Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Effects of species traits and environmental predictors on performance and transferability of ecological niche models


The ability of ecological niche models (ENMs) to produce robust predictions for different time frames (i.e. temporal transferability) may be hindered by a lack of ecologically relevant predictors. Model performance may also be affected by species traits, which may reflect different responses to processes controlling species distribution. In this study, we tested four primary hypotheses involving the role of species traits and environmental predictors in ENM performance and transferability. We compared the predictive accuracy of ENMs based upon (1) climate, (2) land-use/cover (LULC) and (3) ecosystem functional attributes (EFAs), and (4) the combination of these factors for 27 bird species within and beyond the time frame of model calibration. The combination of these factors significantly increased both model performance and transferability, highlighting the need to integrate climate, LULC and EFAs to improve biodiversity projections. However, the overall model transferability was low (being only acceptable for less than 25% of species), even under a hierarchical modelling approach, which calls for great caution in the use of ENMs to predict bird distributions under global change scenarios. Our findings also indicate that positive effects of species traits on predictive accuracy within model calibration are not necessarily translated into higher temporal transferability.


Ecological niche models (ENMs) based on correlative species-environment relationships are widely used to assess the impact of past and future global change on biodiversity1,2. Despite the long-standing and important role of ENMs in global change research3, these correlative approaches present important shortcomings that challenge their applicability in a changing, highly dynamic world (see e.g.4,5,6). One such limitation is the low transferability (i.e. model extrapolation) of parameterised models to other regions or times beyond the range of data used for model fitting (see7, for a review). However, the spatial and temporal transferability of ENMs is rarely evaluated prior to predicting species distribution in different regions or times.

Previous studies have provided evidence that model transferability is influenced by the modelling technique, as different algorithms show different sensitivities to spatial dimensionality and correlation8,9,10,11,12. Ensemble modelling and consensus methods have been proposed as an alternative to avoid overreliance on a single technique and reduce uncertainty across individual models13. Moreover, excessively complex models risk overfitting the data used for calibration, ultimately leading to predictions that can be too specific to the reference system to be transferable14. Most parsimonious models – built upon a small set of predictors – are expected to lead to greater transferabilities15. Thus, one critical step to ensure that models are ecologically meaningful, and therefore may support transferability14, is the choice of predictor variables16,17. However, many studies appear to use only environmental data that are readily available, while failing to consider other variables that may be ecologically relevant to the species distribution (i.e. causal rather than substitute predictors)18, and often missing important environmental drivers. New research on how the integration of key environmental drivers affects the transferability of ENMs is urgently needed given the increasing consensus on the importance of addressing the combined impacts of future climate and land-use/cover changes on biodiversity19,20,21. In this regard, the incorporation of less common but equally important predictors is now increasingly feasible, as remote sensing data are more readily available22. This broad range of available predictors can help improve the performance of ENMs and, in turn, their transferability14,23,24. This is the case for the intra- and inter-annual variability and seasonal dynamics of vegetation indices obtained from monthly images captured by satellite sensors. Several metrics summarising such dynamics are considered robust, integrative descriptors of ecosystem functioning25,26 and have recently been found to be meaningful predictors for species distributions and their temporal dynamics27,28,29. However, to the best of our knowledge, there is only one study that has assessed the role of these remotely-sensed ecosystem functional variables on model transferability30, although without considering the potential advantages of combining those variables with other relevant environmental drivers such as climate and land use change.

To advance on the application of model extrapolation in time and space, a better understanding of the processes and conditions that affect transferability is needed14. The ecology of species has also been demonstrated to impact model performance31,32,33 and predictor ranking34. Indeed, the effects of species traits (i.e., characteristics or qualities of the organisms of a species influencing performance or fitness)35 on the predictive ability of ENMs were found to override any differences in modelling technique, as these traits may reflect the different responses of species to processes that control species distributions36. For instance, a global meta-analysis has showed that species traits such as body size, dispersal model and trophic position may be good indicators of predictability37. Other studies found that non-endemic species with greater dispersal capacity, intermediate levels of prevalence, and little fire adaptation held higher model transferability than endemic species with limited dispersal capacity38. In addition, ENMs based upon climate, land cover and soil type variables performed less well for highly mobile species with large ranges39. A recent study also highlighted that the predictive ability of ENMs differs with regard to life history characteristics such as range, migration, habitat and rarity of a species40.

Despite the growing body of research and evidence, the relationship between specific traits, predictor variables and model transferability still remains poorly understood7,14. To address this caveat, here we tested four primary hypotheses involving the role of species traits and environmental predictors in model performance and transferability (see Table 1 for a detailed description):

Table 1 Hypotheses and sub-hypotheses, expectations arising from each hypothesis, and corresponding comparisons of model accuracy.

(H1) Model transferability hypothesis: Model accuracy will be higher within than beyond the model calibration time frame7.

(H2) Environmental-predictor hypothesis: Model accuracy, both within and beyond the model calibration time period, will increase with the combination of different types of relevant environmental predictors related to climate, habitat availability and ecosystem functioning conditions6,7.

(H3) Hierarchical-integration hypothesis: Model accuracy will improve, particularly beyond the model calibration time period, by hierarchically integrating environmental drivers whose effects are more evident and relevant at regional scales (i.e. climate) with those found to be more important at local scales (i.e. land use/cover and ecosystem functioning)41,42,43.

(H4) Species-traits hypothesis: Model performance and transferability will be influenced by species traits31,38,40,44 (see Table 1 for expectations and sub-hypotheses).

To test these hypotheses, we compared the predictive accuracy of ENMs based upon (1) climate, (2) land-use/cover and (3) ecosystem functional attributes, and (4) the combination of these factors, for 27 bird species within the model calibration period (year 2000) and in a different time frame (year 2010), in a highly dynamic landscape of NW Iberia (Gerês-Xurés Biosphere Reserve). The predictive accuracy was aggregated by species traits related to biogeographic origin, migratory status and habitat preference and specialization.


Study site

The study area is the Baixa Limia-Serra do Xurés Natural Park (c. 29 345 ha), located in the northwest of the Iberian Peninsula (Fig. 1A). It is included in the Gerês-Xurés transboundary UNESCO Biosphere Reserve (Fig. 1B) and protected under EU legislation (Habitat and Bird Directives). The zone is a mountain range, with elevation gradient between 323 and 1529 m a.s.l., located in the transition between the Mediterranean and Eurosiberian biogeographic regions and in the proximity of the Atlantic coast45, making it particularly susceptible to climate change. The landscape is mainly dominated by shrubland (69%) and forest (21%)46. The area has been greatly affected by the abandonment of traditional agricultural and livestock activities, which has favoured vegetation encroachment and recent forest expansion46,47. The area is also subjected to a high frequency of human-caused fires − linked to long-standing socio-economic difficulties in rural communities − and to a fast recovery of vegetation after fire48, resulting in an unstable and highly dynamic system46.

