Page 1

Copyright © 2008 by the author(s). Published here under license by the Resilience Alliance.

Scott, J. M., F. L. Ramsey, M. Lammertink, K. V. Rosenberg, R. Rohrbaugh, J. A. Wiens, and J. M. Reed.

2008. When is an “extinct” species really extinct? Gauging the search efforts for Hawaiian forest birds and

the Ivory-billed Woodpecker. Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2):

3. [online] URL: http://www.ace-eco.org/vol3/iss2/art3/

Research Papers

When is an “Extinct” Species Really Extinct? Gauging the Search Efforts

for Hawaiian Forest Birds and the Ivory-Billed Woodpecker

Quand une espèce "disparue" l'est-elle vraiment? Évaluation des efforts

de recherche sur les oiseaux forestiers hawaïens et le Pic à bec ivoire

J. Michael Scott1, Fred L. Ramsey2, Martjan Lammertink3, Kenneth V. Rosenberg3, Ron Rohrbaugh3,

John A. Wiens4, and J. Michael Reed5

ABSTRACT. Rare species, particularly those in inaccessible habitat, can go years without being observed. If we are

to allocate conservation resources appropriately to conserving such species, it is important to be able to distinguish

“rare” from “extinct.” Criteria for designating extinction, however, tend to be arbitrary or vaguely defined. This

designation should not be made unless the search effort has been sufficient to yield a high degree of confidence that

the species is in fact absent. We develop models to assess the probability of extinction and the search effort necessary

to detect an individual in a small population. We apply these models to searches for nine potentially extinct Hawaiian

forest birds and for the Ivory-billed Woodpecker (Campephilus principalis) in intensively searched areas in Arkansas.

The Hawaiian forest bird survey was extensive, providing excellent information on population sizes and habitat

associations of species encountered during the survey. Nonetheless, we conclude that the survey effort was not

sufficient to conclude extinction (p > 0.90) for populations of 10 or fewer individuals for those species that were not

encountered during surveys. In contrast, our analysis for Ivory-billed Woodpeckers suggests that, unless there were

actually two or fewer birds present, the search effort was sufficient to conclude (p > 0.95) that Ivory–billed

woodpeckers were not present in the intensively searched area. If one assumes distributions other than uniform, there

is a greater chance that Ivory-billed Woodpeckers may persist in the intensively searched areas. Conclusions regarding

occupancy of suitable habitat throughout the rest of the former range will require similarly intensive survey efforts.

The degree of confidence in the absence of the Ivory-billed Woodpecker depended in part on our assumptions about

the distribution of birds in the search area. For species with limited detection distance and small populations, a

massive search effort may be required to conclude with confidence that a species is unlikely to be extant.

RÉSUMÉ. Les observations d’espèces rares peuvent être séparées par plusieurs années, particulièrement lorsque

leur habitat est inaccessible. Si des ressources sont consacrées à la conservation de ces espèces, il est important de

pouvoir établir des critères propres aux espèces « rares » et « disparues ». Par contre, les critères utilisés pour

déterminer l’extinction tendent à être arbitraires et plutôt vagues. Cette désignation ne devrait jamais être appliquée,

à moins que les efforts de recherche aient été suffisants pour affirmer avec confiance que l’espèce est effectivement

absente. Nous avons développé des modèles afin d’estimer la probabilité d’extinction ainsi que l’effort nécessaire

pour la détection d’un individu provenant d’une petite population. Nous avons appliqué ces modèles à neuf espèces

d’oiseaux forestiers potentiellement disparus de Hawaï ainsi qu'au Pic à bec ivoire (Campephilus principalis) dans

des sites de recherche intensive de l’Arkansas. L’inventaire des oiseaux forestiers d’Hawaï était extensif, ce qui nous

a procuré de l’excellente information sur l’effectif des populations ainsi que l’association aux habitats des espèces

rencontrées durant l’inventaire. Néanmoins, nous estimons que l’effort d’inventaire était insuffisant pour conclure

à l’extinction (p > 0.90) chez les populations comprenant 10 individus ou moins parmi les espèces non détectées

durant l’inventaire. À l’opposé, notre analyse pour le Pic à bec ivoire suggère qu'à moins que la population ne compte

que deux individus ou moins, l’effort de recherche était suffisant (p > 0.95) pour conclure à l’absence de l'espèce

dans le site de recherche intensive. Lorsqu’on suppose que la distribution de l’espèce n’est pas uniforme, la probabilité

de persistance du Pic à bec ivoire est plus élevée dans les sites de recherche intensive. Les conclusions sur l’occupation

de l’habitat favorable dans le reste de l’aire de répartition exigeront des efforts d’inventaire d’intensité comparable.

Le degré de confiance quant à l’absence du Pic à bec ivoire dépendait en partie de nos suppositions par rapport à la

1University of Idaho, 2Oregon State University, 3Cornell University,

4PRBO Conservation Science, 5Department of Biology, Tufts University

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

distribution de l’espèce dans l’aire d’étude. Pour les espèces dont la distance de détection est limitée ainsi que pour

les petites populations, un effort de recherche intensif peut être requis afin de conclure avec confiance que la présence

de l’espèce est improbable.

Key Words: Endangered Species Act; extinction; Hawaii; monitoring; survey effort

INTRODUCTION

“Extinction is forever.” True enough. For some

species, the moment of extinction is known with

certainty. The last three individuals of the Laysan

Honeyeater (Himatione sanguinea frathii) perished

in a dust storm in 1923 (Bailey 1956) and the last

Passenger Pigeon (Ectopistes migratorius) died in

the Cincinnati Zoological Garden in 1914

(Blockstein 2002). More often, however, species are

declared “extinct” because recorded sightings

became less frequent and then ceased altogether.

