Testing methods in species distribution modelling using virtual species: what have we learnt and what are we missing?

Species distribution models (SDMs) have become one of the major predictive tools in ecology. However, multiple methodological choices are required during the modelling process, some of which may have a large impact on forecasting results. In this context, virtual species, i.e. the use of simulations involving a fictitious species for which we have perfect knowledge of its occurrence–environment relationships and other relevant characteristics, have become increasingly popular to test SDMs. This approach provides for a simple virtual ecologist framework under which to test model properties, as well as the effects of the different methodological choices, and allows teasing out the effects of targeted factors with great certainty. This simplification is therefore very useful in setting up modelling standards and best practice principles. As a result, numerous virtual species studies have been published over the last decade. The topics covered include differences in performance between statistical models, effects of sample size, choice of threshold values, methods to generate pseudo‐absences for presence‐only data, among many others. These simulations have therefore already made a great contribution to setting best modelling practices in SDMs. Recent software developments have greatly facilitated the simulation of virtual species, with at least three different packages published to that effect. However, the simulation procedure has not been homogeneous, which introduces some subtleties in the interpretation of results, as well as differences across simulation packages. Here we 1) review the main contributions of the virtual species approach in the SDM literature; 2) compare the major virtual species simulation approaches and software packages; and 3) propose a set of recommendations for best simulation practices in future virtual species studies in the context of SDMs.     

The effect of positional error on fine scale species distribution models increases for specialist species

Species occurrences inherently include positional error. Such error can be problematic for species distribution models (SDMs), especially those based on fine-resolution environmental data. It has been suggested that there could be a link between the influence of positional error and the width of the species ecological niche. Although positional errors in species occurrence data may imply serious limitations, especially for modelling species with narrow ecological niche, it has never been thoroughly explored. We used a virtual species approach to assess the effects of the positional error on fine-scale SDMs for species with environmental niches of different widths. We simulated three virtual species with varying niche breadth, from specialist to generalist. The true distribution of these virtual species was then altered by introducing different levels of positional error (from 5 to 500 m). We built generalized linear models and MaxEnt models using the distribution of the three virtual species (unaltered and altered) and a combination of environmental data at 5 m resolution. The models’ performance and niche overlap were compared to assess the effect of positional error with varying niche breadth in the geographical and environmental space. The positional error negatively impacted performance and niche overlap metrics. The amplitude of the influence of positional error depended on the species niche, with models for specialist species being more affected than those for generalist species. The positional error had the same effect on both modelling techniques. Finally, increasing sample size did not mitigate the negative influence of positional error. We showed that fine-scale SDMs are considerably affected by positional error, even when such error is low. Therefore, where new surveys are undertaken, we recommend paying attention to data collection techniques to minimize the positional error in occurrence data and thus to avoid its negative effect on SDMs, especially when studying specialist species.     

The effect of a gradual response to the environment on species distribution modeling performance

Species distribution models (SDMs) have been widely used in ecology, biogeography, and conservation. Although ecological theory predicts that species occupancy is dynamic, the outputs of SDMs are generally converted into a single occurrence map, and model performance is evaluated in terms of success to predict presences and absences. The aim of this study was to characterize the effects of a gradual response in species occupancy to environmental gradients into the performance of SDMs. First we outline guidelines for the appropriate simulation of artificial species that allows controlling for gradualism and prevalence in the occupancy patterns over an environmental gradient. Second, we derive theoretical expected values for success measures based on presence‐absence predictions (AUC, Kappa, sensitivity and specificity). And finally we used artificial species to exemplify and test the effect of a gradual probabilistic occupancy response to environmental gradients on SDM performance. Our results show that when a species responds gradually to an environmental gradient, conventional measures of SDM predictive success based on presence‐absence cannot be expected to attain currently accepted performance values considered as good, even for a model that recovers perfectly well the true probability of occurrence. A gradual response imposes a theoretical expected value for these measures of performance that can be calculated from the species properties. However, irrespective of the statistical modeling strategy used and of how gradual the species response is, one can recover the true probability of occurrence as a function of environmental variables provided that species and sample prevalence are similar. Therefore, model performance based on presence‐absence should be judged against the theoretical expected value rather than to absolute values currently in use such as AUC > 0.8. Overall, we advocate for a wider use of the probability of occurrence and emphasize the need for further technical developments in this sense.     

