Predicting potential distributions of two rare allopatric sister species, the globally threatened Doliornis cotingas in the Andes

J. Field Ornithol. 81(4):325–339, 2010 DOI: 10.1111/j.1557-9263.2010.00289.x

Predicting potential distributions of two rare allopatricsister species, the globally threatened Doliornis cotingas

in the Andes

Frederic Jiguet,1,3 Morgane Barbet-Massin,1 and Pierre-Yves Henry2

1UMR 7204 MNHN-CNRS-UPMC, Museum National d’Histoire Naturelle, 55 rue Buffon, CP 51,75005 Paris, France

2UMR 7179 MNHN-CNRS, Museum National d’Histoire Naturelle, 1 avenue du Petit Chateau,91800 Brunoy, France

Received 28 October 2009; accepted 8 July 2010

ABSTRACT. Conservation of rare, elusive species is difficult because of limited knowledge of their biology anddistribution. The two species of Doliornis cotingas are known from 7 (Chestnut-bellied Cotinga, D. remseni) and13 (Bay-vented Cotinga, Doliornis sclateri) locations, respectively. Their limited ranges in combination with habitatloss make them vulnerable to extinction. We modeled the potential distribution of these two rare, allopatric sisterspecies separated by an orographic barrier using species distribution modeling with an ensemble forecast approachusing eight modeling techniques. Predicted distributions (with conservative thresholds of probability of presence,i.e., lowest presence thresholds) for these sister species showed virtually no overlap, although their respective nicheswere statistically not dissimilar. Hence, the existence of the recently discovered Chestnut-bellied Cotinga could nothave been predicted from the range of its sister species, unless using very crude distribution models (with extensiveextrapolation). New areas of likely occurrences were identified for both species, and the genus, and will be useful fordirecting future field searches. The estimates of potentially suitable range for both species still qualify both species as“vulnerable to extinction.” Our study illustrates how opportunistic records collected by field ornithologists can beobjectively transformed, with the help of existing software, into information potentially useful in the conservationof rare species.

RESUMEN. Prediccion de las distribuciones potenciales de dos especies hermanasalopatricas, los globalmente amenazados cotingas Doliornis en los Andes

Conservacion de especies escasas es difıcil por falta de documentacion de su biologıa y su distribucion. Las dosespecies de cotingas Doliornis son conocidas de solo 7 (Cotinga de Remsen, D. remseni) y 13 (Cotinga de Sclater,D. sclateri) registros localizados precisamente. Su pequena area de distribucion y desaparicion de su habitathace que son vulnerables a la extincion. Modelamos la distribucion potencial de esas dos especies hermanasalopatricas, separadas por una barrera orografica, utilizando ocho tecnicas de modelo de distribucion con unmetodo de pronostico de ensamble. Las distribuciones predichas (con niveles umbrales de probabilidad de presenciaconservativos, el nivel umbral mınimo de presencia) para cada especie hermana eran casi completamente distintas.Ası, el rango de distribucion de la especie descubierta recientemente, D. remseni, no hubiera podido ser predichocon los registros de su especie hermana, D. sclateri, aunque hayamos utilizado modelos muy basicos (con extensivasobre-prediccion). Nuevas areas de alta probabilidad de ocurrencia fueron identificadas para cada especie, y elgenero, y podran servir para orientar futuros inventarios de campo. Las estimaciones de los rangos potencialmentefavorables para cada especie siguen calificandolas como ‘vulnerables a la extincion’. Este estudio ilustra como registrosmasivamente colectados de manera oportunista por ornitologos de campo pueden ser transformados objetivamente,con programas existentes, en conocimientos utiles para la investigacion y la conservacion de especies escasas.

Key words: allopatric sister taxa, alloprediction, BIOMOD, distribution range, species distribution modeling,tropical montane forest

Field ornithologists are the very first providersof distributional data for rare or elusive species.Following the discovery of new sites of occur-rence, revisions of species distribution ranges areregularly published. Often, only a descriptiveor mapping interpretation of these records is

3Corresponding author. Email: [email protected]

provided. Conclusions are highly sensitive toreporting biases, such as stochastic nondetectionor inaccessibility of potentially suitable areas (acommon problem in the tropics). One methodfor making a parsimonious use of these scat-tered, anecdotal data is to identify probabilisticspecies distributions with models. These modelscombine occurrence data with environmentaldata to statistically identify areas resembling the

C©2010 The Authors. Journal of Field Ornithology C©2010 Association of Field Ornithologists

325

Journal of Field Ornithology

326 F. Jiguet et al. J. Field Ornithol.

locations where the species is known to occur(Araujo and New 2007, Pearson et al. 2007,Thuiller et al. 2009). Such niche modeling isbecoming an important tool in conservationand biogeography, providing spatially explicit,predicted distribution ranges that are robustto the usual lack of previous ecological anddistributional knowledge of poorly known taxa(Guisan et al. 2006). This feature is particularlyuseful for species known from a limited num-ber of locations, including poorly detectable,inconspicuous or low-density species, speciesin remote, poorly investigated habitats, or re-cently discovered species (Raxworthy et al. 2003,Engler et al. 2004, Franklin et al. 2009, Mariniet al., 2010). These models are increasingly usedby field ornithologists to exploit bird distribu-tional data (e.g., Rojas-Soto et al. 2008, Henryand Aznar 2009).

From a conservation perspective, niche mod-eling provides a methodological framework foridentifying spatially optimal conservation strate-gies, such as designing networks of priorityareas (Marini et al. 2009a) or targeting con-servation efforts for threatened species (Petersonand Robins 2003, Thorn et al. 2009, Mariniet al. 2009b) and regional diversity hotspots(Myers et al. 2000, Garcıa 2005). From a biogeo-graphical perspective, one achievement of nichemodeling is the ability to predict the distribu-tion of unknown, still to be discovered, speciesfrom modeled distributions of closely relatedknown species in cases of niche conservatismin evolutionary time (Raxworthy et al. 2003) orniche similarity between phylogenetically closelyrelated species (Warren et al. 2008). Theorysupports this assumption of niche conservatism,or similarity between closely related species,predicting low niche differentiation betweenvicariant species over evolutionary time scales(Ricklefs and Latham 1992). Further em-pirical work documented the conservativenature of ecological niche evolution by suc-cessfully extrapolating distributions of closelyrelated species using data from the other species(Peterson et al. 1999), and phylogenetic infor-mation has been combined with species distribu-tion modeling to explore speciation mechanisms(Graham et al. 2004). Indeed, in practice, forclosely related allopatric species pairs, nicheconservatism can be deduced from areas ofoverprediction (i.e., areas where presence is pre-dicted outside of the actual range) in one-species

models that actually fit the known descriptionrange of the other species (Peterson et al. 1999,Raxworthy et al. 2003, Pearson et al. 2007).