But how can one be sure that a species is really

extinct, instead of persisting undetected for years or

decades? This problem is exacerbated if the species

is cryptic or lives in dense or inaccessible habitat.

Yet the decision about whether to declare a species

extinct vs. “endangered” or “possibly extinct” is

critical. In the United States, for example, legal

protection of a species and its habitat ceases when

the species is declared extinct. If the species is later

discovered to be extant, the loss of critical habitat

in the interim may hasten its eventual demise and

preclude recovery efforts.

Rediscovery of “extinct” species occurs often

enough to give one pause about making premature

pronouncements. The Bermuda Petrel (Petrodroma

cahow) was considered extinct by 1621, but a small

breeding colony was discovered in 1951 (Murphy

and Mowbray 1951). Gurney’s Pitta (Pitta gurneyi)

in Myanmar (Eames et al. 2005), a freshwater pearl

mussel (Margaritifera margaritifera) in Portugal

(Reis 2003), the Cherry-throated Tanager (Nemosia

rourei) in Brazil (Bauer et al. 2000), Bulmer’s fruit

bat (Aproteles bulmerae) in Papua New Guinea

(Flannery 1994), Wollemi pine (Wollemia nobilis)

in Australia (Woodford 2002)—all were thought to

be extinct for decades or known only from fossil

remains until rediscovered in recent years. In fact,

the phenomenon of prematurely declaring species

extinct is common enough to have been christened

“Romeo’s error” (Collar 1998) and the “Lazarus

effect” (Keith and Burgman 2004).

The World Conservation Union defines a species as

extinct if there is “no reasonable doubt that the last

individual has died” (IUCN 2001). Determining the

point of “no reasonable doubt,” however, is tricky.

Although the rediscovery of some presumably

extinct species has happened serendipitously, more

often it is associated with a focused search effort.

When systematic searches occur, the effort required

to find a rare species with low detectability can be

substantial (Scott et al. 1986, Solow 1993, Reed

1996). For example, over a 6-week period in 2006,

a team of scientists used sophisticated technology

to search 3500 km of the Yangtze River in China in

an attempt to find the Yangtze River dolphin

(Lipotes vexillifer). Their failure led to the decision

in August 2007 to declare the species officially

extinct (Turvey et al. 2007). But, even though the

search effort was intensive and one would think an

animal the size of a dolphin would be hard to miss,

there is uncertainty about whether the search effort

was adequate to make such a terminal decision. In

fact, a possible sighting (and filming) of a Yangtze

River dolphin was reported later in August 2007 (

http://news.bbc.co.uk/2/hi/science/nature/6969226.

stm). The recently reported rediscovery (Fitzpatrick

et al. 2005) of the Ivory-billed Woodpecker

(Campephilus principalis) has captured the

imagination of the American public and given new

hope for its recovery (Walters and Crist 2005). The

lack of indisputable evidence and alternative

explanations of the original data, however, have

raised doubts about the species’ status (Jackson

2006, Sibley et al. 2006, Stokstad 2007). Rigorous

quantitative estimates of the likelihood of

persistence in a given area would help to clarify the

situation.

This is the problem we address here. The probability

of encountering individuals is determined by their

abundance, conspicuousness, and home-range size.

Other factors, such as habitat complexity, weather,

and observer capabilities, also come into play

(Ramsey and Scott 1981, Scott et al. 1986, Buckland

et al. 2001). Collectively, these factors determine

the likelihood that a species will be detected, given

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

a certain search effort. So what we really want to

know is what search effort is required to conclude

with a high degree of certainty that a species is in

fact extinct, or, conversely, given a certain search

effort, what is the probability that individuals of a

rare species should be recorded? We develop an

approach to answer these questions and illustrate

the approach using data from searches for

potentially extinct Hawaiian forest birds and for the

Ivory-billed Woodpecker in Arkansas.

METHODS

The question we ask is, what effort is required to

realize certain values of the probability of

extinction? If N is population size, we want to

determine the probability that N is zero. With

surveys, one may estimate the probabilities that

populations of a given size are missed, i.e.,

conditional probabilities such as P(miss all | N = 5).

To convert such probabilities to probabilities of

extinction, we use a Bayesian argument. The

population size is unknown. Suppose one can

express, before sampling, a set of probabilities for

the population size denoted by P(N=k), for k = 0, 1,

2,... A convenient form for these probabilities is the

Poisson distribution (Appendix), where,

(1)

The value of m is the mean of the distribution, and

it determines all the probabilities exactly. Of

particular interest is the prior probability that the

population is extinct, which is P(N=0) = e−m. We

can specify the entire set of prior Poisson

probabilities by specifying the prior probability of

extinction. If P(N=0) = Q, then m = –ln(Q). If we

assume a prior probability of extinction of 50%

(inferred as Bayesian posterior probabilities from

Scott et al. (1986: 54–55), that corresponds to m = –

ln(1/2) = 0.6931.

Now, suppose that a survey is conducted that has

total effective area of E in a species’ range that has

a total area of A. If the N members of the population

each have probability E/A of detection and if they

are detected independently over their range, then

the probability that all are missed in the survey (X=0,

where X equals the number of birds detected in a

survey) is

(2)

From this one may determine the effective area (E)

that would need to be surveyed to detect at least one

individual from several population sizes. It is

calculated as:

(3)

Application of Bayes Theorem then produces a set

of posterior probabilities for the population size.

They are

(4)

for k = 0, 1, 2, ... The resulting probability of

extinction is

(5)

To achieve a desired value, V, for the posterior

probability of extinction, starting with a prior

probability, Q, of extinction, requires a total

effective area surveyed of

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

(6)

The area effectively surveyed from a line transect

is 2Lw, where w is the effective half-width or

effective detection distance and Lis the total transect

length. In variable circular plot surveys, E = nπr2,

where r is the effective radius or effective detection

distance and n is the total number of point counts.

