Sensitivity of predictive species distribution models to change in grain size

Predictive species distribution modelling (SDM) has become an essential tool in biodiversity conservation and management. The choice of grain size (resolution) of environmental layers used in modelling is one important factor that may affect predictions. We applied 10 distinct modelling techniques to presence-only data for 50 species in five different regions, to test whether: (1) a 10-fold coarsening of resolution affects predictive performance of SDMs, and (2) any observed effects are dependent on the type of region, modelling technique, or species considered. Results show that a 10 times change in grain size does not severely affect predictions from species distribution models. The overall trend is towards degradation of model performance, but improvement can also be observed. Changing grain size does not equally affect models across regions, techniques, and species types. The strongest effect is on regions and species types, with tree species in the data sets (regions) with highest locational accuracy being most affected. Changing grain size had little influence on the ranking of techniques: boosted regression trees remain best at both resolutions. The number of occurrences used for model training had an important effect, with larger sample sizes resulting in better models, which tended to be more sensitive to grain. Effect of grain change was only noticeable for models reaching sufficient performance and/or with initial data that have an intrinsic error smaller than the coarser grain size.     

Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models

Predictive performance is important to many applications of species distribution models (SDMs). The SDM ‘ensemble’ approach, which combines predictions across different modelling methods, is believed to improve predictive performance, and is used in many recent SDM studies. Here, we aim to compare the predictive performance of ensemble species distribution models to that of individual models, using a large presence–absence dataset of eucalypt tree species. To test model performance, we divided our dataset into calibration and evaluation folds using two spatial blocking strategies (checkerboard-pattern and latitudinal slicing). We calibrated and cross-validated all models within the calibration folds, using both repeated random division of data (a common approach) and spatial blocking. Ensembles were built using the software package ‘biomod2’, with standard (‘untuned’) settings. Boosted regression tree (BRT) models were also fitted to the same data, tuned according to published procedures. We then used evaluation folds to compare ensembles against both their component untuned individual models, and against the BRTs. We used area under the receiver-operating characteristic curve (AUC) and log-likelihood for assessing model performance. In all our tests, ensemble models performed well, but not consistently better than their component untuned individual models or tuned BRTs across all tests. Moreover, choosing untuned individual models with best cross-validation performance also yielded good external performance, with blocked cross-validation proving better suited for this choice, in this study, than repeated random cross-validation. The latitudinal slice test was only possible for four species; this showed some individual models, and particularly the tuned one, performing better than ensembles. This study shows no particular benefit to using ensembles over individual tuned models. It also suggests that further robust testing of performance is required for situations where models are used to predict to distant places or environments.     

A comparative evaluation of presence-only methods for modelling species distribution

In spite of increasing application of presence-only models in ecology and conservation and the growing number of such models, little is known about the relative performance of different modelling methods, and some of the leading models (e.g. GARP and ENFA) have never been compared with one another. Here we compare the performance of six presence-only models that have been selected to represent an increasing level of model complexity [BIOCLIM, HABITAT, Mahalanobis distance (MD), DOMAIN, ENFA, and GARP] using data on the distribution of 42 species of land snails, nesting birds, and insectivorous bats in Israel. The models were calibrated using data from museum collections and observation databases, and their predictions were evaluated using Cohen's Kappa based on field data collected in a standardized sampling design covering most parts of Israel. Predictive accuracy varied between modelling methods with GARP and MD showing the highest accuracy, BIOCLIM and ENFA showing the lowest accuracy, and HABITAT and DOMAIN showing intermediate accuracy levels. Yet, differences between the various models were relatively small except for GARP and MD that were significantly more accurate than BIOCLIM and ENFA. In spite of large differences among species in prevalence and niche width, neither prevalence nor niche width interacted with the modelling method in determining predictive accuracy. However, species with relatively narrow niches were modelled more accurately than species with wider niches. Differences among species in predictive accuracy were highly consistent over all modelling methods, indicating the need for a better understanding of the ecological and geographical factors that influence the performance of species distribution models.     