Our objective was to improve our knowl-edge of two rare sister, tropical humid mon-tane forest bird species, Bay-vented Cotingas(Doliornis sclateri) and Chestnut-bellied Coting-as (D. remseni), by analyzing previously reported(Appendix 1) and new (Henry 2008) occur-rence data. These species are restricted to hu-mid montane forests at elevations ranging from2740—to 3800 m in the central and northernAndes (Ridgely and Tudor 1994, Ridgely andGreenfield 2001, BirdLife International 2010).Bay-vented Cotingas were discovered in 1871at Maraynioc, Peru (Taczanowski 1874; seeAppendix 1) and subsequently reported at 13other locations, all on the east slope of theeastern Peruvian Andes (Appendix 1). Chestnut-bellied Cotingas were first observed in the fieldin 1989 and described as recently as 1994(Robbins et al. 1994), and are known fromonly 10 locations on the eastern slope of theAndes in Ecuador and from the central Andes ofColombia (Appendix 1). These two species areseparated by an orographic barrier, that is, anarea of low elevation (2000–3000 m) and aridhabitats delimiting the northern and central An-des called the North Peru Low (NPL). The NPLis regarded as a major biogeographic barrier, anda transitional area for many sister-species pairsof the high-elevation montane forest (Cracraft1985, Stotz et al. 1996; but see Parker et al.1985, Garcıa-Moreno and Fjeldsa 1999). Wefirst modeled the distribution range of the twosister species, then tested for evolutionary nichesimilarity (Warren et al. 2008) between thetwo species on both sides of the biogeographicbarrier. We also tested the hypothesis that theexistence and distribution of the most recentlydiscovered species (Chestnut-bellied Cotinga;first described by Robbins et al. 1994) couldhave been predicted with location data fromits sister species described 120 yr earlier (Bay-vented Cotinga). We further used all 20 re-ported locations for both species to model thepotential distribution of the genus Doliornis.If a more robust prediction of the overall lin-eage’s potential distribution could be achieved(Graham et al. 2004), it may be possible toidentify areas where one of the two species,or a still unknown congeneric species, mayremain to be discovered. Identifying such gaps

Vol. 81, No. 4 Distributions of Doliornis Cotingas 327

should ensure that most of the remnant patchesof humid montane forest that potentially stillhost a Doliornis cotinga can be identified andconsidered in conservation planning. Our studyillustrates how species distribution modelingcould be used for assessing the distributionand conservation status of rare species knownonly from scattered, opportunistic records col-lected by field ornithologists (Bourg et al. 2005,Peterson and Papes 2007, Rojas-Soto et al. 2008,Marini et al., 2010).

METHODS

Study species and occurrence localities.Both Doliornis species are rare to locally un-common, with population sizes in the range of2500–9999 individuals (BirdLife International2010). They are confined to dense thicketsin the paramo-cloud forest ecotone at eleva-tions of 2875–3650 m for Chestnut-belliedCotingas (Appendix 1; contra Ridgely andGreenfield 2001) and 2740–3800 m for Bay-vented Cotingas. The low number of knownlocalities for Doliornis cotingas is likely due totheir low density and their inconspicuous, low-volume calls, lethargic perching behavior, andremote habitats. The biology of both Doliornisspecies is poorly known (Henry 2008, BirdLifeInternational 2010).

Data for presence records (Appendix 1) wereextracted from original publications, comple-mented with direct contact with observers andcollection curators, to assess the precision ofthe published coordinates. For niche modelingto be as robust as possible when the numberof occurrence records is very low, geographicalcoordinates of sites need to be precise andspecies identification verified. Three Chestnut-bellied Cotinga records were therefore discarded:locations 3 and 9 (Appendix 1) because ofimprecise locations, and location 4 because oflack of appropriate habitat in the area and aposteriori identification ( J.F. Freile in Henry2008). Two Bay-vented Cotinga records wereobtained in the same pixel of the study area(locations 2 and 3; Appendix 1) so were notreplicated in the data we used for modeling. Wethus used 13 locations for Bay-vented Cotingasand seven for Chestnut-bellied Cotingas.

Environmental variables. To model thepotential distribution of the two species, weconsidered a geographic area defined by country

borders, including Peru, Ecuador, Colombia,and Venezuela. Within this area, we obtaineddata concerning 11 environmental variablesobtained at a resolution of 2.5 s (i.e., abouta 4.5 km × 4.5 km pixel). We restricted thestudy area to pixels with an average altitude over2000 m to consider only the altitudinal rangewhere both species may occur. We extractedthree topographical variables from the Hydro-1K dataset of the USGS (http://edc.usgs.gov/products/elevation/gtopo30/hydro/index.html),namely slope, aspect, and the compoundtopographic index. The slope describes themaximum change in elevation between each celland its eight neighbors. The slope is expressedin integer degrees between 0 and 90. Theaspect describes the direction of maximum rateof change in the elevations between each celland its eight neighbors and gives the slopedirection, measured in positive integer degreesfrom 0 to 360 (clockwise from north). Thecompound topographic index, commonlyreferred to as the wetness index, is a function ofthe upstream contributing area and the slope ofthe landscape. We further used eight climaticvariables from the BioClim database (http://cres.anu.edu.au/outputs/anuclim/doc/bioclim.html)that are integrative annual or seasonal climaticvariables: annual mean temperature (BIO1),temperature seasonality (BIO4), maximumtemperature of the warmest month (BIO5),minimum temperature of the coldest month(BIO6), annual precipitation (BIO12),precipitation seasonality (BIO15), and amountof precipitation during the wettest (BIO13) anddriest (BIO14) months. We did not considervegetation layers in the modeling runs becauseof a large temporal mismatch between the birdoccurrence data (mostly collected in the 1970s,but one in the 19th century) and availablevegetation layers (like the Leaf Area Index orthe NDVI, for the 2000s; see Myneni et al.2002, Parra et al. 2004). Massive deforestationoccurred between these two periods (Stotzet al. 1996, Renjifo et al. 2002, BirdLifeInternational 2010). As a result, this temporalmismatch could lead to, at best, a weak and, atworst, a false modeled effect of vegetation onbird occurrence. We wanted, first, to objectivelyidentify the potential distribution areas withrelatively stationary ecoclimatic variables and,second, to include a posteriori the relationshipto current, suitable habitat availability (cf.


section “Extents of occurrence and vegetationcoverage”), hence identifying areas where thespecies may have disappeared due to habitatloss.

Species distribution modeling. We usedeight different niche-based modeling techniqueswith the BIOMOD computational framework(detailed by Thuiller et al. 2009): (1) generalizedlinear model (GLM), a regression method withpolynomial terms for which a stepwise proce-dure is used to select the most significant vari-ables, (2) generalized additive model (GAM),another regression method with four degrees offreedom and a stepwise procedure to select themost parsimonious model, (3) classification treeanalysis (CTA), a classification method runninga 50-fold cross-validation to select the best trade-off between the number of leaves of the treeand the explained deviance, (4) artificial neuralnetworks (ANN), a machine-learning method,with the mean of three runs used to providepredictions and projections, as each simula-tion gives slightly different results, (5) mixturediscriminant analysis (MDA), a classificationmethod based on mixture models, (6) mul-tivariate adaptive regression splines (MARS),a regression method, (7) generalized boostingmodel (GBM), a machine-learning method thatcombines a boosting algorithm and a regressiontree algorithm to construct an “ensemble” oftrees, and (8) Random Forest (RF), a machine-learning method that is a combination of treepredictors such that each tree depends on the val-ues of a random vector sampled independentlyand with the same distribution for all trees inthe forest. All models require information aboutpresence and absence to be able to determinesuitable conditions for a given species, so a set of10,000 pseudo-absences was selected at randomfrom the overall area to calibrate the models.The models compute a probability distributionbased on environmental variables spread overthe entire study area and assign a probability ofsuitability to each cell in the study area.

Presence threshold and Jackknifetesting. To validate and interpret models,distinguishing suitable from unsuitable areas isnecessary and is achieved by setting a decisionthreshold above which the model output isinterpreted as predicting presence (Pearsonet al. 2004). Different approaches have beenemployed for setting thresholds (Liu et al.2005), and they must be chosen carefully

because receiver operating characteristic (ROC)tests can be weak at discerning models ableto predict species’ distributions in broad,unsampled areas from those that cannot(Peterson et al. 2007), and are difficult toimplement when presence data are scarce. Thethreshold we used was the lowest predictedvalue associated with any one of the observedpresence records, called the “lowest presencethreshold” (LPT). This approach can beinterpreted ecologically as identifying pixelspredicted to be at least as suitable as thosewhere a species has been recorded as present; itis thus conservative, identifying the minimumpredicted area possible while maintainingzero omission error in the data set of knownoccurrences (Pearson et al. 2007). Such athreshold may be too conservative for speciesknown from very few locations, that is, whereavailable records may not represent the wholerange of ecoclimatic conditions where thespecies might occur. In that case, the variabilityof the output among the different modelingtechniques can be used to further assess therobustness of the predictions of species presence.

