Are comparisons of species distribution models biased? Are they biologically meaningful?

A major problem in ecology is to understand how environmental requirements change over space and time. To this end, numerous authors have attempted to use comparisons of species’ distributions as a surrogate for comparisons of environmental requirements. Unfortunately, it is currently unclear when comparisons of species’ distributions produce reliable inferences about changes in environmental requirements. To address this problem, I develop an analytic model that identifies the conditions under which a comparison of species’ distribution models can serve as surrogate for a comparison of environmental requirements. This work demonstrates that 1) comparisons of species’ distributions typically produce biased comparisons of environmental requirements, 2) assuming distribution models are fit appropriately, it is possible to compare environmental requirements of distinct taxa, 3) there are multiple biologically relevant questions we can address using comparisons of distribution models, with each question corresponding to a distinct measure of the difference between distribution models. By developing an analytic model for comparisons of species’ distributions this work helps to clarify and remedy poorly understood sources of error associated with existing methods.     

Effects of biotic interactions on modeled species' distribution can be masked by environmental gradients

A fundamental goal of ecology is to understand the determinants of species' distributions (i.e., the set of locations where a species is present). Competition among species (i.e., interactions among species that harms each of the species involved) is common in nature and it would be tremendously useful to quantify its effects on species' distributions. An approach to studying the large‐scale effects of competition or other biotic interactions is to fit species' distributions models (SDMs) and assess the effect of competitors on the distribution and abundance of the species of interest. It is often difficult to validate the accuracy of this approach with available data. Here, we simulate virtual species that experience competition. In these simulated datasets, we can unambiguously identify the effects that competition has on a species' distribution. We then fit SDMs to the simulated datasets and test whether we can use the outputs of the SDMs to infer the true effect of competition in each simulated dataset. In our simulations, the abiotic environment influenced the effects of competition. Thus, our SDMs often inferred that the abiotic environment was a strong predictor of species abundance, even when the species' distribution was strongly affected by competition. The severity of this problem depended on whether the competitor excluded the focal species from highly suitable sites or marginally suitable sites. Our results highlight how correlations between biotic interactions and the abiotic environment make it difficult to infer the effects of competition using SDMs.     

No‐analog climates and shifting realized niches during the late quaternary: implications for 21st‐century predictions by species distribution models

Empirically derived species distributions models (SDMs) are increasingly relied upon to forecast species vulnerabilities to future climate change. However, many of the assumptions of SDMs may be violated when they are used to project species distributions across significant climate change events. In particular, SDM's in theory assume stable fundamental niches, but in practice, they assume stable realized niches. The assumption of a fixed realized niche relative to climate variables remains unlikely for various reasons, particularly if novel future climates open up currently unavailable portions of species’ fundamental niches. To demonstrate this effect, we compare the climate distributions for fossil‐pollen data from 21 to 15 ka bp (relying on paleoclimate simulations) when communities and climates with no modern analog were common across North America to observed modern pollen assemblages. We test how well SDMs are able to project 20th century pollen‐based taxon distributions with models calibrated using data from 21 to 15 ka. We find that taxa which were abundant in areas with no‐analog late glacial climates, such as Fraxinus, Ostrya/Carpinus and Ulmus, substantially shifted their realized niches from the late glacial period to present. SDMs for these taxa had low predictive accuracy when projected to modern climates despite demonstrating high predictive accuracy for late glacial pollen distributions. For other taxa, e.g. Quercus, Picea, Pinus strobus, had relatively stable realized niches and models for these taxa tended to have higher predictive accuracy when projected to present. Our findings reinforce the point that a realized niche at any one time often represents only a subset of the climate conditions in which a taxon can persist. Projections from SDMs into future climate conditions that are based solely on contemporary realized distributions are potentially misleading for assessing the vulnerability of species to future climate change.     

Used‐habitat calibration plots: a new procedure for validating species distribution,resource selection,and step‐selection models

‘Species distribution modeling’ was recently ranked as one of the top five ‘research fronts’ in ecology and the environmental sciences by ISI's Essential Science Indicators, reflecting the importance of predicting how species distributions will respond to anthropogenic change. Unfortunately, species distribution models (SDMs) often perform poorly when applied to novel environments. Compounding on this problem is the shortage of methods for evaluating SDMs (hence, we may be getting our predictions wrong and not even know it). Traditional methods for validating SDMs quantify a model's ability to classify locations as used or unused. Instead, we propose to focus on how well SDMs can predict the characteristics of used locations. This subtle shift in viewpoint leads to a more natural and informative evaluation and validation of models across the entire spectrum of SDMs. Through a series of examples, we show how simple graphical methods can help with three fundamental challenges of habitat modeling: identifying missing covariates, non‐linearity, and multicollinearity. Identifying habitat characteristics that are not well‐predicted by the model can provide insights into variables affecting the distribution of species, suggest appropriate model modifications, and ultimately improve the reliability and generality of conservation and management recommendations.     

