

(1990) to nest in 100 ha of Am azon ian forest. Because

Terborgh et al. (1990) determined that 26% of the

nesting species had population densities of1 pair

per 100 ha, we assumed 400 ha to completely containt h e h o m e r a n g e o f 0 . 5 p a i r o f t h e l o w e s t - d e n s i t y

breeding species.

From the list of consecutive observations, the p ro-

g r a m co m p i le d 2 0- sp e cie s l is t s fo llo w in g M a c-

Kinnon and Phillips (1993): the first list consists of

the first 20 species observed, the second list includes

the next 20 species and may contain sp ecies already

found on the first l ist, and so on. Original informa-

tion on the abu nda nces of species was maintained on

the 20-species lists. Poulsen et al. (1997a) found that

20-species lists were too long in species-poor com-

mu nities and su ggested u sing 10-species lists. There-

fore, we used list sizes of 5, 10, and 20 species. Dif-

ferent list sizes obviously signify different sample

sizes, so a given number ofm 1-species lists is thu s not

directly comparable to the same num ber of m 2-spe-

cies lists. For the statistical analysis with ESTI-

MATES, each species list was treated as a separate

sample.

To evaluate estimator p erforma nce, we varied the

following para meters: (1) species richness (pools of50, 100, 250, 500 sp ecies), (2) list siz e (5, 10, 20 species

per list), and (3) num ber of lists (10, 20, 50, 100; 100

only for 5-species lists). For each combination of pa-

rameters, we compu ted 40 replications that were an-

alyzed with nine richness estimators included in the

pr ogra m ESTIMATES 5.0.1 (Colwell 1997; see also

Colwell and Codd ington 1994): seven nonp aram etric

statistics (ACE: Chao et al. 1993, Chazd on e t al. 1998;

ICE: Lee and Chao 1994, Chazd on et al. 1998; Chao

1: Chao 1984; Chao 2: Chao 1987; jackknife 1: Burn-

ham and Overton 1978, 1979; Heltshe an d Forrester

1983; Smith a nd van Belle 1984; jackknife 2:Bur nh am

and Overton 1978, 1979; Smith and van Belle 1984;Palmer 1991; bootstrap: Smith and van Belle 1984)

and two statistics that extrapolate species accumu-

lation curves (MMRuns, MMMean: Raaijmakers

1987). ACE and Chao 1 are abundance-based esti-

mators, whereas all other statistics are based on the

incidence of species in samples. For details on all es-

timators, including equations of the seven nonpara-

metric statistics, see Colwell (1997). MMRuns and

MMMean are based on the Michaelis-Menten model

(Raaijma kers 1987), w hich was evaluated recently by

Keating and Quinn (1998). For comparison we also

included S obs, which is the raw species accumula-tion curve that i tself is an estimator with a strong,

negative bias (Colwell and Codd ington 1994), in the

analysis of estimator accuracy.

ESTIMATES parameters were set to the default

values (50 randomized runs, random number seed

17, 10 incidence classes for ICE, 10 abundance classes

for ACE). When compiling species accumulation

curves and computing richness estimates, ESTI-

MATES randomizes the ord er of species lists, so the

original sample ord er is irrelevant to all analyses.

To evaluate estimator biases with respect to com-

munity structure, we calculated richness estimates as

described above with a uniform distribution of spe-cies (i.e. all species were equally abundant) for the

following parameter combinations: (1) pools of 50,


5/21


100, 250, and 500 species; (2) lists of 5 and 20 species;

and (3) 10, 20, and 50 lists. These results were then

compared to those obtained from the exponential

decline distribution. Although a uniform species-

abund ance distribution has not been docum ented for

natural bird commu nities, determining estimator

performance under such extreme, unrealistic condi-

tions is nonetheless helpful for a general un derstand -

ing of estimator beh avior w ithin the natu ral range of

circumstances.

The accuracy of all richness estimates for both ex-

ponential decline and even distributions was calcu-

lated as the mean (SD) of 40 replications for each

p a r am e te r co m bi na tio n a n d e st im a to r, a n d e x-

p r e ss ed a s p e r ce n t ag e o f t h e p r e d e te r m in e d t o ta l

species richness.

To determine effectiveness of list size, sets of 400

consecutive observations were drawn from 50- and

500-species p ools conforming to th e exponential d e-cline mod el and th en sub divided into 5-, 10-, and 20-

species lists. The resulting number of lists for each

list size and species pool was averaged (SD) over

20 replications.

EMPIRICAL D ATA FOR ESTIMATION TESTS

Forty-five forest localities were su rveyed by S.K.H .

and M.K. in the Bolivian Andes and adjacent low-

l an d a r e as w i t h t h e s p e ci es -l is t m e t h od t o t al in g

400 person days of data collection from 1995 to

2000. From that we derived our recommendations for

standa rdizing th e species-list method, wh ich are de-tailed in the below. Data from two localities, the Bo-

tanical Garden of Santa Cruz de la Sierra (departa-

m en to Sa nt a Cr u z) a nd C er ro A su n ta Pat a/

Calabatea (departamento La Paz), were analyzed

here with th e same nine estimators used on the com-

puter-simulated data sets (see above) for a quanti-

tative test of the method. The avifauna of both sites

was inventoried intensively by a number of field

workers and the resu lting species lists were used for

comparison with data obtained during rapid-assess-

ment surveys using the species-list method.

Surveys were conducted from dawn to mid-dayand often again from late afternoon until after dusk.

While walking slowly and quietly along existing

roads or trails and randomly through the habitat

wh erever feasible, all visual an d acoustical observa-

tions of birds within 50 m of the observer (Schieck

1997) were recorded continuously, includ ing the

number of individuals per species for each encoun-

ter; observations of birds at camp sites also were re-

corded. Tape recordings of dawn choruses, mixed-

species flocks, and individual birds were made to

supplement visual observations and for later identi-

fication of un know n voices (Parker 1991; see also Ha -

selmayer and Quinn 2000). The observers rate of movement depended largely on the level of bird ac-

tivity. When n ecessary, an hour or more w as spent in

one sp ot to observe m ixed-species flocks or high vo-

cal activity at dawn. In such cases and during occa-

sional resampling of the same trail area, repeated

c ou n t s o f o b vi ou s ly t er r it o ri al i n d iv id u a l s w e r e

avoided . The resulting master list of temp orally con-

secutive observations from each site was then sub-

divided into 5-, 10-, and 20-species lists as ou tlined

above. All tape recordings were treated like other ob-

servations and were integrated into the master list.