A new method for dealing with residual spatial autocorrelation in species distribution models

Species distribution modelling (SDM) is a widely used tool and has many applications in ecology and conservation biology. Spatial autocorrelation (SAC), a pattern in which observations are related to one another by their geographic distance, is common in georeferenced ecological data. SAC in the residuals of SDMs violates the ‘independent errors’ assumption required to justify the use of statistical models in modelling species’ distributions. The autologistic modelling approach accounts for SAC by including an additional term (the autocovariate) representing the similarity between the value of the response variable at a location and neighbouring locations. However, autologistic models have been found to introduce bias in the estimation of parameters describing the influence of explanatory variables on habitat occupancy. To address this problem we developed an extension to the autologistic approach by calculating the autocovariate on SAC in residuals (the RAC approach). Performance of the new approach was tested on simulated data with a known spatial structure and on strongly autocorrelated mangrove species’ distribution data collected in northern Australia. The RAC approach was implemented as generalized linear models (GLMs) and boosted regression tree (BRT) models. We found that the BRT models with only environmental explanatory variables can account for some SAC, but applying the standard autologistic or RAC approaches further reduced SAC in model residuals and substantially improved model predictive performance. The RAC approach showed stronger inferential performance than the standard autologistic approach, as parameter estimates were more accurate and statistically significant variables were accurately identified. The new RAC approach presented here has the potential to account for spatial autocorrelation while maintaining strong predictive and inferential performance, and can be implemented across a range of modelling approaches.     

Predicting the species composition of Nardus stricta communities by logistic regression modelling

Question: Predictive models in plant ecology usually deal with single species or community types. Little effort has so far been made to predict the species composition of a community explicitly. The modelling approach presented here provides a conceptual framework on how to achieve this by combining habitat models for a large number of species to an additive community model. Our approach is exemplified by Nardus stricta communities (acidophilous, low‐productive grassland). Location: Large areas of Germany, 0–2040 m a.s.l. Methods: Logistic regression is applied for individual species models which are subsequently combined for an explicit prediction of species composition. Several parameters reflecting soil, management and climatic conditions serve as predictor variables. For validation, bootstrap and jackknife resampling procedures are used as well as ordination techniques (DCA, CCA). Results: We calculated significant models for 138 individual species. The predictions of species composition and species richness yield good agreements with the observed data. DCA and CCA results show that the community model preserves the main patterns in floristic space. Conclusions: Our approach of predicting species composition is an effective tool that can be applied in nature conservation, e.g. to assess the effects of different site conditions and alternative management scenarios on species composition and richness.     

Do traits of plant species predict the efficacy of species distribution models for finding new occurrences?

Species distribution models (SDMs) are used to test ecological theory and to direct targeted surveys for species of conservation concern. Several studies have tested for an influence of species traits on the predictive accuracy of SDMs. However, most used the same set of environmental predictors for all species and/or did not use truly independent data to test SDM accuracy. We built eight SDMs for each of 24 plant species of conservation concern, varying the environmental predictors included in each SDM version. We then measured the accuracy of each SDM using independent presence and absence data to calculate area under the receiver operating characteristic curve (AUC) and true positive rate (TPR). We used generalized linear mixed models to test for a relationship between species traits and SDM accuracy, while accounting for variation in SDM performance that might be introduced by different predictor sets. All traits affected one or both SDM accuracy measures. Species with lighter seeds, animal‐dispersed seeds, and a higher density of occurrences had higher AUC and TPR than other species, all else being equal. Long‐lived woody species had higher AUC than herbaceous species, but lower TPR. These results support the hypothesis that the strength of species–environment correlations is affected by characteristics of species or their geographic distributions. However, because each species has multiple traits, and because AUC and TPR can be affected differently, there is no straightforward way to determine a priori which species will yield useful SDMs based on their traits. Most species yielded at least one useful SDM. Therefore, it is worthwhile to build and test SDMs for the purpose of finding new populations of plant species of conservation concern, regardless of these species’ traits.     

