Contributed Paper

Detecting Extinction Risk from Climate Changeby IUCN Red List CriteriaDAVID A. KEITH,∗† MICHAEL MAHONY,‡ HARRY HINES,§ JANE ELITH,∗∗ TRACEY J. REGAN,††JOHN B. BAUMGARTNER,†† DAVID HUNTER,‡‡ GEOFFREY W. HEARD,††NICOLA J. MITCHELL,§§ KIRSTEN M. PARRIS,†† TRENT PENMAN,∗∗∗ BEN SCHEELE,†††CHRISTOPHER C. SIMPSON,† REID TINGLEY,†† CHRISTOPHER R. TRACY,‡‡‡††††MATT WEST,∗∗††§§§ AND H. RESIT AKCAKAYA∗∗∗∗∗Centre for Ecosystem Sciences, School of Biological, Earth and Environmental Sciences, University of New South Wales, KensingtonNSW 2052, Australia, email david.keith@environment.nsw.gov.au†NSW Office of Environment and Heritage, P.O. Box 1967, Hurstville NSW 2220, Australia‡University of Newcastle, Callaghan NSW 2308, Australia§Queensland Parks and Wildlife Service, Brisbane Qld 4000, Australia∗∗ARC Centre of Excellence for BioSecurity Risk Analysis, School of Botany, University of Melbourne, Parkville, Victoria 3010, Australia††ARC Centre of Excellence for Environmental Decisions, School of Botany, University of Melbourne, Parkville, Victoria 3010,Australia‡‡NSW Office of Environment and Heritage, P.O. Box 733, Queanbeyan NSW 2620, Australia§§School of Animal Biology, University of Western Australia, Crawley, Western Australia 6009, Australia∗∗∗Centre for Environmental Risk Management of Bushfires, School of Biological Sciences, University of Wollongong, NSW 2522,Australia†††Fenner School of Environment and Society, Australian National University, Canberra, ACT 0200, Australia‡‡‡Department of Zoology, University of Melbourne, Parkville, Victoria 3010, Australia§§§Wildlife Conservation and Science, Zoos Victoria, Victoria 3052, Australia∗∗∗∗Department of Ecology and Evolution, Stony Brook University, Stony Brook, NY 11794, U.S.A.††††Department of Biological Science, California State University Fullerton, 800 N. State College Boulevard, Fullerton, CA 92831,U.S.A.

Abstract: Anthropogenic climate change is a key threat to global biodiversity. To inform strategic actionsaimed at conserving biodiversity as climate changes, conservation planners need early warning of the risksfaced by different species. The IUCN Red List criteria for threatened species are widely acknowledged as usefulrisk assessment tools for informing conservation under constraints imposed by limited data. However, doubtshave been expressed about the ability of the criteria to detect risks imposed by potentially slow-acting threatssuch as climate change, particularly because criteria addressing rates of population decline are assessed overtime scales as short as 10 years. We used spatially explicit stochastic population models and dynamic speciesdistribution models projected to future climates to determine how long before extinction a species wouldbecome eligible for listing as threatened based on the IUCN Red List criteria. We focused on a short-livedfrog species (Assa darlingtoni) chosen specifically to represent potential weaknesses in the criteria to allowdetailed consideration of the analytical issues and to develop an approach for wider application. The criteriawere more sensitive to climate change than previously anticipated; lead times between initial listing in athreatened category and predicted extinction varied from 40 to 80 years, depending on data availability. Weattributed this sensitivity primarily to the ensemble properties of the criteria that assess contrasting symptomsof extinction risk. Nevertheless, we recommend the robustness of the criteria warrants further investigationacross species with contrasting life histories and patterns of decline. The adequacy of these lead times forearly warning depends on practicalities of environmental policy and management, bureaucratic or politicalinertia, and the anticipated species response times to management actions.

Paper submitted April 17, 2013; revised manuscript accepted October 3, 2013.

810Conservation Biology, Volume 28, No. 3, 810–819C© 2014 Society for Conservation BiologyDOI: 10.1111/cobi.12234

Keith et al. 811

Keywords: amphibian, frog, population model, risk assessment, species distribution model, threatenedspecies

Deteccion del Riesgo de Extincion a partir del Cambio Climatico por medio del Criterio de la Lista Roja de la UICNKeith et al.

Resumen: El cambio climatico antropogenico es una amenaza clave para la biodiversidad global. Parainformar a las acciones estrategicas dirigidas a la conservacion de la biodiversidad conforme cambia el clima,los planeadores de la conservacion necesitan advertencias tempranas de los riesgos enfrentados por especiesdiferentes. El criterio de la Lista Roja de la UICN para las especies amenazadas se reconoce ampliamente comouna herramienta util de estudio de riesgo para informar a la conservacion bajo las restricciones impuestaspor datos limitados. Sin embargo, se han expresado dudas sobre la habilidad del criterio para detectar riesgosimpuestos por potenciales amenazas de accion lenta, como el cambio climatico, particularmente porque elcriterio para dirigirse a las tasas de declinacion poblacional se estudian sobre escalas de tiempo tan cortascomo 10 anos. Usamos modelos explıcitamente espaciales de poblacion estocastica y modelos de dinamicasde distribucion de especies proyectados a climas futuros para determinar cuanto tiempo antes de la extincionuna especie serıa elegible para enlistarse como amenazada basandonos en el criterio de la Lista Roja de laUICN. Nos enfocamos en una especie de rana de vida corta (Assa darlingtoni) escogida especıficamente pararepresentar las debilidades potenciales en el criterio para permitir la consideracion detallada de los problemasanalıticos y para desarrollar un acercamiento de aplicacion mayor. El criterio fue mas sensible al cambioclimatico que como se anticipo previamente; los tiempos mayores entre el listado inicial en una categorıade amenaza y la extincion predicha vario entre 40 y 80 anos, dependiendo de la disponibilidad de datos.Atribuimos esta sensibilidad principalmente a las propiedades conjuntas del criterio que estudian sıntomascontrastantes del riesgo de extincion. Sin embargo, recomendamos que lo robusto del criterio garantiza unainvestigacion mas detallada sobre las especies con historias de vida contrastantes y patrones de declinacion.La adecuacion de estos tiempos mayores para una alerta temprana depende de lo practico de la polıticaambiental y el manejo, la inercia polıtica o burocratica y la respuesta anticipada de las especies a lasacciones de manejo.

