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.     

Assessment of spatial habitat heterogeneity by coupling data-driven habitat suitability models with a 2D hydrodynamic model in small-scale streams

Habitat assessment considering habitat quality and quantity is a key approach in conservation and restoration works for biodiversity and ecosystems. In this regard, application of hydrodynamic model for modeling instream habitat conditions and machine learning (ML) methods for modeling habitat suitability of a target species can contribute to better modeling practices in ecohydraulics. Despite the importance of small streams for aquatic ecosystems, previous studies in ecohydraulics have been conducted mainly in medium to large rivers, often disregarding small-scale streams such as agricultural canals. The aim of this study is to demonstrate the applicability of a coupled use of ML and a two-dimensional (2D) hydrodynamic model for assessing spatial habitat heterogeneity in small-scale agricultural canals in Japan. Using abundance data of Japanese medaka (Oryzias latipes), four ML methods, namely artificial neural networks (ANNs), classification and regression trees (CARTs), random forests (RF) and support vector machines (SVMs), were applied to develop habitat suitability models considering water depth and flow velocity. A 2D hydrodynamic model was developed based on field surveys in two types of agricultural canals, namely earthen and concrete-lined canals. Information entropy was used for assessing the spatial heterogeneity of instream habitat conditions. As a result, the hydrodynamic models could model instream habitat conditions in a reasonable accuracy. Despite the differences in accuracies in habitat modeling, the four ML methods illustrated similar habitat suitability information for Japanese medaka. The coupled ecohydraulics modeling approach could quantify habitat quality and its spatial heterogeneity, based on which the differences between the earthen and concrete-lined canals were quantitatively assessed. This study demonstrated the applicability of ML-based habitat suitability evaluation and a 2D hydrodynamic model for modeling the spatial distribution of habitat suitability and assessing its spatial heterogeneity. Further study, assessing the spatial heterogeneity in various types of flows including natural/artificial and small/large streams, can contribute to establish quantitative criteria for an ecologically sound habitat and improved ecofriendly construction works in small-scale rivers and streams.     

How to deal with ground truthing affected by human‐induced habitat change?: Identifying high‐quality habitats for the Critically Endangered Red Siskin

Species distribution models (SDM) can be valuable for identifying key habitats for conservation management of threatened taxa, but anthropogenic habitat change can undermine SDM accuracy. We used data for the Red Siskin (Spinus cucullatus), a critically endangered bird and ground truthing to examine anthropogenic habitat change as a source of SDM inaccuracy. We aimed to estimate: (1) the Red Siskin's historic distribution in Venezuela; (2) the portion of this historic distribution lost to vegetation degradation; and (3) the location of key habitats or areas with both, a high probability of historic occurrence and a low probability of vegetation degradation. We ground‐truthed 191 locations and used expert opinion as well as landscape characteristics to classify species' habitat suitability as excellent, good, acceptable, or poor. We fit a Random Forest model (RF) and Enhanced Vegetation Index (EVI) time series to evaluate the accuracy and precision of the expert categorization of habitat suitability. We estimated the probability of historic occurrence by fitting a MaxLike model using 88 presence records (1960–2013) and data on forest cover and aridity index. Of the entire study area, 23% (20,696 km²) had a historic probability of Red Siskin occurrence over 0.743. Furthermore, 85% of ground‐truthed locations had substantial reductions in mean EVI, resulting in key habitats totaling just 976 km², in small blocks in the western and central regions. Decline in Area of Occupancy over 15 years was between 40% and 95%, corresponding to an extinction risk category between Vulnerable and Critically Endangered. Relating key habitats with other landscape features revealed significant risks and opportunities for proposed conservation interventions, including the fact that ongoing vegetation degradation could limit the establishment of reintroduced populations in eastern areas, while the conservation of remaining key habitats on private lands could be improved with biodiversity‐friendly agri‐ and silviculture programs.     

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.     