Without quality presence–absence data,discrimination metrics such as TSS can be misleading measures of model performance

The discriminating capacity (i.e. ability to correctly classify presences and absences) of species distribution models (SDMs) is commonly evaluated with metrics such as the area under the receiving operating characteristic curve (AUC), the Kappa statistic and the true skill statistic (TSS). AUC and Kappa have been repeatedly criticized, but TSS has fared relatively well since its introduction, mainly because it has been considered as independent of prevalence. In addition, discrimination metrics have been contested because they should be calculated on presence–absence data, but are often used on presence‐only or presence‐background data. Here, we investigate TSS and an alternative set of metrics—similarity indices, also known as F‐measures. We first show that even in ideal conditions (i.e. perfectly random presence–absence sampling), TSS can be misleading because of its dependence on prevalence, whereas similarity/F‐measures provide adequate estimations of model discrimination capacity. Second, we show that in real‐world situations where sample prevalence is different from true species prevalence (i.e. biased sampling or presence‐pseudoabsence), no discrimination capacity metric provides adequate estimation of model discrimination capacity, including metrics specifically designed for modelling with presence‐pseudoabsence data. Our conclusions are twofold. First, they unequivocally impel SDM users to understand the potential shortcomings of discrimination metrics when quality presence–absence data are lacking, and we recommend obtaining such data. Second, in the specific case of virtual species, which are increasingly used to develop and test SDM methodologies, we strongly recommend the use of similarity/F‐measures, which were not biased by prevalence, contrary to TSS.     

The iterative ensemble modelling approach increases the accuracy of fish distribution models

Methodological absences, i.e. when a species is not detected although it is actually present, are known to reduce the prediction accuracy of species distribution models (SDMs). To deal with this problem, we assessed whether a new iterative ensemble modelling (IEM) approach better predicts the spatial distribution of a set of 31 freshwater fish species, exhibiting a wide range of prevalence and methodological absences. Model efficiency was compared using one threshold‐independent (AUC) and three threshold‐dependent indicators of model predictive performance: the percentage of misclassified sites; the Kappa index; and the True Skill Statistic. We then reconstructed species assemblages from individual species predictions and compared observed assemblages to those predicted using EM and IEM using the Jaccard index. Compared to an EM approach, IEM improved model predictive performance for most difficult‐to‐detect species. The iterative approach outperformed EM at modelling the distribution of difficult‐to‐detect species, provided that presence data are representative of the niche of the species. At the assemblage level, the discrepancy between observed and IEM predicted assemblages was significantly lower than that between observed and EM predicted assemblages, showing that IEM can be used to predict the distribution of entire species assemblages. The IEM approach provides a way to consider difficult‐to‐detect species in species distribution models.     

Species traits affect the performance of species distribution models for plants in southern California

Questions: To what extent do plant species traits, including life history, life form, and disturbance response characteristics, affect the degree to which species distributions are determined by physical environmental factors? Is the strength of the relationship between species distribution and environment stronger in some disturbance‐response types than in others? Location: California southwest ecoregion, USA. Methods: We developed species distribution models (SDMs) for 45 plant species using three primary modeling methods (GLMs, GAMs, and Random Forests). Using AUC as a performance measure of prediction accuracy, and measure of the strength of species–environment correlations, we used regression analyses to compare the effects of fire disturbance response type, longevity, dispersal mechanism, range size, cover, species prevalence, and model type. Results: Fire disturbance response type explained more variation in model performance than any other variable, but other species and range characteristics were also significant. Differences in prediction accuracy reflected variation in species life history, disturbance response, and rarity. AUC was significantly higher for longer‐lived species, found at intermediate levels of abundance, and smaller range sizes. Models performed better for shrubs than sub‐shrubs and perennial herbs. The disturbance response type with the highest SDM accuracy was obligate‐seeding shrubs with ballistic dispersal that regenerate via fire‐cued germination from a dormant seed bank. Conclusions: The effect of species characteristics on predictability of species distributions overrides any differences in modeling technique. Prediction accuracy may be related to how a suite of species characteristics co‐varies along environmental gradients. Including disturbance response was important because SDMs predict the realized niche. Classification of plant species into disturbance response types may provide a strong framework for evaluating performance of SDMs.     