Because of the importance of model assess-ment and small numbers of locations for bothspecies, the predictive performances of our mod-els were tested with a jackknife (or “leave-one-out”) procedure (Pearson et al. 2007). Eachobserved location was removed from the dataonce, and a model was built using the remainingn − 1 localities. Predictive performance wasthen assessed based on the ability of each modelto predict the single location excluded fromthe data set. We tested the statistical predictiveperformance of a model with the program ofPearson et al. (2007). For each modeling tech-nique and each species, this program returns aP value that is the probability of predicting ran-domly the known locations, according to the to-tal area predicted as suitable by the model usinga given threshold and the success in predictingthe single excluded location. This procedure wasconducted for all eight modeling techniques foreach data set (Bay-vented Cotinga, Chestnut-bellied Cotinga, and Doliornis sp.) using theLPT as threshold.

Ensemble forecast and contribution ofvariables. Ensemble forecast techniques at-tempt to account for variability among speciesdistribution models to determine the centraltendency (Araujo and New 2007). For each


taxon, the consensus distribution was obtainedby selecting the outputs of those modeling tech-niques validated by the jackknife tests (leave-one-out procedure), that is, those modelingtechniques with P-values < 0.05. We then pro-duced maps reporting the unweighted averagesuitability of the selected models (Marmion et al.2009). Contributions of the variables to themodels were obtained with the BIOMOD com-puter platform. Independently, we characterizedthe contributions of the variables to the defini-tion of the bioclimatic niche of each species andthe genus by running a principal componentanalysis (PCA) and plotting coordinates of the20 independent Doliornis locations along thefirst and second principal components.

Interpredictivity and genus distributionmodel. To assess ecological similarity betweenthe two sister species, we used one of the statis-tical tests of niche conservatism described byWarren et al. (2008). We tested the equivalencyof the niches of the two species by looking at thesimilarity between the modeled niches. We cal-culated two similarity indices, D and I (rangingfrom 0 when there is no overlap to 1 when nichemodels are identical), that allow the plausibilityof niche equivalency between the two species(i.e., no statistical difference between alternativeniche models) to be assessed. To know if thesesimilarity measures were different from whatcould be obtained randomly, we created 100null distributions, as proposed by Warren et al.(2008), by considering 100 randomly partitionsof the 20 locations in groups of 13 and sevenlocations for which the same similarity indiceswere estimated. We then compared the D and Ivalues obtained for the species-specific data andthe distributions of those obtained for the 100null distributions.

Finally, we used a data set containing all20 locations from both species to model thepotential distribution of the genus Doliornis as awhole, potentially helping to identify new areasof occurrence for both species or undescribedrelated taxa. This model was evaluated using thesame leave-one-out procedure, and we producedthe same ensemble forecast maps.

Extents of occurrence and vegetationcoverage. The extent of occurrence is de-fined as the area within the shortest continuousimaginary boundary that can be drawn to en-compass all known, inferred, or projected sitesof present occurrence of a species, excluding

cases of vagrancy (IUCN 2001). We estimatedthe potential extent of occurrence using theensemble forecast predictions for each speciesin the study area. We summed the suitabilityvalues of all pixels where a given species waspredicted as present by models retained in theensemble forecast prediction, considering onlysuitability values above the LPT. By multiplyingthese sums by the pixel area (ca. 4.6 km ×4.6 km, or 21.16 km2, in the latitudinal range10◦N to 10◦S), we obtained an estimate of thepotential extent of occurrence of each species inkm2. These estimates were compared to previ-ous estimates of extent of occurrence (BirdLifeInternational 2010).

To account for actual habitat suitability, weused land-cover data from the vegetation mapfor South America published by Eva et al.(2002). This cover is a layer of 73 vegetationtypes at a 1 km × 1 km scale. In a con-servative approach, we retained as suitable allvegetation types that could include some of thehabitat of Doliornis cotingas at the pixel level,including mosaic agriculture/degraded forests,closed shrublands, closed and open montanegrasslands, and all montane forests (>1000 m,the only available elevation threshold) exceptclosed and open deciduous forests. Several ofthese habitats are not suitable for Doliorniscotingas, but, given the square kilometer scaleused, their habitat (paramo-cloud forest eco-tone) might occur within squares dominated bythese vegetation types. To obtain estimates ofpotential distribution sizes corrected for habitatsuitability, we first redefined pixel size of thedistribution output with a bilinear interpolationto fit that of the vegetation map (so a 1 km ×1 km scale). We then recalculated the estimateof extent of occurrence for each species for pixelsthat (1) had a suitable habitat (from the vegeta-tion map), and (2) were considered suitable for aspecies given the threshold (LPT).

RESULTS

Jackknife model testing. The jackknifetests of the ability of the different models topredict excluded localities were implementedwith prediction thresholds set at the LPT. Theresults of the tests are presented in Table 1,with the success rate and associated P-valuesfor each modeling technique per species andfor the genus. To produce the ensemble forecast


Table 1. Jackknife tests of the ability of the different models to predict excluded localities, and performedto validate or not the modeling techniques for each data set. Models in bold are those used in the ensembleforecasts. See the “Methods” section for descriptions of the different models.

Group GLM GAM CTA ANN MDA MARS GBM RF

D. sclateri Success rate 0.85 0.77 0.69 0.77 1 0.85 0.62 0P value <0.001 <0.001 <0.001 <0.001 0.46 <0.001 <0.001 1

D. remseni Success rate 0.71 1 0.86 0.14 0.71 0.43 0.14 0P value 0.006 1 <0.001 0.42 1 0.88 1 1

Doliornis sp. Success rate 0.84 0.84 0.95 0.79 0.95 0.89 0.47 0P value <0.001 <0.001 <0.001 <0.001 0.48 <0.001 <0.001 1

predictions, we first retained the modeling tech-niques with significant P-values (P < 0.05),then restrained the subset to the four modelswith the best success rates: GLM, GAM, ANNand MARS for Bay-vented Cotingas, GLM andCTA for Chestnut-bellied Cotingas, and GLM,GAM, CTA and MARS for Doliornis sp.

Predicted distribution ranges. For bothspecies, variables that contributed most to themodels were maximum temperature of the

Fig. 1. Graphic output of a principal component analysis (the two axes are the first two principal components)based on 11 environmental variables (see Methods for details) for the 20 locations where D. remseni andD. sclateri have been observed (7 for D. remseni—DR, in red; 13 for D. sclateri—DS, in blue). Ellipses of95% confidence interval are shown for each species.

hottest month (BIO5) and minimum tem-perature of the coldest month (BIO6). ForBay-vented Cotingas, precipitation seasonality(BIO15) contributed to the same extent. Thegraphic output of the PCA conducted on thevalues of the 11 environmental variables for the20 locations (7 for Chestnut-bellied Cotinga—DR, and 13 for Bay-vented Cotinga—DS) isprovided in Figure 1, and reveals that locationsof the two species already segregate along the first


Fig. 2. Predicted distribution ranges of (A) Chestnut-bellied Cotingas (Doliornis remseni), (B) Bay-ventedCotingas (Doliornis sclateri), and (C) the genus Doliornis sp. (combining the data from the two species) inthe Andes as the average suitability across models contributing to the ensemble forecast predictions and forthose areas where the suitability values overpass the mean lowest presence threshold (LPT). The color gradient(light, medium, and dark) represents the increasing average niche suitability (ranging from 0 to 1). Alsodepicted are record locations used in modeling (dots and squares) and country and province borders.

axis (51.5% of explained variance) characterizedby a positive association with precipitation(BIO12-13-14) and a negative associationwith temperature (BIO1-4-5). Chestnut-belliedCotingas is positively associated with this axis,whereas Bay-vented Cotingas loads negatively.The distribution maps obtained by the ensembleforecast predictions are presented in Figure 2A(Chestnut-bellied Cotinga) and Figure 2B (Bay-vented Cotinga). The ensemble forecast predic-tions were also obtained for the complete setof all Doliornis locations (N = 20) to modelthe potential distribution range of the genus(Fig. 2C).