Figure 1

Location of the study area (Gerês-Xurés Mountains) in the Iberian Peninsula. (A) 10-km squares in the Iberian Peninsula. (B) Location of the study area in the Gerês-Xurés transboundary UNESCO Biosphere Reserve. (C) Spatial distribution of the point counts used for sampling bird communities: 5-min point counts for 2000 (i.e. calibration dataset) (red dots), 5-min point counts for 2010 (i.e. ‘Internal TT’ dataset) (blue dots), and 20-min point counts for 2010 (i.e. ‘External TT’ dataset) (green dots). Maps generated with the QGIS 2.16.2

Bird data

We used occurrence (presence/absence) data for bird species at two different scales (both spatial extent and resolution): (1) Iberian extent at 10-km resolution and (2) Gerês-Xurés Mountains at 230-m resolution (Fig. 1). At the Iberian scale, bird data were obtained from the combination of the Second Spanish and Portuguese Atlases of Breeding Birds49,50. This dataset documents the occurrence of breeding bird species in 6212 grid cells of 10-km resolution (Fig. 1A), with field data mostly collected between 1998 and 2005. At the Gerês-Xurés scale, bird data was surveyed in 2000 and 2010 using two different sampling methodologies. A set of 344 5-min point counts with unlimited distance were carried out in 2000 (hereafter: ‘calibration dataset’) (Fig. 1C). This initial dataset was partly replicated in 2010 by re-sampling a subset of 204 5-min point counts to evaluate the temporal transferability of ENMs (i.e. internal temporal transferability assessment; hereafter: ‘Internal TT’ dataset). Simultaneously, another set of 384 20-min point counts with a limited distance of 80 m was surveyed twice during the spring of 2010 to form a spatially and temporally independent data set (i.e. external temporal transferability assessment; hereafter: ‘External TT’ dataset) (Fig. 1C). The censuses were undertaken during the breeding season (May–June). To avoid possible detection biases caused by the time of survey, wind speed or cloud cover, all censuses were carried out during the 4 h after sunrise (peak vocal activity) and under uniform weather conditions (days without marked rainfall or wind). In the ‘External TT’ survey, the point counts were resampled in the afternoon during the 4 h before sunset (peak vocal activity).

From the initial set of species, we only selected those with more than 15 presences to ensure a minimum of 5 occurrences per predictor (i.e. to avoid overfitting the models)51 (see Table 2 for a list of target species). These species were grouped in different guilds based upon their ecological traits related to biogeographical origin (Eurosiberian/Mediterranean), phenology (sedentary/migrant), and habitat specialization (generalist/specialist) and preference (forest/open habitat), information extracted from different sources, from continental to local atlases49,50,52,53,54,55 (Table 2).

Table 2 Traits related to biogeographic origin, habitat preference and specialization, and phenology of each target bird species.

Environmental data

Climate variables

Climate variables were calculated from meteorological stations in Spain and Portugal by using multiple regression techniques (see details in56). In particular, 19 climate variables were computed for each year from monthly time series of air temperatures (mean maximum and mean minimum temperature) and total precipitation by using the function ‘biovars’ available in the R package ‘dismo’, version 1.1–457. From the initial set of 19 climate variables, we selected the two most ecologically relevant for the bird species according to expert knowledge and scientific literature58,59: (1) the maximum temperature of the warmest month, and (2) the average annual precipitation. A third predictor variable − (3) seasonality of precipitation (i.e. coefficient of variation) − was also selected after ensuring no problems related to collinearity with the other climate variables (see Supplementary Information S1 for correlations between pairs of predictors, description and visualization). Climate variables were resampled from their original resolution of 200 m to 230 m in order to match the spatial resolution of the EFAs.

Land-use/cover variables

The land-use/cover variables were derived from optical and thermal multispectral bands of Landsat TM and ETM + images acquired over the same temporal sequence as the bird sampling was carried out (20 March 2000, 8 June 2000, 24 June 2000, 19 May 2010 and 30 July 2010). The land-cover classification was obtained using a hybrid classification procedure, which combines unsupervised and supervised strategies47 (see details in Supplementary Information S2). Land-use/cover types were then aggregated in broad categories to describe the landscape composition, and three main categories were selected to describe habitat preference for the target bird species: (1) scrubland, (2) forest and (3) cropland. Land-use/cover variables represent the percentage of these three main categories in each geographic unit (i.e. grid cell), and they were resampled from their original resolution of 30 m to the spatial resolution of the EFAs (~230 m).

Ecosystem functioning variables

Ecosystem functional attributes (EFAs) were derived from the Enhanced Vegetation Index (EVI) obtained from 16-day maximum value composite images captured by the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor at a spatial resolution of 230 m. EVI is related to the fraction of photosynthetically active radiation intercepted by vegetation. EVI proved to have better sensitivity in high biomass regions and lower influence from the canopy background and atmospheric noise than other vegetation indices60. Three independent metrics of the EVI seasonal dynamics were calculated for 2001 (MODIS complete year closest to 2000) and 2010: (1) EVI annual mean, an estimator of annual primary production, which is one of the most integrative descriptors of ecosystem functioning (productivity indicator); (2) EVI seasonal standard deviation, a descriptor of the difference in carbon gains between seasons (seasonality indicator); and (3) the date of the maximum EVI value, an indicator of the growing season peak, as it indicates the most productive month during the year (phenology indicator) (see26 for details).

Modelling framework

We first developed three sets of trivariate models based exclusively upon each type of environmental predictor (henceforth ‘individual’ models): (1) Climate, (2) Land cover, and (3) EFAs (see Fig. 2 for a flow diagram of the modelling approach and steps). We also developed an additional set of models by using the environmental suitability predicted from the ‘individual’ models as predictor variables to integrate in a balanced way all possible combinations of these three environmental drivers within the modelling framework (henceforth ‘combined’ models) (see e.g.30,48,49): (4) Climate + Land cover, (5) Climate + EFAs; (6) Land cover + EFAs, and (7) Climate + Land cover + EFAs (Fig. 2). Both individual and combined ENMs were fitted with a maximum number of three variables to control model complexity, increase model transferability17,61, and make model performance and transferability comparable61 (Fig. 2). Climate models were calibrated at local (Gerês-Xurés) scale at 230 m resolution (henceforth ‘local climate models’) and then combined with the environmental suitability predicted from ENMs based upon land cover and/or EFAs (henceforth ‘non-hierarchical approach’). Climate models were also calibrated at the Iberian Peninsula scale (henceforth ‘regional climate models’) at 10-km resolution. These regional climate models were then directly projected (‘direct downscaling approach’, see62,63,64) to the extent and resolution of Gerês-Xurés Mountains for both 2000 and 2010. These local climate projections were used as predictor variables into combined models to hierarchically integrate climate suitability calibrated at regional scale with the environmental suitability predicted from ENMs based upon land cover and/or EFAs at local scale (henceforth ‘hierarchical approach’, sensu41,42).

Figure 2

Flow diagram of the modelling approach and steps (see Material and Methods for a detailed description of each step).

All ENMs were calibrated using seven modelling algorithms available in the R package ‘Biomod2’, version 3.3–765: generalized linear models (GLM), generalized additive models (GAM), generalized boosted regression models (GBM), random forest (RF), factorized distribution algorithm (FDA), multivariate adaptive regression splines (MARS) and artificial neural networks (ANN). For each technique, we used the default settings in biomod2 because these settings are optimized for SDMs (see66). Presence-background modelling techniques such as MaxEnt or GARP were not considered as they are designed to cope with the lack of absence data67, which is not the case, and did not show significant improvement in terms of predictive accuracy. We applied a 30-fold split-sample procedure by randomly selecting 70% of the data for calibration and the remaining 30% for model evaluation (hereafter ‘crossvalitation’). Our results for the area under the curve (AUC) of the receiver-operating characteristic (ROC), true skill statistic (TSS), the Cohen’s kappa coefficient were highly correlated (see Supplementary Table S3.1), and consequently, we only present results on AUC − a threshold-independent method − as a means of evaluating model performance68. In order to quantify model ability to accurately predict presences, we also computed sensitivity (i.e. the proportion of presences that were correctly predicted) as indicator of omission errors68. We applied the weighted average approach for computing a consensus (hereafter ‘ensemble model’) of the single models by calculating the weighted average of all single models with AUC > 0.65, using AUC values as model weights. This consensus method provides significantly more robust predictions than all the single-models and other consensus methods69. Models with AUC higher than 0.65 (instead of the more usual 0.7)70 were considered in the consensus method in order to obtain an ensemble model for all species, thereby ensuring a meaningful set of results to test our hypotheses and avoiding misleading conclusions that might result from considering only AUC values for those species with the highest model accuracies.

