Evaluating sampling bias correction methods for invasive species distribution modeling in Maxent

As globalization continues, the spread of invasive species is accelerating, posing a severe threat to native biodiversity. To manage such species, reduce their negative impact on native biota and utilize management costs efficiently, a profound understanding of their geographical distribution pattern is mandatory. In this study, the species distribution model Maxent was used to predict the potential spatial distribution of U. europaeus. To account for sampling bias, three bias correction methods were applied, including a novel approach to increase the number of presence points by sampling occurrences based on satellite images. Furthermore, a decision structured process was used to evaluate and select optimal Maxent parameterization and account for limitations of single evaluation criteria. The currently suitable area of U. europaeus is primarily distributed in the coastal and central regions of Chilean natural region Zona Sur in south-central Chile. Annual mean temperature (bio1), annual precipitation (bio12), and precipitation seasonality (bio15) were the most important environmental variables that affected the distribution of U. europaeus. The sampling of additional presence points could effectively correct for sampling bias in species occurrence data. The use of a decision structured process for model evaluation proved to be useful in determining optimal model parameterization for decreased model complexity. This study highlights the importance of optimized Maxent calibrations to yield results as accurately as possible. The predicted suitable habitats can inform nature conservation planners and landscape managers to guide and prioritize conservation measures.     

Bunching up the background betters bias in species distribution models

Sets of presence records used to model species’ distributions typically consist of observations collected opportunistically rather than systematically. As a result, sampling probability is geographically uneven, which may confound the model's characterization of the species’ distribution. Modelers frequently address sampling bias by manipulating training data: either subsampling presence data or creating a similar spatial bias in non‐presence background data. We tested a new method, which we call ‘background thickening’, in the latter category. Background thickening entails concentrating background locations around presence locations in proportion to presence location density. We compared background thickening to two established sampling bias correction methods – target group background selection and presence thinning – using simulated data and data from a case study. In the case study, background thickening and presence thinning performed similarly well, both producing better model discrimination than target group background selection, and better model calibration than models without correction. In the simulation, background thickening performed better than presence thinning when the number of simulated presence locations was low, and vice versa. We discuss drawbacks to target group background selection, why background thickening and presence thinning are conservative but robust sampling bias correction methods, and why background thickening is better than presence thinning for small sample sizes. Particularly, background thickening is advantageous for treating sampling bias when data are scarce because it avoids discarding presence records.     

Community versus single‐species distribution models for British plants

Aim Species distribution models are increasingly used to predict the impacts of global change on whole ecological communities by modelling the individualistic niche responses of large numbers of species. However, it is not clear whether this single‐species ensemble approach is preferable to community‐wide strategies that represent interspecific associations or shared responses to environmental gradients. Here, we test the performance of two multi‐species modelling approaches against equivalent single‐species models. Location Great Britain. Methods Single‐ and multi‐species distribution models were fitted for 701 native British plant species at a 10‐km grid scale. Two machine learning methods were used – classification and regression trees (CARTs) and artificial neural networks (ANNs). The single‐species versions are widely used in ecology but their multivariate extensions are less well known and have not previously been evaluated against one another. We compared their abilities to predict species distributions, community compositions and species richness in an independent geographical region reserved from model‐fitting. Results The single‐ and multi‐species models performed similarly, although the community models gave slightly poorer predictive accuracy by all measures. However, from the point of view of the whole community they were much simpler than the array of single‐species models, involving orders of magnitude fewer parameters. Multi‐species approaches also left greater residual spatial autocorrelation than the individualistic models and, contrary to expectation, were relatively less accurate for rarer species. However, the fitted multi‐species response curves had lower tendency for pronounced discontinuities that are unlikely to be a feature of realized niche responses. Main conclusions Although community distribution models were slightly less accurate than single‐species models, they offered a highly simplified way of modelling spatial patterns in British plant diversity. Moreover, an advantage of the multi‐species approach was that the modelling of shared environmental responses resolved more realistic response curves. However, there was a slight tendency for community models to predict rare species less accurately, which is potentially disadvantageous for conservation applications. We conclude that multi‐species distribution models may have potential for understanding and predicting the structure of ecological communities, but were slightly inferior to single‐species ensembles for our data.     

Transferability of trait‐based species distribution models

The need for reliable prediction of species distributions dependent upon traits has been hindered by a lack of model transferability testing. We tested the predictive capacity of trait‐SDMs by fitting hierarchical generalised linear models with three trait and four environmental predictors for 20 eucalypt taxa in a reference region. We used these models to predict occurrence for a much larger set of taxa and target areas (82 taxa across 18 target regions) in south‐eastern Australia. Median predictive performance for new species in target regions was 0.65 (area under receiver operating curve) and 1.24 times random (area under precision recall curve). Prediction in target regions did not worsen with increasing geographic, environmental or community compositional distance from the reference region, and was improved with reliable trait–environment relationships. Transfer testing also identified trait–environment relationships that did not transfer. These results give confidence that traits and transfer testing can assist in the hard problem of predicting environmental responses for new species, environmental conditions and regions.     

