Simulated Species Distributions Research Articles

Recommendations for quantifying and reducing uncertainty in climate projections of species distributions.

Projecting the future distributions of commercially and ecologically important species has become a critical approach for ecosystem managers to strategically anticipate change, but large uncertainties in projections limit climate adaptation planning. Although distribution projections are primarily used to understand the scope of potential change-rather than accurately predict specific outcomes-it is nonetheless essential to understand where and why projections can give implausible results and to identify which processes contribute to uncertainty. Here, we use a series of simulated species distributions, an ensemble of 252 species distribution models, and an ensemble of three regional ocean climate projections, to isolate the influences of uncertainty from earth system model spread and from ecological modeling. The simulations encompass marine species with different functional traits and ecological preferences to more broadly address resource manager and fishery stakeholder needs, and provide a simulated true state with which to evaluate projections. We present our results relative to the degree of environmental extrapolation from historical conditions, which helps facilitate interpretation by ecological modelers working in diverse systems. We found uncertainty associated with species distribution models can exceed uncertainty generated from diverging earth system models (up to 70% of total uncertainty by 2100), and that this result was consistent across species traits. Species distribution model uncertainty increased through time and was primarily related to the degree to which models extrapolated into novel environmental conditions but moderated by how well models captured the underlying dynamics driving species distributions. The predictive power of simulated species distribution models remained relatively high in the first 30 years of projections, in alignment with the time period in which stakeholders make strategic decisions based on climate information. By understanding sources of uncertainty, and how they change at different forecast horizons, we provide recommendations for projecting species distribution models under global climate change.

Identifying changing interspecific associations along gradients at multiple scales using wavelet correlation networks.

Identifying interspecific associations is very important for understanding the community assembly process. However, most methods provide only an average association and assume that the association strength does not vary along the environmental gradient or with time. The scale effects are generally ignored. We integrated the idea of wavelet and network topological analysis to provide a novel way to detect nonrandom species associations across scales and along gradients using continuous or presence-absence ecological data. We first used a simulated species distribution data set to illustrate how the wavelet correlation analysis builds an association matrix and demonstrates its statistical robustness. Then, we applied the wavelet correlation network to a presence-absence data set of soil invertebrates. We found that the associations of invertebrates varied along an altitudinal gradient. We conclude by discussing several possible extensions of this method, such as predicting community assembly, utility in the temporal dimension, and the shifting effects of highly connected species within a community. The combination of the multiscale decomposition of wavelet and network topology analysis has great potential for fostering an understanding of the assembly and succession of communities, as well as predicting their responses to future climate change across spatial or temporal scales.

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.

Cost-effective spatial sampling designs for field surveys of species distribution

High-quality biodiversity inventories are key tools to develop effective conservation strategies, but financial resources devoted to systematic species inventories are usually limited. Different sampling strategies have been proposed to efficiently allocate such limited resources (i.e. accessibility-based, stratified random and grid samplings), but their effectiveness may depend on the aim of the survey. Our aim was to assess which approach can provide the best trade-off between sampling costs and accuracy in estimating both single species distribution and species set composition. We generated simulated species distribution data to compare costs and performances of the three sampling methods in assessing species distribution. When we aim at measuring species range (i.e. area of occupancy or extent of occurrence), or obtaining baseline ecological data for conservation assessments (i.e. niche breadth), grid sampling usually provided the best trade-off between performance and costs at both the single- and multi-species levels. Otherwise, the stratified random sampling outperformed the other methods when we are interested in assessing the relative rarity (i.e. species frequency) of the species across the study area. Low quality distribution data can lead to heavily biased conclusions on biodiversity trends or impacts of environmental changes; our findings highlight that selecting the right sampling strategy is essential to obtain reliable estimates of both single species distribution and species set composition.