Effect of species rarity on the accuracy of species distribution models for reptiles and amphibians in southern California

Aim Several studies have found that more accurate predictive models of species’ occurrences can be developed for rarer species; however, one recent study found the relationship between range size and model performance to be an artefact of sample prevalence, that is, the proportion of presence versus absence observations in the data used to train the model. We examined the effect of model type, species rarity class, species’ survey frequency, detectability and manipulated sample prevalence on the accuracy of distribution models developed for 30 reptile and amphibian species. Location Coastal southern California, USA. Methods Classification trees, generalized additive models and generalized linear models were developed using species presence and absence data from 420 locations. Model performance was measured using sensitivity, specificity and the area under the curve (AUC) of the receiver‐operating characteristic (ROC) plot based on twofold cross‐validation, or on bootstrapping. Predictors included climate, terrain, soil and vegetation variables. Species were assigned to rarity classes by experts. The data were sampled to generate subsets with varying ratios of presences and absences to test for the effect of sample prevalence. Join count statistics were used to characterize spatial dependence in the prediction errors. Results Species in classes with higher rarity were more accurately predicted than common species, and this effect was independent of sample prevalence. Although positive spatial autocorrelation remained in the prediction errors, it was weaker than was observed in the species occurrence data. The differences in accuracy among model types were slight. Main conclusions Using a variety of modelling methods, more accurate species distribution models were developed for rarer than for more common species. This was presumably because it is difficult to discriminate suitable from unsuitable habitat for habitat generalists, and not as an artefact of the effect of sample prevalence on model estimation.     

A Statistical Comparison between Less and Common Applied Models to Estimate Geographical Distribution of Endangered Species (<Emphasis Type="Italic">Felis margarita</Emphasis>) in Central Iran

Species distribution in space is important in habitat conservation and biodiversity protection, so gaining knowledge about species range would be worthwhile to rescue endangered species and plan conservation policy. This study evaluates and compares the performance of an array of Species Distribution Models (SDMs), namely RF, SVM, MaxEnt, GLMNET, and MARS, in predicting rare sand cat distribution across a large unprotected sand dune area in central Iran. Due to absence of reliable data and difficulties in recording the species itself, the SDMs were challenged by limited data including 55 absence (background) and 40 presence points as well as nine climatic and geological parameters that influence on species distribution, including humidity, maximum, minimum and mean temperature, precipitation, amount of sunshine, ground water level, aspect, and DEM. Moreover, each model was replicated 20 times and the statistics including TSS, AUC, COR and Deviance were computed. Then, based on computed statistics, the model performances were evaluated by TUKEY and ANOVA. Finally, ensemble map was obtained by weighted approach using AUC. The results of this study showed that complex machine learning methods, like SVM, RF, and MaxEnt are more outperformed to predict the distribution of rare species.     