APPLICATIONS

Ivory-billed Woodpecker

The decline of the Ivory-billed Woodpecker is

widely attributed to loss of habitat, primarily large

continuous forest with large mature trees and

numerous snags. In addition, the opening up of these

large tracts of land for logging is also thought to

have increased collection and poaching of Ivory-

billed Woodpecker (U.S. Fish and Wildlife Service

2006a, Snyder 2007).

Systematic surveys for Ivory-billed Woodpeckers

using line transects and variable circular plots

(VCPs) were concentrated in a 19 700-ha tract along

White River (WR) and a 9500-ha tract at Bayou de

View (BDV) in Arkansas, USA. This area was

imbedded within a larger area of potential Ivory-

billed Woodpecker habitat of 233 896 ha (Fig. 1)

that was surveyed during the winter months of

2004–2005 and 2005–2006 (Fitzpatrick et al. 2005,

Rohrbaugh et al. 2006). The survey was focused on

areas with recent reported sightings of Ivory-billed

Woodpeckers and nearby best available habitat—

those areas thought to have the highest probability

of detecting the species. Transects were placed

systematically 50 m apart in these areas following

Universal Transverse Mercator (UTM)-based

transect lines. Stations on which variable circular

counts were conducted were placed throughout the

target area. Our calculations are based on those with

a minimum of 60 m separation from the next nearest

station. We used an effective detection distance of

25 m for linear transects and 50 m for VCPs.

Variable circular plots have a larger effective

detection distance because they are often situated at

sites with a wider view than line transects and

because a stationary observer is likely to see more

than would an observer who is moving. These

detection distances are similar to those known for

Pileated Woodpeckers (Dryocopus pileatus), a

large woodpecker that is common in the Arkansas

study area (Fitzpatrick et al. 2005). The population

size for the Ivory-billed Woodpecker is unknown

but assumed to be small.

Hawaiian Forest Birds

Many Hawaiian forest birds, particularly the

honeycreepers (Drepanidinae), became extinct

following first human contact (Banko et al. 2001).

Other species may be extinct (Scott et al. 1986, Pratt

2005), but determining this with certainty is

problematic. What we can do is conduct surveys

that allow us to state the probability that at least one

bird would have been detected from a population of

stipulated size. The Hawaii Forest Bird Survey

(HFBS) was a 6-year effort covering the forested

areas of five principal islands in the Hawaiian

Archipelago. The HFBS used VCPs (Reynolds et

al. 1980) conducted at 9940 stations during 20 789

8-min count periods over 1401 km of transects (Fig.

2) (Scott et al. 1986) to document distribution and

abundance of the endemic Hawaiian avifauna.

Transects were placed 1.6 or 3.2 km apart located

at right angles to elevational contours. Distances

between stations were more than two times the

anticipated effective detection distances of the most

conspicuous species.

Several of the putatively extinct Hawaiian species

occupied densely vegetated habitat in treacherous

and inaccessible terrain when last recorded, making

surveys and detection difficult. To analyze the

likelihood that these species actually are extinct, we

use information for a suite of Hawaiian forest birds

surveyed by the HFBS (Fig. 2; Scott et al. 1986).

We consider nine endemic species, two of which

were on each of two islands (Table 1). Each of these

species (or island populations) is thought to have

become recently extinct (U.S. Fish and Wildlife

Service 2006b) and had been proposed for removal

from the Endangered Species List (Draft Ruling, U.

S. Fish and Wildlife Service, Washington Office;

pers. comm. from E. VanderWerf, U.S. Fish and

Wildlife Service, 6 April 2006). Effective detection

distances for most of these species are low (≤75 m),

so a large number of survey stations is required to

cover a species’ historic range (Table 1).

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Fig. 1. Map of the target search area for Ivory-billed Woodpeckers in Arkansas, USA, including the

extent of search coverage for the Bayou de View and the White River sites in one or both years.

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Fig. 2. Map showing the transect areas for the Hawaiian Forest Bird Surveys. From Scott et al. (1986).

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Table 1. Search effort required to make statement of probability of absence (extinction) for eight species

of Hawaiian forest birds. All have been suggested for delisting by the U.S. Fish and Wildlife Service and

were reported as presumed extinct in a 2003–2004 report to Congress (U.S. Fish and Wildlife Service

2006b).

Number of point counts required

for three levels of confidence in

absence

Species

Island Range

(km2)

Effective d-

etection dis-

tance (m)

Effective area

surveyed /

station (ha)

Number of

point counts

done

0.90

0.95

0.99

Lesser ’Akialoa

Hemignathus ellisianus

obscurus

Hawaii 1045

32

0.32

2319

275 462 300 800

320 128

Moloka’i Creeper

Paroreomyza flammea

Molokai 573

28

0.25

131

197 280 215 427

229 270

Kamao Myadestes

myadestinus

Kauai

25

60

1.13

140

1875

2047

2179

Oloma’o M. lanaiensis

Molokai 16

23

0.17

120

8165

8916

9488

Kaua’i O’o Moho braccatus Kauai

25

66

1.37

140

1550

1692

1801

Greater ’Akialoa

Hemignathus ellisianus

procerus

Kauai

25

32

0.32

140

6590

7197

7659

O’u Psittirostra psittacea

Hawaii 145

66

1.37

357

8986

9812

10 443

Nukupu’u Hemignathus

lucidus

Kauai

Kauai

25

25

66

39

1.37

0.48

140

140

1550

4437

1692

4845

1801

5157

Maui Akepa Loxops

coccineus

Maui

Maui

7

23

39

34

0.48

0.36

35

84

1243

5371

1357

5865

1444

6242

RESULTS

Ivory-billed Woodpecker

Assuming an effective detection distance for visual

detections of 25 m for line-transect surveys and 50

m for VCPs, 48%–54% of BDV and 44%–59% of

WR were effectively searched in the 2 years (Table

2). The probabilities of visually detecting a single

remaining individual were approximately 0.5 for

each year of survey, increasing to a probability of

roughly 0.75 of detecting a single individual in at

least one of the two survey years (Table 3). At least

one individual from a population of two or more

individuals should have had a roughly 95% or

greater chance of being detected over the 2 years

(Table 3).