Palabras Clave: anfibio, especies amenazadas, estudio de riesgo, modelo de distribucion de especies, modelode poblacion, rana

Introduction

Anthropogenic climate change is one of the major globalthreats to biodiversity (Parry et al. 2007; Hannah 2012);more than a million terrestrial species are potentially“committed to extinction” (Thomas et al. 2004) bymid-century. Accurate predictions of the risks of suchlosses are vital to motivate and inform effective reme-dial actions at global and local scales and to focus spe-cific efforts on species and locations that would ben-efit most. For these reasons, species at substantial riskof extinction as a result of climate change ought to beincluded on listings of threatened species (Westoby &Burgman 2006).

The IUCN Red List of Threatened Species (e.g., Vieet al. 2009) is widely recognized as an authoritativeand reliable global listing in which species are assignedto categories representing different levels of extinc-tion risk (IUCN 2001; Mace et al. 2008). The IUCNRed List is based on 5 quantitative assessment crite-ria, derived from population theory, that are used toassign species to any of 3 different threat categories(critically endangered, endangered, or vulnerable) or toassign taxa not identified as threatened to an appro-priate category (extinct in the wild, near threatened,least concern, or data deficient) (IUCN 2001). Empiri-

cal evaluations show that the red list protocol is consis-tent and predicts extinction risk accurately (e.g., Keithet al. 2004).

Climate change has been identified as a threat to a rel-atively small fraction of species currently listed as threat-ened in Australia (Westoby & Burgman 2006) and globally(Vie et al. 2009), yet distribution modeling (Thomas et al.2004; Hof et al. 2011) and trait analyses (Foden et al.2013) suggest more taxa are at risk. For example, one-quarter of the world’s bird and amphibian taxa and halfof all coral taxa possess traits that make them suscepti-ble to climate change, but these taxa are not currentlylisted as threatened (Foden et al. 2013). There couldbe several explanations for this discrepancy. First, manytaxa that are potentially susceptible to climate changemay currently appear safe from extinction because oflimited exposure to climate change or lagged responses(Foden et al. 2013). Second, assessors may be reluctantto consider climate change as a threat to specific taxabecause of high levels of uncertainty about both the mag-nitude of future climate change and its ecological effects(Westoby & Burgman 2006). Third, climate change mayact in concert with other, more readily identified threats(Hof et al. 2011). Finally, the red list criteria may notbe well suited to detection of climate change threats(Hannah 2012).

Conservation BiologyVolume 28, No. 3, 2014

812 Detecting Extinction Risk from Climate Change

One of the IUCN Red List’s most important roles is as anearly-warning system that identifies species at most imme-diate risk to inform priorities for conservation action (Vieet al. 2009). To perform this role effectively, the red listcriteria should identify species at risk of extinction withsome lead time in advance of the expected extinctionevent, irrespective of its cause. The red list criteria assessdeclines over time horizons that are scaled by generationlength to account for the fact that long-lived species areat greater risk of extinction when exposed to elevatedannual mortality rates than short-lived species (Maceet al. 2008). Many vertebrate and invertebrate species thathave generation lengths <7 years are therefore assessedover time horizons of 10–20 years.

In contrast, the majority of studies assessing poten-tial impacts of climate change on species are based onprojections that range from 50 to 100 years into thefuture (Cameron 2012). This has led to concerns thatthe red list criteria are “poorly suited to assessing threatssuch as climate change, which happen now but haveeffects years or decades in the future” (Hannah 2012).Thomas et al. (2004) suggest the red list criteria assessdeclines with time scales that “are not suited to evaluatethe consequences of slow-acting but persistent threats.”In essence, for many taxa (especially those with shortgeneration lengths), the assessment time frames withinthe red list criteria may be too short for climate relateddeclines to be apparent, even though past greenhouse gasemissions have already determined the fate of the speciesin the absence of timely remedial action (Akcakaya et al.2006).

Several studies have used methods based on the IUCNRed List criteria to identify species threatened by climatechange (reviewed by Akcakaya et al. 2006; also Levinskyet al. 2007; Thuiller et al. 2008). These studies applyonly some of the red list criteria (typically only criterionA, which addresses population reduction) and involvemodifications to the time horizons or the thresholds ofdecline. As Akcakaya et al. (2006) point out, these ar-bitrary modifications make the criteria less consistentacross different symptoms of extinction risk and acrossdifferent taxonomic groups and consequently make thecriteria less capable of identifying the species at greatestrisk of extinction.

Despite the concerns and speculations about the per-formance of the red list criteria for assessing extinctionrisks under climate change, we could find no empiricalrisk assessments that directly address the issue. To in-form the debate and demonstrate how the problem maybe investigated, we used all 5 IUCN Red List criteria ina detailed assessment of extinction risk. Specifically, weasked how quickly the red list criteria identify a species asthreatened by climate change and which criteria are cru-cial for early detection of risk. As an example, we chose ashort-lived amphibian species for which climate changeappears to be the major current threat. We used a coupled

modeling approach capable of addressing species life his-tory and habitat responses to projected climate scenarios(Keith et al. 2008; Fordham et al. 2012) to carry out therisk assessment.