Model transferability

The temporal transferability of the ENMs (i.e. the ability of the model to predict species distribution in a different time frame) was tested by comparing the environmental suitability predicted for 2010 − derived from ensemble models calibrated for 2000 − against observed occurrence data available for 2010 from two independent sources: (1) a dataset obtained from the replication in 2010 of a subset of the initial point counts used to calibrate the models, which enables a temporally (but not spatially) independent assessment of model transferability (i.e. ‘Internal TT’); and (2) another dataset obtained from a distinct set of locations and using a different sampling methodology, which offers a more independent − spatially and temporally − evaluation of the model transferability (i.e. ‘External TT’).

As described above, the AUC of the ROC and Sensitivity were also used as accuracy metrics to evaluate model transferability68. We also calculated the number of species with AUC higher than 0.7 as a complementary indicator. To compare the spatial predictions for years 2000 and 2010, derived from the different ENMs, we calculated the Schoener’s D metric, the most common measure of niche overlap71,72. In particular, we computed niche overlap between spatial predictions derived from ‘individual’ and ‘combined’ models for each species using the function ‘nicheOverlap’ available in the R package ‘dismo’, version 1.1–457. Then, the Schoener’s D metric values were summarised by computing the average across species for each type of model.

Hypothesis testing

Each hypothesis and sub-hypothesis was tested by comparing the AUC values (see Table 1 for a detailed description) between the following:

(H1) crossvalidation and temporal transferability assessments (i.e. ‘Internal TT’ and ‘External TT’),

(H2) model types (i.e. ‘individual’ vs ‘combined’; e.g. ‘climate’ vs ‘climate + land cover’),

(H3) modelling approaches (i.e. ‘hierarchical’ vs ‘non-hierarchical’), and

(H4) species traits (e.g. ‘sedentary’ vs ‘migratory’).

We used box plots to display the variability of predictive accuracy obtained from crossvalidation, ‘Internal TT’ and ‘External TT’ assessments, across all predictor combinations (i.e. from ‘individual’ to ‘combined’ models), the two modelling approaches (i.e. ‘hierarchical’ and ‘non-hierarchical’) and all species traits (see Table 2). All plots were constructed with R software and the ggplot2 package73.

In addition, the effect of environmental predictor, modelling approach and species traits on AUC values were estimated using GLMs with a Gaussian error distribution and the ‘identity’ link function74 separately for crossvalidation, ‘TT Internal’ and ‘TT External’. In particular, we fitted a set of competing models and applied the multimodel inference approach75. We applied a data dredging analysis (the dredge function, available in the R package MuMIn; R Core Team 2015) to run the GLMs for all (valid) combinations of environmental predictors (interactions between predictor, modelling approach and species trait not included). For each model, we calculated the Akaike Information Criterion (AIC) and ∆I, where ∆I = AICi−AICminimum. All the models with ∆I < 7 were considered to have support75. The importance of each predictor was obtained by adding the Akaike weights (Wi) to the models in which that variable is present76. The addition of the weights of each variable was considered consequential when ΣWi > 0.5, meaning that half or more of the total Akaike weight for the model set was represented by models that contained that variable77. The AUC values for species traits were compared by a Wilcoxon signed rank test for paired samples. To be conservative, we only considered significant changes associated with p-values smaller than 0.001.


Model transferability hypothesis (H1)

All ENMs performed well during the crossvalidation procedure (AUCmean = 0.923 ± 0.081). However, the predictive accuracy of their temporal projections was much lower when tested against temporally (i.e. ‘Internal TT’; AUCmean = 0.615 ± 0.071) and spatiotemporally (i.e. ‘External TT’; AUCmean = 0.601 ± 0.067) independent datasets (Fig. 3). The low temporal transferability (AUC was higher than 0.7 for less than 25% of species) indicates that the ability of the models to predict distributional shifts is limited.

Figure 3

Model performance and transferability measured by AUC and sensitivity values for each set of predictors, modelling approach, and type of evaluation. For all box plots, lower and upper whiskers encompass the 95% interval, lower and upper hinges indicate the first and third quartiles, and the central black line indicates the median value.

Environmental-predictor hypothesis (H2)

ENMs developed with the three sets of predictors were all useful for describing the distribution of the target species (AUCClimate = 0.889 ± 0.104; AUCEFA = 0.867 ± 0.117; AUCLCT = 0.873 ± 0.071; Fig. 3). The high niche overlap between projections indicates that the different ENMs produce congruent predictions for our target species distributions (Fig. 4). However, spatial dissimilarities were also found, which confirms the complementary spatial information that ENMs based upon the different types of predictors can provide (Schoener’s D values ranging from 0.60 to 0.82, Fig. 4). Models based upon EFAs showed higher niche overlap with combined models (Schoener’s D = 0.69–0.79) than land cover (Schoener’s D = 0.67–0.73) or climate models (Schoener’s D = 0.63–0.70; Fig. 4). The gradual integration of information derived from ENMs based upon the different sets of predictors substantially increased model performance, both within (AUCmean up to 0.98; Fig. 3; ΣWPRED = 0.99; Table 3) but also beyond (Fig. 3; ΣWPRED = 0.98; see ‘Internal TT’ in Table 3; ΣWPRED = 0.99; see ‘TT External’ in Table 3) the calibration time frame. However, despite the enhanced model performance, the overall accuracy of temporal projections remained very low (Fig. 3).

Figure 4

The Schoener’s D values estimated from the spatial predictions derived from the different ENMs for both year 2000 and 2010.

Table 3 Model ranking according to ∆AIC (delta Akaike Information Criterion; only models ∆AIC < 7 are shown) for each evaluation type.

Hierarchical-integration hypothesis (H3)

The hierarchical integration of climate suitability predicted from regional climate models (calibrated at Iberian level) with the environmental suitability predicted from local ENMs (calibrated at Gerês-Xurés level, based upon land use/cover and ecosystem functioning variables) significantly improved our predictions within (AUCmean up to 0.97; Fig. 3; ΣWAPP = 0.90; Table 3), but not beyond the model calibration time period (Fig. 3; ΣWAPP = 0.26; see ‘Internal TT’ in Table 3; ΣWAPP = 0.28; see ‘External TT’ in Table 3).

Species-traits hypothesis (H4)

The species traits significantly affected our ability to accurately describe species distribution within the model calibration time period (Fig. 5; ΣWTRAIT = 0.99; Table 3) and also to predict distributional shifts over time (Fig. 5; ΣWTRAIT = 0.99; see ‘External TT’ in Table 3). However, this result was not consistent when temporal projections were tested against temporally (but not spatially) independent datasets (ΣWTRAIT = 0.04; see ‘Internal TT’ in Table 3).

Figure 5

Model performance and transferability measured by AUC values aggregated by species traits for each set of predictors, modelling approach, and type of evaluation. For all box plots, lower and upper whiskers encompass the 95% interval, lower and upper hinges indicate the first and third quartiles, and the central black line indicates the median value.