Ensemble modelling of species distribution: the effects of geographical and environmental ranges

The aim of this study was to analyse the effects of species geographical and environmental ranges on the predictive performances of species distribution models (SDMs). We explored the usefulness of ensemble modelling approaches and tested whether species attributes influenced the outcomes of such approaches. Eight SDMs were used to model the current distribution of 35 fish species at 1110 stream sections in France. We first quantified the consensus among the resulting set of predictions for each fish species. Next, we created an average model by taking the average of the individual model predictions and tested whether the average model improved the predictive performances of single SDMs. Lastly, we described the ranges of fish species along four gradients: latitudinal, thermal, stream gradient (i.e. upstream‐downstream) and elevation. After accounting for the effects of phylogenetic relatedness and species prevalence, these four species attributes were related to the observed variations in both consensus among SDMs and predictive performances by using generalized estimation equations. Our results highlight the usefulness of ensemble approaches for identifying geographical areas of agreement among predictions. Although the geographical extent of species had no effect on the performances of SDMs, we demonstrated that more consensual and accurate predictions were obtained for species with low thermal and elevation ranges, validating the hypothesis that specialist species yield models with higher accuracy than generalist ones. We emphasized that significant improvements in the accuracy of SDMs can be achieved by using an average model. Furthermore, these improvements were higher for species with smaller ranges along the four gradients studied. The geographical extent and ranges of species along environmental gradients provide promising insights into our understanding of uncertainties in species distribution modelling.     

Impacts of imperfect reference data on the apparent accuracy of species presence–absence models and their predictions

Aim To explore the impacts of imperfect reference data on the accuracy of species distribution model predictions. The main focus is on impacts of the quality of reference data (labelling accuracy) and, to a lesser degree, data quantity (sample size) on species presence–absence modelling. Innovation The paper challenges the common assumption that some popular measures of model accuracy and model predictions are prevalence independent. It highlights how imperfect reference data may impact on a study and the actions that may be taken to address problems. Main conclusions The theoretical independence of prevalence of popular accuracy measures, such as sensitivity, specificity, true skills statistics (TSS) and area under the receiver operating characteristic curve (AUC), is unlikely to occur in practice due to reference data error; all of these measures of accuracy, together with estimates of species occurrence, showed prevalence dependency arising through the use of a non‐gold‐standard reference. The number of cases used also had implications for the ability of a study to meet its objectives. Means to reduce the negative effects of imperfect reference data in study design and interpretation are suggested.     

Creating individual accessible area hypotheses improves stacked species distribution model performance

Aim

Stacked species distribution models (SDMs) are an important step towards estimating species richness, but frequently overpredict this metric and therefore erroneously predict which species comprise a given community. We test the idea that developing hypotheses about accessible area a priori can greatly improve model performance. By integrating dispersal ability via accessible area into SDM creation, we address an often‐overlooked facet of ecological niche modelling.

Innovation

By limiting the training and transference areas to theoretically accessible areas, we are creating more accurate SDMs on the basis of a taxon's explorable environments. This limitation of space and environment is a more accurate reflection of a taxon's true dispersal properties and more accurately reflects the geographical and environmental space to which a taxon is exposed. Here, we compare the predictive performance of stacked SDMs derived from spatially constrained and unconstrained training areas.

Main conclusions

Restricting a species’ training and transference areas to a theoretically accessible area greatly improves model performance. Stacked SDMs drawn from spatially restricted training areas predicted species richness and community composition more accurately than non‐restricted stacked SDMs. These accessible area‐based restrictions mimic true dispersal barriers to species and limit training areas to the suite of environments to those which a species is exposed to in nature. Furthermore, these restrictions serve to ‘clip’ predictions in geographical space, thus removing overpredictions in adjacent geographical regions where the species is known to be absent.     