Further d etails on the sur vey methodology are given

in the below.

Th e Sa n ta C ru z Bo ta n ica l G ar d en (1747S,

6304W, 450 m elevation) was a r emn ant 187 ha fra g-

ment of tall lowland semidecidu ous forest and chaco

t h o rn s cr u b s u r r ou n d e d b y u r b a n a n d a g r ic u lt u r a l

areas 13 km east of Santa Cruz city (Parker et al.

1993). It w as inventoried over a 38 month period by

c on s t an t -e ffo r t m i s t n e t ti n g ( si x d a y s p e r m o n t h

from September 1995 to December 1997, three d aysp e r m o n t h f r om Ja n u a r y t o O ct o be r 1 99 8), n e st

searches, and opportunistic observations by S. E. Da-

vis (pers. comm.) and coworkers of the Museo de

Historia Natural Noel Kempff Mercado, Universidad

Auton oma Gabriel Rene Moreno, Santa Cru z. Ad di-

t io n a ll y, t a p e r e co r d in g s a n d o b se r va t io n s w e r e

made during repeated visits by S. Mayer and S.K.H.

(see Mayer 2000, Herzog and Kessler 2002). The re-

sulting cumulative species list for the area was pro-

vided by S. E. Davis (unpubl. data). The rapid as-

sessment data analyzed here were obtained from 17

to 20 August 1999 by S.K.H. in an area of 45 ha in-

clud ing the two m ajor habitat typ es. Because the cu-mulative locality list contained data from all seasons,

we extracted, in collaboration with S. E. Davis, those

species known to occur in the forest and thorn scrub

area during the mid- to late dry season (July to Sep-

tember). Very rare species of u ncertain status that

probably repr esent vagrants w ere also exclud ed. The

resu lting list of 140 species was consid ered the ar eas

actual species richness du ring the sur vey period and

was u sed to calibrate resu lts of the statistical ana lysis

of our rapid assessment data.

Cerro Asunta Pata (1503S, 6829, 8501,500 m)

a n d C a la b at e a (1 459S, 68

28W, 1,3001,600 m)comprise an ar ea of evergreen monta ne forest on the

southwest and northeast side, respectively, of the Ro

Yuyo (850 m) along a dirt road from Charazani to

Apolo. The areas vegetation consisted of up to 30 m

tall evergreen forest in a transition from lowland to

montane elevational belts (Parker and Bailey 1991).

Some sm all-scale forest clearing h ad occurr ed along

the road , but in general the area was covered by pris-

tine forest. Calabatea was surveyed (tape recording

and observation) by T. A. Parker from 7 to 12 June

1990 (Parker an d Bailey 1991, Parker et a l. 1991). Cer-

ro Asunta Pata was inventoried (specimen collection,

tape recording) by a team of the Museum of NaturalScience, Louisiana State University, Baton Rouge,

from 11 July t o 17 Augu st 1993, totaling 10,500 net -


6/21


7/21


empirical data, see below), whereas that pat-

tern was r eversed in the r emaining estimators.

Comparing mean overall accuracy for each of

the four species pools, ICE was the least sen-

sitive to species richn ess (10.8% d ifference be-

tween highest and lowest mean overall accu-r a cy ), fo llo w ed b y AC E (13.2% ), C h ao 2

(17.0%), Chao 1 (18.9%), MMMean (21.2%),

jack kn ife 2 (21.9%), jack kn ife 1 (26.1%),

MMRun s (27.7%), an d bootstr ap (29.3%).

Sensitivity to commun ity structure. All esti-

mators consistently computed higher estimates

for pools w ith equally abund ant species (Ap-

p e n d i x 2 ) t h a n fo r p o o ls co n fo r m in g t o t h e

exp o n en t ia l d e clin e m o d el (A p p en d ix 1).

MMRuns and MMMean overestimated species

richness considerably for most parameter com-binations, especially in 250- an d 500-species

pools, but overestimation decreased with in-

creasing sample size. Contrarily, with 20-spe-

cies lists Chao 1 and ACE (and to a lesser de-

gree Chao 2 and ICE) prod uced h ighly accurate

estimates with little or no sensitivity to sample

size, which contrasted with the performance of

those estimators in simulated exponential de-

cline comm u nities. Jackknife 1 and 2 and boot-

strap performed well in 50-species pools, but

were mod erately to h ighly sensitive to samp le

size in the remaining species pools. Thus, all

estimators were sensitive to d ifferences in com-

munity structure. That was most pronounced

in estimates of MMRuns and MMMean, which,

for the same parameter combination, were as

much as 3.5 and 3.1 higher, respectively, in

evenly distributed species pools (10 5-species

lists in 250-species pools for MMRuns, 10 5-

sp ecies lists in 100-sp ecies p ools for MMMean ).

Effect iveness of list siz e. For 50-species p ools,

68.2 2.6 5-species lists, 25.9 1.1 10-species

lists, and 6.7 0.7 20-species lists were com-piled from 400 consecutive observations. For

500-species pools, the respective values were

79.4 0.3, 39.3 0.2, and 19.2 0.1. Because

100 5-species lists achieved accuracies very

similar t o th ose of 50 10-species lists and 10 or

20 20-species lists for any given species pool

and estimator (Appendix 1), 5-species lists

made the most effective use of the raw data

without being more biased than either 10- or

20-species lists when excluding the extremely

sma ll samp le size of 10 5-species lists.Figure 1 illustrates estimator performance

for 10-species lists. An ideal estimator would

have a level surface with accuracy close to

100%. Surfaces of bootstrap, jackknife 1, and

jack kn ife 2 clo se ly p aralle led th at of Sobs

but at

high er accura cy levels and , for jackkn ife 2, w ith

less sensitivity to species richness, illustrating

the h igh sensitivity to sample size of those es-timators. The similar surfaces of ICE, ACE,

Chao 1, and Chao 2 had their largest deviation

from the ideal estimator along the transition

from small to moderately species-rich com-

mu nities. MMRun s and MMMean had the per-

h a p s m o st lev el s u r fa ce s, b u t w i th a p r o-

nounced tendency towards low accuracy in

small species pools for small sample size.