Of the overall 233–896-ha Arkansas study area

(Figs. 1 and 3), 29 200 ha were intensively searched

through 2006. Given the detection probabilities we

used, this survey effort yielded roughly a 93%

chance of detecting at least one individual from a

population of 20 birds, a 74% chance from a

population of 10, and only a 49% chance if five

individuals were actually present in the study area

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Table 2. Effective areas surveyed (EAS) for Ivory-billed Woodpeckers by visual means during 2 years for

tracts along White River (WR) and Bayou de View (BDV), Arkansas (Fig. 2), assuming effective detection

distances of 25 m for line transect (LT) surveys and 50 m for variable circular plot (VCP) surveys.

2004–2005

2005–2006

Survey Site

Area (ha)

LT (km)

VCP (#)

EAS (ha)

LT (km)

VCP (#)

EAS (ha)

BDV

9500

900

300

4592

1000

613

5168

WR

19 700

2 280

518

11 559

1660

1552

8725

as a whole (Fig. 3). The chance of detecting a

population consisting of a single bird is only 12%.

Thus, given this search effort, if more than a very

few Ivory-billed Woodpeckers were present in the

search area during the surveys, their presence should

have been detected.

We also calculated Bayes’ posterior probabilities of

extinction for the Ivory-billed Woodpecker in those

areas that were intensively surveyed, assuming that

there were no valid detections. Beginning with prior

probabilities of extinction of 50% in both BDV and

WR, the surveys result in posterior probabilities of

70% in BDV and 75% in WR. Considering the large

search effort, one might anticipate that not detecting

any birds would lead to higher posterior

probabilities of extinction than these. That there

remains considerable doubt of extinction results

from the moderate probabilities of missing

populations consisting of only one or two

individuals.

How do home-range size and the assumed

distribution affect our estimates of the likelihood of

detecting Ivory-billed Woodpeckers? Suppose that

a single Ivory-billed Woodpecker occupies a home

range of area H, that it can be at any position within

the home range at the survey time, and that it does

not venture outside its home range. In a target region

with area A, the probability that any given individual

will be missed by a survey that effectively covers

the fraction p of the total area is (1 – fp), where f =

H/A. Therefore, it follows that the probability that

all individuals of a population of size N will be

missed is P(n = 0 | N) = (1 − fp)N. Figure 4 shows

the posterior probabilities of extinction given that

there were no valid visual detections in either winter.

It assumes a geometric prior distribution with prior

probability of extinction of 0.5. The results are

graphed against a range of home-range sizes. With

very large home-range sizes, home ranges will

naturally overlap. When the size of the home range

exceeds the area of the target region, one arrives at

the “uniform distribution” situation. The effectiveness

of a survey is measured by the difference between

the posterior probability of extinction and the prior

probability of extinction. As the graph indicates, the

home range must be large for the survey to have

been moderately effective. The surveys as

conducted would produce posterior probabilities of

extinction exceeding 80% only if the home range of

an individual Ivory-billed Woodpecker exceeds

5000 ha.

Home-range size is unknown for Ivory-billed

Woodpeckers. During his studies, Tanner (1942)

followed birds and found that, during the nesting

season, individuals often traveled >2.0 km from the

nest, with one record of a bird traveling 4.0 km from

its nest in the course of daily activities. Based on

anecdotal information, winter ranges (when current

Ivory-billed Woodpecker surveys are done) were

much larger than nesting-season ranges (Tanner

1942). From this, if winter home ranges are

equivalent to a circle with a 4.0 km radius, home

range size would be 50 km2 (5000 ha), which would

produce posterior probabilities of extinction slightly

under 80% in both BDV and WR. From this

analysis, we conclude that much greater search

efforts will be required to obtain a level of 90% or

greater for posterior probabilities of extinction.

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Table 3. Probabilities of sighting at least one Ivory-billed Woodpecker for populations of different sizes,

given areas surveyed in 2004–2005 and 2005–2006 in an estimated 19 700-ha tract along White River

(WR) and a 9500-ha tract at Bayou de View (BDV). Also included are joint probabilities of detection in

at least 1 year, assuming independent probabilities between years.

Population Size

Survey Area

1

2

5

10

20

BDV

2004–2005

0.483

0.733

0.963

0.999

1.000

2005–2006

0.544

0.792

0.980

1.000

1.000

combined

0.764

0.944

0.999

1.000

1.000

WR

2004–2005

0.587

0.829

0.988

1.000

1.000

2005–2006

0.443

0.690

0.946

0.997

1.000

combined

0.770

0.947

0.999

1.000

1.000

Hawaiian Forest Birds

Based on our calculations from the HFBS data, none

of the endangered Hawaiian forest birds we assessed

was surveyed sufficiently to yield a greater than

90% probability of detecting at least one bird from

a population of 10 or more individuals. In most

cases, we conclude that the HFBS survey effort was

1–2 orders of magnitude less than that required to

assert extinction with a high degree of certainty.

Choice of Prior

The choice of prior distribution and prior probability

of extinction will influence the sampling effort

required to produce a desired level of certainty of

extinction. To illustrate the range of effects of such

choices, consider the Oloma’o (scientific names of

species are given in Table 1) of Molokai, Hawaii.