How different are species distribution model predictions?—Application of a new measure of dissimilarity and level of significance to giant panda Ailuropoda melanoleuca

Species distribution models (SDMs) are widely used for predicting species' spatial distributions. Different model setup and data input however can lead to variable model predictions. Existing studies on quantifying SDM dissimilarity primarily rely on partitioning the variability in SDM-produced community level metrics such as species richness and turnover rate which are threshold-dependent and are generated with binary range maps of multiple species. Most existing measurements of spatial dissimilarity constitute geometric comparisons, which is limited compared to a more information-theoretic application of statistical dissimilarity measures using SDM predictions as direct input without renormalization. We introduce a novel method to quantify the degree of dissimilarity and its level of significance between unscaled SDM predictions of a single species. We apply the method to giant panda Ailuropoda melanoleuca data as well as pairs of simulated species distributions. We utilize a pixel-based Bhattacharyya distance to quantify the degree of dissimilarity among predictions of giant panda habitat of different combinations of model types, Global Climate Models (GCMs) and Representative Concentration Pathways (RCPs). Comparisons are also made between pairs of simulated species with different degrees of dissimilarity in spatial distribution. To evaluate the level of significance, the observed dissimilarity measure is compared against a null distribution that captures the level of dissimilarity caused by small and random variations. Specific pairs of climate scenarios (HadGEM2-ES with HadGEM2-AO and HadGEM2-AO with MIROC5) consistently produce statistically similar predictions of giant panda habitat; the highest level of RCP tends to result in more similar predictions, suggesting a convergence of model predictions. The simulated scenarios also show that the proposed method is able to effectively differentiate a range of artificial species with varying degree of dissimilarity in their resource selection preference. Our method can also reflect the dissimilarity that cannot be quantified by traditional metrics that rely on geometric comparisons. The proposed method supplements existing studies by utilizing a novel application of statistical comparisons to measure dissimilarity between user-defined pairs of models. It provides a robust way to construct the null distribution of dissimilarity that contrasts the degree of the observed dissimilarity with the intrinsic model variability. Our study provides useful insight to facilitate building more computationally efficient and robust ensemble SDMs, and it lends a practical tool to help understand the processes that contribute to prediction variability among SDMs.

Detecting outliers in species distribution data

AbstractAimSpecies distribution data play a pivotal role in the study of ecology, evolution, biogeography and biodiversity conservation. Although large amounts of location data are available and accessible from public databases, data quality remains problematic. Of the potential sources of error, positional errors are critical for spatial applications, particularly where these errors place observations beyond the environmental or geographical range of species. These outliers need to be identified, checked and removed to improve data quality and minimize the impact on subsequent analyses. Manually checking all species records within large multispecies datasets is prohibitively costly. This work investigates algorithms that may assist in the efficient vetting of outliers in such large datasets.LocationWe used real, spatially explicit environmental data derived from the western part of Victoria, Australia, and simulated species distributions within this same region.MethodsBy adapting species distribution modelling (SDM), we developed a pseudo‐SDM approach for detecting outliers in species distribution data, which was implemented with random forest (RF) and support vector machine (SVM) resulting in two new methods: RF_pdSDM and SVM_pdSDM. Using virtual species, we compared eight existing multivariate outlier detection methods with these two new methods under various conditions.ResultsThe two new methods based on the pseudo‐SDM approach had higher true skill statistic (TSS) values than other approaches, with TSS values always exceeding 0. More than 70% of the true outliers in datasets for species with a low and intermediate prevalence can be identified by checking 10% of the data points with the highest outlier scores.Main conclusionsPseudo‐SDM‐based methods were more effective than other outlier detection methods. However, this outlier detection procedure can only be considered as a screening tool, and putative outliers must be examined by experts to determine whether they are actual errors or important records within an inherently biased set of data.

Identifying Reliable Opportunistic Data for Species Distribution Modeling: A Benchmark Data Optimization Approach

The purpose of this study is to increase the number of species occurrence data by integrating opportunistic data with Global Biodiversity Information Facility (GBIF) benchmark data via a novel optimization technique. The optimization method utilizes Natural Language Processing (NLP) and a simulated annealing (SA) algorithm to maximize the average likelihood of species occurrence in maximum entropy presence-only species distribution models (SDM). We applied the Kruskal–Wallis test to assess the differences between the corresponding environmental variables and habitat suitability indices (HSI) among datasets, including data from GBIF, Facebook (FB), and data from optimally selected FB data. To quantify uncertainty in SDM predictions, and to quantify the efficacy of the proposed optimization procedure, we used a bootstrapping approach to generate 1000 subsets from five different datasets: (1) GBIF; (2) FB; (3) GBIF plus FB; (4) GBIF plus optimally selected FB; and (5) GBIF plus randomly selected FB. We compared the performance of simulated species distributions based on each of the above subsets via the area under the curve (AUC) of the receiver operating characteristic (ROC). We also performed correlation analysis between the average benchmark-based SDM outputs and the average dataset-based SDM outputs. Median AUCs of SDMs based on the dataset that combined benchmark GBIF data and optimally selected FB data were generally higher than the AUCs of other datasets, indicating the effectiveness of the optimization procedure. Our results suggest that the proposed approach increases the quality and quantity of data by effectively extracting opportunistic data from large unstructured datasets with respect to benchmark data.

Selecting thresholds for the prediction of species occurrence with presence‐only data

AbstractAimSpecies distribution models have been widely used to tackle ecological, evolutionary and conservation problems. Most species distribution modelling techniques produce continuous suitability predictions, but many real applications (e.g. reserve design, species invasion and climate change impact assessment) and model evaluations require binary outputs, and thresholds are needed for these transformations. Although there are many threshold selection methods for presence/absence data, it is unclear whether these are suitable for presence‐only data. In this paper, we investigate mathematically and empirically which of the existing threshold selection methods can be used confidently with presence‐only data.LocationWe used real spatially explicit environmental data derived from the western part of the state of Victoria, south‐eastern Australia, and simulated species distributions within this area.MethodsThirteen existing threshold selection methods were investigated mathematically to see whether the same threshold can be produced using either presence/absence data or presence‐only data. We further adopted a simulation approach, created many virtual species with differing prevalences in a real landscape in south‐eastern Australia, generated data sets with different proportions of pseudo‐absences, built eight types of models with four modelling techniques, and investigated the behaviours of four threshold selection methods in these situations.ResultsThree threshold selection methods were not affected by pseudo‐absences, including max SSS (which is based on maximizing the sum of sensitivity and specificity), the prevalence of model training data and the mean predicted value of a set of random points. Max SSS produced higher sensitivity in most cases and higher true skill statistic and kappa in many cases than the other methods. The other methods produced different thresholds from presence‐only data to those determined from presence/absence data.Main conclusionsMax SSS is a promising method for threshold selection when only presence data are available.

Selecting pseudo‐absences for species distribution models: how, where and how many?

Summary1. Species distribution models are increasingly used to address questions in conservation biology, ecology and evolution. The most effective species distribution models require data on both species presence and the available environmental conditions (known as background or pseudo‐absence data) in the area. However, there is still no consensus on how and where to sample these pseudo‐absences and how many.2. In this study, we conducted a comprehensive comparative analysis based on simple simulated species distributions to propose guidelines on how, where and how many pseudo‐absences should be generated to build reliable species distribution models. Depending on the quantity and quality of the initial presence data (unbiased vs. climatically or spatially biased), we assessed the relative effect of the method for selecting pseudo‐absences (random vs. environmentally or spatially stratified) and their number on the predictive accuracy of seven common modelling techniques (regression, classification and machine‐learning techniques).3. When using regression techniques, the method used to select pseudo‐absences had the greatest impact on the model’s predictive accuracy. Randomly selected pseudo‐absences yielded the most reliable distribution models. Models fitted with a large number of pseudo‐absences but equally weighted to the presences (i.e. the weighted sum of presence equals the weighted sum of pseudo‐absence) produced the most accurate predicted distributions. For classification and machine‐learning techniques, the number of pseudo‐absences had the greatest impact on model accuracy, and averaging several runs with fewer pseudo‐absences than for regression techniques yielded the most predictive models.4. Overall, we recommend the use of a large number (e.g. 10 000) of pseudo‐absences with equal weighting for presences and absences when using regression techniques (e.g. generalised linear model and generalised additive model); averaging several runs (e.g. 10) with fewer pseudo‐absences (e.g. 100) with equal weighting for presences and absences with multiple adaptive regression splines and discriminant analyses; and using the same number of pseudo‐absences as available presences (averaging several runs if few pseudo‐absences) for classification techniques such as boosted regression trees, classification trees and random forest. In addition, we recommend the random selection of pseudo‐absences when using regression techniques and the random selection of geographically and environmentally stratified pseudo‐absences when using classification and machine‐learning techniques.

Scaling down distribution maps from atlas data: a test of different approaches with virtual species

AbstractAim Conservation managers are increasingly looking for modelled projections of species distributions to inform management strategies; however, the coarse resolution of available data usually compromises their helpfulness. The aim of this paper is to delineate and test different approaches for converting coarse‐grain occurrence data into high‐resolution predictions, and to clarify the conceptual circumstances affecting the accuracy of downscaled models.Location We used environmental data from a real landscape, southern Africa, and simulated species distributions within this landscape.Methods We built 10 virtual species at a resolution of 5 arcmin, and for each species we simulated atlas range maps at four decreasing resolutions (15, 30, 60, 120 arcmin). We tested the ability of three downscaling strategies to produce high‐resolution predictions using two modelling techniques: generalized linear models and generalized boosted models. We calibrated reference models with high‐resolution data and we compared the relative reduction of predictive performance in the downscaled models by using a null model approach. We also estimated the applicability of downscaling procedures to different situations by using distribution data for Mediterranean reptiles.Results All reference models achieved high performance measures. For all strategies, we observed a reduction of predictive performance proportional to the degree of downscaling. The differences in evaluation indices between reference models and downscaled projections obtained from atlases at 15 and 30 arcmin were never statistically significant. The accuracy of projections scaled down from 60 arcmin largely depended on the combination of approach and algorithm adopted. Projections scaled down from 120 arcmin gave misleading results in all cases.Main conclusions Moderate levels of downscaling allow for reasonably accurate results, regardless of the technique used. The most general effect of scaling down coarse‐grain data is the reduction of model specificity. The models can successfully delineate a species’ environmental association up until a 12‐fold downscaling, although with an increasing approximation that causes the overestimation of true distributions. We suggest appropriate procedures to mitigate the commission error introduced by downscaling at intermediate levels (approximately 12‐fold). Reductions of grain size > 12‐fold are discouraged.

Development of species composition in long term simulations with an individual-tree growth simulator

The spruce-fir-beech dominated forest stands in Litschau in the Austrian part of the Bohemian Massif were converted by former forest management practices into pure Norway spruce stands and are now discussed to be reconverted into the potential natural vegetation type. The targeted potential natural vegetation type is usually defined by experts in vegetation sciences. Because meanwhile individual-tree growth simulators are a well acknowledged tool for predicting future forest stand development, in this study we investigate if PROGNAUS can also be used to predict the redevelopment of managed forest ecosystems into natural forest ecosystems regarding species composition. The development of 23 stands in Litschau has been simulated over 1,000 years under the “no-management” option. Generally, the simulated species distribution agrees quite well with the expectations of the potential natural vegetation type. However, the predicted amounts of silver fir and maple species are lower than expected, which probably is due to browsing and management effects represented in the parameterization data for PROGNAUS.