Intraspecific variation buffers projected climate change impacts on Pinus contorta

Species distribution modeling (SDM) is an important tool to assess the impact of global environmental change. Many species exhibit ecologically relevant intraspecific variation, and few studies have analyzed its relevance for SDM. Here, we compared three SDM techniques for the highly variable species Pinus contorta. First, applying a conventional SDM approach, we used MaxEnt to model the subject as a single species (species model), based on presence–absence observations. Second, we used MaxEnt to model each of the three most prevalent subspecies independently and combined their projected distributions (subspecies model). Finally, we used a universal growth transfer function (UTF), an approach to incorporate intraspecific variation utilizing provenance trial tree growth data. Different model approaches performed similarly when predicting current distributions. MaxEnt model discrimination was greater (AUC – species model: 0.94, subspecies model: 0.95, UTF: 0.89), but the UTF was better calibrated (slope and bias – species model: 1.31 and −0.58, subspecies model: 1.44 and −0.43, UTF: 1.01 and 0.04, respectively). Contrastingly, for future climatic conditions, projections of lodgepole pine habitat suitability diverged. In particular, when the species'' intraspecific variability was acknowledged, the species was projected to better tolerate climatic change as related to suitable habitat without migration (subspecies model: 26% habitat loss or UTF: 24% habitat loss vs. species model: 60% habitat loss), and given unlimited migration may increase amount of suitable habitat (subspecies model: 8% habitat gain or UTF: 12% habitat gain vs. species model: 51% habitat loss) in the climatic period 2070–2100 (SRES A2 scenario, HADCM3). We conclude that models derived from within-species data produce different and better projections, and coincide with ecological theory. Furthermore, we conclude that intraspecific variation may buffer against adverse effects of climate change. A key future research challenge lies in assessing the extent to which species can utilize intraspecific variation under rapid environmental change.     

Habitat suitability modelling for species at risk is sensitive to algorithm and scale: A case study of Blanding's turtle,Emydoidea blandingii,in Ontario,Canada

Species distribution modelling (SDM) can help conservation by providing information on the ecological requirements of species at risk. We developed habitat suitability models at multiple spatial scales for a threatened freshwater turtle, Emydoidea blandingii, in Ontario as a case study. We also explored the effect of background data selection and modelling algorithm selection on habitat suitability predictions. We used sighting records, high-resolution land cover data (25 m), and two SDM techniques: boosted regression trees; and maximum entropy modelling. The area under the receiver characteristic operating curve (AUC) for habitat suitability models tested on independent data ranged from 0.878 to 0.912 when using random background and from 0.727 to 0.741 with target-group background. E. blandingii habitat suitability was best predicted by air temperature, wetland area, open water area, road density, and cropland area. Habitat suitability increased with increasing air temperature and wetland area, and decreased with increasing cropland area. Low road density and open water increased habitat suitability, while high levels of either variable decreased habitat suitability. Robust habitat suitability maps for species at risk require using a multi-scale and multi-algorithm approach. If well used, SDM can offer insight on the habitat requirements of species at risk and help guide the development of management plans. Our results suggest that E. blandingii management plans should promote the protection of terrestrial habitat surrounding residential wetlands, halt the building of roads within and adjacent to currently occupied habitat, and identify movement corridors for isolated populations.     

Species distribution models as a tool to estimate reproductive parameters: a case study with a passerine bird species

1.?Correlative species distribution models (SDMs) assess relationships between species distribution data and environmental features, to evaluate the environmental suitability (ES) of a given area for a species, by providing a measure of the probability of presence. If the output of SDMs represents the relationships between habitat features and species performance well, SDM results can be related also to other key parameters of populations, including reproductive parameters. To test this hypothesis, we evaluated whether SDM results can be used as a proxy of reproductive parameters (breeding output, territory size) in red-backed shrikes (Lanius collurio). 2.?The distribution of 726 shrike territories in Northern Italy was obtained through multiple focused surveys; for a subset of pairs, we also measured territory area and number of fledged juveniles. We used Maximum Entropy modelling to build a SDM on the basis of territory distribution. We used generalized least squares and spatial generalized mixed models to relate territory size and number of fledged juveniles to SDM suitability, while controlling for spatial autocorrelation. 3.?Species distribution models predicted shrike distribution very well. Territory size was negatively related to suitability estimated through SDM, while the number of fledglings significantly increased with the suitability of the territory. This was true also when SDM was built using only spatially and temporally independent data. 4.?Results show a clear relationship between ES estimated through presence-only SDMs and two key parameters related to species' reproduction, suggesting that suitability estimated by SDM, and habitat quality determining reproduction parameters in our model system, are correlated. Our study shows the potential use of SDMs to infer important fitness parameters; this information can have great importance in management and conservation.     