Its range is A = 1600 ha. With its effective detection

distance of 23 m (Scott et al. 1986), the effective

area covered by a single station count is f = 0.166

ha. Again, let Qbe the prior probability of extinction

and V be the desired posterior probability of

extinction following a survey of nstations that result

in no detections. The Poisson prior distribution

loads most of its probabilities of N > 0 on very low

population sizes, where the probabilities of missing

them are high. With a Poisson prior distribution, the

required number of stations (n) is:

(7)

We might also consider the prior probability of

extinction as a Geometric probability distribution,

where P(N = k) = (1-p) pk , for k = 0, 1, 2, ...

(Appendix). The Geometric distribution spreads its

probabilities of N > 0 over a wider range of

population sizes than does the Poisson distribution.

It is less likely that these larger populations would

be missed by the survey effort. The posterior

probabilities of extinction (given that there were no

confirmed detections) are, therefore, somewhat

higher for the Geometric prior distribution. With a

Geometric prior distribution, the required number

of stations (n) is:

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Fig. 3. The area that must be effectively surveyed (EAS) relative to the probability of detection of at

least one member of a randomly distributed population of Ivory-billed Woodpeckers across a 233 896-

ha study area for actual populations of different sizes.

(8)

The sampling effort under the different assumptions

of prior distributions (Table 4) was less than 10%

greater for the Poisson prior distribution. Yet the

message is the same: it requires very intensive

efforts to conclude with a high degree of certainty

that the species is extinct.

DISCUSSION

Using a combination of biological data and

statistical calculations, we show, on the basis of the

Hawaiian Forest Bird Survey in the 1980s, it would

be premature to declare that the Hawaiian forest

birds we evaluated are in fact extinct. More recent

intensive survey efforts, however, suggest that the

Greater ’Akialoa and Molokai Creeper, two species

that were not detected during the HFBS, and three

species that were detected, the O’u on Kauai, Kauai

O’o, and Kama’o, have high likelihoods (p > 0.95)

of being extinct (Reynolds and Snetsinger 2001).

Reynolds and Snetsinger’s analysis used data from

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Fig. 4. The relationship between home-range size and posterior probability of extinction given no valid

visual detections and the search efforts at White River (WR) and Bayou de View (BDV). We assume a

geometric prior distribution with prior probability of extinction being 0.5. The arrow indicates a home

range as large as the surveyed area; this gives the same results as having birds uniformly distributed

throughout the region.

targeted searches in 1994–1996. These searches

differed from the HFBS in survey methods (they

defined a single visit as 10–20 h of auditory and

visual searching, whereas the HFBS single survey

was an 8-min point count), in what was calculated

(they determined the probability of detecting at least

one individual from a population of 10 across the

species range, whereas we determined the

probability that N=0), and in detection distances

(they assumed greater detection distances for the

Kaua’i O’o [150 m using song playback] and

Greater ’Akioloa [39 m]). Regardless, this is an

example of the intensive search effort required to

have a statistically strong certainty of the likelihood

of extinction.

The search effort required to have a high degree of

certainty that a species occurs (or does not occur)

in an area becomes extremely large for populations

of 10 or fewer individuals. Our analysis of the search

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Table 4. Sampling effort (number of variable circular plot (VCP) stations) required for three levels of

statistical confidence of extinction (i.e., desired posterior probabilities) (V) across a range of prior

probabilities of extinction (Q) for two prior probability distributions.

Poisson prior

Q =

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

V = 0.90:

9,188

8,998

8,786

8,522

8,165

7,643

6,785

5,083

−

0.95:

9,414

9,322

9,218

9,090

8,916

8,662

8,244

7,416

4,942

0.99:

9,587

9,568

9,548

9,523

9,489

9,439

9,357

9,195

8,710

Geometric prior

Q =

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

V = 0.90:

8,559

8,425

8,253

8,024

7,703

7,222

6,419

4,815

−

0.95:

9,094

9,027

8,941

8,826

8,666

8,425

8,024

7,222

4,815

0.99:

9,522

9,508

9,491

9,468

9,436

9,388

9,308

9,147

8,666

effort for Ivory-billed Woodpeckers in the White

River and Bayou de View, assuming a uniform

distribution of birds if they are present, suggest that

if birds were present the actual number of

individuals was very low (N<2). With this

assumption it is unlikely that birds were still present

in the intensively searched area but not seen during

surveys. However, below, we present distributions

other than the uniform, which suggest higher

probabilities that the Ivory-billed Woodpeckers

may still persist in White River and Bayou de View.

Based on historical information, Ivory-billed

Woodpeckers could be heard over much greater

distances than they could be seen (Tanner 1942). If

the greater auditory detection distances are used in

our analysis, the survey effort in intensively

searched regions of Arkansas would have virtually

guaranteed detection if even a single bird was

present. In fact, there are auditory records during

these surveys that may be Ivory-billed

Woodpeckers (Charif et al. 2005) (see also http://w

ww.birds.cornell.edu/ivory, accessed 2007; but see

Jones et al. 2007). The auditory records, if they are

of Ivory-billed Woodpeckers, are consistent with

the presence of a few individuals in the area.

The Ivory-billed Woodpecker surveys we

considered included only ~12% of the forested

habitat in the overall search area in Arkansas, so our

analysis does not enable us to assess the likelihood

that Ivory-billed Woodpeckers are extinct in the

wider search area, much less across the currently

presumably suitable habitat within the species’

historic range. Based on a statistical assessment

using time series of historic sightings, Roberts

(2006) argued that a declaration of extinction is

premature for this species. There are large forested

areas of the historic range of the Ivory-billed

Woodpecker, such as the Apalachicola and Chipola

Rivers in Florida (ca. 73 000 ha) and the Atchafalaya

Swamp in Louisiana (ca. 536 000 ha) that include

possible habitat. The effort required to extend the

search for Ivory-billed Woodpeckers to the large

tracts of potential habitat will require hundreds of

thousands of count periods or thousands of

kilometers of transects. That process has already

begun in other portions of the species’ historic range

in an attempt to locate birds and suitable habitat (Hill

et al. 2006, Rohrbaugh et al. 2007) (see also http://

www.fws.gov/ivorybill/ and http://www.birds.cornell.

edu/ivory/latest/0708summary/document_view, a-

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

ccessed 2007). Since 2007, a tool for broadcasting

imitations of double-knock drums of Ivory-billed

Woodpeckers is being widely used in surveys.