Unraveling climate influences on the distribution of the parapatric newts <Emphasis Type="Italic">Lissotriton vulgaris meridionalis</Emphasis> and <Emphasis Type="Italic">L. italicus</Emphasis>

Background

Climate is often considered as a key ecological factor limiting the capability of expansion of most species and the extent of suitable habitats. In this contribution, we implement Species Distribution Models (SDMs) to study two parapatric amphibians, Lissotriton vulgaris meridionalis and L. italicus, investigating if and how climate has influenced their present and past (Last Glacial Maximum and Holocene) distributions. A database of 901 GPS presence records was generated for the two newts. SDMs were built through Boosted Regression Trees and Maxent, using the Worldclim bioclimatic variables as predictors.

Results

Precipitation-linked variables and the temperature annual range strongly influence the current occurrence patterns of the two Lissotriton species analyzed. The two newts show opposite responses to the most contributing variables, such as BIO7 (temperature annual range), BIO12 (annual precipitation), BIO17 (precipitation of the driest quarter) and BIO19 (precipitation of the coldest quarter). The hypothesis of climate influencing the distributions of these species is also supported by the fact that the co-occurrences within the sympatric area fall in localities characterized by intermediate values of these predictors. Projections to the Last Glacial Maximum and Holocene scenarios provided a coherent representation of climate influences on the past distributions of the target species. Computation of pairwise variables interactions and the discriminant analysis allowed a deeper interpretation of SDMs’ outputs. Further, we propose a multivariate environmental dissimilarity index (MEDI), derived through a transformation of the multivariate environmental similarity surface (MESS), to deal with extrapolation-linked uncertainties in model projections to past climate. Finally, the niche equivalency and niche similarity tests confirmed the link between SDMs outputs and actual differences in the ecological niches of the two species.

Conclusions

The different responses of the two species to climatic factors have significantly contributed to shape their current distribution, through contractions, expansions and shifts over time, allowing to maintain two wide allopatric areas with an area of sympatry in Central Italy. Moreover, our SDMs hindcasting shows many concordances with previous phylogeographic studies carried out on the same species, thus corroborating the scenarios of potential distribution during the Last Glacial Maximum and the Holocene emerging from the models obtained.

     

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.     

How Can Model Comparison Help Improving Species Distribution Models?

Today, more than ever, robust projections of potential species range shifts are needed to anticipate and mitigate the impacts of climate change on biodiversity and ecosystem services. Such projections are so far provided almost exclusively by correlative species distribution models (correlative SDMs). However, concerns regarding the reliability of their predictive power are growing and several authors call for the development of process-based SDMs. Still, each of these methods presents strengths and weakness which have to be estimated if they are to be reliably used by decision makers. In this study we compare projections of three different SDMs (STASH, LPJ and PHENOFIT) that lie in the continuum between correlative models and process-based models for the current distribution of three major European tree species, Fagus sylvatica L., Quercus robur L. and Pinus sylvestris L. We compare the consistency of the model simulations using an innovative comparison map profile method, integrating local and multi-scale comparisons. The three models simulate relatively accurately the current distribution of the three species. The process-based model performs almost as well as the correlative model, although parameters of the former are not fitted to the observed species distributions. According to our simulations, species range limits are triggered, at the European scale, by establishment and survival through processes primarily related to phenology and resistance to abiotic stress rather than to growth efficiency. The accuracy of projections of the hybrid and process-based model could however be improved by integrating a more realistic representation of the species resistance to water stress for instance, advocating for pursuing efforts to understand and formulate explicitly the impact of climatic conditions and variations on these processes.     

