Performance Of Species Distribution Models Research Articles

Boosting biodiversity monitoring using smartphone-driven, rapidly accumulating community-sourced data.

Comprehensive biodiversity data is crucial for ecosystem protection. The Biome mobile app, launched in Japan, efficiently gathers species observations from the public using species identification algorithms and gamification elements. The app has amassed >6 million observations since 2019. Nonetheless, community-sourced data may exhibit spatial and taxonomic biases. Species distribution models (SDMs) estimate species distribution while accommodating such bias. Here, we investigated the quality of Biome data and its impact on SDM performance. Species identification accuracy exceeds 95% for birds, reptiles, mammals, and amphibians, but seed plants, molluscs, and fishes scored below 90%. Our SDMs for 132 terrestrial plants and animals across Japan revealed that incorporating Biome data into traditional survey data improved accuracy. For endangered species, traditional survey data required >2000 records for accurate models (Boyce index ≥ 0.9), while blending the two data sources reduced this to around 300. The uniform coverage of urban-natural gradients by Biome data, compared to traditional data biased towards natural areas, may explain this improvement. Combining multiple data sources better estimates species distributions, aiding in protected area designation and ecosystem service assessment. Establishing a platform for accumulating community-sourced distribution data will contribute to conserving and monitoring natural ecosystems.

Unravelling the impact of soil data quality on species distribution models of temperate forest woody plants

Soil properties influence plant physiology and growth, playing a fundamental role in shaping species niches in temperate forest ecosystems. Here, we investigated the impact of soil data quality on the performance of species distribution models (SDMs) of 41 woody plant species in Swiss forests. We compared models based on measured soil properties with those based on digitally mapped soil properties on regional (Swiss Forest Soil Maps) and global scales (SoilGrids). We first calibrated topo-climatic SDMs with measured soil data and plant species presences and absences from mature temperate forest stand plots. We developed further models using the same soil predictors, but with values extracted from digital soil maps at the nearest neighbouring plots of the Swiss National Forestry Inventory. The predictive power of SDMs without soil information compared to those with soil information, as well as measured soil information vs digitally mapped, was evaluated with metrics of model performance and variable contribution. On average, models with measured and digitally mapped soil properties performed significantly better than those without soil information. SDMs based on measured and Swiss Forest Soil Maps showed higher performance, especially for species with an ‘extreme’ niche position (e.g., preference for high or low pH), compared to those using SoilGrids. Nevertheless, if no regional soil maps are available, SoilGrids should be tested for their potential to improve SDMs. Moreover, among the tested soil predictors, pH, and clay content of the topsoil layers most improved the predictive power of SDMs for forest woody plants. In conclusion, we demonstrate the value of regional soil maps for predicting the distribution of woody species across strong environmental gradients in temperate forests. The improved accuracy of SDMs and insights into drivers of distribution may support forest managers in strategies supporting e.g. biodiversity conservation, or climate adaptation planning.

Adaptive sampling by citizen scientists improves species distribution model performance: A simulation study

Abstract Volunteer recorders generate large amounts of biodiversity data through citizen science which is used in conservation planning and policy decision‐making. Unstructured sampling, where the volunteer can record what they want, where they want, leads to spatial unevenness in these data. While there are many statistical techniques to account for the resulting biases, it may be possible to improve datasets by directing a subset of recorders to sample in the most informative locations, known as adaptive sampling. We investigated the potential for adaptive sampling to improve the performance of species distribution models built on citizen science data using simulated ecological communities. We simulated ecological assemblages across Great Britain based on current butterfly data and modelled the distributions of each species. We then simulated the sampling of new data based on five adaptive sampling methods (one empirical method based simply on gap‐filling, and four model‐based methods using various measures from the model outputs) and one non‐adaptive method (a method in which recording continued in the current pattern), and re‐ran the species distribution models. In these, we also varied the rate of recording effort that was distributed according to adaptive sampling. The model predictions using the original and adaptively sampled data were compared to true species distributions to evaluate the performance of each method. We found that all adaptive sampling approaches improved model performance, with greatest improvement for model‐based approaches compared to the empirical sampling method (i.e. simple gap‐filling). All four model‐based adaptive sampling approaches provided similar benefits for model outputs. Improvements in model performance were greatest when the amount of adaptive sampling changed from no uptake to 1% uptake, indicating that only a small amount of change in recorder behaviour is needed to improve model performance. Directing volunteer recorders to places where records are most needed, based on information from model outputs, can improve species distribution models built on citizen science data, even with minimal uptake of suggested locations. Our results therefore suggest that adaptive sampling by recorders could be beneficial for real‐world citizen science datasets.