EMPIRICAL DAT A

At San ta Cr u z, 91 species (65.0% of the actu al

species richness) were observed and 85 5-spe-

cies lists, 41 10-species lists, and 18 20-species

lists were compiled. At Asunta Pata, 224 spe-

cies (74.2% of the actual species richn ess) were

recorded and 324 5-species lists, 157 10-species

lists, and 74 20-species lists were comp iled.

Jackknife 2 overestimated richness by 6.3% at

Asunta Pata, but all other estimators underes-

timated actual species richness to varying de-

grees (Table 1). Overall accuracy (mean SD

of all estimates from Table 1 for each statistic)

was (1) MMRuns, 76.6 9.1%; (2) jackknife, 2

76.3 20.8%; (3) MMMean, 74.6 8.8%; (4)

Ch ao, 2 71.6 15.5%; (5) ICE, 70.4 14.0%; (6)

jack kn ife, 1 67.1 20.3%; (7) ACE, 59.5

17.0%; (8) Cha o, 1 59.4 17.8%; (9) bootstrap,

57.7 19.4%; and (10) S obs, 49.5 18.2%.

Figures 2 and 3 illustrate the performance of

seven estimator s for 10-species lists. Curves of

MMRun s (except for an initial spike) and Ch ao

1 resembled those of MMMean and ACE, re-

spectively, and are not shown. ACE, jackknife 1and 2, and bootstrap mostly paralleled the S

obs

curve (Figs. 2A and 3A), illustrating their high

sensitivity to samp le size, wh ich w as invariably

higher at Asu nta Pata. Here, both jackknife sta-

tistics and bootstrap performed worse than Sobs

,

increasing in accuracy from 10 lists to maxi-

mum sample size by 65%, 55%, and 40%

with 5-, 10-, and 20-species lists, respectively

(Table 1). Those estimators also were the most

sensitive at Sant a Cru z with only slightly better

performance than S obs. MMRuns and MMMeanwere by far the least sensitive to sample size at

both sites; for 5-species lists at Santa Cru z, ac-


8/21


FIG . 1. Performance of (A) Sobs

and (BJ) nine sta-

t is t ics i n e s ti m at in g s p e ci es r ich n e ss u s i n g t h e

MacKinnon a nd Phillips (1993) species-list ap pr oach

(10 species per list) for four computer-simulated spe-cies pools (50, 100, 250, 500 species) with species-

abund ance distributions conforming to the exponen-

tial decline model. Accuracy values are percentage of

the predetermined species richness. Open circles

represent raw data points (mean of 40 replicationswith 50 randomizations each). Surfaces were fitted

by d istance-weighted least-squares smooth ing.

curacy d ecreased by 13.6 and 5.0%, resp ective-

ly, from 10 to 85 lists. At Asunta Pata, MMRuns

p erformed similarly w ith 5-species lists, except

that after dropping to a minimum at interme-

diate sample size accuracy increased again,

w h e re as M MMe an a ccu r a cy g r ad u a lly in -creased with sample size (Table 1). With 10-

and 20-species lists, both MM estimators were

highly insensitive to sample size at Santa Cruz

(see Fig. 2B for MMMean); at Asunta Pata,

however, accuracy increased by approximately

2025% from 10 lists to maximum sample size

(see Fig. 3B for MMMean). ICE and Chao 2 p er-

formed better than their respective abundan ce-

based counterpart, but both had inconsistent

estimates at small sample size, were overall

more sensitive to sample size than MMMean,

and ICE overestimated species r ichness after

t w o a n d t h r ee l is t s a t Sa n t a C r u z (Fi gs . 2 B

and 3B).

The MM statistics outperformed all other es-

timators also with respect to sensitivity to sp e-

cies richness, computing nearly identical accu-

racy at both sites for maximum sample size,

whereas the remaining estimators had consid-

erably higher final accuracy at Asunta Pata

(Table 1). Contrarily, for small and mediu m

samp le size, all estimators compu ted h igher ac-

curacy at Santa Cr uz . Averaging th e three max-imu m samp le size accuracy values for each site

and estimator, ICE (19.9% higher accuracy at

Asun ta Pata), jackknife 2 (18.9%), and ACE

(18.2%) had highest sensitivity to species rich-

ness. List size had little or no influence on pre-

dicted sp ecies richness for m aximum sample

size except in Ch ao 2 and jackknife 2 at Santa

Cruz. Here, 10-species lists resulted in 5.7 and

4.3% h igher accur acy, r espectively, th an 5-spe-

cies lists.

Standard deviations of r ichness estimateswere compu ted by ESTIMATES only for Sobs

and five estimators (Table 1). Standar d devia-

tions mostly d ecreased with increasing sam ple

size, and Sobs

, ACE, and ICE returned no stan-

dard deviation at m aximum sample size. Sobs

and jackknife 1 computed low values, whereas


9/21


those of Chao 2 were h ighest. ICE and Chao 2

had higher standard deviations than their re-

spective abund ance-based counterpar t.

We consider MMMean the overall most ro-

bust estimator (see below), but two contrasting

patterns in its performance for simulated ver-s u s e m p ir ica l d a t a r e qu i r e fu r t h er a n a ly s is .

First, accuracy steadily increased w ith sam ple

size in small species pools but quickly reached

an asymptote in large species pools for simu-

lated d ata set s (Fig. 1J, App end ix 1). For emp ir-

ical data, t he situation was reversed: accuracy

quickly reached an asymptote at Santa Cruz

(Fig. 2B), but still ascended at maximum sam-

ple size at Asunta Pata (Fig. 3B). Second, ac-

curacy increased with species-pool size for

simu lated d ata (Fig. 1J, App end ix 1) and hencetended to inflate true differences in species

richness. For empirical data, respective esti-

mates w ere lower for th e more sp ecies-rich data

set for small to intermediate sample size; for

maximu m sam ple size, however, MMMean had

a slightly higher accuracy at Asu nta Pata than

at Santa Cru z. Becau se accuracy still increased

at maximum sample size (Fig. 3B), continued

surveying at Asun ta Pata pr obably would have

resulted in a still higher final accuracy value.