Niche similarity between sister species.The PCA already revealed that locations ofboth species largely diverged for at least fourbioclimatic variables. With the conservativeLPT threshold, species-specific models couldnot predict any occurrence of the other species.When considering the ensemble forecast modelcalibrated with the data for Chestnut-bellied

Cotingas, the average suitability of Chestnut-bellied Cotinga locations was 0.806, whereas theaverage suitability of the 13 Bay-vented Cotingalocations was 0.102. Similarly, when consideringthe ensemble forecast model calibrated withthe data for Bay-vented Cotingas, the averagesuitability of Bay-vented Cotinga locations was0.873 and the average suitability of the sevenChestnut-bellied Cotinga locations was 0.051.Finally, the two indices of niche similarity hadlow values (D = 0.241 and I = 0.526), smallerthan any of the values obtained for the 100null distributions. Hence, the niches of the twospecies can be considered dissimilar.

Estimated extents of occurrence. Esti-mated extents of occurrence—the sum of suit-ability values for pixels with values above theLPT as predicted by the ensemble forecast—were 22 498 km2 for Bay-vented Cotingas and62 616 km2 for Chestnut-bellied Cotingas. Afteraccounting for habitat suitability within theseprobabilistic ranges (i.e., discarding pixels with


Fig. 3. Predicted distribution of D. remseni restrictedto suitable vegetation cover, overlaid on the BirdLifeInternational (2010) range map (in red). The bluecolor gradient (light, medium, and dark) indicatesthe increasing average niche suitability (ranging from0 to 1) within the ensemble forecast predictions.Also shown are country (black) and province (gray)borders, and species records, including those usedin modeling (gray dots), imprecise locations (blackovals), and a doubtful record (black circle, in Henry2008). The gray star outside the BirdLife Interna-tional’s range is a new location (Henry 2008).

predicted presence, but unsuitable habitat andfurther restricting the range to the north orsouth of the NPL according to the species),predicted extents of occurrence were downto 37 672 km2 for Chestnut-bellied Cotingas(Fig. 3) and 14 790 km2 for Bay-vented Cotingas(Fig. 4).

Fig. 4. Predicted distribution of D. sclateri restrictedto suitable vegetation cover, overlaid on the Nature-Serve (Young et al. 2009) range map (in red). The bluecolor gradient (light, medium, and dark) indicatesthe increasing average niche suitability (ranging from0 to 1) within the ensemble forecast predictions.Also shown are country (black) and province (gray)borders, and species records used in modeling (graysquares).

DISCUSSION

Modeling performance and predicteddistributions. When trying to predict thedistribution ranges of rare species, having asmall number of occurrence locations avail-able for modeling is a problem. However, forthe rare cotingas in our study, the predicteddistributions were statistically powerful, thatis, they predicted distributions better than ex-pected even with few locations. This couldbe due to the highly specific habitat used byDoliornis cotingas, that is, mainly treeline ofAndean humid montane forest (Henry 2008).This habitat is confined to a very narrow eco-climatic zone, making delimitation with cli-matic variables easy and robust (Parra et al.2004). Robustness of models is illustrated whenprojecting known localities that were excludedfrom the analysis onto the predicted distribu-tion range of Chestnut-bellied Cotingas. Thisspecies was known to occur at two addi-tional locations (3 and 9 in Appendix 1),


but precise locations for these records werelacking. Approximate locations (Fig. 3), owingto location names, indeed point to large areas ofcontinuous suitable areas (over 0.80 probabilityof suitability for the species in darker blue). Therecord from location 4 (see Appendix 1) was con-sidered doubtful because of a lack of appropriatehabitat in surrounding areas ( J.F. Freile in Henry2008). This is supported by the intermediate(0.5) probability of bioclimatic suitability inthis area, mostly below the LPT (Fig. 2A), andthe actual lack of suitable habitat (Fig. 3). ForBay-vented Cotingas, we compared our modeloutput with those of the NatureServe project(Young et al. 2009; Fig. 4). Again, the core areasof the distribution range of Bay-vented Cotin-gas are congruent among both distribution-modeling projects. The comparison also revealsmajor western overpredictions by our model,even with the LPT. Note that thresholds forthe NatureServe project were defined a poste-riori, according to expert judgment. Reviewerswere asked to choose the predicted distributionmaps that looked the most reasonable. Thus,by construction, overpredictions are stronglyminimized in such predicted distribution rangescompared to our models’ output.

Based on the current state of knowledge of Do-liornis distributions (Ridgely and Tudor 1994,Ridgely and Greenfield 2001, Henry 2008),these species avoid the western slopes of theAndes (except at locations 2 and 5; see Ap-pendix 1). However, our models predict theiroccurrence further west than actually known(e.g., western Colombia, west side of the cen-tral inter-Andean valley in Ecuador, and westof the Rıo Maranon valley in Peru). Severalfactors suggest that these western extrapola-tions are unwarranted (unless future inven-tories provide supportive evidence for them).First, according to field visits by J. Fjeldsa(pers. comm.), climatic conditions seem inap-propriate for Doliornis cotingas in these west-ern areas. At similar elevation, the habitat ismuch drier on the western slope than theeastern slope. In addition, no Doliornis wasrecorded in these western areas despite sev-eral intensive investigations (e.g., Imbabura,Pichincha, and Azuay provinces in Ecuador;Krabbe et al. 1998, Ridgely and Greenfield2001; several departments in Peru, J. Fjeldsa,pers. comm.). Also, if we consider the occurrenceof vegetation types (compare Figs. 2A—and B

and 3–4), the western part of the predicted rangeturns out to be largely unsuitable for Doliornis.And, finally, for Chestnut-bellied Cotingas, thewestern part of the predicted distribution ap-pears highly fragmented, with few continuousareas of high occurrence probability. This is par-ticularly evident when comparing fragmentationof the eastern versus western parts of the range.This suggests that the western overpredictionareas are unlikely to host Doliornis populations.Even though these areas may have hosted suit-able humid montane forest in the past, humansmay have largely modified these habitats, ren-dering them unsuitable for Doliornis (Stotz et al.1996, Renjifo et al. 2002, BirdLife International2010). Actually, the western overprediction areasfit well with zones of dense, traditional, ruralhuman populations (cf. Fig. 7b in Fjeldsa et al.1999).

Environmental niches of allopatric sisterspecies. We found a common ecologicalinfluence of environmental components whendefining the ecological niches of the two sis-ter species, corresponding to the hottest andthe coldest temperatures of the year. Overall,observed locations of one species could not bepredicted by the niche model developed for thesister species. Hence, the existence of the recentlydescribed Chestnut-bellied Cotinga (Robbinset al. 1994) could not have been predicted byextrapolation from a species distribution modelcalibrated with occurrence data for its previouslydescribed sister species, the Bay-vented Cotinga.This is also supported by the nonequivalence ofthe niches of the two species.