Underprediction of extirpation and colonisation following climate and land‐use change using species distribution models

AbstractAimTo evaluate the performance of species distribution models in predicting observed colonisations, persistences and extirpations in response to changes in climate and land use over a multi‐decadal period.LocationSweden.MethodsWe use historical (early 20th century) land use and climate data to build species distribution models for 84 plant species across three provinces of Sweden. Model performance was then evaluated internally using a subset of the historical data for cross‐validation, as well as by using the models to project occurrences to the modern day and validating them with observed occurrences from 1990 to 2020. We then analysed predicted and observed occurrences in the modern period in terms of persistence, extirpation (local extinction) and colonisation in relation to species' habitat and climate associations.ResultsWe found overall high agreement between evaluation methods, although internal evaluation gave consistently higher values for model performance (using true skill statistic, TSS). Overall, extirpations were worst predicted, with on average fewer than one‐third of each species' extirpations being foreseen by the models. Colonisations were better predicted, while persistences were relatively well‐predicted. Predictive accuracy of colonisations was higher for species with relatively warmer temperature associations (climate‐driven expansion), while extirpations were better predicted in cool‐related species (retractions at cool edges). Colonisations of forest‐associated species were more common than predicted (underpredicted), despite widespread patterns of afforestation. Assessing grid‐cell level turnover, we found that in grid cells that experienced the largest changes in terms of climate and land use, predicted extirpations were less likely to have happened.Main ConclusionsWe found that commonly applied modelling approaches have limited ability to predict observed changes in species occurrences, especially extirpations. This suggests that we should take predictions of future biodiversity loss very seriously. However, the ability for species to (at least temporarily) persist in unsuitable conditions could be an opportunity for biodiversity conservation.

The influence of scale-dependent geodiversity on species distribution models in a biodiversity hotspot.

Improving models of species' distributions is essential for conservation, especially in light of global change. Species distribution models (SDMs) often rely on mean environmental conditions, yet species distributions are also a function of environmental heterogeneity and filtering acting at multiple spatial scales. Geodiversity, which we define as the variation of abiotic features and processes of Earth's entire geosphere (inclusive of climate), has potential to improve SDMs and conservation assessments, as they capture multiple abiotic dimensions of species niches, however they have not been sufficiently tested in SDMs. We tested a range of geodiversity variables computed at varying scales using climate and elevation data. We compared predictive performance of MaxEnt SDMs generated using CHELSA bioclimatic variables to those also including geodiversity variables for 31 mammalian species in Colombia. Results show the spatial grain of geodiversity variables affects SDM performance. Some variables consistently exhibited an increasing or decreasing trend in variable importance with spatial grain, showing slight scale-dependence and indicating that some geodiversity variables are more relevant at particular scales for some species. Incorporating geodiversity variables into SDMs, and doing so at the appropriate spatial scales, enhances the ability to model species-environment relationships, thereby contributing to the conservation and management of biodiversity. This article is part of the Theo Murphy meeting issue 'Geodiversity for science and society'.