Adaptable gene‐specific dye bias correction for two‐channel DNA microarrays

DNA microarray technology is a powerful tool for monitoring gene expression or for finding the location of DNA‐bound proteins. DNA microarrays can suffer from gene‐specific dye bias (GSDB), causing some probes to be affected more by the dye than by the sample. This results in large measurement errors, which vary considerably for different probes and also across different hybridizations. GSDB is not corrected by conventional normalization and has been difficult to address systematically because of its variance. We show that GSDB is influenced by label incorporation efficiency, explaining the variation of GSDB across different hybridizations. A correction method (Gene‐ And Slide‐Specific Correction, GASSCO) is presented, whereby sequence‐specific corrections are modulated by the overall bias of individual hybridizations. GASSCO outperforms earlier methods and works well on a variety of publically available datasets covering a range of platforms, organisms and applications, including ChIP on chip. A sequence‐based model is also presented, which predicts which probes will suffer most from GSDB, useful for microarray probe design and correction of individual hybridizations. Software implementing the method is publicly available.     

Space‐time tradeoffs in the development of precipitation‐based isoscape models for determining migratory origin

Precipitation stable isotope patterns over continental scales provide a fundamental tool for tracking origins of migratory species. Hydrogen isotopes from rain and environmental waters are assimilated into animal tissues and may thereby reveal the location where tissues were synthesized. Predictive isotopic maps (or isoscapes) of stable hydrogen isotope values in precipitation (δ²H_p) are typically generated by time‐averaging observations from a global network of stations that have been sampled irregularly in space and time. We previously demonstrated that restricting the temporal range in δ²H_p isoscapes to biologically relevant time frames did not improve predictions of geographic origin for two migratory species in North America and Europe; rather, it decreased the accuracy of assignment. Here, we examined whether the reduction in assignment accuracy stemmed from a decrease in the number of sampling stations available to support isoscape development for shorter time periods. Multiple regression models were used to predict the hydrogen isotope composition in precipitation using isotopic measurements from each station along with a suite of independent variables. The reduction in the number of stations with δ²H_p measurements used to estimate isoscape model parameters did not alter the accuracy and precision of assignments consistently. We also examined accuracy across a range of reduced station numbers and found that mean accuracy was affected only at very low numbers of stations, indicating that the spatial isotopic patterns in precipitation that are useful for assignment applications can be characterized with data from relatively limited data stations. The number and spatial distribution of stations may have more influence when geostatistical models are used to generate isoscapes, as they incorporate spatial correlation in the dataset. The results can be used to guide future research in understanding how data availability and constraints in creating δ²H_p isoscapes may affect predictions of geographic origins.     

Niche shift in four non‐native estrildid finches and implications for species distribution models

Non‐native species can have severe impacts on ecosystems. Therefore, predictions of potentially suitable areas that are at risk of the establishment of non‐native populations are desirable. In recent years, species distribution models (SDMs) have been widely applied for this purpose. However, the appropriate selection of species records, whether from the native area alone or also from the introduced range, is still a matter of debate. We combined analyses of native and non‐native realized climate niches to understand differences between models based on all locations, as well as on locations from the native range only. Our approach was applied to four estrildid finch species that have been introduced to many regions around the world. Our results showed that SDMs based on location data from native areas alone may underestimate the potential distribution of a given species. The climatic niches of species in their native ranges differed from those of their non‐native ranges. Niche comparisons resulted in low overlap values, indicating considerable niche shifts, at least in the realized niches of these species. All four species have high potential to spread over many tropical and subtropical areas. However, transferring these results to temperate areas has a high degree of uncertainty, and we urge caution when assessing the potential spread of tropical species that have been introduced to higher latitudes.     

A niche for biology in species distribution models

Why species are found where they are is a central question in biogeography. The most widely used tool for understanding the controls on distribution is species distribution modelling. Species distribution modelling is now a well‐established method in both the theoretical and applied ecological literature. In this special issue we examine the current state of the art in species distribution modelling and explore avenues for including more biological processes in such models. In particular we focus on physiological, demographic, dispersal, competitive and ecological‐modulation processes. This overview highlights opportunities for new species distribution model concepts and developments, as well as a statistical agenda for implementing such models.     

Accounting for preferential sampling in species distribution models

Species distribution models (SDMs) are now being widely used in ecology for management and conservation purposes across terrestrial, freshwater, and marine realms. The increasing interest in SDMs has drawn the attention of ecologists to spatial models and, in particular, to geostatistical models, which are used to associate observations of species occurrence or abundance with environmental covariates in a finite number of locations in order to predict where (and how much of) a species is likely to be present in unsampled locations. Standard geostatistical methodology assumes that the choice of sampling locations is independent of the values of the variable of interest. However, in natural environments, due to practical limitations related to time and financial constraints, this theoretical assumption is often violated. In fact, data commonly derive from opportunistic sampling (e.g., whale or bird watching), in which observers tend to look for a specific species in areas where they expect to find it. These are examples of what is referred to as preferential sampling, which can lead to biased predictions of the distribution of the species. The aim of this study is to discuss a SDM that addresses this problem and that it is more computationally efficient than existing MCMC methods. From a statistical point of view, we interpret the data as a marked point pattern, where the sampling locations form a point pattern and the measurements taken in those locations (i.e., species abundance or occurrence) are the associated marks. Inference and prediction of species distribution is performed using a Bayesian approach, and integrated nested Laplace approximation (INLA) methodology and software are used for model fitting to minimize the computational burden. We show that abundance is highly overestimated at low abundance locations when preferential sampling effects not accounted for, in both a simulated example and a practical application using fishery data. This highlights that ecologists should be aware of the potential bias resulting from preferential sampling and account for it in a model when a survey is based on non‐randomized and/or non‐systematic sampling.     