Peterson et al. (1999) showed conservatism inecological niches evolution for 37 pairs of sisterbird, mammal, and butterfly species isolatedon either side of a lowland barrier in south-ern Mexico. In contrast, niche divergence wasrevealed in some sympatric sister species (Sattleret al. 2007). Indeed, conservatism or divergenceof ecological niches depends on the speciationmechanisms involved. Strict vicariant speciationdepends simply on geographic isolation, andniche conservation is expected (e.g., Petersonet al. 1999). Other scenarios, such as the pe-ripheral isolates model of speciation and manymodels of sympatric speciation (Futuyma 1997),invoke invasion of novel ecological situationsas part of the speciation process. In the caseof Doliornis, sister taxa are found in similarhabitats (east slope, humid montane forest) on


both sides of a low-elevation orographic barrier,the NPL, proposed as a major biogeographicbarrier for high-elevation montane forest species(Cracraft 1985, Parker et al. 1985, Garcıa-Moreno and Fjeldsa 1999). Their phenotypesare similar, with limited differentiation (Rob-bins et al. 1994). Along with simple vicariantisolation and low phenotypic differentiation,niche conservatism was expected. Moleculardata suggest that the Doliornis clade (Ohlsonet al. 2007) dates back to the mid-Miocene(Garcıa-Moreno and Fjeldsa 1999), and vi-cariance of the Doliornis species would datefrom the mid-Pleistocene (Garcıa-Moreno andFjeldsa 1999). Thus, vicariance would be justposterior to the late Miocene uplift of thenorthern Andes and to the formation of theNPL (Garcıa-Moreno and Fjeldsa 1999). Toresolve the incongruence between vicariance andobserved niche divergence, we hypothesize thatniches differentiated after the isolation event,while species adapted to north–south variationsin bioclimatic conditions of their habitat, theparamo-cloud forest ecotone. To test the gen-erality of niche conservatism/divergence in theAndes, similar patterns of ecological divergencecould be tested among other sister allopatric taxaoccurring north and south of the NPL,for exam-ple, with the 21 species pairs identified by Parkeret al. (1985) and Garcıa-Moreno and Fjeldsa(1999).

Distributions and conservation. Ac-cording to IUCN (2001) Red List criteria,Chestnut-bellied Cotingas is classified as vul-nerable because of its small, decreasing popula-tion, and restricted, declining extent of occur-rence, area of occupancy, and area/extent/qualityof habitat (BirdLife International 2010). Bay-vented Cotingas also qualified as vulnerable forthe same reasons, but less evidence of criticalhabitat loss is available for this species (BirdLifeInternational 2010). The potential extent of oc-currences that we estimated (about 37 000 km2

for Chestnut-bellied Cotingas and 15 000 km2

for Bay-vented Cotingas), with the LPT thresh-old and after accounting for habitat availabil-ity, are different from the estimated Extentsof Occurrence provided by BirdLife Interna-tional (2010; 10 900 km2 for Chestnut-belliedCotingas, 13 100 km2 for Bay-vented Cotingas).For Chestnut-bellied Cotingas, the new recordlocations (Henry 2008) explain this differenceto some extent. However, our estimate includes

large, northern areas of predicted occurrence(in Colombia) that are supported by only onelocation and, therefore, may be unwarranted.Hence, the actual extent of occurrence maystill be close to the 20 000-km2 threshold usedwhen evaluating extinction risk (IUCN 2001).For Bay-vented Cotingas, our estimated valueis similar to the formerly available estimate.Thus, our estimates support a Red List statusas “vulnerable to extinction” for both species(BirdLife International 2010, IUCN 2010).

Our models suggest a relatively continuousdistribution of Chestnut-bellied Cotingas northof the NPL where habitat is suitable. The south-ern part of the range is highly fragmented. Fitbetween the predicted distribution range and therange provided by BirdLife experts is relativelygood in the highly fragmented, outer parts ofthe range (cf. northern and southern margins).The major discrepancy is the large, continuousarea of high probability of occurrence in centralEcuador. This omission by Birdlife Internationalis due to the lack of precise occurrence data inthe intervening area, data first documented afterBirdLife International had produced their map(Henry 2008). Our maps of potential distribu-tion will be important for guiding searches foreach species in areas where they are currently notknown to occur.

Species distribution modeling, using locationrecords and readily available environmental vari-ables, provides objective biogeographic infor-mation for poorly known tropical landscapes,and offers an innovative tool for the refinementof species distribution ranges and the searchof unknown species (Raxworthy et al. 2003,Engler et al. 2004, Bourg et al. 2005, Guisanet al. 2006, Peterson and Papes 2007). Modelscalibrated for the two Doliornis species, in ourstudy, revealed areas of likely occurrence thatremain to be validated in the field. The rangeof Bay-vented Cotingas may actually extendfurther to the southeast, reaching the PeruvianDepartment of Cusco, and to the northwest,with an area of high probability of occurrencein the Department of Cajamarca. Chestnut-bellied Cotingas is likely to be found betweenthe Colombian provinces of Cauca and Huila,as well as in several parts of Ecuador and north-ernmost Peru. Two areas of overprediction inthe northern Andes deserve specific comments.First, a Doliornis cotinga is predicted to occur,with an extended area of high probability, in the


eastern Andes of Columbia, south of Bogota.High Andean bird species are poorly differenti-ated among the three Colombian Andean Ridges(Cracraft 1985, Ridgely and Tudor 1994) and, ifa Doliornis sp. was to be found there, we suggestit would be Chestnut-bellied Cotingas and not anew species. Second, the models of Bay-ventedCotingas predict the occurrence of the speciesin the Andes of Venezuela (La Culata or NevadaSierras), whereas models of the closest species(Chestnut-bellied Cotinga) do not predict thispopulation. It suggests that, if a Doliornis sp.was to be discovered in this area, it may bea third species, with bioclimatic requirementscloser to Bay-vented Cotingas than Chestnut-bellied Cotingas.

To conclude, by modeling the ecological nicheof two allopatric, rare Andean endemics, weidentified the most suitable areas where thesespecies should occur based on climatic andtopographic modeling, some of which have beenconfirmed as occupied by one species. Suchspecies distribution modeling can play a majorrole in identifying the probable distributions ofrare species and helping to focus searches forthem and inform conservation planning. Ourstudy illustrates how the observations of fieldornithologists can be used objectively, with thehelp of existing statistical software, to trans-form anecdotal records into knowledge of thedistribution, ecology, and conservation of rarespecies.

ACKNOWLEDGMENTS

We are grateful to collection curators N. Rice, S.Cardiff, P. Sweet, and D. Willard for specimen labelsinformation, J. F. Freile, T. Santander, G. Engblom, R.Ahlman, F. Schmitt, and T. Davis for precise informa-tion about localities and geographical coordinates, M.Balman for providing shapefiles for BirdLife Interna-tional’s distribution maps, and B. Young for the Nature-Serve’s maps. A.T. Peterson kindly provided the USGSHydro-1K dataset. S. Butchart, J. Fjeldsa, and J. F. Freileprovided invaluable comments on an earlier draft of thispaper.

LITERATURE CITED

ARAUJO, M. B., AND M. NEW. 2007. Ensemble forecast-ing of species distributions. Trends in Ecology andEvolution 22: 42–47.

BIRDLIFE INTERNATIONAL [online]. 2010. BirdLife’s on-line World Bird Database: the site for bird conser-vation. Available at <http://www.birdlife.org> (28June 2010).

BOURG, N. A., W. J. MCSHEA, AND D. E. GILL. 2005.Putting a cart before the search: successful habitatprediction for a rare forest herb. Ecology 86: 2793–2804.

CRACRAFT, J. 1985. Historical biogeography and patternsof differentiation within the South American avi-fauna: areas of endemism. In: Neotropical ornithol-ogy (P. A. Buckley, M. S. Foster, E. S. Morton,R. S. Ridgely, and F. G. Buckley, eds.), pp. 49–84. Ornithological Monograph No. 36, AmericanOrnithologists’ Union, Washington, D.C.

CRESSWELL, W., R. MELLANBY, S. BRIGHT, P. CATRY,J. CHAVES, J. F. FREILE, A. GABELA, M. HUGHES,H. MARTINEAU, R. MACLEOD, F. MCPHEE, N. AN-DERSON, S. HOLT, S. BARABAS, C. CHAPEL, ANDT. SANCHEZ. 1999. Birds of the Guandera Biolog-ical Reserve, Carchi province, north-east Ecuador.Cotinga 11: 55–63.

ENGLER, R., A. GUISAN, AND L. RECHSTEINER. 2004. Animproved approach for predicting the distributionof rare and endangered species from occurrence andpseudo-absence data. Journal of Applied Ecology 41:263–274.