Such contrasts in performance could be re-

lated to differences in commun ity structure.

The analysis of simulated pools w ith evenly

abund ant species shows that all statistics, and

especially the MM estimators, are sensitive to

the data sets und erlying species d istribution.

Because MMMean estimates are based on the

incidence of sp ecies in samples, we plotted

rank-frequency curves for 10-species lists for

three simulated and both empirical data sets

(Fig. 4A, B). Commu nity structure at Santa

Cruz was intermediate between the structure

of mod el commun ities of 100 and 250 sp ecies,but it approximated that of 250-species pools

more than that of 100-species pools (Fig. 4A).

Because estimates quickly reached an asymp-

tote for both Santa Cru z and 250-species po ols,

it could be conclud ed th at MMMean is sensitive

to sample size when d ata sets have many high-

frequency an d few low-frequency sp ecies, such

as in 100-species pools (Fig. 4A). However, the

Asunta Pata curve (Fig. 4B) closely matched

that of 250-species pools, except for a longer

tail of low-frequency species, but despitethat similarity, Asunta Pata estimates did not

reach an asymptote quickly. A long tail of low-

f re qu e n cy s p e cie s a lo n e d o e s n o t a p p e a r t o

cause the poorer estimator performance be-

cause such a tail also characterized the 500-spe-

cies-pool curve (Fig. 4B).

DISCUSSION

ST ANDARDIZ AT ION OF T HE SPECIES-LIST

M ETHOD

The species-list method is a useful technique

for rapid assessments of sp ecies r ichness in

trop ical bird comm u nit ies. Fjeldsa (1999) u sed

a s im i la r r a nd o m w a lk a p p r oa ch in h u m i d

montane forests of Tanzania an d found highly

significant correlations between random walk

and point-count data sets and concluded that

rand om walking d oes not appear to be more bi-a se d t h a n o t h er o b se rva tio n a l m e t h od s . It s

main advantages are time efficiency (the entire

daylight period may be used to generate data)

and relative observer independ ence compared

to any timed species-count method (e.g. point

cou n ts ; Fjeld sa 1999).

We agr ee w ith Pou lsen et al. (1997a) that the

method is suitable for appraising (1) the mag-

nitude of the species richness, (2) when a lo-

cality h as been ad equately samp led, and (3) for

determining the relative abundance of eachspecies. However, comparisons of relative

abundances should be made only within spe-

cie s (e.g . a cr o ss s it es o r s ea so n s) b e cau s e

across-species comparisons are hampered by

considerable differences in detectability be-

tw een sp ecies (Karr 1971, 1981, 1990; Boulin ier

et al. 1998). For the same reason, opposed to

Poulsen et al. (1997a), we consider data gath-

ered w ith the species-list method as un suitable

for calculating ind ices of-diversity. As a cau-

tionary note, we recommend further testing of

the accuracy of relative abundance estimatesobtained with the species-list method in com-

p a r is on w i t h p o i n t-co u n t a n d s p o t -m a p p i n g

data especially in Neotropical forests, and we

consider that only spot-mapp ing (Kendeigh

1944, Bibby et al. 2000) during longer field ses-

sions will yield reliable measurements of ab-

solute abundance for most tropical forest birds

(see Terborgh et al. 1990, Robinson et al. 2000).

A number of recommendations for the stan-

dard ization of the species-list method are sug-

gested here (see also above). The method is byno m eans fool proof and a certain level of ex-

perience with visual and vocal identification of


10/21


11/21


TABLE 1 . C on t in u e d .

Estimator

Numberof lists

Santa Cruz (140 species)

5-species lists 10-species lists 20-species lists

Asunta Pata (302 species)

5-species lists 10-species lists 20-species lists

50100Ma xa

81.4

80.7

80.7

80.7

65.269.982.1

69.976.882.5

77.882.8

MMMean

102050

100Ma xa

85.785.081.4

80.7

79.380.7

80.7

80.0

80.7

62.364.264.969.982.1

57.361.969.576.882.5

61.367.577.582.8

a Maximum num ber of lists comp iled: Santa Cr uz 85 5-species, 41 10-species, 18 20-species lists; Asun ta Pata 324 5-species, 157 10-species,

74 20-species lists.

most species present is necessary. Data collect-ed by an observer largely unfamiliar with a

given avifauna are not comparable with those

obtained by experienced observers. It is crucial

that names are assigned to species not confi-

dently identified (by sight or voice) at first

(Poulsen et al. 1997a). The extensive use of a

tap e recorder is ind ispensable (Parker 1991); re-

c o r d i n g s s h o u l d b e m a d e o f d a w n c h o r u s e s ,

mixed-species flocks, ind ividu al birds, and all

unfamiliar vocalizations and integrated into

the consecutive master list by an expert after athorough revision using reference recordings.

Five-, 10-, or 20-species lists should be com-

piled only during data analysis to ensure flex-

ibility (see Pou lsen et al. 1997a). Because n ot all

species are active at the same time of day, sur-

veys should cover most of the daylight period

a s w ell a s d aw n , d u s k , a n d e ar ly ev en in g .

Mixed -species flock encoun ters shou ld be treat-

ed like other observations and detected indi-

viduals should be recorded consecutively, al-

though especially in larger flocks the number

of conspecific flock members may not be as-

sessable for som e sp ecies at first. In such cases,

we sug gest assigning ad ditional ind ividuals to

the same species list as the first individu al of

that species.

Because the detectability of most forest bird

vocalizations drops considerably at distances

of 50 m (Schieck 1997), observations beyond

that d istance should be exclud ed from th e ana l-

ysis. In our experience, this results only very

occasionally in a rare but loud species having

to be dropped entirely from the analysis be-cause it never w as recorded within 50 m of the

observer, but it avoids filling species lists with

comm on, noisy birds (e.g. Screaming Piha [Li-paugus vociferans]) and overestimating their rel-

ative abundance. If longer time p eriods are

spent in one spot or wh en resampling a given

section of the stud y area, repeated coun ts of ob-

viously territorial individuals should be avoid-

ed, because this also tend s to overestimate rel-

ative abund ances of those species. In sexually

dimorphic species such as m any antbirds, the

male and the female in a given territory may

each be counted once. Obviously, it will occa-

sionally be difficult to d etermin e wheth er a ter-ritorial bird has already been counted, and we

have no ready solution to that problem except

not to su rvey sections of a stud y area more than

once, but in most cases that will be impractical

or even impossible. Because activity and de-

tectability levels of most species show diu rnal

variations (Blake 2000), resam pling shou ld id e-

ally be carried out at a different time of day

than previous surveys in the same section to

minimize probability of encountering th e same

individuals more than once. When in doubt, it

might be best to adopt a conservative approach

and omit a given observation from th e analysis.