Assessing the applicability of fuzzy neural networks for habitat preference evaluation of Japanese medaka (Oryzias latipes)

Species–environment relationships are key information for the development of planning and management strategies for conservation or restoration of ecosystems. Artificial neural networks (ANNs) are one widely applied type of species distribution model (SDM). Fuzzy neural networks (FNNs), that is, fuzzified ANNs, have been introduced to take into account the uncertainties inherent in fish behaviour and errors in input data. Despite their high predictive ability in modelling complex systems, FNNs cannot describe habitat preference curves (HPCs), although these are the basis for habitat quality assessment. The present study therefore aimed to evaluate the applicability of FNNs for modelling habitat preference and spatial distributions of Japanese medaka (Oryzias latipes), one of the most common freshwater fish in Japan. Three independent data sets were collected during a series of field surveys and used for model development and evaluation of FNNs. A weight decay backpropagation algorithm was additionally introduced, and its effects on the FNNs were evaluated on the basis of model performance and habitat preference information retrieved from the field observation data. Modified sensitivity analysis was applied to derive HPCs of the target fish. Application of weight decay backpropagation markedly reduced the variability of the model structures, improved the generalization ability of the FNNs, and resulted in well-converged and consistent HPCs that were similar to those evaluated by fuzzy habitat preference models. These results support the applicability of FNNs to habitat preference modelling, which can provide useful information on the habitat use by the target fish. Further study should focus on the effects of sources of uncertainty, such as zero abundance, on the SDMs and the resulting habitat preference evaluation.     

Testing the utility of species distribution modelling using Random Forests for a species in decline

Habitat suitability estimates derived from species distribution models (SDMs) are increasingly used to guide management of threatened species. Poorly estimating species’ ranges can lead to underestimation of threatened status, undervaluing of remaining habitat and misdirection of conservation funding. We aimed to evaluate the utility of a SDM, similar to the models used to inform government regulation of habitat in our study region, in estimating the contemporary distribution of a threatened and declining species. We developed a presence‐only SDM for the endangered New Holland Mouse (Pseudomys novaehollandiae) across Victoria, Australia. We conducted extensive camera trap surveys across model‐predicted and expert‐selected areas to generate an independent data set for use in evaluating the model, determining confidence in absence data from non‐detection sites with occupancy and detectability modelling. We assessed the predictive capacity of the model at thresholds based on (1) sum of sensitivity and specificity (SSS), and (2) the lowest presence threshold (LPT; i.e. the lowest non‐zero model‐predicted habitat suitability value at which we detected the species). We detected P. novaehollandiae at 40 of 472 surveyed sites, with strong support for the species’ probable absence from non‐detection sites. Based on our post hoc optimised SSS threshold of the SDM, 25% of our detection sites were falsely predicted as non‐suitable habitat and 75% of sites predicted as suitable habitat did not contain the species at the time of our survey. One occupied site had a model‐predicted suitability value of zero, and at the LPT, 88% of sites predicted as suitable habitat did not contain the species at the time of our survey. Our findings demonstrate that application of generic SDMs in both regulatory and investment contexts should be tempered by considering their limitations and currency. Further, we recommend engaging species experts in the extrapolation and application of SDM outputs.     

Terrestrial or marine species distribution model: Why not both? A case study with seabirds

Species reliant on both the terrestrial and marine realms present a challenge for conventional species distribution models (SDMs). For such species, standard single‐realm SDMs may omit key information that could result in decreased model accuracy and performance. Existing approaches to habitat suitability modeling typically do not effectively combine information from multiple realms; this methodological gap can ultimately hamper management efforts for groups such as seabirds, seals, and turtles. This study, for the first time, jointly incorporates both terrestrial information and marine information into a single species distribution model framework. We do this by sampling nearby marine conditions for a given terrestrial point and vice versa using parameters set by each species’ mean maximum foraging distance and then use standard SDM methods to generate habitat suitability predictions; therefore, our method does not rely on post hoc combination of several different models. Using three seabird species with very different ecologies, we investigate whether this new multi‐realm approach can improve our ability to identify suitable habitats for these species. Results show that incorporating terrestrial information into marine SDMs, or vice versa, generally improves model performance, sometimes drastically. However, there is considerable variability between species in the level of improvement as well as in the particular method that produces the most improvement. Our approach provides a repeatable and transparent method to combine information from multiple ecological realms in a single SDM framework. Important advantages over existing solutions include the opportunity to, firstly, easily combine terrestrial and marine information for species that forage large distances inland or out to sea and, secondly, consider interactions between terrestrial and marine variables.     