EVA, H. D., E. E. DE MIRANDA, C. M. DI BELLA,V. GOND, O. HUBER, M. SGRENZAROLI, S. JONES,A. COUTINHO, A. DORADO, M. GUIMARAES,C. ELVIDGE, F. ACHARD, A. S. BELWARD, E.BARTHOLOME, A. BARALDI, G. DE GRANDI, P. VOGT,S. FRITZ, AND A. HARTLEY. 2002. A vegetation mapof South America. Office for Official Publicationsof the European Communities, European Com-mission, Joint Research Centre, Luxembourg City,Luxembourg.

FJELDSA, J., E. LAMBIN, AND B. MERTENS. 1999. Cor-relation between endemism and local ecoclimaticstability documented by comparing Andean birddistributions and remotely sensed land surface data.Ecography 22: 63–78.

FRANKLIN, J., K. E. WEJNERT, S. A. HATHAWAY, C. J.ROCHESTER, AND R. N. FISHER. 2009. Effect ofspecies rarity on the accuracy of species distributionmodels for reptiles and amphibians in southernCalifornia. Diversity and Distribution 15: 167–177.

FREILE, J. F., AND T. SANTANDER. 2005. Areas importantespara la conservacion de las aves en Ecuador. In: Areasimportantes para la conservacion de las aves en losAndes tropicales: sitios prioritarios para la conser-vacion de la biodiversidad (BirdLife Internationaland Conservation International, eds.), pp. 283–370.Serie de Conservacion de BirdLife No. 14, BirdLifeInternational, Quito, Ecuador.

FUTUYMA, D. J. 1997. Evolutionary biology. SinauerAssociates, Sunderland, MA.

GARCIA, A. 2005. Using ecological niche modelling toidentify diversity hotspots for the herpetofauna ofPacific lowlands and adjacent interior valleys ofMexico. Biological Conservation 130: 25–46.

GARCIA-MORENO, J., AND J. FJELDSA. 1999. Chronologyand mode of speciation in the Andean avifauna.In: Isolated vertebrate communities in the Tropics(G. Rheinwald, ed.), pp. 25–46. Proceedings ofthe 4th International Symposium of ZoologischesForschungsinstitut und Museum A. Koenig, Bonn,Germany.


GRAHAM, C.H., S. R. RON, J. C. SANTOS, C. J.SCHNEIDER, AND C. MORITZ. 2004. Integratingphylogenetics and environmental niche models toexplore speciation mechanisms in Dendrobatid frogs.Evolution 58: 1781–1793.

GUISAN, A., O. BROENNIMANN, R. ENGLER, M. VUST, N.G. YOCCOZ, A. LEHMANN, AND N. E. ZIMMERMANN.2006. Using niche-based models to improve thesampling of rare species. Conservation Biology 20:501–11.

HENRY, P.-Y. 2008. Aves, Cotingidae, Doliornis rem-seni: filling distribution gap, habitat, conservation,Ecuador. Check List 4: 1–4.

———, AND J.-C. AZNAR. 2009. Eco-climatic correlatesof occurrences of the White-throated Hawk (Buteoalbigula) and the Cinereous Harrier (Circus cinereus)in central Ecuador. Ornitologıa Neotropical 20:511–521.

IUCN. 2001. 2001 IUCN Red List Categories and Crite-ria, version 3.1. IUCN, Cambridge, UK.

——— [online]. 2010. IUCN Red List of ThreatenedSpecies. Version 2010.1. Available at <http://www.iucnredlist.org’>, (28 June 2010)

KRABBE, N., F. SKOV, J. FJELDSA, AND I. K. PETERSEN.1998. Avian diversity in the Ecuadorian Andes.DIVA Technical report number 4, Centre for Re-search on Cultural and Biological Diversity of An-dean Rainforests (DIVA), Quito, Ecuador.

LIU, C., P. M. BERRY, T. P. DAWSON, AND R. G. PEARSON.2005. Selecting thresholds of occurrence in the pre-diction of species distributions. Ecography 28: 385–393.

MARINI, M.A., M. BARBET-MASSIN, M. E. LOPES, ANDF. JIGUET. 2009a. Major current and future gapof Brazilian reserves to protect Neotropical sa-vanna birds. Biological Conservation 142: 3039–3050.

———, M. BARBET-MASSIN, M. E. LOPES, AND F. JIGUET.2009b. Predicted climate-driven bird distributionchanges and forecasted conservation conflicts ina Neotropical savanna. Conservation Biology 23:1558–1567.

———, M. BARBET-MASSIN, M. E. LOPES, AND F. JIGUET.2010. Predicting the distribution of rare Brazilianbirds with ecological niche modelling. Journal ofOrnithology 151: 857–866.

MARMION, M., M. PARVIAINEN, M. LUOTO, R. K.HEIKKINEN, AND W. THUILLER. 2009. Evaluationof consensus methods in predictive species distribu-tion modelling. Diversity and Distributions 15: 59–69.

MYERS, N., R. A. MITTERMEIER, C. G. MITTERMEIER, G.A. B. DA FONSECA, AND J. KENT. 2000. Biodiversityhotspots for conservation priorities. Nature 403:853–858.

MYNENI, R.B., S. HOFFMAN, Y. KNYAZIKHIN, J. L.PRIVETTE, J. GLASSY, Y. TIAN, Y. WANG, X. SONG,Y. ZHANGA G. R. SMITH, A. LOTSCH, M. FRIEDL,J. T. MORISETTE, P. VOTAVA, R. R. NEMANI, AND S.W. RUNNING. 2002. Global products of vegetationleaf area and fraction absorbed PAR from year one ofMODIS data. Remote Sensing of Environment 83:214–231.

OHLSON, J. I., R. O. PRUM, AND P. G. P. ERICSON.

2007. A molecular phylogeny of the cotingas (Aves:Cotingidae). Molecular Phylogenetics and Evolution42: 25–37.

PARKER III, T. A., T. S. SCHULENBERG, G. R. GRAVES,AND M. J. BRAUN. 1985. The avifauna of the Huan-cabamba region, northern Peru. In: Neotropical or-nithology (P. A. Buckley, M. S. Foster, E. S. Morton,R. S. Ridgely, and F. G. Buckley, eds.), pp. 169–197. Ornithological Monograph No. 36, AmericanOrnithologists’ Union, Washington, D.C.

PARRA, J. L., C. C. GRAHAM, AND J.F. FREILE. 2004.Evaluating alternative data sets for ecological nichemodels of birds in the Andes. Ecography 27: 350–360.

PEARSON, R. G., T. P. DAWSON, AND C. LIU. 2004.Modelling species distributions in Britain: a hier-archical integration of climate and land-cover data.Ecography 27: 285–298.

———, C. J. RAXWORTHY, M. NAKAMURA, AND A.T. PETERSON. 2007. Predicting species distributionsfrom small numbers of occurrence records: a testcase using cryptic geckos in Madagascar. Journal ofBiogeography 24: 102–117.

PETERSON, A. T., J. SOBERON, AND V. SANCHEZ-CORDERO. 1999. Conservatism of ecological nichesin evolutionary time. Science 285: 1265–1267.

———, AND C. R. ROBINS. 2003. Using ecological-nichemodeling to predict Barred Owl invasions with impli-cations for Spotted Owl conservation. ConservationBiology 17: 1161–1165.

———, AND M. PAPES. 2007. Potential geographic distri-bution of the Bugun Liocichla Liocichla bugunorum,a poorly-known species from north-eastern India.Indian Birds 2: 146–149.

———, M. PAPES, AND M. EATON. 2007. Transferabilityand model evaluation in ecological niche modeling:a comparison of GARP and Maxent. Ecography 30:550–560.

RASMUSSEN, J. F., C. RAHBEK, B. O. POULSEN, M.K. POULSEN, AND H. BLOCH. 1996. Distributionalrecords and natural history notes on threatenedand little known birds of southern Ecuador. Bul-letin of the British Ornithologists’ Club 116: 26–46.