Performance tradeoffs in target‐group bias correction for species distribution models

Species distribution models (SDMs) are often calibrated using presence‐only datasets plagued with environmental sampling bias, which leads to a decrease of model accuracy. In order to compensate for this bias, it has been suggested that background data (or pseudoabsences) should represent the area that has been sampled. However, spatially‐explicit knowledge of sampling effort is rarely available. In multi‐species studies, sampling effort has been inferred following the target‐group (TG) approach, where aggregated occurrence of TG species informs the selection of background data. However, little is known about the species‐ specific response to this type of bias correction. The present study aims at evaluating the impacts of sampling bias and bias correction on SDM performance. To this end, we designed a realistic system of sampling bias and virtual species based on 92 terrestrial mammal species occurring in the Mediterranean basin. We manipulated presence and background data selection to calibrate four SDM types. Unbiased (unbiased presence data) and biased (biased presence data) SDMs were calibrated using randomly distributed background data. We used real and TG‐estimated sampling efforts in background selection to correct for sampling bias in presence data. Overall, environmental sampling bias had a deleterious effect on SDM performance. In addition, bias correction improved model accuracy, and especially when based on spatially‐explicit knowledge of sampling effort. However, our results highlight important species‐specific variations in susceptibility to sampling bias, which were largely explained by range size: widely‐distributed species were most vulnerable to sampling bias and bias correction was even detrimental for narrow‐ranging species. Furthermore, spatial discrepancies in SDM predictions suggest that bias correction effectively replaces an underestimation bias with an overestimation bias, particularly in areas of low sampling intensity. Thus, our results call for a better estimation of sampling effort in multispecies system, and cautions the uninformed and automatic application of TG bias correction.     

Effects of grain size and niche breadth on species distribution modeling

Scale is a vital component to consider in ecological research, and spatial resolution or grain size is one of its key facets. Species distribution models (SDMs) are prime examples of ecological research in which grain size is an important component. Despite this, SDMs rarely explicitly examine the effects of varying the grain size of the predictors for species with different niche breadths. To investigate the effect of grain size and niche breadth on SDMs, we simulated four virtual species with different grain sizes/niche breadths using three environmental predictors (elevation, aspect, and percent forest) across two real landscapes of differing heterogeneity in predictor values. We aggregated these predictors to seven different grain sizes and modeled the distribution of each of our simulated species using MaxEnt and GLM techniques at each grain size. We examined model accuracy using the AUC statistic, Pearson's correlations of predicted suitability with the true suitability, and the binary area of presence determined from suitability above the maximum true skill statistic (TSS) threshold. Habitat specialists were more accurately modeled than generalist species, and the models constructed at the grain size from which a species was derived generally performed the best. The accuracy of models in the homogenous landscape deteriorated with increasing grain size to a greater degree than models in the heterogenous landscape. Variable effects on the model varied with grain size, with elevation increasing in importance as grain size increased while aspect lost importance. The area of predicted presence was drastically affected by grain size, with larger grain sizes over predicting this value by up to a factor of 14. Our results have implications for species distribution modeling and conservation planning, and we suggest more studies include analysis of grain size as part of their protocol.     

Profile or group discriminative techniques? Generating reliable species distribution models using pseudo‐absences and target‐group absences from natural history collections

Aim The presence‐only data stored in natural history collections is the most important source of information available regarding the distribution of organisms. These data and profile techniques can be used to generate species distribution models (SDMs), but pseudo‐absences must be generated to use group discriminative techniques. In this study, we evaluated whether the SDMs generated with pseudo‐absences are reliable and also if there are differences in the results obtained with profile and group discriminative techniques. Location Ecuador, South America. Methods The SDMs were generated with a training data set for each of the five species of Anthurium and six different methods: two profile techniques (BIOCLIM and Gower’s distance index), three group discriminative techniques [logistic multiple regression (LMR), multivariate adaptative regression splines (MARS) and Maxent ] and a mixed modelling approach genetic algorithm for rule‐set production (GARP), which employs a combination of profile and group discriminative techniques and generates its own pseudo‐absences. For LMR, MARS and Maxent , three types of absences were generated: (1) random pseudo‐absences in equal number to presences and excluding a buffer area around presences (except for Maxent , which assumes that this background sample includes presences), (2) a large number (10,000) of random pseudo‐absences, also excluding a buffer area around each presence and (3) ‘target‐group absences’ (TGA), consisting of sites where other species of the group have been collected by the specialist, but not the species being modelled. To compare the predictive performance of the SDMs, the area under the curve statistic was calculated using an independent testing data set for each species. Results MARS, Maxent and LMR produce better results than the profile techniques. The models created with TGA are generally more accurate than those generated with pseudo‐absences. Main conclusions The advantages and disadvantages of different options for using pseudo‐absences and TGA with profile and group discriminative modelling techniques are explained and recommendations are made for the future.     