Model performance and transferability were higher for Mediterranean than for Eurosiberian species (‘H4.1’; crossvalidation: AUCmean = 0.96 vs 0.91, respectively; p < 0.001; ‘External TT’ AUCmean = 0.63 vs 0.59, respectively; p < 0.001), especially for ENMs developed with climate variables (Fig. 5), except for ‘Internal TT’ evaluations (AUCmean = 0.61 vs 0.61, respectively; p = 0.753).

Model performance within crossvalidation was higher for habitat specialist than generalist species (AUCmean = 0.93 vs 0.90, respectively; p < 0.001). However, these species traits did not show any significant effect on temporal transferability (‘H4.2’; ‘Internal TT’ AUCmean = 0.60 vs 0.59, respectively; p = 0.043; ‘External TT’ AUCmean = 0.62 vs 0.60, respectively; p = 0.223; Fig. 5).

No significant differences were found between forest and open-habitat dwelling species, in either the crossvalidation (AUCmean = 0.93 vs 0.90, respectively; p = 0.003) or in the temporal transferability (H4.2; ‘Internal TT’ AUCmean = 0.61 vs 0.62, respectively; p = 0.049; ‘TT External’ AUCmean = 0.60 vs 0.59, respectively; p = 0.378; Fig. 5). In addition, migrant species were more accurately predicted than sedentary species (AUCmean = 0.94 vs 0.91, respectively; p < 0.001), especially with ENMs based upon EFAs (‘H4.3’; Fig. 5), but only within the calibration time period (‘Internal TT’ AUCmean = 0.61 vs 0.60, respectively; p = 0.035; ‘External TT’ AUCmean = 0.62 vs 0.59, respectively; p = 0.001).

Overall, species traits affecting the performance and transferability of ENMs based upon climate and land cover variables also affected the predictive accuracy of ENMs based upon EFAs (‘H4.4’; see Fig. 5).


The present study compares - for the first time - the performance and temporal transferability of ecological niche models (ENMs) based upon climate, land-use/cover and ecosystem functional attributes (‘EFAs’), as well as the combinations of these factors. It is also, to our knowledge, the first study to evaluate the ability of an approach that hierarchically integrates climate with land-use/cover and ecosystem functional variables (cf.30,31) to improve bird distribution predictions beyond the calibration time frame. In addition, this study provides new insights into the relative role of species traits in model performance, being one the few attempts so far to examine the effects on model transferability (but see38,40,78,79).

Integration of predictors and model transferability

Our findings showed that ENMs developed with the three sets of predictors (namely climate, land-use/cover and EFAs) were all useful for describing the distribution of our target species (cf. Figs 3 and S3.1). The integration of these ecologically relevant predictors significantly increased the model performance within the calibration time period, under both the hierarchical and non-hierarchical integration approach (see ‘crossvalidation’ in Fig. 3, Table 3). As expected, the enhanced model performance within the calibration time was also translated into an increase in model transferability (cf. Fig. 3), providing additional support to the relevance of integrating climate and land-use/cover variables to improve our future biodiversity projections19,20,21 (‘H2’). In this sense, the integration of predictor variables related to ecosystem functioning − an often neglected dimension of ecological niche of species (but see27,80) − was found to increase model performance and transferability (cf. Fig. 3), especially for migrant, forest specialist species (cf. Fig. 5). Nonetheless, the overall predictive accuracy beyond the calibration time period was considerably low, being acceptable for less than 25% of modelled species (‘H1’). The lack of model transferability might be due to the existence of non-analogous conditions and, consequently, the need to extrapolate14,81,82. Then, low transferability is almost a consequence of high uncertainty due to extrapolation83. Indeed, a large spatial dissimilarity between years 2000 and 2010 was found for annual precipitation and precipitation seasonality. Theoretically, more temporally stable variables could lead to higher model transferability, but it depends upon the species to be modelled and the relative importance of those variables in the model (see Supplementary Fig. S4.1). Also, land abandonment process and wildfires has been affecting landscape over the last decade (see Supplementary Figs S4.2 and S4.3), with strong impacts on the bird community, as has been already documented for our study area46,47. In this sense, the explicit consideration of ecologically meaningful processes for species (i.e., fire disturbance) and more mechanistic approaches when modelling species distributions might have improved the temporal transferability of our ENMs, as has been illustrated for fire-prone systems84,85. In addition, the development of more appropriate extrapolation methods can help generate more realistic predictions in different spaces or times, allowing higher model transferability and credibility86.

Theoretically, the hierarchical approach should have improved our predictions beyond the calibration time period, as regional climate models (i.e. those calibrated at Iberian scale and projected into the Gerês-Xurés conditions) should be able to capture a wider range of species distributions and, in turn, a wider range of climate conditions than local climate models (i.e. those calibrated at Gerês-Xurés level) (see spatial projections for Eurasian skylark, Alauda arvensis, in Fig. 6). Indeed, the high sensitivity of the downscaled climate projections derived from regional models (see Fig. 3) suggests that they were able to capture the broad climate envelope of the species, being refined with the subsequent integration of ENMs based upon land cover and ecosystem functioning variables at Gerês-Xurés scale (cf. Fig. 3). These results contribute to the mounting evidence that hierarchical approaches effectively capture the intrinsic characteristics of ecological systems41, successfully integrating climate suitability over large scales with habitat characteristics at more local scales (see e.g.48,61,87,88). This multiscale hierarchical approach did not, however, show any advantages in terms of model transferability relative to the classic, non-hierarchical integrative approach (Table 3, Fig. 3), undermining its suitability for making predictions of bird distributions under global change scenarios, at least in our particular case (‘H3’). Here we applied the ‘direct approach’ to obtain fine-grained, downscaled climate suitability from models calibrated at broader scale62,63,64. This approach relies on the assumption that fine-grain species distributions show the same environmental associations as distributions at the coarse grain62,63, which might not be necessarily the case for all species. Alternative hierarchical approaches based on Bayesian modelling framework might help to overcome these limitations89, eventually leading to more accurate predictions. Moreover, niche truncation is one of the most challenging issues when ENMs are used to predict in other times or spaces14,90. For example, niche truncation can result from a limited range of environmental conditions existing in the study region used to calibrate models (i.e., realized environmental space sensu91). In our study, models calibrated at local (Gerês-Xurés) and regional (Iberian Peninsula) scales have likely resulted in different niche truncation. Then, the challenge for predictive ecology is how to deal with predictions into non-analogous conditions (i.e., beyond the point of niche truncation), which can be an important constraint, even having chosen the most appropriate predictors. Recent research has also highlighted that the poor performance of uncorrected (random) cross-validation might be caused by dependence structures in the data that might also persist in model residuals, violating the assumption of independence. The large difference found between model accuracy within and beyond the calibration timeframe draws attention to the need for using independent validation data or, alternatively, block cross-validation methods if the goal is predicting to new data or predictor space92. Last but not least, the lack of repeated temporal sample structure in our bird data (both at local and regional scale), required for accounting statistically for biases related to imperfect detection, is an additional source of variation93,94.

Figure 6

Spatial projections depicting the suitability values predicted for the Eurasian Skylark (Alauda arvensis) from each type of ENM under the hierarchical and non-hierarchical modelling approach for both year 2000 and 2010.