Niche shift can impair the ability to predict invasion risk in the marine realm: an illustration using Mediterranean fish invaders

Climatic niche conservatism, the tendency of species‐climate associations to remain unchanged across space and time, is pivotal for forecasting the spread of invasive species and biodiversity changes. Indeed, it represents one of the key assumptions underlying species distribution models (SDMs), the main tool currently available for predicting range shifts of species. However, to date, no comprehensive assessment of niche conservatism is available for the marine realm. We use the invasion by Indo‐Pacific tropical fishes into the Mediterranean Sea, the world's most invaded marine basin, to examine the conservatism of the climatic niche. We show that tropical invaders may spread far beyond their native niches and that SDMs do not predict their new distributions better than null models. Our results suggest that SDMs may underestimate the potential spread of invasive species and call for prudence in employing these models in order to forecast species invasion and their response to environmental change.     

The use of incidence-based species richness estimators,species accumulation curves and similarity measures to appraise ethnobotanical inventories from South Africa

The incorporation of suitable quantitative methods into ethnobotanical studies enhances the value of the research and the interpretation of the results. Prediction of sample species richness and the use of species accumulation functions have been addressed little in applied ethnobotany. In this paper, incidence-based species richness estimators, species accumulation curves and similarity measures are used to compare and predict species richness, evaluate sampling effort and compare the similarity of species inventories for ethnobotanical data sets derived from the trade in traditional medicine in Johannesburg and Mpumalanga, South Africa. EstimateS was used to compute estimators of species richness (e.g. Jackknife), rarefaction curves, species accumulation curves and complimentarity. Results showed that while the Michaelis–Menten Means estimator appeared to be the best estimator because the curve approached a horizontal asymptote, it was not able to accurately predict species richness for one of the data sets when two of its subsamples were individually tested. Instead, the first-order Jackknife estimator best approximated the known richness.     

On the use of climate covariates in aquatic species distribution models: are we at risk of throwing out the baby with the bath water?

Species distribution models (SDMs) in river ecosystems can incorporate climate information by using air temperature and precipitation as surrogate measures of instream conditions or by using independent models of water temperature and hydrology to link climate to instream habitat. The latter approach is preferable but constrained by the logistical burden of developing water temperature and hydrology models. We therefore assessed whether regional scale, freshwater SDM predictions are fundamentally different when climate data versus instream temperature and hydrology are used as covariates. Maximum entropy (MaxEnt) SDMs were built for 15 freshwater fishes using one of two covariate sets: 1) air temperature and precipitation (climate variables) in combination with physical habitat variables; or 2) water temperature, hydrology (instream variables) and physical habitat. Three procedures were then used to compare results from climate vs instream models. First, equivalence tests assessed average pairwise differences (site‐specific comparisons throughout each species’ range) among climate and instream models. Second, ‘congruence’ tests determined how often the same stream segments were assigned high habitat suitability by climate and instream models. Third, Schoener's D and Warren's I niche overlap statistics quantified range‐wide similarity in predicted habitat suitability from climate vs instream models. Equivalence tests revealed small, pairwise differences in habitat suitability between climate and instream models (mean pairwise differences in MaxEnt raw scores for all species < 3 × 10^–4). Congruence tests showed a strong tendency for climate and instream models to predict high habitat suitability at the same stream segments (median congruence = 68%). D and I statistics reflected a high margin of overlap among climate and instream models (median D = 0.78, median I = 0.96). Overall, we found little support for the hypothesis that SDM predictions are fundamentally different when climate versus instream covariates are used to model fish species’ distributions at the scale of the Columbia Basin.     

A nonparametric estimator of species overlap

For two communities, species overlap has been defined by Smith, Solow, and Preston (1996, Biometrics 52, 1472-1477) as the probability that a randomly selected species is present in both communities given that it is present in at least one community. Species overlap can thus be used to describe the similarity of two communities. In contrast with the parametric estimator of Smith et al., we propose a nonparametric maximum likelihood estimator (NPMLE). We prove that the NPMLE is consistent and asymptotically normally distributed and show that computation of the NPMLE and its standard error is straightforward. We also compare the NPMLE and the estimator of Smith et al. for a variety of situations.     

Prediction of plant species distributions across six millennia

The usefulness of species distribution models (SDMs) in predicting impacts of climate change on biodiversity is difficult to assess because changes in species ranges may take decades or centuries to occur. One alternative way to evaluate the predictive ability of SDMs across time is to compare their predictions with data on past species distributions. We use data on plant distributions, fossil pollen and current and mid-Holocene climate to test the ability of SDMs to predict past climate-change impacts. We find that species showing little change in the estimated position of their realized niche, with resulting good model performance, tend to be dominant competitors for light. Different mechanisms appear to be responsible for among-species differences in model performance. Confidence in predictions of the impacts of climate change could be improved by selecting species with characteristics that suggest little change is expected in the relationships between species occurrence and climate patterns.     