Model-based uncertainty in species range prediction

Aim Many attempts to predict the potential range of species rely on environmental niche (or ‘bioclimate envelope’) modelling, yet the effects of using different niche‐based methodologies require further investigation. Here we investigate the impact that the choice of model can have on predictions, identify key reasons why model output may differ and discuss the implications that model uncertainty has for policy‐guiding applications. Location The Western Cape of South Africa. Methods We applied nine of the most widely used modelling techniques to model potential distributions under current and predicted future climate for four species (including two subspecies) of Proteaceae. Each model was built using an identical set of five input variables and distribution data for 3996 sampled sites. We compare model predictions by testing agreement between observed and simulated distributions for the present day (using the area under the receiver operating characteristic curve (AUC) and kappa statistics) and by assessing consistency in predictions of range size changes under future climate (using cluster analysis). Results Our analyses show significant differences between predictions from different models, with predicted changes in range size by 2030 differing in both magnitude and direction (e.g. from 92% loss to 322% gain). We explain differences with reference to two characteristics of the modelling techniques: data input requirements (presence/absence vs. presence‐only approaches) and assumptions made by each algorithm when extrapolating beyond the range of data used to build the model. The effects of these factors should be carefully considered when using this modelling approach to predict species ranges. Main conclusions We highlight an important source of uncertainty in assessments of the impacts of climate change on biodiversity and emphasize that model predictions should be interpreted in policy‐guiding applications along with a full appreciation of uncertainty.     

Satellite surface reflectance improves habitat distribution mapping: a case study on heath and shrub formations in the Cantabrian Mountains (NW Spain)

Aim We aim to map the distribution of four heath and shrub formations constituting habitats of high conservation priority in Europe, whose occurrence is strongly dependent on human activities. Specifically, we assess whether the use of LANDSAT data in habitat distribution modelling may account for land use management, allowing accurate mapping of real distribution patterns. In particular, we explore whether reflectance values may be a better alternative to other remote sensing data traditionally used in modelling approaches (i.e. spectral vegetation indices and classified land cover maps). Finally, we test whether modelling performance is affected by the ecological traits of the dominant species of the target formations. Location Cantabrian Mountains (NW Spain). Methods We generated maps for the four formations (two specialists vs. two generalists) using MaxEnt. First, we ran the models with environmental predictors only (topography, climate, lithology and human disturbances). Then, we compared the advantages of including, in turn, different data derived from LANDSAT imagery: reflectance values (corresponding to different wavelength channels of the multispectral image), a spectral index and a land cover map. We assessed changes in explanatory power and also in the formation’s predicted distribution patterns. Results Formations dominated by specialist species were accurately mapped on a base of environmental variables only, whereas those dominated by generalists were overpredicted. Average mean temperature, southness and distance to urban areas were the variables contributing most in predictions of environmental models. LANDSAT channels increased the accuracy of all models, but mainly those for formations dominated by generalist species. They showed advantages against other remote sensing data traditionally used in modelling approaches. Main conclusions Habitat distribution models allowed accurate mapping of heath and shrub formations. The use of reflectance values as predictors improved the accuracy of the models, particularly for formations dominated by generalist species, supplying environmental information that was otherwise unavailable.     

Assessing the accuracy of species distribution models to predict amphibian species richness patterns

1. Evaluating the distribution of species richness where biodiversity is high but has been insufficiently sampled is not an easy task. Species distribution modelling has become a useful approach for predicting their ranges, based on the relationships between species records and environmental variables. Overlapping predictions of individual distributions could be a useful strategy for obtaining estimates of species richness and composition in a region, but these estimates should be evaluated using a proper validation process, which compares the predicted richness values and composition with accurate data from independent sources. 2. In this study, we propose a simple approach to estimate model performance for several distributional predictions generated simultaneously. This approach is particularly suitable when species distribution modelling techniques that require only presence data are used. 3. The individual distributions for the 370 known amphibian species of Mexico were predicted using maxent to model data on their known presence (66,113 presence-only records). Distributions were subsequently overlapped to obtain a prediction of species richness. Accuracy was assessed by comparing the overall species richness values predicted for the region with observed and predicted values from 118 well-surveyed sites, each with an area of c. 100 km(2), which were identified using species accumulation curves and nonparametric estimators. 4. The derived models revealed a remarkable heterogeneity of species richness across the country, provided information about species composition per site and allowed us to obtain a measure of the spatial distribution of prediction errors. Examining the magnitude and location of model inaccuracies, as well as separately assessing errors of both commission and omission, highlights the inaccuracy of the predictions of species distribution models and the need to provide measures of uncertainty along with the model results. 5. The combination of a species distribution modelling method like maxent and species richness estimators offers a useful tool for identifying when the overall pattern provided by all model predictions might be representing the geographical patterns of species richness and composition, regardless of the particular quality or accuracy of the predictions for each individual species.     