Species traits and model transferability

Our findings showed that species traits have a significant impact on both model performance and transferability (‘H4’; cf. Fig. 5), despite the large variability among species (see Supplementary Fig. S3). These results confirm the important role of species ecology on model performance, as previously suggested32,38,40,78,79. However, our results also suggest that the effects of species traits on predictive accuracy within the model calibration time period are not necessarily translated into its temporal transferability (cf. Table 3; Fig. 5). Thus, while biogeographic origin was found to affect model performance and transferability, habitat specialization and migratory status only impacted model performance (not transferability), and habitat preference did not impact either model performance or transferability (cf. Table 3; Fig. 5). Interestingly, the results also indicate a trait-predictor relationship as some specific traits seem to be linked to specific environmental predictors. For instance, model performance was higher for habitat specialists than for generalist species95,96 (see ‘crossvalidation’ in Fig. 5). Generalist species often have local subpopulations that differ in ecological characteristics, so modelling all these subpopulations together could overestimate the species ecological breadth and hence reduce model accuracy96. On the contrary, specialist species that inhabit in more restricted environmental conditions are usually more accurately predicted, as long as these conditions are not widely distributed95. Similarly, Mediterranean species showed higher predictive accuracy than Eurosiberian species, especially in ENMs that included climate variables (see ‘External TT’ in Fig. 5; ‘H4.1’), due also to a narrower niche and smaller range size in the study area97. Hence, our ability to properly model species distributions hinges upon the niche breadth–range size relationship97, even though this does not necessarily have to affect model transferability (‘H4.2’; Fig. 5). In fact, some studies found greater model transferability for wide-ranging organisms with broad environmental niches than for narrow-ranging specialists14,40. In addition, our results showed that migrant species were more accurately predicted than sedentary species (‘H4.3’; Fig. 5). This is consistent with recent studies that demonstrated that migrating birds track vegetation dynamics28; i.e., the birds move with the seasonally progressing green-up of vegetation98,99. Food availability is one of the most important resources determining breeding success. On the basis of the well-established principle that resources availability is related to primary productivity100, several studies successfully correlated species distributions with the inter-annual variability and seasonal dynamics of vegetation, measured through satellite-derived indices27,80,101.

It is particularly interesting to note that species traits affecting the performance and transferability of ENMs based upon climate and land cover variables also affected the predictive accuracy of ENMs based upon EFAs (H4.4, Fig. 5). This may be explained by the fact that EFAs can be directly affected by climate and land use/cover change25 and can thus potentially provide information about the integrative response of species to global environmental change27,102. Indeed, ENMs based exclusively upon EFAs showed higher niche overlap with the combined models – that integrate the three types of environmental predictors – than land cover or climate models (Fig. 4). These results suggest that satellite-derived EFAs, in addition to assisting the on-going assessment of essential biodiversity variables (EBVs) related to ecosystem functioning22 (see also Working Group Ecosystem function GEOBON,, can also contribute substantially to near real-time biodiversity monitoring. Our results therefore confirm the ability of EFAs to accurately model bird distribution – especially for migrant, forest specialist species – although they cannot ensure model transferability by themselves (Fig. 5). Their integration with other ecologically relevant factors, such as climate and land use changes39,79, into ENMs will be key to improve temporal predictions20.

In conclusion, the study findings confirm the important role of species traits and environmental predictors in model performance and transferability for bird distributions, especially in highly dynamic systems. The progressive integration of ecologically relevant predictors related to climate, land-use/cover and ecosystem functional conditions significantly enhanced model performance within the calibration time period, under both hierarchical and non-hierarchical integration approaches. These findings provide additional support to the importance of integrating climate, land-use/cover and ecosystem functional variables to improve our future biodiversity projections. Unfortunately, the hierarchical multiscale approach did not show substantial advantages in terms of model transferability in comparison with the classic, non-hierarchical integrative approach, undermining its suitability for predicting bird distribution under global change scenarios. We strongly emphasize the importance of considering model transferability, in addition to traditional measures of model accuracy, and the need for caution when using ENMs to predict shifts in bird distributions as high discriminatory power within the calibration time frame does not guarantee the predictive ability of a model. Our findings also showed that species traits may significantly impact both model performance and transferability. In particular, the accuracy of prediction was highest for Mediterranean, migrant, habitat-specialist species, which provides guidance for the suitability of ENMs as an approach for predicting global change responses for birds in our region. However, the findings also suggest that the positive effects of species traits on predictive accuracy within model calibration are not necessarily translated into higher temporal transferability.

Data Availability

The bird datasets and environmental variables used during this study are available in Supplementary Information Files.


  1. 1.

    Guisan, A. & Zimmermann, N. E. Predictive habitat distribution models in ecology. Ecol. Modell. 135, 147–186 (2000).

    Article  Google Scholar 

  2. 2.

    Austin, M. P. Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecol. Modell. 157, 101–118 (2002).

    Article  Google Scholar 

  3. 3.

    Brotons, L. Species Distribution Models and Impact Factor Growth in Environmental. Journals: Methodological Fashion or the Attraction of Global Change Science. PLoS One 9, e111996 (2014).

    PubMed  PubMed Central  Google Scholar 

  4. 4.

    Araújo, M. B. & Guisan, A. Five (or so) challenges for species distribution modelling. J. Biogeogr. 33, 1677–1688 (2006).

    Article  Google Scholar 

  5. 5.

    Guisan, A. & Thuiller, W. Predicting species distribution: offering more than simple habitat models. Ecol. Lett. 8, 993–1009 (2005).

    Article  Google Scholar 

  6. 6.

    Dormann, C. F. Promising the future? Global change projections of species distributions. Basic Appl. Ecol. 8, 387–397 (2007).

    Article  Google Scholar 

  7. 7.

    Werkowska, W., Márquez, A. L., Real, R. & Acevedo, P. A practical overview of transferability in species distribution modeling. Environ. Rev. 25, 127–133 (2017).

    Article  Google Scholar 

  8. 8.

    Townsend Peterson, A., Papeş, M. & Eaton, M. Transferability and model evaluation in ecological niche modeling: A comparison of GARP and Maxent. Ecography (Cop.). 30, 550–560 (2007).

    Article  Google Scholar 

  9. 9.

    Randin, C. F. et al. Are niche-based species distribution models transferable in space? J. Biogeogr. 33, 1689–1703 (2006).

    Article  Google Scholar 

  10. 10.

    Duque-Lazo, J., van Gils, H., Groen, T. A. & Navarro-Cerrillo, R. M. Transferability of species distribution models: The case of Phytophthora cinnamomi in Southwest Spain and Southwest Australia. Ecol. Modell. 320, 62–70 (2016).

    Article  Google Scholar 

  11. 11.

    Beaumont, L. J. et al. Which species distribution models are more (or less) likely to project broad-scale, climate-induced shifts in species ranges? Ecol. Modell. 342, 135–146 (2016).

    Article  Google Scholar 

  12. 12.

    Heikkinen, R. K., Marmion, M. & Luoto, M. Does the interpolation accuracy of species distribution models come at the expense of transferability? Ecography (Cop.). 35, 276–288 (2012).

    Article  Google Scholar 

  13. 13.

    Zhu, G. P. & Peterson, A. T. Do consensus models outperform individual models? Transferability evaluations of diverse modeling approaches for an invasive moth. Biol. Invasions 19, 2519–2532 (2017).

    Article  Google Scholar 

  14. 14.

    Yates, K. L. et al. Outstanding Challenges in the Transferability of Ecological Models. Trends Ecol. Evol. xx, 1–13 (2018).