RAXWORTHY, C. J., E. MARTINEZ-MEYER, N. HOMING,R. A. NUSSBAUM, G. E. SCHNEIDER, M. A. ORTEGA-HUERTA, AND A. T. PETERSON. 2003. Predictingdistributions of known and unknown reptile speciesin Madagascar. Nature 426: 837–841.

RENJIFO, M. L.M. 1994. First records of the Bay-ventedCotinga Doliornis sclateri in Colombia. Bulletin ofthe British Ornithologists’ Club 114: 101–103.

RENJIFO, L. M., A. M. FRANCO-MAYA, J. D. AMAYA-ESPINEL, G. KATTAN, AND B. LOPEZ-LANUS. 2002.Libro rojo de aves de Colombia. Serie Libros Rojosde Especies Amenazadas de Colombia. Institutode Investigacion de Recursos Biologicos Alexandervon Humboldt y Ministerio del Medio Ambiente,Bogota, Columbia.

RICKLEFS, R. E., AND R.E. LATHAM. 1992. Interconti-nental correlation of geographical ranges suggestsstasis in ecological traits of relict genera of temperateperennial herbs. American Naturalist 139: 1305–1321.


RIDGELY, R. S., AND P. J. GREENFIELD. 2001. The birds ofEcuador. Cornell University Press, Ithaca, NY.

RIDGELY, R. S., AND G. TUDOR. 1994. The birds of SouthAmerica, vol. 1. The oscine passerines. University ofTexas Press, Austin, TX.

ROBBINS, M. B., G. H. ROSENBERG, AND F. SORNOZAMOLINA. 1994. A new species of cotinga (Cotingi-dae: Doliornis) from the Ecuadorian Andes, withcomments on plumage sequences in Doliornis andAmpelion. Auk 111: 1–7.

ROJAS-SOTO, O. R., E. MARTINEZ-MEYER, A. G.NAVARRO-SIGUENZA, A. O. DE ITA, H. G. DE SILVA,AND A. T. PETERSON. 2008. Modeling distributionsof disjunct populations of the Sierra Madre Sparrow.Journal of Field Ornithology 79: 245–253.

SATTLER, T., F. BONTADINA, A. H. HIRZEL, AND R.ARLETTAZ. 2007. Ecological niche modeling of twocryptic bat species calls for a reassessment of theirconservation status. Journal of Applied Ecology 44:1188–1199.

STEPHENS, L., AND M. A. J. TRAYLOR. 1983. Ornithologi-cal gazetteer of Peru. Harvard University, Cambridge,MA.

STOTZ, D. F., J. W. FITZPATRICK, T. A. PARKER III, AND

D. K. MOSKOVITS. 1996. Neotropical birds: ecol-ogy and conservation. University of Chicago Press,Chicago, IL.

TACZANOWSKI, L. 1874. Description des oiseaux nou-veaux du Perou central. Proceedings of the ZoologicalSociety of London IX: 129–140.

THORN, J. S., V. NIJMAN, D. SMITH, AND K. A. I. NEKARIS.2009. Ecological niche modelling as a technique forassessing threats and setting conservation prioritiesfor Asian slow lorises (Primates: Nycticebus). Diversityand Distribution 15: 289–298.

THUILLER, W., B. LAFOURCADE, R. ENGLER, AND M. B.ARAUJO. 2009. BIOMOD—a platform for ensembleforecasting of species distributions. Ecography 32:369–373.

WARREN, D. L., R. E. GLOR, AND M. TURELLI. 2008. En-vironmental niche equivalency versus conservatism:quantitative approaches to niche evolution. Evolu-tion 62: 2868–2883.

YOUNG, B. E., I. FRANKE, P. A. HERNANDEZ, S. K.HERZOG, L. PANIAGUA, A. SOTO, C. TOVAR, AND T.VALQUI. 2009. Using spatial models to predict areasof endemism and gaps in the protection of Andeanslope birds. Auk 126: 554–565.


APP

EN

DIX

1.Lo

cati

ons

whe

reB

ay-v

ente

dC

otin

gas

(D.s

clat

eri;

DS)

and

Che

stnu

t-be

llied

Cot

inga

s(D

.rem

seni

;DR

)ha

vebe

enre

port

ed.

Spec

ies

No.

Loca

tion

Coo

rdin

ates

aE

leva

tion

(mas

l)Fi

rstr

ecor

d,re

fere

nces

,and

com

men

tsb

DS

1N

ear

sour

ceof

Ayn

amay

o[=

Rıo

Vıt

oc],

Mar

ayni

oc,D

epto

.Jun

ın,P

eru

c.11

◦ 22′ S

75◦ 2

4′ W∗

3310

AN

SP,A

MN

H9

June

1871

,typ

elo

calit

y(T

acza

now

ski

1874

);A

pril

1921

,AN

SP,A

MN

HD

S2

Pan

deA

zuca

r,c.

5–10

kmno

rth-

east

ofA

com

ayo,

Dep

to.H

uanu

co,P

eru

c.09

◦ 42′ S

76◦ 0

0′ W∗

–26

Febr

uary

1972

;FM

NH

DS

3B

osqu

eC

utie

ragr

a(o

rC

utir

ragr

a),s

outh

ofH

uayl

aspa

mpa

,Car

pish

Mou

ntai

ns,

Aco

may

ore

gion

,Dep

to.H

uanu

co,P

eru

c.09

◦ 42′ S

76◦ 0

2′ W∗

2740

LSU

MZ

,27

75LS

UM

Z,

3050

AN

SP

10A

ugus

t197

2;FM

NH

,LSU

MZ

,AN

SP

DS

4B

osqu

eSh

aiga

,Dep

to.H

uanu

co,P

eru

––

20Ju

ly19

73;L

SUM

ZD

S5

Bos

que

Qui

ulla

coch

a,ne

arcr

esto

fCar

pish

Mou

ntai

n,c.

10km

nort

hof

Aco

may

o,D

epto

.Hua

nuco

,Per

u

c.09

◦ 44′ S

76◦ 0

6′ W∗

3000

;380

023

June

1974

,LSU

MZ

DS

6B

osqu

eU

ncho

g,pa

ssbe

twee

nC

huru

bam

baan

dH

acie

nda

Paty

,c.1

4km

nort

h-no

rth-

wes

tofA

com

ayo,

Dep

to.

Hua

nuco

,Per

u

09◦ 4

3′ 45′′ S

76◦ 1

0′ 04′′ W

Ahl

man

;09

◦ 45′ 2

1′′ S76

◦ 09′ 5

1′′ WSc

hmit

t

3350

LUSM

Z,

3450

LUSM

Z,

3450

Ahl

man

,34

40Sc

hmit

t

18Ju

ly19

75,L

SUM

Z;R

.Ahl

man

,per

s.co

mm

.,F.

Schm

itt,

pers

.com

m.,

GPS

geol

ocat

ion

DS

7M

ashu

a,ea

stTa

yaba

mba

,on

trai

lfro

mTa

yaba

mba

toO

ngon

,eas

tDep

to.L

aLi

bert

ad,P

eru

c.08

◦ 12′ S

77◦ 1

4′ W∗

3350

23Se

ptem

ber

1979

,LSU

MZ

DS

8Pu

erta

delM

onte

,c.3

0km

nort

h-ea

stof

Los

Alis

os(L

SUM

Z),

Dep

to.S

anM

artı

n,Pe

ru07

◦ 32′ S

77◦ 2

9′ WLS

UM

Z32

505

Aug

ust1

981,

LSU

MZ

,coo

rdin

ates

prov

ided

byLS

UM

ZD

S9

Lagu

naQ

uiul

laco

cha

(or

Qui

lluac

ocha

),no

rth-

nort

h-w

esto

fAco

may

o,C

arpi

shM

ount

ains

,Dep

to.H

uanu

co,P

eru

c.09

◦ 42′ S

76◦ 0

7′ W∗

3600

28Ju

ly19

84,L

SUM

Z

DS

10M

illpo

,2–6

kmdo

wn

ofA

bra

Port

aC

huel

aPa

ss,e

astT

ambo

deV

acas

,Poz

uzo-

Cha

glla

trai

l,D

epto

.Pas

co,P

eru

09◦ 5

7′ S75

◦ 51′ W

3100

Dav

is34

50LS

UM

Z8

July

1985

,LSU

MZ

;T.D

avis

(per

s.co

mm

.)pr

ovid

edde

tails

that

allo

wed

prec

ise

geol

ocat

ion

wit

hG

oogl

eEar

th(<

0◦ 01′ )