Modelling species distributions in Britain: a hierarchical integration of climate and land-cover data

A modelling framework for studying the combined effects of climate and land-cover changes on the distribution of species is presented. The model integrates land-cover data into a correlative bioclimatic model in a scale-dependent hierarchical manner, whereby Artificial Neural Networks are used to characterise species' climatic requirements at the European scale and land-cover requirements at the British scale. The model has been tested against an alternative non-hierarchical approach and has been applied to four plant species in Britain: Rhynchospora alba , Erica tetralix , Salix herbacea and Geranium sylvaticum . Predictive performance has been evaluated using Cohen's Kappa statistic and the area under the Receiver Operating Characteristic curve, and a novel approach to identifying thresholds of occurrence which utilises three levels of confidence has been applied. Results demonstrate reasonable to good predictive performance for each species, with the main patterns of distribution simulated at both 10 km and 1 km resolutions. The incorporation of land-cover data was found to significantly improve purely climate-driven predictions for R. alba and E. tetralix , enabling regions with suitable climate but unsuitable land-cover to be identified. The study thus provides an insight into the roles of climate and land-cover as determinants of species' distributions and it is demonstrated that the modelling approach presented can provide a useful framework for making predictions of distributions under scenarios of changing climate and land-cover type. The paper confirms the potential utility of multi-scale approaches for understanding environmental limitations to species' distributions, and demonstrates that the search for environmental correlates with species' distributions must be addressed at an appropriate spatial scale. Our study contributes to the mounting evidence that hierarchical schemes are characteristic of ecological systems.     

Effects of species' ecology on the accuracy of distribution models

In the face of accelerating biodiversity loss and limited data, species distribution models – which statistically capture and predict species’ occurrences based on environmental correlates – are increasingly used to inform conservation strategies. Additionally, distribution models and their fit provide insights on the broad‐scale environmental niche of species. To investigate whether the performance of such models varies with species’ ecological characteristics, we examined distribution models for 1329 bird species in southern and eastern Africa. The models were constructed at two spatial resolutions with both logistic and autologistic regression. Satellite‐derived environmental indices served as predictors, and model accuracy was assessed with three metrics: sensitivity, specificity and the area under the curve (AUC) of receiver operating characteristics plots. We then determined the relationship between each measure of accuracy and ten ecological species characteristics using generalised linear models. Among the ecological traits tested, species’ range size, migratory status, affinity for wetlands and endemism proved most influential on the performance of distribution models. The number of habitat types frequented (habitat tolerance), trophic rank, body mass, preferred habitat structure and association with sub‐resolution habitats also showed some effect. In contrast, conservation status made no significant impact. These findings did not differ from one spatial resolution to the next. Our analyses thus provide conservation scientists and resource managers with a rule of thumb that helps distinguish, on the basis of ecological traits, between species whose occurrence is reliably or less reliably predicted by distribution models. Reasonably accurate distribution models should, however, be attainable for most species, because the influence ecological traits bore on model performance was only limited. These results suggest that none of the ecological traits tested provides an obvious correlate for environmental niche breadth or intra‐specific niche differentiation.     

Evaluating the reliability of species distribution models with an indirect measure of bird reproductive performance

Measures of fitness such as reproductive performance are considered reliable indicators of habitat quality for a species. Such measures are, however, only available in a restricted number of sites, which prevents them from being used to quantify habitat quality across landscapes or regions. Alternatively, species presence records can be used along with environmental variables to build models that predict the distribution of species across larger spatial extents. Model predictions are often used for management purposes as they are assumed to describe the quality of the habitats to support a species. Yet, given that species are often present both in optimal and suboptimal areas, the use of data collected during the breeding season to build these models may potentially result in misleading predictions of habitat quality for the reproduction of the species, with potentially significant conservation consequences. In this study we analysed the relationship between fitness parameters informing on habitat quality for reproduction and predictions of species distribution models at multiple spatial scales using two independent sets of data. For 19 passerine bird species, we compared an indirect measure of reproductive performance (ratio of juveniles‐to‐adults) – obtained from Constant Effort Sites (CES) mist‐netting data in Catalonia – with the predictions of models based on bird presence records collected during the Catalan Breeding Bird Atlas (CBBA). A positive relationship between the predictions derived from species distribution models and the reproductive performance of the species was found for almost half of the species at one or more spatial scales. This result suggests that species distribution models may help to predict habitat quality for some species over some extents. However, caution is needed as this is not consistent for all species at all scales. Further work based on species‐ and scale‐specific approaches is now required to understand in which situations species distribution models provide predictions that are in line with reproductive performance.     