Not seeing the forest for the trees: Generalised linear model out-performs random forest in species distribution modelling for Southeast Asian felids

Species Distribution Models (SDMs) are a powerful tool to derive habitat suitability predictions relating species occurrence data with habitat features. Two of the most frequently applied algorithms to model species-habitat relationships are Generalised Linear Models (GLM) and Random Forest (RF). The former is a parametric regression model providing functional models with direct interpretability. The latter is a machine learning non-parametric algorithm, more tolerant than other approaches in its assumptions, which has often been shown to outperform parametric algorithms. Other approaches have been developed to produce robust SDMs, like training data bootstrapping and spatial scale optimisation. Using felid presence-absence data from three study regions in Southeast Asia (mainland, Borneo and Sumatra), we tested the performances of SDMs by implementing four modelling frameworks: GLM and RF with bootstrapped and non-bootstrapped training data. With Mantel and ANOVA tests we explored how the four combinations of algorithms and bootstrapping influenced SDMs and their predictive performances. Additionally, we tested how scale-optimisation responded to species' size, taxonomic associations (species and genus), study area and algorithm. We found that choice of algorithm had strong effect in determining the differences between SDMs' spatial predictions, while bootstrapping had no effect. Additionally, algorithm followed by study area and species, were the main factors driving differences in the spatial scales identified. SDMs trained with GLM showed higher predictive performance, however, ANOVA tests revealed that algorithm had significant effect only in explaining the variance observed in sensitivity and specificity and, when interacting with bootstrapping, in Percent Correctly Classified (PCC). Bootstrapping significantly explained the variance in specificity, PCC and True Skills Statistics (TSS). Our results suggest that there are systematic differences in the scales identified and in the predictions produced by GLM vs. RF, but that neither approach was consistently better than the other. The divergent predictions and inconsistent predictive abilities suggest that analysts should not assume machine learning is inherently superior and should test multiple methods. Our results have strong implications for SDM development, revealing the inconsistencies introduced by the choice of algorithm on scale optimisation, with GLM selecting broader scales than RF.     

The role of demography,intra‐species variation,and species distribution models in species' projections under climate change

Organisms are projected to shift their distribution ranges under climate change. The typical way to assess range shifts is by species distribution models (SDMs), which predict species’ responses to climate based solely on projected climatic suitability. However, life history traits can impact species’ responses to shifting habitat suitability. Additionally, it remains unclear if differences in vital rates across populations within a species can offset or exacerbate the effects of predicted changes in climatic suitability on population viability. In order to obtain a fuller understanding of the response of one species to projected climatic changes, we coupled demographic processes with predicted changes in suitable habitat for the monocarpic thistle Carlina vulgaris across northern Europe. We first developed a life history model with species‐specific average fecundity and survival rates and linked it to a SDM that predicted changes in habitat suitability through time with changes in climatic variables. We then varied the demographic parameters based upon observed vital rates of local populations from a translocation experiment. Despite the fact that the SDM alone predicted C. vulgaris to be a climate ‘winner’ overall, coupling the model with changes in demography and small‐scale habitat suitability resulted in a matrix of stable, declining, and increasing patches. For populations predicted to experience declines or increases in abundance due to changes in habitat suitability, altered fecundity and survival rates can reverse projected population trends.     

Species Distribution Modelling predicts habitat suitability and reduction of suitable habitat under future climatic scenario for Sclerophrys perreti: A critically endangered Nigerian endemic toad

Sclerophrys perreti is a critically endangered Nigerian native frog currently imperilled by human activities. A better understanding of its potential distribution and habitat suitability will aid in conservation; however, such knowledge is limited for S. perreti. Herein, we used a species distribution model (SDM) approach with all known occurrence data (n = 22) from our field surveys and primary literature, and environmental variable predictors (19 bioclimatic variables, elevation and land cover) to elucidate habitat suitability and impact of climate change on this species. The SDM showed that temperature and precipitation were the predictors of habitat suitability for S. perreti with precipitation seasonality as the strongest predictor of habitat suitability. The following variable also had a significant effect on habitat suitability: temperature seasonality, temperature annual range, precipitation of driest month, mean temperature of wettest quarter and isothermality. The model predicted current suitable habitat for S. perreti covering an area of 1,115 km². However, this habitat is predicted to experience 60% reduction by 2050 owing to changes in temperature and precipitation. SDM also showed that suitable habitat exists in south-eastern range of the inselberg with predicted low impact of climate change compared to other ranges. Therefore, this study recommends improved conservation measures through collaborations and stakeholder's meeting with local farmers for the management and protection of S. perreti.     