The size and, in m oun tains, elevational range

of survey areas should be held constant or, if

difficult logistically, should at least be quanti-

fi ed ; t h e s am e a p p lie s t o h a bit at d i ver sit y

(Remsen 1994, Cohn -Haft et a l. 1997). If, for ex-

ample, species richness in Bolivian dry forests

is to be assessed, data from nonzonal vegeta-

tion ty p es (e.g. gallery forest) mu st be excluded

because nonzonal habitats tend to be repre-

s en t e d u n e ve n ly a t d i ffe re n t s it es a n d t h u si n flu e n ce t o ta l s p e cie s r ich n e ss t o v ar y in g

degrees (Herzog and Kessler 2002). The dis-


12/21


FIG . 2. Performance of seven estimators ofsp ecies

richness for an empirical bird data set collected with

the species-list method at the Botanical Garden of

Santa Cruz de la Sierra, Bolivia. (A) ACE, jackknife

1, jackknife 2, boo tstra p. (B) ICE, Chao 2, MM Mean.Curves of MMRuns (except for an initial spike) and

Chao 1 resembled those of MMMean and ACE, re-

spectively, and are not shown. Values are percentage

of the tot al sp ecies richn ess (140) at the site. The spe-

cies accumulation curve (Sobs

) and estimator curves

indicate accuracy as a function of the nu mber of 10-

species lists. Sample accumu lation order of all cur ves

was randomized 50 times, and each point represents

the mean of the resulting 50 estimates.

FIG . 3. Performan ce of seven estimators of species

richness for an empirical bird data set collected with

the sp ecies-list method at Cerro Asu nta Pata, Bolivia.

(A) ACE, jackknife 1, jackknife 2, bootstrap. (B) ICE,

Chao 2, MMMean . Curves of MMRu ns (except for anin it ia l s p ik e) a nd C ha o 1 r es em b le d t ho se o f

MMMean and ACE, respectively, and are not show n.

Values are percentage of the total species richness

(302) at the site. The species accumu lation curve (Sobs

)

and estimator curves ind icate accuracy as a function

of the num ber of 10-species lists. Samp le accum ula-

tion order of all curves was randomized 50 times,

and each p oint represents the m ean of the resulting

50 estimates.

tinction of the avifauna into core and noncore

species (Remsen 1994) is ideal but often not

achievable to a satisfactory degree unless an

observer is very fam iliar w ith a given habitat or

area. However, if survey du rations are similar,

proportions of noncore species should also be

comp arable. Obvious noncore sp ecies (e.g. a

heron in flight over dry forest) should always

be excluded.

To standardize survey effort and to deter-

mine equal stopp ing p oints for data collection,

we suggest using the Chao 1 estimator in the

field by comparing estimated with observedspecies r ichness. That should be done every

evening until the observed species richness is

90% of the respective Chao 1 estimate. Al-

though not the most robust estimator (see be-

low), Chao 1 has a practical advantage over all

other estimators: it can be applied directly to

the raw d ata without the time consuming sub-

division of observations into species lists, and

it is so simple that it can be calcu lated by hand .

For practical reasons, we include the Chao 1

formula:

2S S F / 2FChao1 obs 1 2 (2)

where Sobs

is the number of species observed, F1

the nu mb er of singletons (species with on ly oneindividual), and F2 the number of doubletons

(species with exactly two individuals) in the


13/21


FIG . 4. Frequency-distribution graphs for 10-spe-

cies lists for simulated (100-, 250-, 500-species pools,

n 50 lists per p ool) and em pirical (Santa Cruz, n

41; Asun ta Pata, n 157) data sets. Simulated data

sets were drawn from a species-abundance distri-bution conforming to the exponential decline model,

and each point represents the mean of 40 replica-

tions. (A) Commu nity stru cture and species richness

at Santa Cruz is intermediate between that of simu-

lated 100- and 250-species p ools. (B) Comm un ity

structure at Asunta Pata is similar to that of simu-

lated 250-species pools except for a longer tail of

low-frequency species.

data set (Chao 1984, Chaz don et al. 1998). This

simple measure can also be applied to other

coun t and censu s method s (e.g. point counts) to

monitor the completeness of a survey.

Finally, the species-list method is entirely

compatible with p oint count s (or line transects)

if point-count observations are recorded in the

same consecutive order as in the species-list

approach. Thus, where logistical conditions

permit, experienced observers familiar with a

given bird commu nity could u se both methods

in conjunction, that is, record point-count data

consecutively to allow su bd ivision into species

lists and use the species-list approach betweenpoint-count intervals. That method combina-

tion w ould gain the advantages of r igid stan-

dardization of point counts while still main-

taining the flexibility of species lists.

U SE OF ESTIMATION M ETHODS

Performan ce varied considerably betweenthe nine estimators and between simulated and

empirical data sets for most estimators. Boot-

strap, jackkn ife 1, and jackknife 2 performed so

poorly with simulated (Fig. 1, Appendices 1

and 2) and empirical (Table 1, Figs. 2 and 3)

data sets that they will not be considered any

furt her. The rema ining estimators basically fall

into two grou ps: the first includ es the nonpara-

metr ic statistics ACE, ICE, Chao 1, and Chao 2

(Fig. 1BE), and the second includes the two es-

timators based on the Michaelis-Menten model(Fig. 1IJ). With simu lated exponent ial decline

and empirical data sets, all estimators of the

first group had lower overall accuracy than the

MM estimators. However, accuracy as su ch is

not the key p arameter because an estimator

with low but constant accuracy regardless of

variation in species r ichness or sample size

would be easy to calibrate. Thus, sensitivity to

both sample size and species richness are more

crucial.

