Data prevalence matters when assessing species' responses using data-driven species distribution models

Multi-objective optimisation of species distribution models for river management

Environmental and measure implementation costs are two key factors to be considered by river managers in decision making. To balance effects and costs of an action, practitioners can rely on diagnostic analysis of presence/absence freshwater species distribution models (SDMs) trained to over- or underestimating species presence. Prevalence-adjusted model training aims to balance under- and overestimation depending on study objectives and training data characteristics. The objective of minimising under- and overestimation is a typical example of multi-objective optimisation (MOO). The aim of this paper is to address, for the first time, the practice of MOO-based prevalence-adjusted SDM training for freshwater decision management. In a numerical experiment, the use of Pareto-based MOO, specifically the non-dominated sorting genetic algorithm II (NSGA-II), is compared to commonly-used single-objective optimisation. SDMs for 11 pollution-sensitive freshwater macroinvertebrate species are trained with a subset of the Limnodata, a large data set holding records in the Netherlands over 30 years at 20,000 locations. An increase of two to four times is observed for the ability to identify a large range distribution of the solutions in the Pareto space, when using NSGA-II counter to repeated single-objective optimisation, this by increasing the average runtime with only four percent for a single run. In addition, the use of NSGA-II is found to be effective to identify reliable SDMs useful for diagnostic analysis. By applying and comparing a broad range of MOO methodologies for prevalence-adjusted model training, we believe a closer collaboration between model developers and freshwater managers can be facilitated and environmental standard limits can be set on a more objective basis. In conclusion, the use of MOO for prevalence-adjusted model training is assessed as a valuable tool to support river - and potentially all environmental - decision making.

Predicting Habitat Suitability and Conserving Juniperus spp. Habitat Using SVM and Maximum Entropy Machine Learning Techniques

Support vector machine (SVM) and maximum entropy (MaxEnt) machine learning techniques are well suited to model the habitat suitability of species. In this study, SVM and MaxEnt models were developed to predict the habitat suitability of Juniperus spp. in the Southern Zagros Mountains of Iran. In recent decades, drought extension and climate alteration have led to extensive changes in the geographical occurrence of this species and its growth and regeneration are extremely limited in this area. This study evaluated the habitat suitability of Juniperus through spatial modeling and predicts appropriate regions for future cultivation and resource conservation. We modeled the natural habitat of Juniperus for an area of 700 ha in Sepidan Area in the Fars province using (1) data regarding the presence of the species (295 samples) collected through field surveys and GPS, (2) habitat soil information and indices derived from 60 soil samples collected in the study area, and (3) climatic and topographic datasets collected from various sources. In total, 15 conditioning factors were used for this spatial modeling approach. Receiver operator characteristic (ROC) curves were applied to estimate the accuracy of the habitat suitability models produced by the SVM and MaxEnt techniques. Results indicated logical and similar area under the curve (AUC)-ROC values for the SVM (0.735) and MaxEnt (0.728) models. Both the SVM and MaxEnt methods revealed a significant relationship between the Juniperus spp. distribution and conditioning factors. Environmental factors played a vital role in evaluating the presence of Juniperus sp. as Max and Min temperatures and annual mean rainfall were the three most important factors for habitat suitability in the study area. Finally, an area with high and very high suitability for the future cultivation of Juniperus sp. and for landscape conservation was suggested based on the SVM model.

Machine learning of large-scale spatial distributions of wild turkeys with high-dimensional environmental data

Species distribution modeling often involves high-dimensional environmental data. Large amounts of data and multicollinearity among covariates impose challenges to statistical models in variable selection for reliable inferences of the effects of environmental factors on the spatial distribution of species. Few studies have evaluated and compared the performance of multiple machine learning (ML) models in handling multicollinearity. Here, we assessed the effectiveness of removal of correlated covariates and regularization to cope with multicollinearity in ML models for habitat suitability. Three machine learning algorithms maximum entropy (MaxEnt), random forests (RFs), and support vector machines (SVMs) were applied to the original data (OD) of 27 landscape variables, reduced data (RD) with 14 highly correlated covariates being removed, and 15 principal components (PC) of the OD accounting for 90% of the original variability. The performance of the three ML models was measured with the area under the curve and continuous Boyce index. We collected 663 nonduplicated presence locations of Eastern wild turkeys (Meleagris gallopavo silvestris) across the state of Mississippi, United States. Of the total locations, 453 locations separated by a distance of ≥2 km were used to train the three ML algorithms on the OD, RD, and PC data, respectively. The remaining 210 locations were used to validate the trained ML models to measure ML performance. Three ML models had excellent performance on the RD and PC data. MaxEnt and SVMs had good performance on the OD data, indicating the adequacy of regularization of the default setting for multicollinearity. Weak learning of RFs through bagging appeared to alleviate multicollinearity and resulted in excellent performance on the OD data. Regularization of ML algorithms may help exploratory studies of the effects of environmental factors on the spatial distribution and habitat suitability of wildlife.