Cost-effectiveness of habitat-suitability maps using low-detailed data for elusive bat species

European bat species are strictly protected by law, and the Member States of the European Union are obliged to record species condition and to contribute to their conservation. Habitat-suitability models are an essential aid in assessing the conservation status and distribution of a species. However, model performance depends on the data quality. This study compares habitat-suitability models that were generated from two data sets that differ in the degree of details included. The first model used data that were low in detail but freely available and the second used data that were very detailed but costly. Three hypotheses were addressed: (1) that the model using low-detailed data is sufficient in its performance to aid the assessment of species distribution and infrastructural planning; (2) the visualisation of actual species distribution is more accurate in the high-detailed model; and (3) habitat-suitability maps can depict species distribution better than species occurrence data alone. To develop models, climate, geographic and roosting data of Myotis bechsteinii were used. Models allowed very good spatial predictions of suitable habitats. However, the model using low-detailed data overestimated suitable habitat. The high-detailed model was more able to predict actual species distribution. These findings were supported by field evaluation where M. bechsteinii could only be detected in areas where both models predicted high habitat suitability. This framework is promising as it resulted in spatially explicit habitat-suitability maps and suggests that similar models may be used to improve the understanding of bat distribution and factors endangering other species of bats.     

Finessing atlas data for species distribution models

Aim The spatial resolution of species atlases and therefore resulting model predictions are often too coarse for local applications. Collecting distribution data at a finer resolution for large numbers of species requires a comprehensive sampling effort, making it impractical and expensive. This study outlines the incorporation of existing knowledge into a conventional approach to predict the distribution of Bonelli’s eagle (Aquila fasciata) at a resolution 100 times finer than available atlas data. Location Malaga province, Andalusia, southern Spain. Methods A Bayesian expert system was proposed to utilize the knowledge from distribution models to yield the probability of a species being recorded at a finer resolution (1 × 1 km) than the original atlas data (10 × 10 km). The recorded probability was then used as a weight vector to generate a sampling scheme from the species atlas to enhance the accuracy of the modelling procedure. The maximum entropy for species distribution modelling (MaxEnt) was used as the species distribution model. A comparison was made between the results of the MaxEnt using the enhanced and, the random sampling scheme, based on four groups of environmental variables: topographic, climatic, biological and anthropogenic. Results The models with the sampling scheme enhanced by an expert system had a higher discriminative capacity than the baseline models. The downscaled (i.e. finer scale) species distribution maps using a hybrid MaxEnt/expert system approach were more specific to the nest locations and were more contrasted than those of the baseline model. Main conclusions The proposed method is a feasible substitute for comprehensive field work. The approach developed in this study is applicable for predicting the distribution of Bonelli’s eagle at a local scale from a national‐level occurrence data set; however, the usefulness of this approach may be limited to well‐known species.     

Testing projected wild bee distributions in agricultural habitats: predictive power depends on species traits and habitat type

Species distribution models (SDM) are increasingly used to understand the factors that regulate variation in biodiversity patterns and to help plan conservation strategies. However, these models are rarely validated with independently collected data and it is unclear whether SDM performance is maintained across distinct habitats and for species with different functional traits. Highly mobile species, such as bees, can be particularly challenging to model. Here, we use independent sets of occurrence data collected systematically in several agricultural habitats to test how the predictive performance of SDMs for wild bee species depends on species traits, habitat type, and sampling technique. We used a species distribution modeling approach parametrized for the Netherlands, with presence records from 1990 to 2010 for 193 Dutch wild bees. For each species, we built a Maxent model based on 13 climate and landscape variables. We tested the predictive performance of the SDMs with independent datasets collected from orchards and arable fields across the Netherlands from 2010 to 2013, using transect surveys or pan traps. Model predictive performance depended on species traits and habitat type. Occurrence of bee species specialized in habitat and diet was better predicted than generalist bees. Predictions of habitat suitability were also more precise for habitats that are temporally more stable (orchards) than for habitats that suffer regular alterations (arable), particularly for small, solitary bees. As a conservation tool, SDMs are best suited to modeling rarer, specialist species than more generalist and will work best in long‐term stable habitats. The variability of complex, short‐term habitats is difficult to capture in such models and historical land use generally has low thematic resolution. To improve SDMs’ usefulness, models require explanatory variables and collection data that include detailed landscape characteristics, for example, variability of crops and flower availability. Additionally, testing SDMs with field surveys should involve multiple collection techniques.     