Ivory-billed Woodpeckers would likely respond to

these imitations in a way that improves their

detectability at distances larger than the effective

detection distances we used for our calculations

based on visual encounters. Area effectively

searched would be higher, and the overall search

effort required for absence inferences lower, with

use of this tool.

Model Assumptions

All modeling analyses make assumptions, of course,

and ours is no exception. In this analysis, we

assumed that each bird in the population would be

detected with a probability equal to the fraction of

the target area that is effectively surveyed. That

would be the case, for example, if the birds were

uniformly distributed; this assumption is also a

reasonable default if we are ignorant of a more

complex spatial distribution. If the birds are

spatially distributed in ways other than uniformly,

survey efforts are likely to be even less effective.

This can be illustrated simply. Suppose that the

proportion of the target area effectively surveyed is

p, but that the population occupies only a fraction

(f) of the target region. In that case, the chance that

an individual will be detected in the survey is fp. As

an example, consider the case where no Ivory-billed

Woodpeckers were confirmed in the 2 years of

survey effort. Starting with a 50% prior probability

of extinction and a geometric prior distribution, the

posterior probability of extinction declines steadily

with f from 88% when f=1 to 55% when f=0.1. So,

if a remaining population is confined to a small sub-

region of the area targeted by the surveys, an even

greater search effort than the enormous effort

already taken would be needed to achieve

substantial gains in the certainty that the Ivory-

billed Woodpecker is extinct in the target region

when no birds are detected. Known differences in

distribution can be handled by stratification of effort

at the design stage and by covariate analysis at the

analysis stage. Standard survey methods allow for

clustering by counting the clusters rather than the

individuals, so even if Ivory-billed Woodpeckers

travel in pairs it does not invalidate the methods and

results we present.

We also assumed a constant detection distance.

Detectability may vary by habitat, by time of day,

by observer, and by other factors. Methods exist for

incorporating such variations into the analysis

(Ramsey et al. 1987, Beavers and Ramsey 1998),

but they depend on having actual detections of the

species in a variety of conditions to calibrate the

detectability functions. Lacking actual detections,

we relied on knowledge of the species’ biology and

that of similar species to set an arbitrary detection

distance. Finally, we made assumptions about

Ivory-billed Woodpecker behaviors. For example,

we assumed that birds are neither attracted to nor

repelled by observers. If this assumption were

violated, detection distances would be different—

increased if birds avoid surveyors (Tanner 1942)

and reduced if birds are attracted to observers—

yielding different detection likelihoods.

Implications

Determining whether or not a species is extinct is a

problem with both biological and economic

consequences. It is not possible to prove with

absolute certainty the absence of a species—

statements about extinction (or about the successful

eradication of introduced or invasive species

[Morrison et al. 2007]) are probabilistic. Achieving

a lack of reasonable doubt requires sufficient survey

efforts, coupled with realistic probability

calculations. As we have shown, assumptions about

the distribution of a species’ influence answers to

how much search effort is required for a specified

level of confidence in the absence of a species. Thus,

if little information is available, we suggest

assuming distributions that are consistent with

biological evidence and modeling outcomes under

varying assumptions of distribution. We suggest

that statements about extinction should be

accompanied by a statement of probability or

uncertainty and sufficient information about how

that conclusion was determined. Declaring a species

as extinct or delisting a species where legal

protection exists is a poor idea unless statistically

sufficient survey efforts have been made.

What should one consider when establishing

sampling protocols for species that are vanishingly

rare so that precious conservation resources are

efficiently spent? We suggest that the first step

involves searching in those areas where there is the

greatest chance of encountering the species of

interest. This determination can be based on

historical records, recent sightings, and/or best

remaining available habitat. Next, it is important

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

that highly skilled observers be chosen, screening

potential observers for hearing, vision, experience,

and physical ability to do the job. Those selected

can then be trained in the methods and identification

issues. Once a search area has been selected, the

sampling method needs to be determined. Line

transects are appropriate when it is possible to

survey while moving, whereas point counts are

appropriate when the observer must traverse

difficult terrain. It is important that sampling points

and/or transects be located in a way that gives all

parts of the target region some quantifiable chance

of being selected. The time of year and day, weather

conditions, etc. should be chosen to increase

chances that species of interest will be detected

(Ralph and Scott 1981, Buckland et al. 2001). The

Ivory-billed Woodpecker surveys have been

conducted during the winter and early spring when

birds are expected to drum and call more frequently

and when deciduous trees lacked leaves, increasing

chances of seeing a bird at a distance in heavily

forested areas. The HFBS was conducted in areas

of remaining habitat when birds were highly vocal

and favorable weather was likely.

Questions will always arise as to how resource

managers should allocate limited funds among

endangered species and habitat protection, and how

the costs of protection should be balanced against

the likelihood that a species is still extant and, if so,

can be managed to recovery (Scott et al. 2005). What

level of certainty should be sufficient to declare a

species extinct and shift resources elsewhere, vs.

mounting what may be a massive surveying effort

to improve our confidence in that declaration?

Ultimately, these are societal, not scientific,

decisions. We argue that, if one wants to protect

against the error of declaring an extant species as

extinct, the threshold for “reasonable doubt” should

be set high, perhaps higher than the commonly used

95% level.