With respect to sample size, MMMean and

when exclud ing the smallest sample size in

500-species poolsMMRuns again performed

better t han either ACE, ICE, Chao 1, or Chao 2

with both simulated exponential decline (Fig.

1, Appendix 1) and empirical (Table 1, Figs. 2

and 3) data sets. In contrast, regarding sensi-

tivity to species richness, ACE, ICE, Chao 1,

and Chao 2 obtained better results for simu lat-

ed d at a t ha n b ot h M M est im at or s, a nd

M M Me an p e r fo r m ed b e tt er t h a n M M Ru n s .

However, the MM estimators clearly outper-

formed all other statistics with empirical data(Table 1, Figs. 2 and 3).

Thus, with simulated exponential decline and

empirical data sets MMMean and MMRu ns were

the overall most robust of the nine estimators

tested h ere. Although both generally obtained

very similar results, MMRuns was m ore sen-

sitive to species richness with simulated data

than MMMean and also tended to be more sen-

sitive to sam ple size for 5-species lists with spe-

cies-rich data sets (Appendix 1). We therefore

consider MMMean as the overall least biased,preferable estimator, despite its poor perfor-

m a n ce i n m o d e l co m m u n i tie s w i t h e qu a l ly


14/21


15/21


communities (Keating and Quinn 1998), on the

basis of our resu lts we nonetheless consider it

a useful and promising technique for analyzing

data gathered in diverse tropical bird commu-

nities with the species-list method. This ap -

proach to assessing species richness is doubt-lessly much more standard ized and valuable

than simple, one-dimensional locality lists ac-

companied by some measure of survey effort

(e.g. see Remsen 1994, Cohn-Haft et al. 1997).

The lack of a stan dar d deviation, however, may

be problematic when absolute values of esti-

mated species richn ess have to be comp ared d i-

rectly between sites or b etween su rveys of the

same site (e.g. before and after selective log-

ging). In this case, Chao 2 may be used provid-

ed that sample size is similar.However, two basic aspects need to be con-

s id e r ed t o e n su r e co m p a r ab il it y o f r e su l t s.

First, when comparing MMMean (or Chao 2)

estimates between sites, a close inspection of

curve shap es is required. If some or all curves

do not r each a n asy mp tote qu ickly (i.e. after 10

to 15 lists), a standardized cut-off point must

be determined to control for the confounding

effects of samp ling effort (see Gotelli and Col-

well 2001). Rather than taking the estimated

richness after a given number of individuals

sampled as proposed by Gotelli and Colwell

(2001; sample-based rarefaction with a rescaled

x-axis from samples to individuals), we sug-

gest determining the cut-off point from the re-

lation between the Sobs

and the MMMean curve

by expressing each Sobs

value as th e prop ortion

of the respective MMMean value. Our Santa

Cru z d ata set contained 41 10-species lists (Ta-

ble 1), and after 41 lists the Sobs

value constitut-

ed 80.2% of the MMMean value. At Asunta

Pata, the equivalent cut-off point is found after

78 10-species lists, where the S obs value consti-tut ed 80.1% of the MMMean value. The resp ec-

tive richn ess estim ates were 113 species at San-

ta Cruz and 224 at Asunta Pata, or 80.7 and

74.2% of each sites actual species richness, re-

spectively. Although a smaller difference be-

tween the two values is desirable (ideally they

should be equal), it is lower than most Chao 2

s t an d a r d d e v ia tio n s fo r t h e e m p ir ica l d a t a

(Table 1).

In this particular case, determining the cut-

off point based on a given number of individ-uals after rescaling th e x-axis from samples to

individuals (Gotelli and Colwell 2001) obtains

a nearly identical result (estimates of 113 spe-

cies at Santa Cruz and 223 species at Asunta

Pata). However, tropical habitats often contain

one to several species occurring in large num -

b er s i n i n tr a sp e ci fic a g g re ga tio n s, s u ch a s

flocks of parakeets or swifts. The presence ofsuch species can lead to a rapid and dramatic

i ncr ea se i n t h e n u m b er o f in d i vid u a ls o b-

served, which in turn will bias the rarefaction

based on a rescaled x-axis causing an artificial

inflation of survey effort. Preliminary analyses

indicate that the more abund ant the most com-

mon species, the lower will be the estimate ob-

tained by individu al-based rarefaction when

compared to rarefaction based on the relation

between the Sobs

and the MMMean curve, and

that relationship ap pears to be significant.Second, we strongly advocate the use of a

standard list size unless it can be shown that

that leads to m ore biased resu lts than other list

sizes for a given data set. In species-poor hab-

itats, 20-species lists often are too long (Poulsen

et al. 1997a), and samp le size can b e very sma ll

with 20-species lists when the size of the sur-

vey area is limited (S. K. Herzog pers. obs.). Es-

timator curves for 5-species lists h ad several

basic shapes (asymptotic, gradual increase,

minimum at intermediate sample size, initialspike), whereas curves based on 10-sp ecies lists

were less variable. Ten-species lists therefore

appear to be a good intermediate solution.

Conclusion. We b el iev e t h at w i t h a m i n i-

m u m d e g r ee o f s t a n d a rd i z at io n a n d ca r efu l

data analysis, the species-list method and the

quantitative comparison of the resulting spe-

cies accumulation curves are useful tools for

both conservation assessment and the stud y of

avian sp ecies richn ess pattern s in the tr opics. It

can be argued that it simply is the nature of es-timation and extrapolation that obtained values

are highly sp eculative and that comparisons of

estimates are not reliable. From a conservation

viewp oint it mu st be emphasized, though, that

comp lete avifaunal inventories in the tropics

simp ly cann ot be cond ucted as often as they are

required and that standardized survey meth-

ods employed in temperate regions are often

difficult or impossible to apply. Rapid assess-

ments are a necessity dictated by the urge for

conservation action and by limitations of time,personnel, and fund ing. Here, the u se of statis-

tical estimators will certainly aid in comparing


16/21


different stud ies and in mak ing more informed

conservation decisions.

ACKNOWLEDGMENTS

This paper benefited from discussions with R. K.