Revisiting probabilistic neural networks: a comparative study with support vector machines and the microhabitat suitability for the Eastern Iberian chub (Squalius valentinus)

Probabilistic Neural Networks (PNNs) and Support Vector Machines (SVMs) are flexible classification techniques suited to render trustworthy species distribution and habitat suitability models. Although several alternatives to improve PNNs' reliability and performance and/or to reduce computational costs exist, PNNs are currently not well recognised as SVMs because the SVMs were compared with standard PNNs. To rule out this idea, the microhabitat suitability for the Eastern Iberian chub (Squalius valentinus Doadrio & Carmona, 2006) was modelled with SVMs and four types of PNNs (homoscedastic, heteroscedastic, cluster and enhanced PNNs); all of them optimised with Differential Evolution. The fitness function and several performance criteria (correctly classified instances, true skill statistic, specificity and sensitivity) and partial dependence plots were used to assess respectively the performance and reliability of each habitat suitability model. Heteroscedastic and enhanced PNNs achieved the highest performance in every index but specificity. However, these two PNNs rendered ecologically unreliable partial dependence plots. Conversely, homoscedastic and cluster PNNs rendered ecologically reliable partial dependence plots. Thus, Eastern Iberian chub proved to be a eurytopic species, presenting the highest suitability in microhabitats with cover present, low flow velocity (approx. 0.3 m/s), intermediate depth (approx. 0.6 m) and fine gravel (64–256 mm). PNNs outperformed SVMs; thus, based on the results of the cluster PNN, which also showed high values of the performance criteria, we would advocate a combination of approaches (e.g., cluster & heteroscedastic or cluster & enhanced PNNs) to balance the trade-off between accuracy and reliability of habitat suitability models.     

Dealing with non-equilibrium bias and survey effort in presence-only invasive Species Distribution Models (iSDM); predicting the range of muntjac deer in Britain and Ireland

Invasive species managers utilise species records to inform management. These data can also be used in Species Distribution Models (SDM) to predict future spread or potential invasion of new areas. However, issues with non-equilibrium (also called disequilibrium) can cause difficulties in modelling invasive species that have not fully colonised their potential distribution and, in addition, sampling bias can result from a lack of information on survey effort, a particular issue for presence only modelling techniques. Geographical confounds are unavoidable when building iSDMs but there are methods that allow prediction to be optimised. We used maximum entropy (Maxent) to model suitable habitat for invasive Reeve's muntjac deer (Muntiacus reevesi) throughout Great Britain and Ireland comparing several methods that aimed to address invasive Species Distribution Modelling (iSDM) bias including spatial filtering, weighted background points and targeted background points built at varying spatial extents. Model evaluation metrics suggested that the model, which explicitly failed to account for non-equilibrium at the full extent of Great Britain and Ireland using random background points, predicted the species' current invasive range best. This highlighted that negative environmental relationships are likely to represent uncolonised areas rather than habitat selection and thus, low predicted suitability of uncolonised areas was misleading. Of the models that dealt with non-equilibrium conceptually best, by restricting the training extent to their current invasive range or core range, and utilised targeted background points accounting for survey effort (cells with other deer species recorded as present yet with no records for muntjac) as the best model evaluation metric, yielded relatively poor predictive performance. This implied limited habitat selectivity or avoidance within the colonised range which, when spatially extrapolated, suggested virtually all regions in Great Britain and Ireland may be vulnerable to future muntjac invasion.     

Prospective sampling based on model ensembles improves the detection of rare species

Identifying the geographic distribution of populations is a basic, yet crucial step in many fundamental and applied ecological projects, as it provides key information on which many subsequent analyses depend. However, this task is often costly and time consuming, especially where rare species are concerned and where most sampling designs generally prove inefficient. At the same time, rare species are those for which distribution data are most needed for their conservation to be effective. To enhance fieldwork sampling, model‐based sampling (MBS) uses predictions from species distribution models: when looking for the species in areas of high habitat suitability, chances should be higher to find them. We thoroughly tested the efficiency of MBS by conducting an important survey in the Swiss Alps, assessing the detection rate of three rare and five common plant species. For each species, habitat suitability maps were produced following an ensemble modeling framework combining two spatial resolutions and two modeling techniques. We tested the efficiency of MBS and the accuracy of our models by sampling 240 sites in the field (30 sites×8 species). Across all species, the MBS approach proved to be effective. In particular, the MBS design strictly led to the discovery of six sites of presence of one rare plant, increasing chances to find this species from 0 to 50%. For common species, MBS doubled the new population discovery rates as compared to random sampling. Habitat suitability maps coming from the combination of four individual modeling methods predicted well the species' distribution and more accurately than the individual models. As a conclusion, using MBS for fieldwork could efficiently help in increasing our knowledge of rare species distribution. More generally, we recommend using habitat suitability models to support conservation plans.     