The influence of the number and distribution of background points in presence-background species distribution models

Species distribution models (SDMs), which relate recorded observations (presences) and absences or background points to environmental characteristics, are powerful tools used to generate hypotheses about the biogeography, ecology, and conservation of species. Although many researchers have examined the effects of presence and background point distributions on model outputs, they have not systematically evaluated the effects of various methods of background point sampling on the performance of a single model algorithm across many species. Therefore, a consensus on the preferred methods of background point sampling is lacking. Here, we conducted presence-background SDMs for 20 vertebrate species in North America under a variety of background point conditions, varying the number of background points used, the size of the buffer used to constrain the background points around the occurrences, and the percentage of background points sampled within the buffer (“spatial weighting”). We evaluated the accuracy and transferability of the models using Boyce index, overlap with expert-generated range maps, and area overpredicted and underpredicted by the SDM (and AUC for comparability with other studies).SDM performance is highly dependent on the species modelled but is affected by the number and spread of background points. Models with little spatial weighting had high accuracy (overlap values), but extreme extrapolation errors and overprediction. In contrast, SDMs with high transferability (high Boyce index values and low overprediction) had moderate-to-high spatial weighting. These results emphasize the importance of both background points and evaluation metric selection in SDMs. For other, more successful metrics, using many background points with spatial weighting may be preferred for models with large extents. These results can assist researchers in selecting the background point parameters most relevant for their research question, allowing them to fine-tune their hypotheses on the distribution of species through space and time.

Niche shifts undermine the prediction performance of species distribution models: estimating potentially suitable areas for Myriophyllum aquaticum at the global scale

Invasive alien species (IAS) can undergo niche shifts upon establishment in novel environments that differ from their native ranges. The validity of species distribution models (SDMs) depends on the underlying assumption of niche conservatism that IAS can only successfully establish populations in environments that closely resemble their native regions. This has led us to examine how ecologists can use SDMs to predict potentially suitable areas for invasive alien species in the context of niche shifts. Here, we compared the niches of Myriophyllum aquaticum in its native and invasive ranges to quantify the shifts that occurred during its global invasion using the centroid shift, overlap, unfilling, and expansion (COUE) framework. We also aimed to investigate the effects of niche shift on the prediction performance of SDMs. The SDMs were calibrated using data from three different sources: the native range, the invasive range, and a combination of both. The study examined how these different data sources influenced the accuracy of the SDM predictions. The findings indicated the niche expansion between the invasive and native ranges. The invasive (based on the occurrence data of invasive ranges) and global (based on the occurrence data of global ranges) models underestimated the potentially suitable areas (PSAs) of M. aquaticum in the Southern Hemisphere. The native model (based on the occurrence data of native ranges) underestimated the PSAs of M. aquaticum in the Northern Hemisphere. We combined the results of the invasive and native range models to estimate the global PSAs of M. aquaticum, addressing the predictive limitations of a single model in a complementary form. The PSAs of M. aquaticum were predominantly distributed in eastern Asia, western Europe, southern North America, southern South America, and southern Oceania and expanded into higher latitudes under future climate scenarios. Our study has applied a simple yet effective method for predicting the global PSAs of M. aquaticum given niche shifts, which confirmed the observations of the previous studies and provided a theoretical basis for the early warning, prevention and control of M. aquaticum.

Thinning occurrence points does not improve species distribution model performance

AbstractSpatial biases are an intrinsic feature of occurrence data used in species distribution models (SDMs). Thinning species occurrences, where records close in the geographic or environmental space are removed from the modeling procedure, is an approach often used to address these biases. However, thinning occurrence data can also negatively affect SDM performance, given that the benefits of removing spatial biases might be outweighed by the detrimental effects of data loss caused by this approach. We used real and virtual species to evaluate how spatial and environmental thinning affected different performance metrics of four SDM methods. The occurrence data of virtual species were sampled randomly, evenly spaced, and clustered in the geographic space to simulate different types of spatial biases, and several spatial and environmental thinning distances were used to thin the occurrence data. Null datasets were also generated for each thinning distance where we randomly removed the same number of occurrences by a thinning distance and compared the results of the thinned and null datasets. We found that spatially or environmentally thinned occurrence data is no better than randomly removing them, given that thinned datasets performed similarly to null datasets. Specifically, spatial and environmental thinning led to a general decrease in model performances across all SDM methods. These results were observed for real and virtual species, were positively associated with thinning distance, and were consistent across the different types of spatial biases. Our results suggest that thinning occurrence data usually fails to improve SDM performance and that the use of thinning approaches when modeling species distributions should be considered carefully.