BIOMOD – optimizing predictions of species distributions and projecting potential future shifts under global change

A new computation framework (BIOMOD: BIOdiversity MODelling) is presented, which aims to maximize the predictive accuracy of current species distributions and the reliability of future potential distributions using different types of statistical modelling methods. BIOMOD capitalizes on the different techniques used in static modelling to provide spatial predictions. It computes, for each species and in the same package, the four most widely used modelling techniques in species predictions, namely Generalized Linear Models (GLM), Generalized Additive Models (GAM), Classification and Regression Tree analysis (CART) and Artificial Neural Networks (ANN). BIOMOD was applied to 61 species of trees in Europe using climatic quantities as explanatory variables of current distributions. On average, all the different modelling methods yielded very good agreement between observed and predicted distributions. However, the relative performance of different techniques was idiosyncratic across species, suggesting that the most accurate model varies between species. The results of this evaluation also highlight that slight differences between current predictions from different modelling techniques are exacerbated in future projections. Therefore, it is difficult to assess the reliability of alternative projections without validation techniques or expert opinion. It is concluded that rather than using a single modelling technique to predict the distribution of several species, it would be more reliable to use a framework assessing different models for each species and selecting the most accurate one using both evaluation methods and expert knowledge.     

LIVES: a new habitat modelling technique for predicting the distribution of species’ occurrences using presence-only data based on limiting factor theory

Predictive modelling techniques using presence-only data have attracted increasing attention because they can provide information on species distributions and their potential habitat for conservation and ecosystem management. However, the existing predictive modelling techniques have several limitations. Here, we propose a novel predictive modelling technique, Limiting Variable and Environmental Suitability (LIVES), for predicting the distributions and potential habitats of species using presence-only data. It is based on limiting factor theory, which postulates that the occurrence of a species is only determined by the factor that most limits its distribution. LIVES predicts the suitability of a candidate grid cell for a species in terms of limiting environmental factor. It also predicts the most limiting factor or the potential limiting factor at the grid cell. The environmental factors can be climatic, geological, biological and any other relevant environmental factors, whether quantitative or qualitative. The predicted habitats consist of the current distribution of the species and the potentially suitable areas for the species where there is currently no record of occurrence. We also compare several properties of LIVES and other predictive modelling techniques. On the basis of 1,000 simulations, the average predictions of LIVES are more accurate than the two other commonly used modelling techniques (BIOCLIM and DOMAIN) for presence-only data.     

Ecological traps and species distribution models: a challenge for prioritizing areas of conservation importance

Species distribution models analyse how species use different types of habitats. Their spatial predictions are often used to prioritize areas for conservation. Individuals may, however, prefer settling in habitat types of low quality compared to other available habitats. This ecological trap phenomenon is usually studied in a small number of habitat patches and consequences at the landscape level are largely unknown. It is therefore often unclear whether the spatial pattern of habitat use is aligned with the behavioural decisions made by the individuals during habitat selection or reflects actual variation in the quality of different habitat types. As species distribution models analyse the pattern of occurrence in different habitats, there is a conservation interest in examining what their predictions mean in terms of habitat quality when ecological traps are operating. Previous work in Belgium showed that red-backed shrikes Lanius collurio are more attracted to newly available clear-cut habitat in plantation forests than to the traditionally used farmland habitat. We developed models with shrike distribution data and compared their predictions with spatial variation in shrike reproductive performance used as a proxy for habitat quality. Models accurately predicted shrike distribution and identified the preferred clear-cut patches as the most frequently used habitat, but reproductive performance was lower in clear-cut areas than in farmland. With human-induced rapid environmental changes, organisms may indeed be attracted to low-quality habitats and occupy them at high densities. Consequently, the predictions of statistical models based on occurrence records may not align with variation in significant population parameters for the maintenance of the species. When species expand their range to novel habitats, such models are useful to document the spatial distribution of the organisms, but data on population growth rates are worth collecting before using model predictions to guide the spatial prioritization of conservation actions.     