Refugee species: which historic baseline should inform conservation planning?

Understanding species' historical ranges can provide important information for conservation planning in the face of environmental change. Cromsigt et al. (this issue) comment on our recent European bison (Bison bonasus) range reconstruction, suggesting that bison were already 8000 years ago a refugee species (i.e. restricted to marginal habitat due to past human pressure) and that species distribution models (SDM) are generally of limited use for refugee species conservation. While we welcome this discussion, we find no evidence for the claim that human pressure prior to 8000 BP determined where bison occurred. More importantly, as human pressure is generally high and increasing, attempts to restore species across their former range may fail where the factors that relegated species into refugee status are still at play or where their optimal habitat has vanished. Identifying areas where human pressure is low and where refugee species have persisted over the last millennia is crucial, and SDM based on historical data are important for doing so. Refugee species suffer from the shifting baseline syndrome, but careful reality checks are needed and all available data should be considered before determining the baseline that should inform conservation planning.     

Niche variability and its consequences for species distribution modeling

When species distribution models (SDMs) are used to predict how a species will respond to environmental change, an important assumption is that the environmental niche of the species is conserved over evolutionary time-scales. Empirical studies conducted at ecological time-scales, however, demonstrate that the niche of some species can vary in response to environmental change. We use habitat and locality data of five species of stream fishes collected across seasons to examine the effects of niche variability on the accuracy of projections from Maxent, a popular SDM. We then compare these predictions to those from an alternate method of creating SDM projections in which a transformation of the environmental data to similar scales is applied. The niche of each species varied to some degree in response to seasonal variation in environmental variables, with most species shifting habitat use in response to changes in canopy cover or flow rate. SDMs constructed from the original environmental data accurately predicted the occurrences of one species across all seasons and a subset of seasons for two other species. A similar result was found for SDMs constructed from the transformed environmental data. However, the transformed SDMs produced better models in ten of the 14 total SDMs, as judged by ratios of mean probability values at known presences to mean probability values at all other locations. Niche variability should be an important consideration when using SDMs to predict future distributions of species because of its prevalence among natural populations. The framework we present here may potentially improve these predictions by accounting for such variability.     

Can species distribution modelling provide estimates of population densities? A case study with jaguars in the Neotropics

Aim To test the prediction that environmental suitability derived from species distribution modelling (SDM) could be a surrogate for jaguar local population density estimates. Location Americas. Methods We used 1409 occurrence records of jaguars to model the distribution of the species using 11 SDM methods. We tested whether models’ suitability is linearly correlated with jaguar population densities estimated from 37 different locations. We evaluated whether the relationship between density and suitability forms a constraint envelope, in which higher densities are found mainly in regions with high suitability, whereas low densities can occur in regions with variable suitability. We tested this using heteroscedasticity test and quantile regressions. Results A positive linear relationship between suitability and jaguar density was found only for four methods [bioclimatic envelope (BIOCLIM), genetic algorithm for rule set production (GARP), maximum entropy (Maxent) and generalized boosting models (GBM)], but with weak explanatory power. BIOCLIM showed the strongest relationship. Variance of suitability for lower densities values was larger than for higher values for many of the SDM models used, but the quantile regression was significantly positive only for BIOCLIM and random forests (RF). RF and GBM provided the most accurate models when measured with the standard SDM evaluation metrics, but possess poor relationship with local density estimates. Main conclusions Results indicate that the relationship between density and suitability could be better described as a triangular constraint envelope than by a straight positive relationship, and some of the SDM methods tested here were able to discriminate regions with high or low local population densities. Low jaguar densities can occur in areas with low or high suitability, whereas high values are restricted to areas where the suitability is greater. In high suitability areas but with low jaguar density estimates, we discuss how extrinsic factors driving abundance could act at local scales and then prevent higher densities that would be expected by the favourable regional environmental conditions.