Predicting species distributions: a critical comparison of the most common statistical models using artificial species

AbstractAim To test statistical models used to predict species distributions under different shapes of occurrence–environment relationship. We addressed three questions: (1) Is there a statistical technique that has a consistently higher predictive ability than others for all kinds of relationships? (2) How does species prevalence influence the relative performance of models? (3) When an automated stepwise selection procedure is used, does it improve predictive modelling, and are the relevant variables being selected?Location We used environmental data from a real landscape, the state of California, and simulated species distributions within this landscape.Methods Eighteen artificial species were generated, which varied in their occurrence response to the environmental gradients considered (random, linear, Gaussian, threshold or mixed), in the interaction of those factors (no interaction vs. multiplicative), and on their prevalence (50% vs. 5%). The landscape was then randomly sampled with a large (n = 2000) or small (n = 150) sample size, and the predictive ability of each statistical approach was assessed by comparing the true and predicted distributions using five different indexes of performance (area under the receiver‐operator characteristic curve, Kappa, correlation between true and predictive probability of occurrence, sensitivity and specificity). We compared generalized additive models (GAM) with and without flexible degrees of freedom, logistic regressions (general linear models, GLM) with and without variable selection, classification trees, and the genetic algorithm for rule‐set production (GARP).Results Species with threshold and mixed responses, additive environmental effects, and high prevalence generated better predictions than did other species for all statistical models. In general, GAM outperforms all other strategies, although differences with GLM are usually not significant. The two variable‐selection strategies presented here did not discriminate successfully between truly causal factors and correlated environmental variables.Main conclusions Based on our analyses, we recommend the use of GAM or GLM over classification trees or GARP, and the specification of any suspected interaction terms between predictors. An expert‐based variable selection procedure was preferable to the automated procedures used here. Finally, for low‐prevalence species, variability in model performance is both very high and sample‐dependent. This suggests that distribution models for species with low prevalence can be improved through targeted sampling.

Spectral analysis of simulated species distribution maps provides insights into metapopulation dynamics

Modelling the spatial dynamics of populations is a basic approach in ecology, in order to understand their observed spatial and temporal patterns, which can be diverse and complex. From the metapopulation perspective, the spatial distribution of populations results from colonization–extinction random process over a network of suitable habitat cells. Hence, evaluating such dynamic is an important issue for the follow-up of populations. Our aim here is to demonstrate that Fourier spectral analysis of population distribution maps can provide insights into metapopulation dynamics in a heterogeneous habitat. We simulated metapopulation dynamics in spatially structured habitat maps and investigated the steady spatial occupancy patterns using Fourier analysis. We showed that there were separable spectral signatures of habitat structure and of population dynamics. Fourier spectral analysis thus provides a promising tool for inferring independent characteristics of metapopulation dynamics and habitat structure from species occurrence data.

The University of Illinois, Urbana‐Champaign three‐dimensional stratosphere‐troposphere general circulation model with interactive ozone photochemistry: Fifteen‐year control run climatology

A new University of Illinois at Urbana‐Champaign 24‐layer troposphere‐stratosphere general circulation model with interactive photochemistry is presented. We document here the results of a 15‐year‐long control run of the model and an evaluation of its performance. The ability of the model to simulate the Northern Hemisphere polar vortex has been improved because of the modification of the gravity wave drag parameterization. The extension and update of the chemical routine improved the simulated species distributions. As a result, the agreement of the simulated fields with observations is within 20% for total ozone and other species, 5 K for temperature, and 10 m s−1 for zonal wind over most of the atmosphere. The model simulates well the appearance of polar stratospheric clouds and the “ozone hole” over Antarctica. However, in some particular points the model results are unrealistic. These are (1) a cold bias (∼25 K) in the Southern Hemisphere lower stratosphere during austral summer, (2) a warm bias (∼20 K) in the middle stratosphere during austral winter, (3) a warm bias (∼10 K) near the stratopause in the tropics, and (4) a persistent cold bias of up to 8 K in the lowermost stratosphere almost everywhere over the globe. These problems may be due to the lack of a nonorographic gravity wave drag parameterization in the model, the influence of the top boundary conditions, and/or some problems with model treatment of the zonal wind‐planetary wave interaction in the stratosphere. Despite these few shortcomings, the model is now ready to be applied to several climate‐chemistry issues such as future ozone layer changes due to increasing greenhouse gas concentrations, the atmospheric influence of volcanic eruptions, and sun‐climate relations.