Methods

Study Species

Frogs are an extinction-prone group (Houlahan et al.2000; Stuart et al. 2004) and thought to be particularly sus-ceptible to future climatic changes (Pounds et al. 2006).Our study species, Assa darlingtoni (hip-pocket frog,family Myobatrachidae), occurs between the ConondaleRange (26°42’S) and the Dorrigo Plateau (30°20’S) in coolmoist rainforest or eucalypt forest within several disjunctmontane areas of the coastal escarpment, mostly above600 m, in central eastern Australia. Females lay 8–18 rela-tively large eggs in leaf litter and tadpoles emerge after ashort period of embryonic development and wriggle intolateral pouches of the adult male which cares for them forabout 40 d before emergence as subadults (Tyler 1985;Ehmann & Swan 1985). Breeding occurs in the spring andsummer months, usually after moderate to heavy rainfall.Males call from secreted positions under leaf litter. Dis-persal appears to be very localized; few animals are likelyto move more than a few hundred meters through theleaf litter.

Climate change is the most serious plausible threat tothe persistence of A. darlingtoni, given that it is restrictedto cool moist montane areas that are projected to warmover the coming decades. Although the species’ rangemay have been reduced previously by forest clearing andlogging, almost all the remaining habitat in its distributionis protected. The species co-occurs with several stream-dependent frogs whose decline is attributed to diseasecaused by chytrid fungus. Few individuals of A. darling-toni have been tested for susceptibility to the disease, butits populations are apparently not declining. Its limitedreliance on water bodies may limit exposure to chytrid,although some species with similar life histories are sus-ceptible (Bell et al. 2004). We did not model the effectsof chytrid on the persistence of A. darlingtoni. Otherthreats to the species are localized, including timber ex-traction, road building, weed invasion, grazing, and fre-quent burning (NSW Environment and Heritage 2013). A.darlingtoni is currently classified as Least Concern on theIUCN Red List (Table 1) “in view of its wide distribution,presumed large population, and because it is unlikely tobe declining fast enough to qualify for listing in a morethreatened category” (Hero et al. 2004).

Assessment

We assessed the status of A. darlingtoni using IUCN RedList categories and criteria (IUCN 2001). The 5 criteria

Conservation BiologyVolume 28, No. 3, 2014

Keith et al. 813

Table 1. Current conservation status of Assa darlingtoni in different geographic domains.

Domain List Status Source

Global IUCN Red List of threatened species least concern http://www.iucnredlist.org/details/full/41130/0,downloaded 14 January 2013

National Australian Environment Protectionand Biodiversity Conservation Act1992

not listed http://www.environment.gov.au/cgi-bin/sprat/public/publicthreatenedlist.pl?wanted= fauna, downloaded 14 January 2013

State Queensland Nature Conservation Act1992

near threatened http://www.ehp.qld.gov.au/wildlife/threatened-species/near-threatened/near_threatened_animals.html,downloaded 14 January 2013

State NSW Threatened SpeciesConservation Act 1995

vulnerable http://www.environment.nsw.gov.au/threatenedspeciesapp/profile.aspx?id = 10070,downloaded 14 January 2013

(A–E, respectively) assess population reductions over10 years or 3 generations, whichever is longer; geo-graphic range size in combination with severe fragmen-tation, number of locations, continuing declines, and ex-treme fluctuations; population size in combination withcontinuing declines, population structure, and extremefluctuations; population size and range size in isolationof other factors; and quantitative estimates of extinctionrisk. The criteria were interpreted in accordance withIUCN guidelines (IUCN 2011).

Although the data permit a limited assessment of cur-rent status, they were insufficient for comprehensive as-sessments of the criteria at multiple temporal referencepoints. We therefore used a population model to esti-mate all the variables required for red list assessment. Themodel was constructed and parameterized from availabledata and expert knowledge. We incorporated uncertaintyby constructing fuzzy estimates of population variablescomprising best estimates (mean of all simulations) withplausible upper and lower bounds drawn from the 5thand 95th percentiles of the model output (Akcakaya et al.2000). We used RAMAS Red List (Akcakaya et al. 2007) tocalculate the risk categories from the fuzzy estimates. Toexamine the robustness of assessment outcomes to miss-ing data, we re-assessed the overall status after excludingthe criterion that returned the highest category of risk(Keith et al. 2000).

In implementing red list assessments, RAMAS Red Listallows users to specify their attitude to risk and uncer-tainty by setting values to represent their risk toleranceand dispute tolerance (Akcakaya et al. 2000). Values ofrisk tolerance range from 0 (extremely risk-averse precau-tionary attitude to risk and uncertainty) to 1 (extremelyrisk-prone evidentiary attitude). For all assessments, weset risk tolerance to 0.45, a slightly more precautionaryattitude than the balance of evidence, consistent withIUCN’s (2001) recommendation that assessors take a“precautionary but realistic” attitude to risk and uncer-tainty. Values of dispute tolerance range from 0 for in-clusion of all estimates to 1 for inclusion of only theconsensus estimates (in this case the best estimate). For

all assessments, we set dispute tolerance at 0.5, excludingthe most extreme estimates of each variable (Akcakayaet al. 2000, 2007).

Predictive Modeling

Locality data were assembled from survey records andspecimens held by the NSW Office of Environment andHeritage, Queensland Environmental Protection Author-ity, and Australian and Queensland Museums. All recordswere checked by experts (HH, MM) to remove erroneousand unreliable records. The 758 reliable records wererandomly thinned so that each location was separatedby a minimum distance of 1.5 km to reduce spatial de-pendence and survey bias. A further 7 records were dis-carded because they were located at sites outside themapped area of native forest, on which this species iswholly dependent. This is likely to have resulted fromsmall positional inaccuracies rather than forest loss af-ter the species was recorded. Remaining records weretransposed onto a 9 arc-second (approximately 285 m)grid, producing a final data set for modeling that included127 presence records. A set of 9100 background samplepoints (Phillips et al. 2006) was generated on the same285 m grid from within a region defined by 4 biore-gions (South Eastern Queensland, NSW North Coast, NewEngland, and Nandewar) and 2 subregions (Banana-Auburn Ranges and Eastern Darling Downs) within theBrigalow Belt South bioregion (Department of the Envi-ronment 2013). Background sites were only taken fromsites with native vegetation. The use of bioregions wasintended to constrain the absence records to ecologicallyplausible regions of species occurrence under presentday or future climatic conditions, thereby sharpening thepredictions of habitat area.

