A supervised learning algorithm to evaluate occurrence records in virtual species

Digital and open access of occurrence data have encouraged the development of tools to improve biodiversity conservation and management. In this study, we proposed a methodology to evaluate point-occurrence records based on expert knowledge. We firstly generated virtual data to test our methodology without confounding factors by simulating geographical distributions, virtual sampling, and expert checking of occurrence records. We used a set of non-linear bioclimatic variables and principal component analysis (PCA) to define a duality function between niche and biotope spaces. Subsequently, a supervised-learning model was fit to classify records between true and doubtful presence based on the virtual expert checking. We then tested our methodology using three virtual species and 10-fold cross validation. Also, we evaluated the prediction performance of the supervise model compared with the virtual observer using a virtual external database of occurrence data.

Bottom Communities and Abiotic Factors: Analysis of Statistical Relationship Using the Instability Index and Virtual Species Distribution

Statistical procedures for quantifying the relationships between the community structure and abiotic variables start with selecting a minimum set of uncorrelated environmental factors that determine the ecological conditions essential for each of the species. This is especially important when constructing models of spatial distribution of species which are key to ecology of communities and conservation of nature. The aim of the study is to explore whether some applications of information theory can be used to rank environmental factors with respect to their contribution to the formation of the ecological structure of aquatic communities. We consider the applicability of the instability index, which is a special case of the Kullback-Leibler entropy divergence and reflects the information gain from the displacement of a particular realization of a random variable relative to its mean value. Using of instability indices allows to reduce multidimensional data sets on species structure of communities and abiotic factors to lower dimension sets of commensurate standardized variables and to explore the relationships between the latter. The initial data we used were the results of the long-term (1990–2019) hydrobiological survey of benthic communities in small and medium-sized rivers in the Middle and Lower Volga regions. We consider the indices of instability calculated for each of 147 taxa of macrozoobenthos and 8 geophysical and hydrochemical indicators. Based on these data, we constructed random forest regression models and calculated potential weights of environmental factors that determine ecological preferences of species. The most significant explanatory variables were used to construct distribution maps of «virtual species», which were compared with the corresponding empirical data. A habitat suitability map of chironomids (Diptera, Chironomidae), the Prodiamesinae subfamily, is presented. Instability indices can be effectively used for exploratory analysis of various ecosystems, e. g. ranking habitats according to the degree of environmental instability and / or species associations, selecting the most informative abiotic variables that determine the population density of the taxa, etc.

Inhomogeneous Poisson Point Process for Species Distribution Models: Relative Performance of Methods Accounting for Sampling Bias and Imperfect Detection

Species distribution models have become tools of great importance in ecology since the advanced knowledge of suitable habitat of species is needed in the process of the world's biodiversity conservation. Models that use presence-only data are of great interests and are widely used in ecology due to their easy access. However, these models do not estimate accurately the true spatial species distribution based solely on presence-only data since they do not account for biases induced by the sampling techniques used and imperfect detection. To address this gap, Hierarchical integrated models have been recently introduced. Through this study, we assessed the relative performance of these new SDMs models using simulated data. The performance of the models was tested by comparing the estimates of parameters of the distribution models they provide with parameters used to simulate the distribution of the virtual species. The best model was the one whose estimates were close to the true distribution parameters of the virtual species. Results showed that analyzing Presence-only data in conjunction with Point-counts data through the Dorazio's Hierarchical model produced estimates of the coecients of the species intensity models with high precision and less bias while the Koshkina integrated model showed poor performance. Site-occupancy data, being not informative of species abundance, did not allow reducing biases in Presence-only data. The Dorazio's Hierarchical model produced estimates with high precision even with low detection probability. We have also found that the species rarity tends to in ate the variability of the models' estimates making modelling abundant species to be more accurate than modelling less abundant species. Hence, to model the species distribution with high precision based on Presence-only data, additional Point-counts data are required to account for sampling bias and imperfect detection.

Plotting receiver operating characteristic and precision-recall curves from presence and background data

1. The receiver operating characteristic (ROC) and precision-recall (PR) plots have been widely used to evaluate the performances of species distribution models. Plotting ROC/PR curves requires a traditional test set with both presence and absence data (namely PA approach), but species absence data are usually not available in reality. Plotting ROC/PR curves from presence-only data while treating background data as pseudo absence data (namely PO approach) may provide misleading results. 2. In this study we propose a new approach to calibrate the ROC/PR curves from presence and background data with user-provided information on a constant c, namely PB approach. An estimate of c can also be derived from the PB-based ROC/PR plots given that a model with good ability of discrimination is available. We used three virtual species and a real aerial photography to test the effectiveness of the proposed PB-based ROC/PR plots. Different models (or classifiers) were trained from presence and background data with various samples sizes. The ROC/PR curves plotted by PA approach were used to benchmark the curves plotted by PO and PB approaches. 3. Experimental results show that the curves and areas under curves by PB approach are more similar to that by PA approach as compared with PO approach. The PB-based ROC/PR plots also provide highly accurate estimations of c in our experiment. 4. We conclude that the proposed PB-based ROC/PR plots can provide valuable complements to existing model assessment methods, and they also provide an additional way to estimate the constant c (or species prevalence) from presence and background data.

Assessing the reliability of species distribution projections in climate change research

AimForecasting changes in species distribution under future scenarios is one of the most prolific areas of application for species distribution models (SDMs). However, no consensus yet exists on the reliability of such models for drawing conclusions on species distribution response to changing climate. In this study we provide an overview of common modelling practices in the field and assess model predictions reliability using a virtual species approach.LocationGlobalMethodsWe first provide an overview of common modelling practices in the field by reviewing the papers published in the last 5 years. Then, we use a virtual species approach and three commonly applied SDM algorithms (GLM, MaxEnt and Random Forest) to assess the estimated (cross-validated) and actual predictive performance of models parameterized with different modelling settings and violations of modelling assumptions.ResultsOur literature review shows that most papers that model species distribution under climate change rely on single models (65%) and small samples (< 50 presence points, 62%), use presence-only data (85%), and binarize models’ output to estimate range shift, contraction or expansion (74%). Our virtual species approach reveals that the estimated predictive performance tends to be over-optimistic compared to the real predictive performance. Further, the binarization of predicted probabilities of presence reduces models’ predictive ability considerably. Sample size is one of the main predictors of real accuracy, but has little influence on estimated accuracy. Finally, the inclusion of irrelevant predictors and the violation of modelling assumptions increases estimated accuracy but decreases real accuracy of model projections, leading to biased estimates of range contraction and expansion.Main conclusionsOur study calls for extreme caution in the application and interpretation of SDMs in the context of biodiversity conservation and climate change research, especially when modelling a large number of species where species-specific model settings become impracticable.