    Google Scholar 

  15. 15.

    Wenger, S. J. & Olden, J. D. Assessing transferability of ecological models: An underappreciated aspect of statistical validation. Methods Ecol. Evol. 3, 260–267 (2012).

    Article  Google Scholar 

  16. 16.

    Synes, N. W. & Osborne, P. E. Choice of predictor variables as a source of uncertainty in continental-scale species distribution modelling under climate change. Glob. Ecol. Biogeogr. 20, 904–914 (2011).

    Article  Google Scholar 

  17. 17.

    Petitpierre, B., Broennimann, O., Kueffer, C., Daehler, C. & Guisan, A. Selecting predictors to maximize the transferability of species distribution models: lessons from cross-continental plant invasions. Glob. Ecol. Biogeogr. 26, 275–287 (2017).

    Article  Google Scholar 

  18. 18.

    Kühn, I. & Dormann, C. F. Less than eight (and a half) misconceptions of spatial analysis. J. Biogeogr. 39, 995–998 (2012).

    Article  Google Scholar 

  19. 19.

    Titeux, N. et al. Biodiversity scenarios neglect future land-use change. Glob. Chang. Biol. 22, 2505–2515 (2016).

    ADS  PubMed  Article  Google Scholar 

  20. 20.

    Titeux, N. et al. Global scenarios for biodiversity need to better integrate climate and land use change. Divers. Distrib. 23, 1231–1234 (2017).

    Article  Google Scholar 

  21. 21.

    Sirami, C. et al. Impacts of global change on species distributions: obstacles and solutions to integrate climate and land use. Glob. Ecol. Biogeogr. 26, 385–394 (2017).

    Article  Google Scholar 

  22. 22.

    Pettorelli, N. et al. Framing the concept of satellite remote sensing essential biodiversity variables: challenges and future directions. Remote Sens. Ecol. Conserv. 2, 122–131 (2016).

    Article  Google Scholar 

  23. 23.

    He, K. S. et al. Will remote sensing shape the next generation of species distribution models? Remote Sens. Ecol. Conserv. 1, 4–18 (2015).

    Article  Google Scholar 

  24. 24.

    Zimmermann, N. E., Edwards, T. C., Moisen, G. G., Frescino, T. S. & Blackard, J. A. Remote sensing-based predictors improve distribution models of rare, early successional and broadleaf tree species in Utah. J. Appl. Ecol. 44, 1057–1067 (2007).

    CAS  PubMed  PubMed Central  Article  Google Scholar 

  25. 25.

    Paruelo, J. M., Burke, I. C. & Lauenroth, W. K. Land-use impact on ecosystem functioning in eastern Colorado, USA J. M. Paruelo, I. C. Burke and W. K. Lauenroth. Glob. Chang. Biol. 7, 631–639 (2001).

    ADS  Article  Google Scholar 

  26. 26.

    Alcaraz, D., Paruelo, J. & Cabello, J. Identification of current ecosystem functional types in the Iberian Peninsula. Glob. Ecol. Biogeogr. 15, 200–212 (2006).

    Article  Google Scholar 

  27. 27.

    Alcaraz-Segura, D. et al. Potential of satellite-derived ecosystem functional attributes to anticipate species range shifts. Int. J. Appl. Earth Obs. Geoinf. 57, 86–92 (2017).

    ADS  Article  Google Scholar 

  28. 28.

    Thorup, K. et al. Resource tracking within and across continents in long-distance bird migrants. Sci. Adv. 3, e1601360 (2017).

    ADS  PubMed  PubMed Central  Article  Google Scholar 

  29. 29.

    Tuanmu, M. N. & Jetz, W. A global, remote sensing-based characterization of terrestrial habitat heterogeneity for biodiversity and ecosystem modelling. Glob. Ecol. Biogeogr. 24 (2015).

  30. 30.

    Tuanmu, M.-N. et al. Temporal transferability of wildlife habitat models: implications for habitat monitoring. J. Biogeogr. 38, 1510–1523 (2011).

    Article  Google Scholar 

  31. 31.

    Pöyry, J., Luoto, M., Heikkinen, R. K. & Saarinen, K. Species traits are associated with the quality of bioclimatic models. Glob. Ecol. Biogeogr. 17, 403–414 (2008).

    Article  Google Scholar 

  32. 32.

    McPherson, J. M. & Jetz, W. Effects of species’ ecology on the accuracy of distribution models. Ecography (Cop.). 30, 135–151 (2007).

    Google Scholar 

  33. 33.

    Hanspach, J., Kühn, I., Pompe, S. & Klotz, S. Predictive performance of plant species distribution models depends on species traits. Perspect. Plant Ecol. Evol. Syst. 12, 219–225 (2010).

    Article  Google Scholar 

  34. 34.

    Vicente, J., Alves, P., Randin, C., Guisan, A. & Honrado, J. What drives invasibility? A multi-model inference test and spatial modelling of alien plant species richness patterns in northern Portugal. Ecography (Cop.). 33, 1081–1092 (2010).

    Article  Google Scholar 

  35. 35.

    Nock, C. A., Vogt, R. J. & Beisner, B. E. Functional Traits. eLS 1–8, (2016).

  36. 36.

    Syphard, A. D. & Franklin, J. Differences in spatial predictions among species distribution modeling methods vary with species traits and environmental predictors. Ecography (Cop.). 32, 907–918 (2009).

    Article  Google Scholar 

  37. 37.

    Soininen, J. & Luoto, M. Predictability in species distributions: A global analysis across organisms and ecosystems. Glob. Ecol. Biogeogr. 23, 1264–1274 (2014).

    Article  Google Scholar 

  38. 38.

    Dobrowski, S. Z. et al. Modeling plant ranges over 75 years of climate change in California, USA: Temporal transferability and species traits. Ecol. Monogr. 81, 241–257 (2011).

    Article  Google Scholar 

  39. 39.

    Eskildsen, A. et al. Testing species distribution models across space and time: high latitude butterflies and recent warming. Glob. Ecol. Biogeogr. 22, 1293–1303 (2013).

    Article  Google Scholar 

  40. 40.

    Wogan, G. O. U. Life history traits and niche instability impact accuracy and temporal transferability for historically calibrated distribution models of North American birds. PLoS One 11, e0151024 (2016).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  41. 41.

    Pearson, R. G., Dawson, T. P. & Liu, C. Modelling species distributions in Britain: a hierarchical integration of climate and land-cover data. Ecography (Cop.). 3, 285–298 (2004).

    Article  Google Scholar 

  42. 42.

    Pearson, R. G. & Dawson, T. P. Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Glob. Ecol. Biogeogr. 12, 361–371 (2003).

    Article  Google Scholar 

  43. 43.

    Vicente, J. R. et al. A framework for assessing the scale of influence of environmental factors on ecological patterns. Ecol. Complex. 20, 151–156 (2014).

    Article  Google Scholar 

  44. 44.

    Syphard, A. D. & Franklin, J. Species traits affect the performance of species distribution models for plants in southern California. J. Veg. Sci. 21, 177–189 (2010).

    Article  Google Scholar 

  45. 45.

    Martínez-Cortizas, A. & Pérez-Alberti, A. Atlas Bioclimático de Galicia. (Xunta de Galicia, 1999).

  46. 46.

    Regos, A., Ninyerola, M., Moré, G. & Pons, X. Linking land cover dynamics with driving forces in mountain landscape of the Northwestern Iberian Peninsula. Int. J. Appl. Earth Obs. Geoinf. 38, 1–14 (2015).