A set of environmental data layers (9 arc-second gridcells) was selected by experts (HH, MM) as potential pre-dictors of habitat for A. darlingtoni (Table 2). A seriesof alternative species distribution models (SDMs) withdifferent subsets of predictors was fitted with a maxi-mum entropy algorithm, MAXENT (Phillips et al. 2006),

Conservation BiologyVolume 28, No. 3, 2014

814 Detecting Extinction Risk from Climate Change

Table 2. Environmental variables used as predictors of habitat for Assa darlingtoni.

Predictor layer Source Best modela

Annual mean temperature BioClim v6.0 XMean diurnal temperature range BioClim v6.0 XTemperature seasonality BioClim v6.0Annual mean moisture index BioClim v6.0Mean moisture index of the lowest quarter BioClim v6.0 XMaximum topographic wetness index XSlope XSubstrate mask (excluding coastal sands) Keith (2011) with additions for SE Qld +Native vegetation (present/absent) Keith (2011) with additions for SE Qld X

aSix (x) of the predictors were included in the best model, whereas a substrate mask (+) was applied to exclude predicted habitat from coastalsands.

which performed well in comparative tests with othermethods (Elith et al. 2006). We used hinge features witha regularisation multiplier of 1.5 to construct modelsthat had reasonably smooth responses that could be ex-trapolated in a biologically realistic manner (Elith et al.2010). Model outputs were evaluated using the area un-der the receiver operating characteristic curve (AUC) forthe training points used in model fitting and subjectivereview by experts familiar with the species’ habitat anddistribution (HH, MM). The most suitable model included7 predictors (Table 2) and was masked to exclude clearedland.

The future distribution of A. darlingtoni was predictedthrough changes in the climatic variables included in theSDM (Table 2). Changes in climatic variables were pro-jected using 4 global circulation models (GCMs; IPCC2007) that perform well for temperature and rainfallanomalies in eastern Australia (Suppiah et al. 2007):CSIRO-Mk3, GDFL-CM2, MPMP-ECHAM5, and UKMO-HADCM3. Two greenhouse gas emission scenarios wereused from each model (A1FI and A2), of which A1FI hasa larger temperature increase that most closely resem-bles the realized trajectory (Peters et al. 2013). In addi-tion to projecting models to baseline (1990) conditions,projections were generated for years 2030, 2050, 2070,and 2100. Projected distributions were then generatedfor each year from 2000 to 2099 by linear interpolation(Keith et al. 2008).

A spatially explicit, stochastic matrix population modelbased on annual time steps was constructed in RAMASMetapop v5 (Akcakaya & Root 2005). Three life historystages were recognized in the model: tadpoles, juveniles,and adults. Only females were modeled because we as-sumed that the availability of males (which can care formultiple clutches) did not limit population growth. Meanrates of survival and fecundity were estimated from liter-ature (Ehmann & Swan 1985) and the authors’ unpub-lished data and expert knowledge on small terrestrialectothermic vertebrates (Table 3). Mean observed clutchsizes (range 8–18, mean = 13 eggs) were doubled toestimate annual fecundity because females are capable ofdepositing 2 clutches in a season.

Table 3. Estimated vital rates (coefficients of variation) for the matrixpopulation model for A. darlingtoni.

Tadpole Juvenile Adult

Tadpole 0 0 5(2%)Juvenile 0.6(2%) 0 0Adult 0 0.2(2%) 0.4(1%)

In the absence of a population census time series, themean rates were assumed to vary yearly because of en-vironmental stochasticity and to have small coefficientsof variation (Table 3), reflecting the relatively stable con-ditions on the rainforest floor and assuming adult sur-vival was less sensitive to environmental stochasticitythan other life history processes. We used a lognormaldistribution to model environmental stochasticity of allvital rates. We assumed variation in survival and fecun-dity were correlated within populations according to adistance function derived from annual rainfall data fromstations in the region (Fordham et al. 2012). Demographicstochasticity was incorporated using a binomial distribu-tion for survival and a Poisson distribution for fecundity(see Akcakaya & Root 2005).

Density dependence in A. darlingtoni was representedby a ceiling model (Akcakaya & Root 2005) because thespecies has a short generation length and lacks territorialbehavior and strong aggressive interactions between indi-viduals. Thus, if carrying capacity was exceeded, the sizeof a population was adjusted to the carrying capacity inthe subsequent year. The carrying capacity of each popu-lation was estimated from the suitability values predictedby the SDM (Keith et al. 2008).

Populations were defined spatially as 10 × 10 km gridcells to enable computationally practical modeling of dis-persal between adjacent cells in the landscape and toaverage local variations in habitat suitability. Dispersalrates were estimated by experts familiar with movementof A. darlingtoni in the field (MM, HH), assuming thatonly juveniles move and that they move only betweenadjacent cells mostly over distances <100 m. Based onthe geometry of the grid, we estimated that the annual

Conservation BiologyVolume 28, No. 3, 2014

Keith et al. 815

Figure 1. Population trajectories of Assa Darlingtoni under alternative future climate projections: (a) stableclimate and CSIRO Mk3, (b) MPMP-ECHAM5, (c) UKMO-HADCM3, and (d) GDFL-CM2. Population size expressedas percentage of initial number of mature females. Global circulation models and emission scenarios defined inIPCC (2007).

probability of a juvenile dispersing to a neighbouring cellwas 0.002.

