Lukáš Gábor prezentoval aktuální výzkum na konferenci Ecological Modelling Globar Conference.
Does Positional Error Affect Fine-Scale Species Distribution Models?
Lukáš Gábor a, Vítězslav Moudrý a, Vincent Lecours b, Marco Malavasi a, Vojtěch Barták a
a Department of Applied Geoinformatics and Spatial Planning, Faculty of Environmental Sciences, Czech University of Life Sciences Prague, Kamycka 129, Praha – Suchdol, 165 00, Czech Republic
b School of Forest Resources and Conservation, University of Florida, Gainesville, Florida, USA
Abstract
Modelling of species-environment relationships using the combination of species occurrences and environmental data is one of frequently used tools in ecological research. A common assumption is that species data are accurate. However, all occurrence data contain some level of positional error; this can be problematic, especially when combined with high-resolution environmental data derived from remote sensing (e.g. LiDAR data). Although such situation may imply serious limitations for the already challenging fine-scale modelling, it has never been thoroughly explored.
Here, we used a virtual species approach to assess the effects of positional error on fine-scale species distribution models (SDMs) for species with various breadths of the ecological niche. We used three virtual species with varying degrees of relative occurrence area (ROA), simulating realistic scenarios of species’ occurrence from rare to common species. To artificially generated occurrences, we subsequently applied GLM and MaxEnt models using the same LiDAR derived environmental variables (canopy height, topography wetness index, elevation) that were used for generating virtual species. Models performance (AUC, TSS) and predicted distributions (Warren I) were compared to assess the effect of the positional error in combination with varying species niches.
Results showed that positional error leads to a decrease in model performance. While the presence of positional error affected performance metric, the predicted species distribution was still in relatively good agreement with known distributions of the virtual species. The positional error affected both (GLM, MaxEnt) in the same way; increasing the sample size did not mitigate this effect. Perhaps most importantly, the effect of positional error differed with respect to the breadth of the species ecological niche. Models of rare species with narrow niche were more affected than those of more common species. These results are important for modelling species-environment relationships as well as for applications of fine-scale SDMs in ecology.