Decisions about allocating resources to threatened

or endangered species are difficult because the

consequences of habitat protection cannot always

be anticipated and some payoffs may be a long time

in coming. Others have suggested that a structured

decision-making process would allow managers to

determine if resources would be better spent

searching for the species than in managing the

habitat (Chades et al. 2008). The search for the

Ivory-billed Woodpecker has sparked conservation

efforts and searches throughout the southeastern

United States. Indeed, we would not be discussing

Ivory-billed Woodpecker conservation and resource

allocation if the area in Arkansas where Ivory-billed

Woodpeckers were reported in 2004 and early 2005

had not been set aside over several decades as a

refuge for other species by the U.S. Fish and

Wildlife Service (Butch 2003) and The Nature

Conservancy.

Responses to this article can be read online at:

http://www.ace-eco.org/vol3/iss2/art3/responses/

Acknowledgments:

We thank E. VanderWerf for information regarding

Hawaiian forest birds and M. Reynolds for

comments on a draft manuscript. For the Ivory-

billed Woodpecker data we thank E. Swarthout for

field leadership and data summarization, M.

Piorkowski for preparing Fig. 1, the U.S. Fish and

Wildlife Service and U.S. Geological Survey for

funding, and The Nature Conservancy for in-kind

support.

LITERATURE CITED

Bailey, A. M. 1956. Birds of Midway and Laysan

Islands.Museum Pictorial No. 12. Denver Museum

of Natural History, Denver, Colorado, USA.

Banko, P. C., R. E. David, J. D. Jacobi, and W. E.

Banko. 2001. Conservation status and recovery

strategies for endemic Hawaiian birds. Studies in

Avian Biology 22:359–376.

Bauer, C. J., J. F. Pacheco, A. C. Venturini, and

B. M. Whitney. 2000. Rediscovery of the Cherry-

throated Tanager Nemosia rourei in southern

Espirito Santo, Brazil. Bird Conservation

International 10:97–108.

Beavers, S., and F. L. Ramsey. 1998. Detectability

analysis in wildlife surveys, with application to sea

turtle populations in the tropical Pacific Ocean.

Journal of Wildlife Management 62:948–957.

Blockstein, D. E. 2002. Passenger pigeon

(Ectopistes migratorius). Number 611 in A. Poole

and F. Gill, editors. The birds of North America.The

Birds of North America, Inc., Philadelphia,

Pennsylvania, USA.

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Buckland, S. T., D. R. Anderson, K. P. Burnham,

and J. L. Laake. 2001. Introduction to distance

sampling: estimating abundance of biological

populations. Oxford University Press, New York,

New York, USA.

Butch, R. D. 2003. America’s national wildlife

refuges: a complete guide. Robert Rhinehart

Publisher, Lineham, Maryland, USA.

Chadés, I., E. McDonald-Madden, M. A.

McCarthy, B. Wintle, M. Linkie, and H. P.

Possingham. 2008. When to stop managing or

surveying cryptic threatened species. Proceedings

of the National Academy of Sciences (USA)

105:13936–13940.

Charif, R. A., K. A. Cortopassi, H. K. Figueroa,

J. W. Fitzpatrick, K. M. Fristrup, M.

Lammertink, M. D. Luneau, Jr., M. E. Powers,

and K. V. Rosenberg. 2005. Notes and double

knocks from Arkansas. Science 309:489–1490.

Collar, N. J. 1998 . Extinction by assumption; or

the Romeo error on Cebu. Oryx 32:239–244.

Eames, J. C., H. Hla, P. Leimgruber, D. S. Kelly,

S. M. Aung, S. Moses, and U. S. N. Tin. 2005. The

rediscovery of Gurney’s Pitta Pitta gurneyi in

Myanmar and an estimate of its population size

based on remaining forest cover. Bird Conservation

International 5:3–26.

Fitzpatrick, J. W., M. Lammertink, M. D.

Luneau, T. W. Gallagher, B. R. Harrison, G. M.

Sparling, K. V. Rosenberg, R. W. Rohrbaugh, E.

C. H. Swarthout, P. H. Wrege, S. B. Swarthout,

M. S. Dantzker, R. A. Charif, T. R. Barksdale, J.

V. Remsen, Jr., S. D. Simon, and D. Zollner.2005.

Ivory-billed Woodpecker (Campephilus principalis)

persists in continental North America. Science

308:1460–1462.

Flannery, T. 1994. The rediscovery of Blumer’s

fruit bat. Bats 12:3–5.

Hill, G. E., D. J. Mennill, B. W. Rolek, T. L. Hicks,

and K. A. Swiston. 2006. Evidence suggesting that

Ivory-billed Woodpeckers (Campephilus principalis)

exist in Florida. Avian Conservation and Ecology

1: 3. [online] URL: http://www.ace-eco.org/vol1/iss3/

art2/.

IUCN. 2001. IUCN red list categories and criteria:

version 3.1. IUCN Species Survival Commission,

IUCN, Gland, Switzerland and Cambridge, UK.

Jackson, J. A. 2006. Ivory-billed woodpecker

(Campephilus principalis): hope, and the interfaces

of science, conservation, and politics. Auk 123:1–

15.

Jones, C. D., J. R. Troy, and L. Y. Pomara. 2007.

Similarities between Campephilus woodpecker

double raps and mechanical sounds produced by

duck flocks. Wilson Journal of Ornithology

119:259–262.

Keith, D. A., and M. A. Burgman. 2004. The

Lazarus effect: can the dynamics of extinct species

lists tell us anything about the status of biodiversity?

Biological Conservation 117:41–48.

Morrison, S. A., N. Macdonald, K. Walker, L.

Lozier, and M. R. Shaw.2007. Facing the dilemma

at eradication’s end: uncertainty of absence and the

Lazarus effect. Frontiers in Ecology and the

Environment 5:271–276.

Murphy, R. C., and L. S. Mowbray. 1951. New

light on the Cahow, Petrodroma cahow. Auk

68:266–280.

Pratt, H. D. 2005. The Hawaiian Honeycreepers.

Oxford University Press, New York, New York,

USA.

Ralph, C. J., and J. M. Scott, editors. 1981.

Estimating numbers of terrestrial birds. Studies in

Avian Biology No. 6, Cooper Ornithological

Society, Boise, Idaho, USA.

Ramsey, F. L., and J. M. Scott. 1981. Analysis of

bird survey data using a modification of Emlen’s

method. Pages 483–487 in C. J. Ralph and J. M.

Scott, editors. Estimating numbers of terrestrial

birds. Studies in Avian Biology No. 6, Cooper

Ornithological Society, Boise, Idaho, USA.

Ramsey, F. L., V. J. Wildman, and J. Engbring.

1987. Covariate adjustments of effective area in

variable area wildlife surveys. Biometrics 43:1–11.

Reis, J. 2003 . The freshwater pearl mussel

[Margaritifera margaritifera (L.)] (Bivalvia,

Unionoida) rediscovered in Portugal and threats to

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

its survival. Biological Conservation 114:447–452.

Reed, J. M. 1996. Using statistical probability to

increase confidence of inferring species extinction.

Conservation Biology 10:1283–1285.

Reynolds, R. T., J. M. Scott, and R. A. Nussbaum.

1980. A variable circular-plot method for estimating

bird numbers. Condor 82:309–313.

Reynolds, M. H., and T. J. Snetsinger. 2001. The

Hawaii rare bird search 1994–1996. Pages 133–142

in J. M. Scott, S. Conant, and C. Van Riper, III,

editors. Evolution, ecology, conservation, and

management of Hawaiian birds: a vanishing

avifauna. Studies in Avian Biology No. 22, Cooper

Ornithological Society, Boise, Idaho, USA.

Roberts, D. L. 2006. Extinct or possibly extinct?

Science 312:997–998.

Rohrbaugh, R., K. Rosenberg, M. Lammertink,

E. Swarthout, R. Charif, S. Barker, and M.

Powers. 2006. Summary and conclusions of the

2005–06 Ivory-billed Woodpecker search in

Arkansas. Report, Cornell Laboratory of Ornithology,

Ithaca, New York, USA. [online] URL: http://www.

birds. cornell.edu/ivory/pdf/FinalReportIBWOOsmall.

pdf.

Rohrbaugh, R., M. Lammertink, K. Rosenberg,

M. Piorkowski, S. Barker, and K. Levenstein.

2007. 2006–07 Ivory-billed Woodpecker surveys

and equipment loan program. Report, Cornell

Laboratory of Ornithology, Ithaca, New York,

USA. [online] URL: http://www.birds. cornell.edu/

ivory/pdf/FinalReportIBWO_071121_TEXT.pdf.

Scott, J. M., D. D. Goble, J. A. Wiens, D. S.

Wilcove, M. Bean, and T. Male. 2005. Recovery

of imperiled species under the Endangered Species

Act: the need for a new approach. Frontiers in

Ecology and the Environment 3:383–389.

Scott, J. M., S. Mountainspring, F. L. Ramsey,

and C. B. Kepler. 1986. Forest bird communities

of the Hawaiian islands: their dynamics, ecology,

and conservation. Studies in Avian Biology No. 9,

Cooper Ornithological Society, Boise, Idaho, USA.

Sibley, D. A., L. R. Bevier, M. A. Patten, and C.

S. Elphick. 2006. Comment on “Ivory-billed

woodpecker (Campephilus principalis) persists in

continental North America.” Science 311:1555.

Snyder, N. F. R. 2007. An alternative hypothesis

for the cause of the Ivory-billed Woodpecker’s

decline. Monographs of the Western Foundation of

Vertebrate Zoology 1:1–58.

Solow, A. R. 1993. Inferring extinction from

sighting data. Ecology 74:962–964.

Stokstad, E. 2007. Gambling on a ghost bird.

Science 317:888–892.

Tanner, J. T. 1942. The Ivory-billed Woodpecker.

National Audubon Society Research Report No. 1.

National Audubon Society, New York, New York,

USA.

Turvey, S. T., R. L. Pitman, B. L. Taylor, J.

Barlow, T. Akamatsu, L. A. Barrett, X. Zhao, R.

R. Reeves, B. S. Stewart, K. Wang, Z. Wei, X.

Zhang, L. T. Pusser, M. Richlen, J. R. Brandon,

and D. Wang. 2007. First human-caused extinction

of a cetacean species? Biology Letters 3:537–540.

U. S. Fish and Wildlife Service. 2006a. Recovery

plan for the Ivory-billed Woodpecker (Campephilus

principalis). U. S. Fish and Wildlife Service,

Atlanta, Georgia, USA.

U. S. Fish and Wildlife Service. 2006b. Report to

Congress on the recovery of threatened and

endangered species Fiscal Years 2003–2004. U. S.

Department of Interior, Washington, D.C., USA.

Walters, J. R., and E. L. Crist. 2005.

Rediscovering the king of woodpeckers: exploring

the implications. Avian Conservation and Ecology

1(1): 6. [online] URL: http://www.ace-eco.org/vol1/

iss1/art6/.

Woodford, J. 2002. The Wollemi pine: the

incredible discovery of a living fossil from the age

of the dinosaurs. The Text Publishing Company,

Melbourne, Australia.

Avian Conservation and Ecology - Écologie et conservation des oiseaux 3(2): 3

http://www.ace-eco.org/vol3/iss2/art3/

Appendix 1. Appendix

Please click here to download file ‘appendix1.pdf’.