Pseudoabsence generation strategies for species distribution models

Background

Species distribution models require selection of species, study extent and spatial unit, statistical methods, variables, and assessment metrics. If absence data are not available, another important consideration is pseudoabsence generation. Different strategies for pseudoabsence generation can produce varying spatial representation of species.

Methodology

We considered model outcomes from four different strategies for generating pseudoabsences. We generating pseudoabsences randomly by 1) selection from the entire study extent, 2) a two-step process of selection first from the entire study extent, followed by selection for pseudoabsences from areas with predicted probability <25%, 3) selection from plots surveyed without detection of species presence, 4) a two-step process of selection first for pseudoabsences from plots surveyed without detection of species presence, followed by selection for pseudoabsences from the areas with predicted probability <25%. We used Random Forests as our statistical method and sixteen predictor variables to model tree species with at least 150 records from Forest Inventory and Analysis surveys in the Laurentian Mixed Forest province of Minnesota.

Conclusions

Pseudoabsence generation strategy completely affected the area predicted as present for species distribution models and may be one of the most influential determinants of models. All the pseudoabsence strategies produced mean AUC values of at least 0.87. More importantly than accuracy metrics, the two-step strategies over-predicted species presence, due to too much environmental distance between the pseudoabsences and recorded presences, whereas models based on random pseudoabsences under-predicted species presence, due to too little environmental distance between the pseudoabsences and recorded presences. Models using pseudoabsences from surveyed plots produced a balance between areas with high and low predicted probabilities and the strongest relationship between density and area with predicted probabilities ≥75%. Because of imperfect accuracy assessment, the best assessment currently may be evaluation of whether the species has been sufficiently but not excessively predicted to occur.     

Assessing multitemporal calibration for species distribution models

Species distribution models (SDMs), the most prominent tool in modern biogeography, rely on the assumptions that (i) species distribution is in equilibrium with the environment and (ii) that climatic niche has been conserved throughout recent geological time. These issues affect the spatial and temporal transferability of SDMs, limiting their reliability for applications such as when studying effects of past climate change on species distribution and extinctions. The integration of paleontological and neontological data for a multitemporal calibration and validation of SDMs has been suggested for improving SDMs flexibility. Here, we provide an empirical test for a multitemporal calibration, employing virtual species (i.e., with perfectly-known distributions) and comparing them directly with monotemporal SDMs (i.e., SDM calibrated in a single time layer). We used 1kyr-interval scenarios throughout the last 22 kyr BP for two ecologically different species in South America (a “hot and wet” species and a “cold and dry” species). Models with multitemporal calibration performed similarly to models with monotemporal calibration, regardless of species, sample sizes, and time frame. However, multitemporal calibration performed better when dealing with non-analogous climates among time layers. By improving the temporal SDMs transferability, multitemporal calibration opens new avenues for integrating fossil and recent occurrence data, which may substantially benefit biogeography and paleoecology.     

Finessing atlas data for species distribution models

Aim The spatial resolution of species atlases and therefore resulting model predictions are often too coarse for local applications. Collecting distribution data at a finer resolution for large numbers of species requires a comprehensive sampling effort, making it impractical and expensive. This study outlines the incorporation of existing knowledge into a conventional approach to predict the distribution of Bonelli’s eagle (Aquila fasciata) at a resolution 100 times finer than available atlas data. Location Malaga province, Andalusia, southern Spain. Methods A Bayesian expert system was proposed to utilize the knowledge from distribution models to yield the probability of a species being recorded at a finer resolution (1 × 1 km) than the original atlas data (10 × 10 km). The recorded probability was then used as a weight vector to generate a sampling scheme from the species atlas to enhance the accuracy of the modelling procedure. The maximum entropy for species distribution modelling (MaxEnt) was used as the species distribution model. A comparison was made between the results of the MaxEnt using the enhanced and, the random sampling scheme, based on four groups of environmental variables: topographic, climatic, biological and anthropogenic. Results The models with the sampling scheme enhanced by an expert system had a higher discriminative capacity than the baseline models. The downscaled (i.e. finer scale) species distribution maps using a hybrid MaxEnt/expert system approach were more specific to the nest locations and were more contrasted than those of the baseline model. Main conclusions The proposed method is a feasible substitute for comprehensive field work. The approach developed in this study is applicable for predicting the distribution of Bonelli’s eagle at a local scale from a national‐level occurrence data set; however, the usefulness of this approach may be limited to well‐known species.