DS

111

and

2km

nort

hwes

tofP

unta

deSa

ria

onPo

zuzo

-Cha

glla

trai

l(LS

UM

Z);

Sari

apun

ta(S

teph

ens

and

Tray

lor

1983

),D

epto

.Pas

co,

Peru

09◦ 4

3′ S75

◦ 54′ W

∗31

00;3

200

4A

ugus

t198

5,LS

UM

Z

DS

12La

Mon

tani

ta,n

ear

Bul

dibo

yo,a

bove

Yacu

abam

ba,s

outh

ofA

bise

oN

P,C

arpi

shM

ount

ains

,Dep

to.L

aLi

bert

ad,P

eru

08◦ 0

6′ 00′′ S

77◦ 1

5′ 30′′ W

3300

;355

010

Aug

ust2

000,

S.A

llen,

http

://

may

bank

.trip

od.c

om/S

outh

Am

eric

a/Pe

ru/P

eru-

07-0

8-20

00.h

tm,G

PSge

oloc

atio

nD

S13

Chu

chun

punt

a,w

ithi

n10

kmof

Aco

may

o,D

epto

.Hua

nuco

,Per

uc.

09◦ 4

7′ S76

◦ 01′ W

∗30

501

Aug

ust1

973,

FMN

H

Con

tinu

ed

Vol. 81, No. 4 Distributions of Doliornis Cotingas 339A

PPE

ND

IX1.

Con

tinu

ed.

Spec

ies

No.

Loca

tion

Coo

rdin

ates

aE

leva

tion

(mas

l)Fi

rstr

ecor

d,re

fere

nces

,and

com

men

tsb

DS

14B

osqu

eU

ncho

g,D

epto

.Hua

nuco

,Per

u09

◦ 39′ 4

1′′ S76

◦ 13′ 4

2′′ W–

29Ju

ly20

06,e

Bir

dda

taba

se,fi

deG

loba

lB

iodi

vers

ity

Info

rmat

ion

Faci

lity

web

site

DR

1C

ajan

uma,

Podo

carp

usN

atio

nalP

ark,

Loja

/Zam

ora-

Chi

nchi

pepr

ov.,

Ecu

ador

04◦ 0

6′ S79

◦ 09′ W

2875

;310

07

Mar

ch19

89,R

obbi

nset

al.(

1994

),R

asm

usse

net

al.(

1996

),pu

blis

hed

coor

dina

tes

DR

2C

anon

delQ

uind

ıoN

atur

eR

eser

ve,Q

uind

ıode

pt.,

wes

tslo

peof

the

Cen

tral

Cor

dille

ra,

Col

ombi

a

04◦ 3

6′ 33′′ N

75◦ 2

7′ 09′′ W

3520

;362

031

Aug

ust1

989;

Ren

jifo

(199

4);R

enjif

oet

al.

(200

2)

DR

3C

ayam

be-C

oca

Eco

logi

calR

eser

ve,E

cuad

orc.

00◦ 1

7′ S78

◦ 12′ W

−0◦ 1

4′ N77

◦ 58′ W

∗∗

–19

90,n

ode

tail

onor

igin

alre

cord

;Fre

ilean

dSa

ntan

der

(200

5),c

oord

inat

epr

ecis

ion:

0◦ 07′

DR

4E

ntra

nce

ofC

otop

axiN

atio

nalP

ark,

Cot

opax

ipr

ov.,

Ecu

ador

c.00

◦ 39′ S

78◦ 3

1′ W∗∗

c.34

00–3

600

July

1991

,M.H

onic

kan

dJ.

C.M

athe

us,

http

://w

ww

.bir

dlife

foru

ms.

org/

,coo

rdin

ate

prec

isio

n:0◦ 0

3′D

R5

3km

sout

h-ea

stof

Impu

eran

,wes

tslo

peof

Cer

roM

ongu

s,so

uth-

east

ofC

arch

ipro

v.,

Ecu

ador

00◦ 2

7′ N77

◦ 52′ W

3550

;357

5;36

50M

arch

1992

;Rob

bins

etal

.(19

94),

publ

ishe

dco

ordi

nate

s

DR

6C

ajam

arca

,eas

tslo

peof

Cor

dille

raLa

sLa

guni

llas,

sout

hof

Zam

ora-

Chi

nchi

pepr

ov.,

Ecu

ador

04◦ 4

7′ S79

◦ 24′ W

3350

28O

ctob

er19

92;R

obbi

nset

al.(

1994

),pu

blis

hed

coor

dina

tes

DR

7G

uand

era

Bio

logi

calR

eser

ve,C

arch

ipro

v.,

Ecu

ador

c.00

◦ 36′ N

77◦ 4

2′ Wc.

3600

–370

019

97;C

ress

wel

leta

l.(1

999)

,coo

rdin

ates

inFr

eile

and

Sant

ande

r(2

005)

,pre

cisi

on:

0◦ 01′

DR

8B

elow

the

pass

ofG

uala

ceo-

Lim

on(M

acas

)ro

ad,w

ithi

n1–

2km

ofth

epa

ss,e

aste

rnsl

ope,

Mor

ona-

Sant

iago

prov

.,E

cuad

or

c.03

◦ 00′ S

78◦ 3

9′ W∗∗

c.35

002

Janu

ary

1999

,G.H

.Ros

enbe

rg(p

ers.

com

m.)

,htt

p://

vire

o.ac

nats

ci.o

rg/,

coor

dina

tes

iden

tifie

dw

ith

am

ap(p

reci

sion

:0◦ 0

2′ )D

R9

Sang

ayN

atio

nalP

ark,

Ecu

ador

c.01

◦ 30′ S

78◦ 2

0′ W−0

2◦ 30′ S

78◦ 5

0′ W∗∗

–20

02,n

ode

tail

onor

igin

alre

cord

;Fre

ilean

dSa

ntan

der

(200

5),c

oord

inat

epr

ecis

ion:

0◦ 30′

DR

10Ya

naco

cha

Lake

,eas

tslo

pe,L

lang

anat

esN

atio

nalP

ark,

Nap

opr

ov.,

Ecu

ador

01◦ 0

5′ 55′′ S

78◦ 1

8′ 27′′ W

3390

23Fe

brua

ry20

06,H

enry

(200

8),G

PSge

oloc

atio

na T

henu

mbe

rof

deci

mal

sof

the

coor

dina

tes

indi

cate

sth

eir

prec

isio

n.∗

and

∗∗in

dica

tege

ogra

phic

alco

ordi

nate

sex

trac

ted

from

Step

hens

and

Tray

lor

(198

3;co

ordi

nate

prec

isio

nno

tspe

cifie

d)or

dedu

ced

from

loca

lity

nam

es(s

eeco

mm

ents

for

deta

ilsab

outc

oord

inat

epr

ecis

ion)

,res

pect

ivel

y.bC

olle

ctio

nsho

stin

gsp

ecim

ens

are

the

Aca

dem

yof

Nat

ural

Scie

nces

ofPh

ilade

lphi

a(A

NSP

),A

mer

ican

Mus

eum

ofN

atur

alH

isto

ry(A

MN

H)

inN

ewYo

rk,

and

Fiel

dM

useu

mof

Nat

ural

His

tory

(FM

NH

)in

Chi

cago

.Web

site

reso

urce

sw

ere

acce

ssed

on28

June

2010

.–

indi

cate

sm

issi

ngda

ta.

Predicting potential distributions of two rare allopatric sister species, the globally threatened Doliornis cotingas in the Andes

Documents