Selecting from correlated climate variables: a major source of uncertainty for predicting species distributions under climate change

Correlative species distribution models are frequently used to predict species’ range shifts under climate change. However, climate variables often show high collinearity and most statistical approaches require the selection of one among strongly correlated variables. When causal relationships between species presence and climate parameters are unknown, variable selection is often arbitrary, or based on predictive performance under current conditions. While this should only marginally affect current range predictions, future distributions may vary considerably when climate parameters do not change in concert. We investigated this source of uncertainty using four highly correlated climate variables together with a constant set of landscape variables in order to predict current (2010) and future (2050) distributions of four mountain bird species in central Europe. Simulating different parameterization decisions, we generated a) four models including each of the climate variables singly, b) a model taking advantage of all variables simultaneously and c) an un‐weighted average of the predictions of a). We compared model accuracy under current conditions, predicted distributions under four scenarios of climate change, and – for one species – evaluated back‐projections using historical occurrence data. Although current and future variable‐correlations remained constant, and the models’ accuracy under contemporary conditions did not differ, future range predictions varied considerably in all climate change scenarios. Averaged models and models containing all climate variables simultaneously produced intermediate predictions; the latter, however, performed best in back‐projections. This pattern, consistent across different modelling methods, indicates a benefit from including multiple climate predictors in ambiguous situations. Variable selection proved to be an important source of uncertainty for future range predictions, difficult to control using contemporary information. Small, but diverging changes of climate variables, masked by constant overall correlation patterns, can cause substantial differences between future range predictions which need to be accounted for, particularly when outcomes are intended for conservation decisions.     

An evaluation of four threshold selection methods in species occurrence modelling with random forest: Case studies with Davidia involucrata and Cunninghamia lanceolata

Aims Predictive species distribution models (SDMs) are increasingly applied in resource assessment, environmental conservation and biodiversity management. However, most SDM models often yield a predicted probability (suitability) surface map. In conservation and environmental management practices, the information presented as species presence/absence (binary) may be more practical than presented as probability or suitability. Therefore, a threshold is needed to transform the probability or suitability data to presence/absence data. However, little is known about the effects of different threshold-selection methods on model performance and species range changes induced by future climate. Of the numerous SDM models, random forest (RF) can produce probabilistic and binary species distribution maps based on its regression and classification algorisms, respectively. Studies dealing with the comparative test of the performances of RF regression and classification algorisms have not been reported.
Methods Here, the RF was used to simulate the current and project the future potential distributions of Davidia involucrata and Cunninghamia lanceolata. Then, four threshold-setting methods (Default 0.5, MaxKappa, MaxTSS and MaxACC) were selected and used to transform modelled probabilities of occurrence into binary predictions of species presence and absence. Lastly, we investigated the difference in model performance among the threshold selection methods by using five model accuracy measures (Kappa, TSS, Overall accuracy, Sensitivity and Specificity). We also used the map similarity measure, Kappa, for a cell-by-cell comparison of similarities and differences of distribution map under current and future climates.
Important findings We found that the choice of threshold method altered estimates of model performance, species habitat suitable area and species range shifts under future climate. The difference in selected threshold cut-offs among the four threshold methods was significant for D. involucrata, but was not significant for C. lanceolata. Species’ geographic ranges changed (area change and shifting distance) in response to climate change, but the projections of the four threshold methods did not differ significantly with respect to how much or in which direction, but they did differ against RF classification predictions. The pairwise similarity analysis of binary maps indicated that spatial correspondence among prediction maps was the highest between the MaxKappa and the MaxTSS, and lowest between RF classification algorism and the four threshold-setting methods. We argue that the MaxTSS and the MaxKappa are promising methods for threshold selection when RF regression algorism is used for the distribution modeling of species. This study also provides promising insights to our understanding of the uncertainty of threshold selection in species distribution modeling.