Colwe ll, B. Fisher , J. Fjeldsa, R. M acLeod , C. Rah bek ,

J. Umbanhowar, and B. A. Walther. We are grateful

to S. W. Car diff, S. E. Davis, and J. V. Remsen for

sharing their u npublished data, to S. Mayer and B.

M. Whitney for help with iden tification of my stery

tape recordings, and to R. K. Colwell for making ES-

TIMATES available to a w ide au d ience. S.K.H. w ould

like to than k F. Bairlein for guid ance and sup por t. F.

Bairlein , J. Fjeld sa , R. Ma cLeod , P. C. Stou ffer, an d a n

anonymous reviewer made valuable comments on

earlier drafts of this paper. Financial support was

provided to S.K.H. by the German Academic Ex-

change Service (DAAD, 1998), Fauna and Flora In-tern ational (1998), the DIVA pr oject u nd er the D an-

i sh En v ir o n m en t a l P r og r a m m e ( 19 97 ), a n d t h e

Gesellschaft fu r Tropen orn ithologie (1996), and to

M.K. by the German Research Council (Deutsche

Forschungsgemeinschaft) and the A.F.W. Schimper-

Stiftung (1995).

LITERATURE CITED

BALTANAS, A. 1992. On the u se of some meth ods for

the estimat ion of sp ecies richness. Oikos 65:484

492.

BIBBY, C. J., N. J. C OLLAR, M. J. C ROSBY, M. F. H EATH ,C. IMBODEN , T . H . JO H N S O N , A. J. LON G, A. J.

STATTERSFIELD, AN D S. J. THIERGOOD. 1992. Pu t-

ting biodiversity on the map: Priority areas for

global conservation. ICBP, Cambrid ge, United

Kingdom.

BIBBY, C. J., D. A. H ILL, N. D. BURGESS, AN D S. MUS-

TO E. 2000. Bird Census Techniques. Academ ic

Press, London .

BLAKE, J. G. 2000. Temp oral var iation in p oint coun ts

of birds in a lowland wet forest in Costa Rica.

Condor 94:265275.

BLAKE

, J. G., F. G. STILES

,AN D

B. A. LOISELLE

. 1990.Bird s of La Selva Biological Station: Hab itat u se,

trophic composition, and migrants. Pages 161

182 in Four Neotropical Forests (A. H. Gentry,

Ed .). Ya le Un iv er si ty P re ss , N ew H a ve n,

Connecticut.

BOULINIER, T., J. D. N ICHOLS, J. R. SAUER, J. E. H INES,

AN D K . H . POLLOCK. 1998. Estimating species

richness: The importance of heterogeneity in

species detectability. Ecology 79:10181028.

BU N G E, J., AN D M. FITZPATRICK. 1993. Estimating t he

n u m b e r o f s p e ci es : A r ev ie w. Jo u r n a l o f t h e

American Statistical Association 88:364373.

BU D N E Y, G. F., AN D R. W. GROTKE. 1997. Techniquesfor audio recording vocalizations of tropical

birds. Ornithological Monographs 48:147163.

BU R N H A M, K. P., AN D W. S. O VERTON . 1978. Estim a-

tion of the size of a closed p opulation when cap -

ture probabilities vary among animals. Biome-

trik a 65:623633.

BU R N H A M, K. P., AN D W. S. O VERTON . 1979. Robust

e st im a t io n o f p o p u l at io n s iz e w h e n c ap t u r e

probabilities vary among animals. Ecology 60:

927936.

CH AO , A. 1984. Non-parametric estimation of the

num ber of classes in a popu lation. Scandinavian

Journal of Statistics 11:265270.

CH AO , A. 1987. Estimating the population size for

capturerecapture data with unequal catchabi-

lity. Biome trics 43:783791.

CH AO , A., M.-C. M A, AN D M. C. K. YAN G. 1993. Stop -

ping ru les and estimation for recapture debug-

ging with unequal failure rates. Biometrika 80:

193201.

CH A Z D O N , R. L., R. K. COLWELL, J. S. D ENSLOW , AN DM. R. GUARIGUATA . 1998. Statistical me thod s for

estimating species richness of woody regenera-

tion in primary an d secondary rain forests of NE

Costa Rica. Pages 285309 in Forest Biodiversity

Research, Monitoring and Modeling: Concep-

tual Background and Old World Case Studies (F.

Dallmeier and J. A. Comiskey, Eds.). Parthenon

Publishing, Paris, France.

COH N -H AF T, M., A. W HITTAKER, AN D P. C. STOUFFER.

1997. A new look at the species-poor central

Amazon : The avifaun a n orth of Manaus, Brazil.

Orn ithological Monogr ap hs 48:205235.COLWELL, R. K. 1997. ESTIMATES: Statistical esti-

mation of species richness and shared species

from samples, version 5. [Online.] Available at

http :/ / viceroy.eeb.uconn.edu / EstimateS.

COLWELL, R. K., AN D J. A. CODDINGTON . 1994. Esti-

mating terrestrial biodiversity through extrap-

olation. Philosophical Transactions of the Royal

Society of London 345:101118.

DAL E, V. H., S. M. PEARSON , H. L. O FFERMAN , AN D R.

V. ON EILL. 1994. Relating patterns of land-use

chang e to faunal biodiversity in the central Am -

azon. Conservation Biology 8:10271036.

FJELDSA , J. 1999. The imp act of hu man forest d istur-

bance on the endem ic avifauna of the Udzu ngwa

Mountains, Tanzania. Bird Conservation Inter-

national 9:4762.

GOTELLI, N. J., AN D R. K. COLWELL. 2001. Quantify-

ing biodiversity: Procedures and pitfalls in the

measurement and comparison of species rich-

ness. Ecology Letters 4:379391.

GOTELLI, N. J., AN D G. R. GRAVES. 1996. Nu ll Models

in Ecology. Smithson ian Institution Press, Wash-

ington, D.C.

GRA M, W. R., AN D J. FAABORG. 1997. The distr ibution

o f N e o t r o p i c a l m i g r a n t b i r d s w i n t e r i n g i n E lCielo Biosphere Reserve, Tamau lipas, Mexico.

Condor 99:658670.


17/21


H ASELMAYER, J., AN D J. S. Q U I N N . 2000. A compari-

son of point counts and sound recording as bird

survey methods in Amazonian southeast Peru.

