Abstract
Species distribution models (SDMs) estimate the geographical distribution of species although with several limitations due to sources of inaccuracy and biases. Empirical tests arose as the most important steps in scientific knowledge to assess the efficiency of model predictions, which are poorly rigorous in SDMs. A good approach to the empirical distribution (ED) of a species can be obtained from comprehensive empirical knowledge, that is, well‐understood distributions gathered from large amount of data generated with appropriate spatial and temporal samples coverage. The aims of this study were to (a) compare different SDMs predictions with an ED; and (b) evaluate if fuzzy global matching (FGM) could be used as an index to compare SDMs predictions and ED. Six algorithms with 5 and 20 variables were used to assess their accuracy in predicting the ED of the venomous snake Bothrops alternatus (Viperidae). Its entire distribution is known, thanks to thorough field surveys across Argentina, with 1,767 records. ED was compared with SDMs predictions using Map Comparison Kit. SDMs predictions showed important biases in all methods used, from 70% sub‐estimation to 40% over‐estimation of ED. BIOCLIM predicted ≈31% of B. alternatus ED. DOMAIN predicted 99% of ED, but over‐estimated 40% of the area. GLM with five variables calculated 75% of ED, while Genetic Algorithm for Rule‐set Prediction showed ≈60% of ED; the last two presenting overpredictions in areas with favorable climatic conditions but not inhabited by the species. MaxEnt and RF were the only methods to detect isolated populations in the southern distribution of B. alternatus. Although SDMs proved useful in making predictions about species distribution, predictions need validation with expert maps knowledge and ED. Moreover, FGM showed a good performance as an index with values similar to True Skill Statistic, so that it could be used to relate ED and SDMs predictions.
Highlights
In recent decades, increased use of GIS and technical tools that quantify species–environment relationships has encouraged the development of algorithms to predict the spatial distribution of species, called species distribution models (SDMs) (Elith & Leathwick, 2009; Guisan & Zimmermann, 2000)
The second group of algorithms is composed of methods that characterize the background with a sample, such as Genetic Algorithm for Rule-set Prediction (GARP) and Maximum Entropy (MaxEnt), or that sometimes use pseudo-absences and/or presence data, like several general lineal models (GLM) regression approaches and Random Forest (RF)
For those models that needed presence/absence data, we generated 1,000 random absence records outside the distribution area (ED) (Tognelli, Roig-junent, Marvaldi, Flores, & Lobo, 2009), where the evidence in the last 100 years showed that B. alternatus does not presently occur but rather became true absences, which are based on reliable field evidence of nonoccurrence (Figure 1c) (Saupe et al, 2012)
Summary
In recent decades, increased use of GIS and technical tools that quantify species–environment relationships has encouraged the development of algorithms to predict the spatial distribution of species, called species distribution models (SDMs) (Elith & Leathwick, 2009; Guisan & Zimmermann, 2000). SDMs relate species occurrence data with a set of variables selected under the assumption that they could be related to the distribution of the species (Guisan & Zimmermann, 2000). They are being increasingly used to assess conservation applications and climate change studies, and predict both ecological ranges and the potential of invasive species and explicit predictions about species environmental suitability (Bosso et al, 2017; Chen, Zhang, Jiang, Nielsen, & He, 2017; Law et al, 2017). Unbiased species distribution information is important to make robust conservation management decisions (Guisan et al, 2013)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have