A simple method to estimate the probable distribution of species

Species distribution models (SDMs) are broadly used to predict species distributions from available presence data. However, SDMs results have been criticized for several reasons mainly related to two basic characteristics of most SDMs: 1) general lack of reliable species absence information, 2) the frequent use of an arbitrary geographical extent (GE) or accessible area of the species. These impediments have motivated us to generate a procedure called niche of occurrence (NOO). NOO provides the probable distribution of species (realized niche) relying solely on partial information about presence of species. It operates within a natural geographical extent delimited by available observations and avoids using misleading thresholds to obtain binary presence–absence estimations when the species prevalence is unknown. In this study the main characteristics of NOO are presented, comparing its performance with other recognized and more complex SDMs by using virtual species to avoid the omnipresent error sources of real data sets.     

Delimiting the geographical background in species distribution modelling

Aim The extent of the study area (geographical background, GB) can strongly affect the results of species distribution models (SDMs), but as yet we lack objective and practicable criteria for delimiting the appropriate GB. We propose an approach to this problem using trend surface analysis (TSA) and provide an assessment of the effects of varying GB extent on the performance of SDMs for four species. Location Mainland Spain. Methods Using data for four well known wild ungulate species and different GBs delimited with a TSA, we assessed the effects of GB extent on the predictive performance of SDMs: specifically on model calibration (Miller’s statistic) and discrimination (area under the curve of the receiver operating characteristic plot, AUC; sensitivity and specificity), and on the tendency of the models to predict environmental potential when they are projected beyond their training area. Results In the training area, discrimination significantly increased and calibration decreased as the GB was enlarged. In contrast, as GB was enlarged, both discriminatory power and calibration decreased when assessed in the core area of the species distributions. When models trained using small GBs were projected beyond their training area, they showed a tendency to predict higher environmental potential for the species than those models trained using large GBs. Main conclusions By restricting GB extent using a geographical criterion, model performance in the core area of the species distribution can be significantly improved. Large GBs make models demonstrate high discriminatory power but are barely informative. By delimiting GB using a geographical criterion, the effect of historical events on model parameterization may be reduced. Thus purely environmental models are obtained that, when projected onto a new scenario, depict the potential distribution of the species. We therefore recommend the use of TSA in geographically delimiting the GB for use in SDMs.     

Causal inference and large-scale expert validation shed light on the drivers of SDM accuracy and variance

Aim

To develop a causal understanding of the drivers of Species distribution model (SDM) performance.

Location

United Kingdom (UK).

Methods

We measured the accuracy and variance of SDMs fitted for 518 species of invertebrate and plant in the UK. Our measure of variance reflects variation among replicate model fits, and taxon experts assessed model accuracy. Using directed acyclic graphs, we developed a causal model depicting plausible effects of explanatory variables (e.g. species' prevalence, sample size) on SDM accuracy and variance and quantified those effects using a multilevel piecewise path model.

Results

According to our model, sample size and niche completeness (proportion of a species' niche covered by sampling) directly affect SDM accuracy and variance. Prevalence and range completeness have indirect effects mediated by sample size. Challenging conventional wisdom, we found that the effect of prevalence on SDM accuracy is positive. This reflects the facts that sample size has a positive effect on accuracy and larger sample sizes are possible for widespread species. It is possible, however, that the omission of an unobserved confounder biased this effect. Previous studies, which reported negative correlations between prevalence and SDM accuracy, conditioned on sample size.