Controlled comparison of species- and community-level models across novel climates and communities

Species distribution models (SDMs) assume species exist in isolation and do not influence one another''s distributions, thus potentially limiting their ability to predict biodiversity patterns. Community-level models (CLMs) capitalize on species co-occurrences to fit shared environmental responses of species and communities, and therefore may result in more robust and transferable models. Here, we conduct a controlled comparison of five paired SDMs and CLMs across changing climates, using palaeoclimatic simulations and fossil-pollen records of eastern North America for the past 21 000 years. Both SDMs and CLMs performed poorly when projected to time periods that are temporally distant and climatically dissimilar from those in which they were fit; however, CLMs generally outperformed SDMs in these instances, especially when models were fit with sparse calibration datasets. Additionally, CLMs did not over-fit training data, unlike SDMs. The expected emergence of novel climates presents a major forecasting challenge for all models, but CLMs may better rise to this challenge by borrowing information from co-occurring taxa.     

Incorporating physiology into species distribution models moderates the projected impact of warming on selected Mediterranean marine species

Species distribution models (SDMs) correlate species occurrences with environmental predictors, and can be used to forecast distributions under future climates. SDMs have been criticized for not explicitly including the physiological processes underlying the species response to the environment. Recently, new methods have been suggested to combine SDMs with physiological estimates of performance (physiology-SDMs). In this study, we compare SDM and physiology-SDM predictions for select marine species in the Mediterranean Sea, a region subjected to exceptionally rapid climate change. We focused on six species and created physiology-SDMs that incorporate physiological thermal performance curves from experimental data with species occurrence records. We then contrasted projections of SDMs and physiology-SDMs under future climate (year 2100) for the entire Mediterranean Sea, and particularly the ‘warm’ trailing edge in the Levant region. Across the Mediterranean, we found cross-validation model performance to be similar for regular SDMs and physiology-SDMs. However, we also show that for around half the species the physiology-SDMs substantially outperform regular SDM in the warm Levant. Moreover, for all species the uncertainty associated with the coefficients estimated from the physiology-SDMs were much lower than in the regular SDMs. Under future climate, we find that both SDMs and physiology-SDMs showed similar patterns, with species predicted to shift their distribution north-west in accordance with warming sea temperatures. However, for the physiology-SDMs predicted distributional changes are more moderate than those predicted by regular SDMs. We conclude, that while physiology-SDM predictions generally agree with the regular SDMs, incorporation of the physiological data led to less extreme range shift forecasts. The results suggest that climate-induced range shifts may be less drastic than previously predicted, and thus most species are unlikely to completely disappear with warming climate. Taken together, the findings emphasize that physiological experimental data can provide valuable supplemental information to predict range shifts of marine species.     

Species distributions models may predict accurately future distributions but poorly how distributions change: A critical perspective on model validation

Aim

Species distribution models (SDMs) are widely used to make predictions on how species distributions may change as a response to climatic change. To assess the reliability of those predictions, they need to be critically validated with respect to what they are used for. While ecologists are typically interested in how and where distributions will change, we argue that SDMs have seldom been evaluated in terms of their capacity to predict such change. Instead, typical retrospective validation methods estimate model's ability to predict to only one static time in future. Here, we apply two validation methods, one that predicts and evaluates a static pattern, while the other measures change and compare their estimates of predictive performance.

Location

Fennoscandia.

Methods

We applied a joint SDM to model the distributions of 120 bird species in four model validation settings. We trained models with a dataset from 1975 to 1999 and predicted species' future occurrence and abundance in two ways: for one static time period (2013–2016, ‘static validation’) and for a change between two time periods (difference between 1996–1999 and 2013–2016, ‘change validation’). We then measured predictive performance using correlation between predicted and observed values. We also related predictive performance to species traits.

Results

Even though static validation method evaluated predictive performance as good, change method indicated very poor performance. Predictive performance was not strongly related to any trait.

Main Conclusions

Static validation method might overestimate predictive performance by not revealing the model's inability to predict change events. If species' distributions remain mostly stable, then even an unfit model can predict the near future well due to temporal autocorrelation. We urge caution when working with forecasts of changes in spatial patterns of species occupancy or abundance, even for SDMs that are based on time series datasets unless they are critically validated for forecasting such change.     