Condor 102:887893.

H ELTSHE , J., AN D N. E. FORRESTER. 1983. Estimating

species richness using the jackknife procedure.

Biom etr ics 39:111.

H ERZOG, S. K., AN D M. KESSLER. 2002. Biogeogr ap hy

and composition of dry forest bird commu nities

in Bolivia. Jou rn al fu r Or nit holog ie 143:171204.

H OLMES, R. T., T. W. SHERRY, AN D F. W. STURGES.

1986. Bird community dynamics in a temperate

deciduous forest: Long-term trends at Hubbard

Brook. Ecological Monographs 56:201220.

H USTON , M. A. 1994. Biological Diversity. Cam -

bridge University Press, Cambridge, United

Kingdom.

JE N N I, L., M. LEUENBERGER, AN D F. RA M P A Z Z I. 1996.

Capture efficiency of m ist nets w ith commentson th eir role in th e assessment of passerine hab-

itat use. Journal of Field Ornithology 67:263274.

KAR R, J. R. 1971. Structure of avian commu nities in

selected Panam a and Illinois habitats. Ecological

Monogra ph s 41:207231.

KAR R, J . R. 1981. Surveying birds with mist nets.

Studies in Avian Biology 6:6267.

KAR R, J. R. 1990. Birds of tropical rainforest: com-

par ative biogeography and ecology. Pages 215

228 in Biogeography and Ecology of Forest Bird

Communities (A. Keast, Ed.). SPB Academic

Publishing, The Hague, The Netherlands.

KEATING , K. A., AN D J. F. Q U I N N . 1998. Estimatingspecies richness: The Michaelis-Menten model

revisited. Oikos 81:411416.

KENDEIGH , S. C. 1944. Measuremen t of bird popu la-

tions. Ecological Monog rap hs 14:67106.

LEE, S.-M., AN D A. CH AO . 1994. Estimating popu la-

tion size via sample coverage for closed cap-

turerecapture models. Biometrics 50:8897.

MAC KI N N O N , S., AN D K. P HILLIPPS. 1993. A Field

Guide to the Birds of Borneo, Sumatra, Java and

Bali. Oxford University Press, Oxford.

MAYER, S. 2000. Bird s of Bolivia 2.0. [CD-ROM.] Bird

Songs International B.V., Westernieland , TheNetherlands.

O NIKI, Y., AN D E. O. WILLIS. 1982a. Breeding records

of birds from Manaus, Brazil: Formicariidae to

Pipridae. Revista Brasileira de biologia 42:563

569.

O NIKI, Y., AN D E. O. W ILLIS. 1982b. Breeding records

of birds from Manaus, Brazil, II . Apodidae to

Furnariidae. Revista Brasileira de biologia 42:

745752.

O NIKI, Y., AN D E. O. W ILLIS. 1983. Breeding records

of birds from Manau s, Brazil, IV. Tyrann idae to

Vireonidae. Revista Brasileira de biologia 43:45

54.PALMER, M. W. 1990. The estim ation of species rich-

ness by extrapolation. Ecology 71:11951198.

PALMER, M. W. 1991. Estimating species richness:

The second-ord er jackknife reconsidered . Ecol-

ogy 72:15121513.

PARKER, T. A., III. 1991. On t he u se of tap e record ers

in avifaun al su rveys. Auk 108:443444.

PARKER, T. A., III, AN D B. BAILEY, EDS. 1991. A bio-

logical assessment of the Alto Madidi region

and adjacent areas of northwest Bolivia, May

18June 15, 1990. Conservation International,

RAP Working Papers, no. 1.

PARKER, T. A., III, A. C ASTILLO U., M. GELL -M AN N ,

AN D O. RO C H A . 1991. Records of new and un-

usual birds from northern Bolivia. Bulletin of

the British Orn ithologists Club 111:120131.

PARKER, T. A., III, A. H. GENTRY, R. B. FOSTER, L. H.

EM M O N S, AN D J. V. REMSEN , JR. 1993. The low-

land dry forests of Santa Cruz, Bolivia: A global

conservation priority. Conservation Internation-

al, RAP Working Papers, no. 4.PARKER, T. A., III, D. F. STOTZ, AN D J. W. FITZPATRICK.

1996. Ecological and distributional databases.

Pages 131436 in Neotropical Birds: Ecology and

Conservation (D. F. Stotz, J. W. Fitzpatrick, T. A.

Parker III, and D. K. Moskovits, Eds.). University

of Chicago Press, Chicago.

POULSEN , B. O. 1994. Mist-netting as a census meth-

od for determining richness and abundances in

an Andean cloud forest bird community. Ger-

faut 84:3949.

POULSEN , B. O., AN D N. KRABBE. 1998. Avifaunal di-

versity of five high-altitude cloud forests on th e

Andean western slope of Ecuador: Testing a rap-id assessment method . Journal of Biogeography

25:8393.

POULSEN , B. O., N. KRABBE, A . FRLANDER, M . H I-

NOJOSA B., AN D C. Q UIROGA O. 1997a. A rapid

assessment of Bolivian an d Ecuad orian montan e

avifaun as usin g 20-species lists: Efficiency, bi-

ases and data gathered. Bird Conservation Inter-

national 7:5367.

POULSEN , B. O., N. KRABBE, A . FRLANDER, M . H I-

NOJOSA B., AN D C. Q UIROGA O. 1997b. A note on

20-species lists. Bird Conservation International

7:293.RAAIJMAKERS, J. G. W. 1987. Statist ical an alysis of th e

Michaelis-Menten equation. Biometrics 43:793

803.

REMSEN , J. V., JR. 1994. Use and misuse of bird lists

in commu nity ecology and conservation. Auk

111:225227.

REMSEN , J. V., JR., AN D D. A. GOO D. 1996. Misuse of

data from mist-net captures to assess relative

abundance in bird populations. Auk 113:381

398.

REMSEN , J. V., JR., AN D T. A. P ARKER. III 1983. Con-

tribution of river-created habitats to bird species

richness in Am azon ia. Biotropica 15:223231.ROBINSON , W. D., J. D. BRAWN , AN D S. K. ROBINSON .

2000. Forest bird community structure in central


18/21


19/21


20/21


21/21

Documents

20 Species List