Flood susceptibility mapping to improve models of species distributions

As significant ecosystem disturbances flooding events are expected to increase in both frequency and severity due to climate change, underscoring the critical need to understand their impact on biodiversity. In this study, we employ advanced remote sensing and machine learning methodologies to investigate the effects of flooding on biodiversity, from individual species to broader ecological communities. Specifically, we utilized Sentinel-1 synthetic aperture radar (SAR) images and an ensemble of machine-learning algorithms to derive a flood susceptibility indicator. Our primary objective is to investigate the potential benefits of incorporating flood susceptibility, as a proxy for flood risk, into species distribution models (SDMs). By doing so, we aim to improve the performance of SDMs and gain deeper insights into the consequences of floods to biodiversity. Within the biodiverse landscape of the Zagros Mountains, a crucial Irano-Anatolian biodiversity hotspots, we examined the sensitivity of mammals, amphibians, and reptiles’ distributions to flooding. Our analysis compared the performance of models that combined flood susceptibility with climate variables against models relying solely on climate variables. The results indicate that the inclusion of flood susceptibility significantly improves the capacity of models to explain and map species distributions for 67% of the species in our study region. Notably, amphibians and mammals are more profoundly affected by flooding compared to reptiles. The study highlights the importance of incorporating flood susceptibility as a predictor variable in species distribution models to improve the baseline characterization of potential species distributions. The importance of this variable will obviously depend on the regional context and the species studied but its relevance is likely to increase with climate change. In summary, our research demonstrates the integration of remote sensing and machine learning as a potent approach to advance biodiversity data science, monitoring, and conservation in the face of climate-induced flooding.

The dimensionality reductions of environmental variables have a significant effect on the performance of species distribution models.

How to effectively obtain species-related low-dimensional data from massive environmental variables has become an urgent problem for species distribution models (SDMs). In this study, we will explore whether dimensionality reduction on environmental variables can improve the predictive performance of SDMs. We first used two linear (i.e., principal component analysis (PCA) and independent components analysis) and two nonlinear (i.e., kernel principal component analysis (KPCA) and uniform manifold approximation and projection) dimensionality reduction techniques (DRTs) to reduce the dimensionality of high-dimensional environmental data. Then, we established five SDMs based on the environmental variables of dimensionality reduction for 23 real plant species and nine virtual species, and compared the predictive performance of those with the SDMs based on the selected environmental variables through Pearson's correlation coefficient (PCC). In addition, we studied the effects of DRTs, model complexity, and sample size on the predictive performance of SDMs. The predictive performance of SDMs under DRTs other than KPCA is better than using PCC. And the predictive performance of SDMs using linear DRTs is better than using nonlinear DRTs. In addition, using DRTs to deal with environmental variables has no less impact on the predictive performance of SDMs than model complexity and sample size. When the model complexity is at the complex level, PCA can improve the predictive performance of SDMs the most by 2.55% compared with PCC. At the middle level of sample size, the PCA improved the predictive performance of SDMs by 2.68% compared with the PCC. Our study demonstrates that DRTs have a significant effect on the predictive performance of SDMs. Specifically, linear DRTs, especially PCA, are more effective at improving model predictive performance under relatively complex model complexity or large sample sizes.

A Benford’s law-based framework to determine the threshold of occurrence sites for species distribution modelling from ecological monitoring databases

The use of data-based species distribution models (SDMs) has increased significantly in recent years. However, studies of determining the minimum requirements of occurrence sites from ecological monitoring datasets used in species distribution modelling remain insufficient. Therefore, this study proposed a framework to determine the threshold of minimum occurrence sites for SDMs by assessing compliance with Benford’s law. The compliance test verified that the national-scale freshwater fish monitoring dataset was natural and reliable. Results derived from true skill statistics (TSS) determined the minimum number of occurrence sites for reliable species distribution modelling was 20 with a TSS value of 0.793 and an overall accuracy of 0.804. The Benford compliance test has shown to be a useful tool for swift and efficient evaluation of the reliability of species occurrence datasets, or the determination of the threshold of occurrence sites before species distribution modelling. Further studies regarding the evaluation of this method’s transferability to other species and validation using SDM performance are required. Overall, the framework proposed in this study demonstrates that Benford compliance test applied to species monitoring datasets can be used to derive a universal and model-independent minimum occurrence threshold for SDMs.

Modelling the effects of topographic heterogeneity on distribution of Nitraria tangutorum Bobr. species in deserts using LiDAR-data

Microclimate ecology is attracting renewed attention because of its fundamental importance in understanding how organisms respond to climate change. Many hot issues can be investigated in desert ecosystems, including the relationship between species distribution and environmental gradients (e.g., elevation, slope, topographic convergence index, and solar insolation). Species Distribution Models (SDMs) can be used to understand these relationships. We used data acquired from the important desert plant Nitraria tangutorum Bobr. communities and desert topographic factors extracted from LiDAR (Light Detection and Ranging) data of one square kilometer in the inner Mongolia region of China to develop SDMs. We evaluated the performance of SDMs developed with a variety of both the parametric and nonparametric algorithms (Bioclimatic Modelling (BIOCLIM), Domain, Mahalanobi, Generalized Linear Model, Generalized Additive Model, Random Forest (RF), and Support Vector Machine). The area under the receiver operating characteristic curve was used to evaluate these algorithms. The SDMs developed with RF showed the best performance based on the area under curve (0.7733). We also produced the Nitraria tangutorum Bobr. distribution maps with the best SDM and suitable habitat area of the Domain model. Based on the suitability map, we conclude that Nitraria tangutorum Bobr. is more suited to southern part with 0–20 degree slopes at an elevation of approximately 1010 m. This is the first attempt of modelling the effects of topographic heterogeneity on the desert species distribution on a small scale. The presented SDMs can have important applications for predicting species distribution and will be useful for preparing conservation and management strategies for desert ecosystems on a small scale.

Interannual climate variability improves niche estimates for ectothermic but not endothermic species

Climate is an important limiting factor of species’ niches and it is therefore regularly included in ecological applications such as species distribution models (SDMs). Climate predictors are often used in the form of long-term mean values, yet many species experience wide climatic variation over their lifespan and within their geographical range which is unlikely captured by long-term means. Further, depending on their physiology, distinct groups of species cope with climate variability differently. Ectothermic species, which are directly dependent on the thermal environment are expected to show a different response to temporal or spatial variability in temperature than endothermic groups that can decouple their internal temperature from that of their surroundings. Here, we explore the degree to which spatial variability and long-term temporal variability in temperature and precipitation change niche estimates for ectothermic (730 amphibian, 1276 reptile), and endothermic (1961 mammal) species globally. We use three different species distribution modelling (SDM) algorithms to quantify the effect of spatial and temporal climate variability, based on global range maps of all species and climate data from 1979 to 2013. All SDMs were cross-validated and accessed for their performance using the Area under the Curve (AUC) and the True Skill Statistic (TSS). The mean performance of SDMs using only climatic means as predictors was TSS = 0.71 and AUC = 0.90. The inclusion of spatial variability offers a significant gain in SDM performance (mean TSS = 0.74, mean AUC = 0.92), as does the inclusion of temporal variability (mean TSS = 0.80, mean AUC = 0.94). Including both spatial and temporal variability in SDMs shows the highest scores in AUC and TSS. Accounting for temporal rather than spatial variability in climate improved the SDM prediction especially in ectotherm groups such as amphibians and reptiles, while for endothermic mammals no such improvement was observed. These results indicate that including long term climate interannual climate variability into niche estimations matters most for ectothermic species that cannot decouple their physiology from the surrounding environment as endothermic species can.

Spatial resolution impacts projected plant responses to climate change on topographically complex islands

AbstractAimUnderstanding how grain size affects our ability to characterize species responses to ongoing climate change is of crucial importance in the context of an increasing awareness for the substantial difference that exists between coarse spatial resolution macroclimatic data sets and the microclimate actually experienced by organisms. Climate change impacts on biodiversity are expected to peak in mountain areas, wherein the differences between macro and microclimates are precisely the largest. Based on a newly generated fine‐scale environmental data for the Canary Islands, we assessed whether data at 100 m resolution is able to provide more accurate predictions than available data at 1 km resolution. We also analysed how future climate suitability predictions of island endemic bryophytes differ depending on the grain size of grids.LocationCanary Islands.Time periodPresent (1979–2013) and late‐century (2071–2100).TaxaBryophytes.MethodsWe compared the accuracy and spatial predictions using ensemble of small models for 14 Macaronesian endemic bryophyte species. We used two climate data sets: CHELSA v1.2 (~1 km) and CanaryClim v1.0 (100 m), a downscaled version of the latter utilizing data from local weather stations. CanaryClim also encompasses future climate data from five individual model intercomparison projects for three warming shared socio‐economic pathways.ResultsSpecies distribution models generated from CHELSA and CanaryClim exhibited a similar accuracy, but CanaryClim predicted buffered warming trends in mid‐elevation ridges. CanaryClim consistently returned higher proportions of newly suitable pixels (8%–28%) than CHELSA models (0%–3%). Consequently, the proportion of species predicted to occupy pixels of uncertain suitability was higher with CHELSA (3–8 species) than with CanaryClim (0–2 species).Main conclusionsThe resolution of climate data impacted the predictions rather than the performance of species distribution models. Our results highlight the crucial role that fine‐resolution climate data sets can play in predicting the potential distribution of both microrefugia and new suitable range under warming climate.

Performance of inhomogeneous Poisson point process models under different scenarios of uncertainty in species presence-only data

Haphazard and opportunistic species occurrence (PO) data are widely used in species distribution models (SDMs) instead of high-quality species data gathered using appropriate and structured sampling methods, which is expensive and often spatially limited. Despite their widespread use in ecology, PO data are prone to errors and uncertainties, such as imperfect detectability, positional imprecision, and spatial niche truncation, which make their use analytically challenging for effective and adaptive biodiversity management and conservation. Using simulated data, this study investigates the effects of these uncertainties on the performance of spatial point process based presence-only and integrated SDMs. We investigated three SDMs in this study, one that ignores imperfect detectability: the presence-only model (PO model), and two that account for it: the thinned presence-only model (THINPO model) and the integrated model (PBPC model). The ability of these SDMs to produce accurate maximum likelihood estimates of intensity model coefficients and reliable predictions of species distributions under different data quality scenarios was investigated. The results show that SDMs that account for imperfect detectability (THINPO or PBPC models) are not applicable in situations of high detectability. In this situation, the PO model produces the most accurate maximum likelihood estimates of the models’ coefficients ({hat{beta }}_k), and consequently the most accurate predictions of species distributions ({hat{lambda }}(s)). The effects of positional uncertainty and spatial niche truncation on this SDM output are minimal. However, in situations of low detectability, it is preferable to use the PBPC model. Positional uncertainty and spatial niche truncation have negligible effects on the output of this SDM, except when positionally uncertain PO data are analyzed along with truncated PC data. These minimal effects of spatial niche truncation on SDM outputs demonstrate the transferability of SDMs. However, the effects of all these uncertainties may depend on the characteristics of the species. Prior to modeling species distributions, a multivariate environmental similarity surface analysis should be performed to test the similarity between data from the restricted region to be used for model calibration and data from the entire range. If this analysis reveals dissimilarities, larger spatial and ecological scales should be considered to address the issue of spatial niche truncation. Further efforts could address the effects of species characteristics on SDMs performance and assess the effects of species-specific uncertainties.

Correcting for the effects of class imbalance improves the performance of machine-learning based species distribution models

Numerous methods have been developed to combat the unwanted effects of imbalanced training data on the performance of machine-learning based predictive models. These methods attempt to balance model sensitivity and specificity. However, the effects of specific imbalance-correction methods on the performance of different machine-learning algorithms are not well understood for ecological data. In this study, we used four machine-learning algorithms (random forest, artificial neural network, gradient boosting, support vector machine) and five imbalance-correction methods (base algorithm = no correction, cutoff, up-sampling, down-sampling, weighting) to produce species distribution models for 15 freshwater macroinvertebrate genera that varied from 2.5 to 29.0% in prevalence. All imbalance-correction methods substantially improved average model performance (true skill statistic) over the base machine-learning algorithms, except when up-sampling was applied to random forest models. Choice of machine-learning algorithm had little effect on model performance, although gradient boosting performed better than other algorithms on the most imbalanced datasets. Our results suggest that the performance of species distribution models built with presence/absence data can generally be improved by correcting for imbalanced data.

Building use-inspired species distribution models: Using multiple data types to examine and improve model performance.

Species distribution models (SDMs) are becoming an important tool for marine conservation and management. Yet while there is an increasing diversity and volume of marine biodiversity data for training SDMs, little practical guidance is available on how to leverage distinct data types to build robust models. We explored the effect of different data types on the fit, performance and predictive ability of SDMs by comparing models trained with four data types for a heavily exploited pelagic fish, the blue shark (Prionace glauca), in the Northwest Atlantic: two fishery dependent (conventional mark-recapture tags, fisheries observer records) and two fishery independent (satellite-linked electronic tags, pop-up archival tags). We found that all four data types can result in robust models, but differences among spatial predictions highlighted the need to consider ecological realism in model selection and interpretation regardless of data type. Differences among models were primarily attributed to biases in how each data type, and the associated representation of absences, sampled the environment and summarized the resulting species distributions. Outputs from model ensembles and a model trained on all pooled data both proved effective for combining inferences across data types and provided more ecologically realistic predictions than individual models. Our results provide valuable guidance for practitioners developing SDMs. With increasing access to diverse data sources, future work should further develop truly integrative modeling approaches that can explicitly leverage the strengths of individual data types while statistically accounting for limitations, such as sampling biases.

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

AbstractAimTo develop a causal understanding of the drivers of Species distribution model (SDM) performance.LocationUnited Kingdom (UK).MethodsWe 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.ResultsAccording 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 conclusionsOur 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.

Data stochasticity and model parametrisation impact the performance of species distribution models: insights from a simulation study

Species distribution models (SDM) are widely used to describe and explain how species relate to their environment and predict their spatial distributions. As such, they are the cornerstone of most of spatial planning efforts worldwide. SDM can be implemented with a wide array of data types (presence-only, presence-absence, count...), which can either be point- or areal-based, and use a wide array of environmental conditions as predictor variables. The choice of the sampling type as well as the resolution of environmental conditions to be used are recognized as of crucial importance, yet we lack any quantification of the effects these decisions may have on SDM reliability. In the present work, we fill this gap with an unprecedented simulation procedure. We simulated 100 possible distributions of two different virtual species in two different regions. Species distribution were modelled using either segment- or areal-based sampling and five different spatial resolutions of environmental conditions. The SDM performances were inspected by statistical metrics, model composition, shapes of relationships and prediction quality. We provided clear evidence of stochasticity in the modelling process (particularly in the shapes of relationships): two dataset from the same survey, species and region could yield different results. Sampling type had stronger effects than spatial resolution on the final model relevance. The effect of coarsening the resolution was directly related to the resistance of the spatial features to changes of scale: SDM failed to adequately identify spatial distributions when the spatial features targeted by the species were diluted by resolution coarsening. These results have important implications for the SDM community, backing up some commonly accepted choices, but also by highlighting some up-to-now unexpected features of SDM (stochasticity). As a whole, this work calls for carefully weighted decisions in implementing models, and for caution in interpreting results.

Species Distribution Models: the delicate balance between signal and noise

A recommendation of: Charlotte Lambert, Auriane Virgili Data stochasticity and model parametrisation impact the performance of species distribution models: insights from a simulation study https://doi.org/10.1101/2023.01.17.524386