Main conclusions

Our model explicates the causal basis of previously reported correlations between SDM performance and species/data characteristics. It also suggests that niche completeness has similarly large effects on SDM accuracy and variance as sample size. Analysts should consider niche completeness, or proxies thereof, in addition to sample size when deciding whether modelling is worthwhile.     

Predicting species distributions based on incomplete survey data: the trade‐off between precision and scale

Systematic species surveys over large areas are mostly not affordable, constraining conservation planners to make best use of incomplete data. Spatially explicit species distribution models (SDM) may be useful to detect and compensate for incomplete information. SDMs can either be based on standardized, systematic sampling in a restricted subarea, or – as a cost‐effective alternative – on data haphazardly collated by “volunteer‐based monitoring schemes” (VMS), area‐wide but inherently biased and of heterogeneous spatial precision. Using data on capercaillie Tetrao urogallus, we evaluated the capacity of SDMs generated from incomplete survey data to localise unknown areas inhabited by the species and to predict relative local observation density. Addressing the trade‐off between data precision, sample size and spatial extent of the sampling area, we compared three different sampling strategies: VMS‐data collected throughout the whole study area (7000 km²) using either 1) exact locations or 2) locations aggregated to grid cells of the size of an average individual home range, and 3) systematic transect counts conducted within a small subarea (23.8 km²). For each strategy, we compared two sample sizes and two modelling methods (ENFA and Maxent), which were evaluated using cross‐validation and independent data. Models based on VMS‐data (strategies 1 and 2) performed equally well in predicting relative observation density and in localizing “unknown” occurrences. They always outperformed strategy 3‐models, irrespective of sample size and modelling method, partly because the VMS‐data provided the more comprehensive clues for setting the discrimination‐threshold for predicting presence or absence. Accounting for potential errors due to extrapolation (e.g. projections outside the environmental domain or potentially biasing variables) reduced, but did not fully compensate for the observed discrepancies. As they cover a broader range of species‐habitat relations, the area‐wide data achieved a better model quality with less a‐priori knowledge. Furthermore, in a highly mobile species like capercaillie a sampling resolution corresponding to an individuals' home range can lead to equally good predictions as the use of exact locations. Consequently, when a trade‐off between the sampling effort and the spatial extent of the sampling area is necessary, less precise data unsystematically collected over a large representative region are preferable to systematically sampled data from a restricted region.     

Spatial conservation priorities are highly sensitive to choice of biodiversity surrogates and species distribution model type

Pressure to conserve biodiversity with limited resources has led to increasing use of species distribution models (SDMs) for spatial conservation prioritization. Published spatial prioritization exercises often focus on well‐studied groups, with data compiled from on‐line databases of ad‐hoc collections. Conservation plans generally aim to protect all components of biodiversity, and it is implied that the species used in prioritization act as surrogates. Here, we assess the sensitivity of spatial priorities to model and surrogate choice using a case study from a fragmented agricultural area of south eastern Australia that is poorly represented in the national reserve system. We model the distributions of 30 species of bird, microbat and bee using two types of SDM; generalised linear models based on systematic surveys that yield presence and absence observations, and MaxEnt models based on biodiversity database records. Eight prioritization scenarios were tested using Zonation software, and were based on either the presence–background or presence–absence SDMs and combinations of surrogacy among the three taxa. We found low correlations between SDMs generated for the same species using different modelling frameworks (μ = 0.18, n = 26). Area under the receiver operating characteristic curve (AUC) estimates generated by MaxEnt were optimistic; on average 1.36 times higher than when tested against the systematic survey data. Conservation priorities were sensitive to the choice of surrogate and type of data used to fit SDMs, and though bats and birds formed moderately good surrogates for each other, there was less compelling evidence of surrogacy for bees. Because valid surrogacy is unlikely with most existing data sets, investment in high quality data for less‐surveyed groups prior to planning should still be a priority. If this is not possible, then it is advisable to analyse the sensitivity of conservation plans to the assumed surrogacy and quality of data available.