Combining physiological threshold knowledge to species distribution models is key to improving forecasts of the future niche for macroalgae

Species distribution models (SDM) are a useful tool for predicting species range shifts in response to global warming. However, they do not explore the mechanisms underlying biological processes, making it difficult to predict shifts outside the environmental gradient where the model was trained. In this study, we combine correlative SDMs and knowledge on physiological limits to provide more robust predictions. The thermal thresholds obtained in growth and survival experiments were used as proxies of the fundamental niches of two foundational marine macrophytes. The geographic projections of these species’ distributions obtained using these thresholds and existing SDMs were similar in areas where the species are either absent‐rare or frequent and where their potential and realized niches match, reaching consensus predictions. The cold‐temperate foundational seaweed Himanthalia elongata was predicted to become extinct at its southern limit in northern Spain in response to global warming, whereas the occupancy of southern‐lusitanic Bifurcaria bifurcata was expected to increase. Combined approaches such as this one may also highlight geographic areas where models disagree potentially due to biotic factors. Physiological thresholds alone tended to over‐predict species prevalence, as they cannot identify absences in climatic conditions within the species’ range of physiological tolerance or at the optima. Although SDMs tended to have higher sensitivity than threshold models, they may include regressions that do not reflect causal mechanisms, constraining their predictive power. We present a simple example of how combining correlative and mechanistic knowledge provides a rapid way to gain insight into a species’ niche resulting in consistent predictions and highlighting potential sources of uncertainty in forecasted responses to climate change.     

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.     

Trade-offs in covariate selection for species distribution models: a methodological comparison

Species distribution models (SDMs) are a common approach to describing species’ space-use and spatially-explicit abundance. With a myriad of model types, methods and parameterization options available, it is challenging to make informed decisions about how to build robust SDMs appropriate for a given purpose. One key component of SDM development is the appropriate parameterization of covariates, such as the inclusion of covariates that reflect underlying processes (e.g. abiotic and biotic covariates) and covariates that act as proxies for unobserved processes (e.g. space and time covariates). It is unclear how different SDMs apportion variance among a suite of covariates, and how parameterization decisions influence model accuracy and performance. To examine trade-offs in covariation parameterization in SDMs, we explore the attribution of spatiotemporal and environmental variation across a suite of SDMs. We first used simulated species distributions with known environmental preferences to compare three types of SDM: a machine learning model (boosted regression tree), a semi-parametric model (generalized additive model) and a spatiotemporal mixed-effects model (vector autoregressive spatiotemporal model, VAST). We then applied the same comparative framework to a case study with three fish species (arrowtooth flounder, pacific cod and walleye pollock) in the eastern Bering Sea, USA. Model type and covariate parameterization both had significant effects on model accuracy and performance. We found that including either spatiotemporal or environmental covariates typically reproduced patterns of species distribution and abundance across the three models tested, but model accuracy and performance was maximized when including both spatiotemporal and environmental covariates in the same model framework. Our results reveal trade-offs in the current generation of SDM tools between accurately estimating species abundance, accurately estimating spatial patterns, and accurately quantifying underlying species–environment relationships. These comparisons between model types and parameterization options can help SDM users better understand sources of model bias and estimate error.     

The potential range of <Emphasis Type="Italic">Ailanthus altissima</Emphasis> (tree of heaven) in South Africa: the roles of climate,land use and disturbance

Invasive trees are a major problem in South Africa. Many species are well established whereas others are still in the early stages of invasion. The management of invasive species is most cost effective at the early stages of invasion; it is thus essential to target and contain naturalizing invaders before they spread across the landscape. Multi-scale species distribution models (SDMs) provide useful insights to managers; they combine species-occurrence observations with climatic variables to predict potential distributions of alien species. Applying SDMs in human-dominated ecosystems is complicated because many factors associated with human actions interact in complex ways with climatic and edaphic factors to determine the potential suitability of sites for species. The aim of this study was to determine the degree to which a worldwide invader, A. altissima (Simaroubaceae) has occupied its potential range in South Africa, to identify areas at risk of future invasion. To do this we built a set of SDMs at both global and country scales using climatic, land use and human-footprint data. Climatic data best explained the distribution of A. altissima at the global scale whereas variables reflecting human-mediated disturbances were most influential at the national scale. Our analyses show the importance of human-mediated disturbances at a global scale and human occupancy at a country scale in determining the range limits of A. altissima. Populations of this tree species are already present in most parts of South Africa that are environmentally suitable for the species, and management actions need to focus on preventing increases in density in these areas.