    ADS  Article  Google Scholar 

  47. 47.

    Regos, A. et al. Rural abandoned landscapes and bird assemblages: winners and losers in the rewilding of a marginal mountain area (NW Spain). Reg. Environ. Chang. 16, 199–211 (2016).

    Article  Google Scholar 

  48. 48.

    Torres, J., Gonçalves, J., Marcos, B. & Honrado, J. Indicator-based assessment of post-fire recovery dynamics using satellite NDVI time-series. Ecol. Indic. 89, 199–212 (2018).

    Article  Google Scholar 

  49. 49.

    Equipa A. Atlas das Aves Nidificantes em Portugal ( 1999–2005). Instituto da Conservação da Natureza e da Biodiversidade Sociedade Portuguesa para o Estudo das Aves Parque Natural da Madeira e Secretaria Regional do Ambiente e do Mar. Assírio and Alvim, Lisboa (2008).

  50. 50.

    Martí, R., Del Moral, J. C. & (Eds). Atlas de las Aves Reproductoras de España. Dirección General de Conservación de la Naturaleza-Sociedad Española de Ornitología. Madrid (2003).

  51. 51.

    Araújo, M. B. & Peterson, A. T. Uses and misuses of bioclimatic envelope modeling. Ecology 93, 1527–1539 (2012).

    PubMed  Article  Google Scholar 

  52. 52.

    Voous, K. Atlas of European Birds. Nelson, London (1960).

  53. 53.

    Hagemeijer, E. J. M. & Blair, M. J. The EBCC Atlas of European Breeding Birds: Their Distribution and Abundance. (T & AD Poycer, 1997).

  54. 54.

    Domínguez, J. et al. Atlas de los vertebrados terrestres reproductores en el Parque Natural Baixa Limia-Serra do Xurés y ZEPVN-LIC Baixa Limia. (Xunta de Galicia, 2012).

  55. 55.

    Pimenta, M. & Santarém, M. Atlas das Aves do Parque Nacional da Peneda-Gerês. (Instituto da Conservação da Natureza. Parque Nacional da Peneda-Gerês., 1996).

  56. 56.

    Ninyerola, M., Pons, X. & Roure, J. M. A methodological approach of climatological modelling of air temperature and precipitation through GIS techniques. Int. J. Climatol. 20, 1823–1841 (2000).

    Article  Google Scholar 

  57. 57.

    Hijmans, R. J., Phillips, S., Leathwick, J. & Elith, J. dismo: species distribution modeling. – R package ver. 1.1–4 (2016).

  58. 58.

    Tingley, M. W., Monahan, W. B., Beissinger, S. R. & Moritz, C. Birds track their Grinnellian niche through a century of climate change. Proc. Natl. Acad. Sci. 106, 19637–19643 (2009).

    ADS  CAS  PubMed  Article  Google Scholar 

  59. 59.

    Araújo, M. B., Pearson, R. G., Thuiller, W. & Erhard, M. Validation of species–climate impact models under climate change. Glob. Chang. Biol. 11, 1504–1513 (2005).

    ADS  Article  Google Scholar 

  60. 60.

    Huete, A. et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 83, 195–213 (2002).

    ADS  Article  Google Scholar 

  61. 61.

    Moreno-Amat, E. et al. Impact of model complexity on cross-temporal transferability in Maxent species distribution models: An assessment using paleobotanical data. Ecol. Modell. 312, 308–317 (2015).

    Article  Google Scholar 

  62. 62.

    Araujo, M. B., Thuiller, W., Williams, P. H. & Reginster, I. Downscaling European species atlas distributions to a finer resolution: Implications for conservation planning. Glob. Ecol. Biogeogr. 14, 17–30 (2005).

    Article  Google Scholar 

  63. 63.

    McPherson, J. M., Jetz, W. & Rogers, D. J. Using coarse-grained occurrence data to predict species distributions at finer spatial resolutions—possibilities and limitations. Ecol. Modell. 192, 499–522 (2006).

    Article  Google Scholar 

  64. 64.

    Fernandes, R. F. et al. A novel downscaling approach to predict plant invasions and improve local conservation actions. Biol. Invasions 16, 2577–2590 (2014).

    Article  Google Scholar 

  65. 65.

    Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M. B. BIOMOD - a platform for ensemble forecasting of species distributions. Ecography (Cop.). 32, 369–373 (2009).

    Article  Google Scholar 

  66. 66.

    Thuiller, W., Engler, R. & Breiner, F. T. biomod2: Ensemble Platform for Species Distribution Modeling. R package version 3.3-7, (2016).

  67. 67.

    Elith, J. et al. A statistical explanation of MaxEnt for ecologists. Divers. Distrib. 17, 43–57 (2011).

    Article  Google Scholar 

  68. 68.

    Fielding, A. H. & Bell, J. F. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ. Conserv. 24, 38–49 (1997).

    Article  Google Scholar 

  69. 69.

    Marmion, M., Parviainen, M., Luoto, M., Heikkinen, R. K. & Thuiller, W. Evaluation of consensus methods in predictive species distribution modelling. Divers. Distrib. 15, 59–69 (2009).

    Article  Google Scholar 

  70. 70.

    Swets, J. A. Measuring the accuracy of diagnostic systems. Science 240, 1285–1293 (1988).

    ADS  MathSciNet  CAS  PubMed  MATH  Article  Google Scholar 

  71. 71.

    W. Schoener, T. Nonsynchronous Spatial Overlap of Lizards in Patchy Habitats. Ecology 51, 408 (1970).

    Article  Google Scholar 

  72. 72.

    Broennimann, O. et al. Measuring ecological niche overlap from occurrence and spatial environmental data. Glob. Ecol. Biogeogr. 21, 481–497 (2012).

    Article  Google Scholar 

  73. 73.

    R Core Team. A language and environment for statistical computing. (R Foundation for Statistical Computing, Vienna, Austria., 2015). at,

  74. 74.

    McCullagh, P. & Nelder, J. A. Generalized Linear Models. London, UK: Chapman and Hall (1989).

  75. 75.

    Burnham, K. P. & Anderson, D. R. Model Selection and Multimodel Inference: a Practical Information-theoretic Approach. (Springer, 2002).

  76. 76.

    Burnham, K. P., Anderson, D. R. & Huyvaert, K. P. AIC model selection and multimodel inference in behavioral ecology: Some background, observations, and comparisons. Behav. Ecol. Sociobiol. 65, 23–35 (2011).

    Article  Google Scholar 

  77. 77.

    Taylor, A. & Knight, R. Wildlife responses to recreation and associated visitor perceptions. Ecol. Appl. 13, 951–963 (2003).

    Article  Google Scholar 

  78. 78.

    Kharouba, H. M., Algar, A. C. & Kerr, J. T. Historically calibrated predictions of butterfly species’ range shift using global change as a pseudo-experiment. Ecology 90, 2213–2222 (2009).

    PubMed  Article  Google Scholar 

  79. 79.

    Regos, A. et al. Hindcasting the impacts of land-use changes on bird communities with species distribution models of Bird Atlas data. Ecol. Appl. 28, 1867–1883 (2018).

    PubMed  Article  Google Scholar 

  80. 80.

    Arenas-Castro, S., Gonçalves, J., Alves, P., Alcaraz-Segura, D. & Honrado, J. P. Assessing the multi-scale predictive ability of ecosystem functional attributes for species distribution modelling. PLoS One 13, e0199292 (2018).