The SDM and population model were coupled usingthe procedure described by Keith et al. (2008). Mod-eled 285 m cells with suitability values less than the fifthpercentile of values at training points (0.082) were as-sumed unsuitable and set to zero. These 285 m cells werethen aggregated into a 9975 m (c. 10 km) grid, in whicheach cell was defined as a population unit for modelingpurposes. Based on call data for males recorded in thefield and field sampling of leaf litter indicating a 1:1 sexratio (M. Mahony, unpublished data), optimal habitat wasestimated to support 160 adult female frogs per hectare(1300/285 m cell, aggregating to 1,592,500/9,975 m cell).The carrying capacity of each population was thereforeestimated as ths∗1300, where ths is the sum of suitabilityvalues within respective 9975 m cells. A second thresholdbased on the first percentile of summed suitability (29.7)was applied to exclude 9975 m cells with a very small andpotentially diffuse area of habitat. The remaining 9975 mcells were designated as individual populations.

For each population, we calculated trends in carryingcapacity under climate change with the projections ofthe SDM from each combination of GCM and emissionscenario. These trends were incorporated into the model

simulations. Nine model scenarios were run; 8 GCM-emission combinations and a stable climate scenario inwhich carrying capacities of all populations were heldconstant. We ran simulations over 1000 replicates of100 annual time steps (2000–2099). Mean estimates ofextinction risk and population size (number of adultfemales) and their standard deviations and the spatialconfiguration of populations for each time step wereextracted from the model output to calculate the vari-ables required for assessing the IUCN Red List criteria.Assessments were made every 20 years from 2010 to2090.

Results

The current population of A. darlingtoni was estimatedto include approximately 1.7 million mature females. Thepopulation remained stable for 100 years when modeledunder a stable climate (Fig. 1). Under all modeled fu-ture climate change scenarios, however, the populationremained stable only until about 2040–2050, with subse-quent declines projected to occur at different rates fordifferent combinations of GCM and emission scenario.Under the most severe projection (CSIRO-Mk3 A1FI), the

Conservation BiologyVolume 28, No. 3, 2014

816 Detecting Extinction Risk from Climate Change

Table 4. IUCN Red List status of Assa darlingtoni assessed for 2010 and 4 future dates under a stable climate and on the basis of projections from4 different global circulation models and 2 emission scenarios. Omission shows the overall status if data were omitted for the criterion supportingthe highest status.

A1FI scenarioa A2 scenarioa

Circulation model and year Status criteria Omission Status criteria Omission

Stable climate2010 LC(LC-LC) – LC2030 LC(LC-LC) – LC2050 LC(LC-LC) – LC2070 LC(LC-LC) – LC2090 LC(LC-LC) – LCCSIRO-Mk32010 VU(VU-VU) E LC LC(LC-LC) – LC2030 VU(VU-VU) E LC VU(VU-VU) B1,B2,E VU2050 VU(VU-EN) (A2),B2,E VU EN(EN-EN) B1 VU2070 EN(EN-EN) B1,B2 VU EN(EN-EN) B1,B2 VU2090 CR(CR-CR) A3,A4,B1,B2,E CR CR(CR-CR) A3,A4 ENGDFL-CM22010 LC(LC-LC) – LC LC(LC-LC) – LC2030 LC(LC-LC) – LC LC(LC-LC) – LC2050 VU(VU-VU) B1 LC VU(VU-VU) B1 LC2070 EN(VU-EN) B1 VU VU(VU-VU) B1,B2,E VU2090 EN(EN-EN) B1 VU EN(EN-EN) B1 VUMPMP-ECHAM52010 LC(LC-LC) – LC LC(LC-LC) – LC2030 VU(LC-VU) E LC VU(LC-VU) E LC2050 EN(EN-EN) B1 VU VU(VU-EN) B1,B2,E VU2070 EN(EN-EN) B1,B2 VU EN(EN-EN) B1 VU2090 EN(EN-EN) A2,A3,B1,B2,E EN EN(EN-EN) B1,B2 ENUKMO-HADCM32010 LC(LC-LC) – LC LC(LC-LC) – LC2030 VU(LC-VU) B1,E VU VU(LC-VU) E LC2050 EN(VU-EN) B1 VU VU(VU-VU) B1,B2,E VU2070 EN(EN-EN) B1,B2 EN EN(EN-EN) B1 VU2090 EN(EN-EN) B1,B2,E EN EN(EN-EN) B1,B2 EN

aStatus is based on best estimate with plausible bounds in parentheses. IUCN criteria determining overall status are listed in parentheses if thebest estimate matches overall status, but plausible bounds span lower categories of risk.

species was categorized as extinct in the wild by 2095(Fig. 1). In other projections, the species remained ex-tant, but total population size was reduced by 39% (GDFL-CM2 A2 scenario) to 96% (CSIRO-Mk3 A2 scenario) over100 years (Fig. 1).

According to the red list criteria, A. darlingtoni wouldbecome eligible for listing as a threatened species as earlyas 2010 or as late as 2050, depending on the pattern of de-cline in the population and distribution projected underdifferent GCMs and emission scenarios (Table 4). Undera stable climate scenario, the status of the species wasLeast Concern throughout the 21st century. The speciesqualified for listing in 2010 in only one projection (CSIRO-Mk3 A1FI). In all 8 combinations of GCM and emissionscenario, early eligibility for listing depended on criterionE or B, with both criteria supporting the first listing in 2 ofthe 8 cases (Table 4). The estimates of extinction risk over100 years (required to assess the Vulnerable category un-der criterion E) were uncertain in the assessments carriedout for years later than 2010 because projections wereonly available to 2100. Hence, we used the shape of the

population trajectory to estimate whether extinction wasa plausible outcome over 100-year time frames ending in2110, 2130, 2150, 2170, and 2190. The plausible boundsof these assessment outcomes always included the LeastConcern category, which recognizes the uncertainty ofan extinction outcome. Criterion B and to a lesser extentcriterion A affected the overall status of the species in thesecond half of the century and often over-rode criterionE by 2070 or 2090 (Table 4). Criteria C and D did notdetermine the overall status of A. darlingtonii in anyassessment (Table 4).

In later years, the red list assessments became morerobust because the overall status was generally supportedby more criteria and more subcriteria than in earlieryears (Table 4). Consequently, the overall status alwaysremained threatened when data for the highest risk crite-rion were omitted from assessments carried out in 2070and 2090, but status was reduced to Least Concern forsome combinations of GCM and emission scenario whendata for the highest risk criterion were omitted from as-sessments carried out in 2030 and 2050.

Conservation BiologyVolume 28, No. 3, 2014

Keith et al. 817

Discussion

Early Detection of Extinction Risks Under Climate Change

Despite short time frames for assessing population de-clines, our results for A. darlingtoni suggest that forsome species with short life cycles the IUCN Red Listcriteria may be more sensitive to extinction risks posedby climate change than anticipated. Under the most ex-treme scenario in which extinction occurred in 100% ofsimulations by 2095, A. darlingtoni first qualified undercriterion A (population decline) for listing as threatened(Vulnerable category) in 2050. Under the 4 least severescenarios, A. darlingtoni did not qualify for listing asthreatened under criterion A at any time during the 21stcentury and the risk of extinction by 2100 was estimatedto be zero. Risks could not be quantified beyond 2100;the shapes of population trajectories suggest species’persistence could be certain for several decades into the22nd century.

Criterion A was not the only criterion that identifiedwhen A. darlingtoni was at appreciable risk of extinc-tion. In most cases, A. darlingtoni qualified for threat-ened status under criteria B (distribution size) or E (quan-titative estimates of extinction risk) before it met criterionA, and the overall outcome of assessments never reliedupon criterion A alone. Under the most severe scenario,the species was assessed as Vulnerable under criterionE as early as 2010. By the time it qualified as Vulnera-ble under criterion A in 2050, it also qualified at thatlevel under criterion B. Hence the red list criteria iden-tified the species as being at risk of extinction 85 yearsbefore it actually went extinct and 40 years before itmet criterion A. Under criterion A alone, the speciesqualified for listing 45 years before its modeled extinc-tion. For all other modeled climate change scenarios,A. darlingtoni qualified for threatened listing at least80 years before its inferred date of extinction based onthe shape of its population trajectory in the late 21stcentury.

The time frames for assessing criterion E (10 years or3 generations, 20 years or 5 generations and 100 years,respectively for CR, EN, and VU) are longer than thosefor criterion A (10 years or 3 generations for all threatcategories), and criterion B does not specify explicittime frames (only qualitative evidence of a continuingdecline). These differences between individual criteriaunderpin their ensemble properties and buffer againstextreme sensitivity of risk assessment outcomes to shortgeneration lengths.

By indicating that A. darlingtoni is likely to qualifyfor threatened status 40–80 years before it goes extinct(depending on which criteria are assessable), our resultssuggest that red list criteria perform reasonably well asan early-warning system. How much lead time an early-warning system should give depends on the practicali-

ties of environmental policy and management, bureau-cratic or political inertia, and the anticipated speciesresponse times to various actions. This mix of factorssuggests a need for a minimum of several decades warn-ing between initial listing and extinction, with longerlead times required if a desire for greater certainty orsocio-economic costs motivated a delay in action untilEndangered or Critically Endangered listing. The longlags associated with climate change mitigation and somemanagement measures underscores the need for earlyaction.

Limitations of Analyses

We examined only a single species chosen specifically torepresent an identified potential weakness in the IUCNRed List criteria, namely the short time frame for assess-ing future declines relative to the expected time scaleof climate change impacts. Our focus on A. darlingtoni,with contrasting population trajectories under differentfuture climates, permitted a detailed examination of theassessment processes and development of an analyticalapproach. Our approach will be useful for evaluating thesensitivity of the red list criteria to climate change impactson species with more diverse life histories. Additionalstudies of short-lived species with different temporal pat-terns of decline and of longer lived species with differentclimate-life history dependencies would further advanceunderstanding of how the red list criteria perform as anearly warning system for climate change impacts.

We assumed our study species would not show ap-preciable evolutionary response to climate change thatmay influence its future persistence. However, specieswith short generation lengths, such as A. darlingtonii,may experience mutation and natural selection at ratesthat are rapid enough to enhance fitness under changingclimates (Hoffmann & Sgro 2011). Phenotypic or behav-ioral plasticity may also influence in situ persistence ofpopulations under climate change. Although our capacityto predict which short-lived species might undergo evolu-tionary responses to climate change is currently limited,such effects have been demonstrated in a range of short-lived species (Skelley et al. 2007; Franks & Hoffmann2012). We expect evolutionary responses to delay thetime at which listing in a threatened category is triggered,as well as the time to eventual extinction. This is unlikelyto result in shorter warning times; hence, our conclusionsshould be conservative whether or not evolutionary re-sponses occur. Species with long generation times, how-ever, may be more sensitive to climate change becausegenomic change is less likely to keep pace with climatechange.

For analytical economy, we used averaged estimates ofpopulation size and distribution (with confidence inter-vals) across 1000 simulations for the variables addressedby the criteria. In reality, each of the 1000 replicates

Conservation BiologyVolume 28, No. 3, 2014

818 Detecting Extinction Risk from Climate Change

represents an alternative future scenario that could beassessed individually against the criteria. Averaging mayhave underestimated the trends in some scenarios. Therisk assessment outcomes from averaged estimates mayapproximate modal or median assessments based on in-dividual trajectories, but this remains to be tested.

Although we incorporated uncertainty in our analysesby implementing fuzzy calculations on best estimateswith upper and lower bounds for all input variables(Akcakaya et al. 2000), we are likely to have underesti-mated the uncertainty relative to real-world assessments.This is because our simulated population trajectories didnot incorporate observer error, which is likely to be asignificant source of uncertainty in real-world estimatesof population sizes based on field surveys (Burgman et al.1999; Regan et al. 2002). Furthermore, although we in-corporated model uncertainty by using projections basedon multiple GCMs, we ignored other components ofmodel uncertainty, such as in the demographic modeland SDM. For example, biophysical models may projectdifferent distributions to those produced by correlativeSDMs (Kearney & Porter 2009). More realistic levels ofuncertainty could be achieved by using simulations basedon several plausible demographic models and SDMs andadding a random error to modeled estimates to simulatefield sampling (Zurell et al. 2010).

Factors Influencing Performance of Red List Criteria

Several factors potentially influence the performance ofthe red list criteria for species threatened by climatechange. In our analyses the earliest detection of threatrelied on a single criterion. Often this was criterion E,which is more likely to detect extinction risks than areother criteria (McLean & Wilson 2011), but this criterionis typically more data demanding and time consuming toassess. When criterion E was not evaluated, several sim-ulations showed that a taxon could remain listed as LeastConcern for several decades despite being at appreciable(>10%) risk of extinction within a century. This could beproblematic when data are insufficient to support a quan-titative analysis of extinction risk. Few of the currentlyred-listed taxa have been assessed on the basis of criterionE (IUCN 2011). This may sometimes be because of limitedexpertise or time to construct appropriate models, ratherthan a lack of data. A. darlingtoni, for example, had notpreviously been assessed under criterion E, even thoughsufficient data were available.

Risk assessments for later in the 21st century wererobust to missing data. Nevertheless, our results rein-force IUCN’s recommendation that “each taxon shouldbe evaluated against all the criteria” (IUCN 2001). In somecases this may require engaging additional expertise orresources to support assessments. Our results showedthat the models necessary to assess criterion E also hadthe capacity to produce bounded estimates of population

size, distribution area, and trends in both population anddistribution, enabling more comprehensive assessmentsof criteria A–D than otherwise possible.

Optimal performance of the red list categories andcriteria relies on correct interpretation of the criteriaand supporting concepts. Our risk assessments closelyfollowed the current guidelines to derive quantitativeestimates of the required variables over the appropriatetime frames and spatial scales (IUCN 2011). We also in-corporated uncertainty in the calculations to determinethe range of plausible risk assessment outcomes. Severalassessments of extinction risk under climate change usedmodified or incomplete versions of the criteria (Akcakayaet al. 2006). This may affect the performance and consis-tency of the criteria and their ability to estimate relativerisks.

To inform proactive policy and management as climatechanges and its impacts unfold, an early warning systemrelies upon monitoring of species populations and dis-tributions to support regular risk assessments (Butchartet al. 2010). Responsive management would likely re-quire more frequent assessments than are currently im-plemented. The most recent IUCN Red List assessmentfor A. darlingtoni was conducted a decade ago (Hero etal. 2004).

Acknowledgments

This research was funded by a Linkage grant from the Aus-tralian Research Council (LP0989537). It was designedand were models constructed at 2 workshops funded bythe Australian Centre for Ecological Analysis and Synthe-sis (ACEAS), within the Terrestrial Ecosystem ResearchNetwork. We thank M. Kearney for discussions on designof the models.

Literature Cited

Akcakaya, H. R., S. Ferson, M. A. Burgman, D. A. Keith, G. M. Mace,and C. R. Todd. 2000. Making consistent IUCN classifications underuncertainty. Conservation Biology 14:1001–1013.

Akcakaya, H. R., S. H. M. Butchart, G. M. Mace, S. N. Stuart, and C.Hilton-Taylor. 2006. Use and misuse of the IUCN Red List criteria inprojecting climate change impacts on biodiversity. Global ChangeBiology 12:2037–2043.

Akcakaya, H. R., S. Ferson, and W. R. Root. 2007. RAMAS Redlist. Version3.0. Applied Biomathematics, Setauket.

Akcakaya, H. R., and W. R. Root. 2005. RAMAS Metapop. Version 5.0.Applied Biomathematics, Setauket.

Bell, B. D., S. Carver, N. J. Mitchell, and S. Pledger. 2004. The recentdecline of a New Zealand endemic: How and why did populations ofArchey’s frog Leiopelma archeyi crash over 1996–2001? BiologicalConservation 120:193–203.

Burgman, M. A., D. A. Keith, and T. V. Walshe. 1999. Uncertainty incomparative risk analysis for threatened Australian plant species.Risk Assessment 1:585–598.

Butchart, S. H. M., et al. 2010. Global biodiversity: indicators of recentdeclines. Science 328:1164–1168.

Conservation BiologyVolume 28, No. 3, 2014

Keith et al. 819

Cameron, A. 2012. Refining extinction risk using models. Pages 41–71in L. Hannah, editor. Saving a million species: extinction risk fromclimate change. Island Press, Washington DC.

Department of the Environment. 2013. Australia’s bioregions (IBRA7).Available from http://www.environment.gov.au/parks/nrs/science/bioregion-framework/ibra/index.html#ibra (accessedMarch 2013).

Ehmann, H., and G. Swan. 1985. Reproduction and development in theMarsupial Frog, Assa darlingtoni (Myobatrachidae, Anura). Pages279–285 in G. Grigg, R. Shine, and H. Ehmann, editors. Biology ofAustralasian Frogs and Reptiles. Royal Zoological Society of NewSouth Wales, Sydney.

Elith, J., et al. 2006. Novel methods improve prediction of species’distributions from occurrence data. Ecography 29:129–151.

Elith, J., M. Kearney, and S. J. Phillips. 2010. The art of modellingrange-shifting species. Methods in Ecology and Evolution 1:330–342.

Foden, W. B., et al. 2013. Identifying the world’s most climate changevulnerable species: a systematic trait-based assessment of all birds,amphibians and corals. PLoS ONE 8:e65427.

Fordham, D. A., et al. 2012. Plant extinction risk under climate change:Are forecast range shifts alone a good indicator of species vul-nerability to global warming? Global Change Biology 18:1357–1371.

Franks, S. J., and A. A. Hoffmann. 2012. Genetics of climate changeadaptation. Annual Review of Genetics 46:185–208.

Hannah, L. 2012. Are a million species at risk? Pages 3–9 in L. Hannah,editor. Saving a million species: extinction risk from climate change.Island Press, Washington DC.

Hero, J.-M., J. Clarke, and E. Meyer. 2004. Assa darlingtoni. In: IUCN2012. IUCN Red List of Threatened Species. Version 2012.2.<www.iucnredlist.org>. Downloaded on 17 January 2013.

Hof, C., M. B. Araujo, W. Jetz, and C. Rahbek. 2011. Additive threatsfrom pathogens, climate and land-use change for global amphibiandiversity. Nature 480:516–519.

Hoffmann, A. A., and C. M. Sgro. 2011. Climate change and evolutionaryadaptation. Nature 470:479-485.

Houlahan, J. E., C. S. Findlay, B. R. Schmidt, A. H. Meyer, and S. Kuzmin.2000. Quantitative evidence for global amphibian population de-clines. Nature 404:752–755.

IPCC. 2007. Climate change 2007: synthesis report. Contribution ofWorking Groups I, II and III to the Fourth Assessment Report of theIntergovernmental Panel on Climate Change. IPCC, Geneva.

IUCN. 2001. IUCN Red List categories and criteria. Version 3.1. IUCN,Gland.

IUCN. 2011. Guidelines for application of IUCN Red List Criteria.Version 9. IUCN Species Survival Commission. IUCN, Gland &Cambridge.

Kearney, M., and W. Porter. 2009. Mechanistic niche modelling: com-bining physiological and spatial data to predict species’ ranges.Ecology Letters 12:334–350.

Keith, D. A. 2011. Relationships between geodiversity and vegetation insouth-eastern Australia. Proceedings of the Linnean Society of NewSouth Wales 132:5–26.

Keith, D. A., T. D. Auld, M. K. J. Ooi, and B. D. E. Mackenzie. 2000.Sensitivity analyses of decision rules in World Conservation Union(IUCN) Red List criteria using Australian plants. Biological Conser-vation 94:311–319.

Keith, D. A., et al. 2004. Protocols for listing threatened species canforecast extinction. Ecology Letters 7:1101–1108.

Keith, D. A., et al. 2008. Predicting extinction risks under climatechange: coupling stochastic population models with dynamic bio-climatic habitat models. Biology Letters 4:560–563.

Levinsky, I., F. Skov, J. C. Svenning, and C. Rahbek. 2007. Potentialimpacts of climate change and land-use change on the distributionand diversity patterns of European mammals. Biodiversity and Con-servation 16:3803–3816.

Mace, G. M., et al. 2008. Quantification of extinction risk: IUCN’s systemfor classifying threatened species. Conservation Biology 22:1424–1442.

McLean, I. M. D., and R. J. Wilson. 2011. Recent ecological re-sponses to climate change support predictions of high extinctionrisk. Proceedings of the National Academy of Science 108:12337–12342.

NSW Environment and Heritage. 2013. Pouched frog threatenedspecies profile. Available from http://www.environment.nsw.gov.au/threatenedspeciesapp/profile.aspx?id=10070 (accessedMarch 2013).

Parry, M. L., O. F. Canziani, J. P. Palutikof, P. J. van der Linden, andC. E. Hanson. 2007. Climate change 2007: impacts, adaptation andvulnerability. Contribution of Working Group II to the Fourth As-sessment Report of the Intergovernmental Panel on Climate Change.Cambridge University Press, Cambridge.

Peters, G. P., et al. 2013. The challenge to keep global warming below2 °C. Nature Climate Change 3:4–5.

Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximumentropy modelling of species geographic distributions. EcologicalModelling 190:231–259.

Pounds, J. A., et al. 2006. Widespread amphibian extinctions from epi-demic disease driven by global warming. Nature 439:161–167.

Regan, H. M., M. Colyvan, and M. A. Burgman. 2002. A taxonomy ofand treatment of uncertainty for ecology and conservation biology.Ecological Applications 12:618–628.

Skelley, D. K., et al. 2007. Evolutionary responses to climate change.Conservation Biology 21:1353–1355.

Stuart, S. N., et al. 2004. Status and trends of amphibian declines andextinctions worldwide. Science 306:1783–1786.

Suppiah, R., et al. 2007. Australian climate change projections derivedfrom simulations performed for the IPCC 4th Assessment Report.Australian Meteorological Magazine 56:131–152.

Thomas, C. D., et al. 2004. Extinction risk from climate change. Nature427:145–148.

Thuiller, W., et al. 2008. Predicting global change impacts on plantspecies’ distributions: future challenges. Perspectives in Plant Ecol-ogy Evolution and Systematics 9:137–152.

Tyler, M. J. 1985. Reproductive modes in Australian Amphibia. Pages265–267 in G. Grigg, R. Shine and H. Ehmann, editors. Biology ofAustralasian Frogs and Reptiles. Royal Zoological Society of NewSouth Wales, Sydney.

Vie, J-C., C. Hilton-Taylor, and S. N. Stuart. 2009. Wildlife in a changingworld—an analysis of the 2008 IUCN Red List of threatened species.IUCN, Gland.

Westoby, M., and M. A. Burgman. 2006. Climate change as a threateningprocess. Austral Ecology 31:549–550.

Zurell, D., et al. 2010. The virtual ecologist approach: simulating dataand observers. Oikos 119:622–635.

Conservation BiologyVolume 28, No. 3, 2014