    PubMed  PubMed Central  Article  CAS  Google Scholar 

  81. 81.

    Fitzpatrick, M. C. et al. How will climate novelty influence ecological forecasts? Using the Quaternary to assess future reliability. Glob. Chang. Biol. 24, 3575–3586 (2018).

    ADS  PubMed  Article  Google Scholar 

  82. 82.

    Roberts, D. R. & Hamann, A. Predicting potential climate change impacts with bioclimate envelope models: a palaeoecological perspective. Glob. Ecol. Biogeogr. 21, 121–133 (2012).

    Article  Google Scholar 

  83. 83.

    Qiao, H. et al. An evaluation of transferability of ecological niche models. Ecography, (2018).

  84. 84.

    Vallecillo, S., Brotons, L. & Thuiller, W. Dangers of predicting bird species distributions in response to land-cover changes. Ecol. Appl. 19, 538–549 (2009).

    PubMed  Article  Google Scholar 

  85. 85.

    De Cáceres, M. & Brotons, L. Calibration of hybrid species distribution models: the value of general-purpose vs. targeted monitoring data. Divers. Distrib. 18, 977–989 (2012).

    Article  Google Scholar 

  86. 86.

    Guevara, L., Gerstner, B. E., Kass, J. M. & Anderson, R. P. Toward ecologically realistic predictions of species distributions: A cross-time example from tropical montane cloud forests. Glob. Chang. Biol. 24, 1511–1522 (2018).

    ADS  PubMed  Article  Google Scholar 

  87. 87.

    Regos, A. et al. Predicting the future effectiveness of protected areas for bird conservation in Mediterranean ecosystems under climate change and novel fire regime scenarios. Divers. Distrib. 22, 83–96 (2016).

    Article  Google Scholar 

  88. 88.

    Regos, A., Clavero, M., D’Amen, M., Guisan, A. & Brotons, L. Wildfire-vegetation dynamics affect predictions of climate change impact on bird communities. Ecography 41, 982–995 (2018).

    Article  Google Scholar 

  89. 89.

    Keil, P., Belmaker, J., Wilson, A. M., Unitt, P. & Jetz, W. Downscaling of species distribution models: a hierarchical approach. Methods Ecol. Evol. 4, 82–94 (2013).

    Article  Google Scholar 

  90. 90.

    Tapia, L., Regos, A., Gil-Carrera, A. & Dominguez, J. Assessing the temporal transferability of raptor distribution models: Implications for conservation. Bird Conserv. Int. 28, 375–389 (2018).

    Article  Google Scholar 

  91. 91.

    Jackson, S. T. & Overpeck, J. T. Paleontological Society Responses of Plant Populations and Communities to Environmental Changes of the Late Quaternary Responses of plant populations and communities to environmental changes of the late Quaternary 26, 194–220 (2000).

  92. 92.

    Roberts, D. R. et al. Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography (Cop.). 40, 913–929 (2017).

    Article  Google Scholar 

  93. 93.

    Kéry, M. Towards the modelling of true species distributions. J. Biogeogr. 38, 617–618 (2011).

    Article  Google Scholar 

  94. 94.

    Lahoz-Monfort, J. J., Guillera-Arroita, G. & Wintle, B. a. Imperfect detection impacts the performance of species distribution models. Glob. Ecol. Biogeogr. 23, 504–515 (2014).

    Article  Google Scholar 

  95. 95.

    Seoane, J., Carrascal, L. M., Alonso, C. L. & Palomino, D. Species-specific traits associated to prediction errors in bird habitat suitability modelling. Ecol. Modell. 185, 299–308 (2005).

    Article  Google Scholar 

  96. 96.

    Brotons, L., Herrando, S. & Pla, M. Updating bird species distribution at large spatial scales: applications of habitat modelling to data from long-term monitoring programs. Divers. Distrib. 13, 276–288 (2007).

    Article  Google Scholar 

  97. 97.

    Slatyer, Ra, Hirst, M. & Sexton, J. P. Niche breadth predicts geographical range size: a general ecological pattern. Ecol. Lett. 16, 1104–1114 (2013).

    PubMed  Article  Google Scholar 

  98. 98.

    Shariatinajafabadi, M. et al. Migratory herbivorous waterfowl track satellite-derived green wave index. PLoS One 9, 1–11 (2014).

    Article  CAS  Google Scholar 

  99. 99.

    Najafabadi, M. S. et al. Satellite-versus temperature-derived green wave indices for predicting the timing of spring migration of avian herbivores. Ecol. Indic. 58, 322–331 (2015).

    Article  Google Scholar 

  100. 100.

    Pettorelli, N. et al. The Normalized Difference Vegetation Index (NDVI): Unforeseen successes in animal ecology. Climate Research. 46, 15–27, (2011).

  101. 101.

    Requena-Mullor, J. M. et al. Modeling spatial distribution of European badger in arid landscapes: an ecosystem functioning approach. Landsc. Ecol. 29, 843–855 (2014).

    Article  Google Scholar 

  102. 102.

    Nagendra, H. et al. Remote sensing for conservation monitoring: Assessing protected areas, habitat extent, habitat condition, species diversity, and threats. Ecol. Indic. 33, 45–59 (2013).

    Article  Google Scholar 

  103. 103.

    Murray, J. V., Low Choy, S., Mcalpine, C. A., Possingam, H. P. & Goldizen, A. W. Evaluating model transferability for a threatened species to adjacent areas: Implications for rock-wallaby conservation. Austral Ecol. 36, 76–89 (2011).

    Article  Google Scholar 

  104. 104.

    Thuiller, W., Brotons, L., Araujo, M. B. & Lavorel, S. Effects of restricting environmental range of data to project current and future species distributions. Ecography (Cop.). 27, 165–172 (2004).

    Article  Google Scholar 

  105. 105.

    Grinnell, J. The Niche-Relationships of the California Thrasher. Auk 34, 427–433 (1917).

    Article  Google Scholar 

  106. 106.

    Grinnell, J. Field tests of theories concerning distribution control. Am. Nat. 51, 115–128 (1917).

    Article  Google Scholar 

Download references


This research was developed as part of the project ECOPOTENTIAL, which received funding from the European Union’s Horizon 2020 Research and Innovation Programme under agreement No. 641762. We thank everyone who contributed to the fieldwork: Xosé Pardavila, Adrián Lamosa (Sorex, Ecoloxía e Medio Ambiente SL), Marta Arenas, Alberto Toupa and Fernando Martínez-Freiría. Field surveys were funded by the project INTERREG-POCTEC (‘NATURA Xurés-Gerês’). A.R. was financially supported by the Xunta de Galicia (post-doctoral fellowship ED481B2016/084-0). Miquel Ninyerola and Meritxell Batalla (UAB) generated climate variables from data provided by the Spanish Meteorological Agency and the Spanish Ministry of Marine and Rural Environment within the MONTES-Consolider project (CSD2008-00040).

Author information




A.R., D.A.-S. and J.H. conceived the ideas and designed methodology; J.D. designed and coordinated fieldwork; A.R., L.G. and D.A.-S. analysed the data; A.R. led the writing of the manuscript. All authors contributed critically to the draft and gave final approval for publication.

Corresponding author

Correspondence to Adrián Regos.

Ethics declarations

Competing Interests

The authors declare no competing interests.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Regos, A., Gagne, L., Alcaraz-Segura, D. et al. Effects of species traits and environmental predictors on performance and transferability of ecological niche models. Sci Rep 9, 4221 (2019).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing