Effects of Dispersal-Related Factors on Species Distribution Model Accuracy for Boreal Lake Ecosystems

Simon Hallstan,Richard Johnson,Leonard Sandin

doi:10.3390/d5020393

Simon Hallstan, Richard Johnson + Show 1 more

Open Access

https://doi.org/10.3390/d5020393

Copy DOI

Abstract

Species distribution modeling is used in applied ecology; for example in predicting the consequences of global change. However, questions still remain about the robustness of model predictions. Here we estimate effects of landscape spatial configuration and organism flight ability—factors related to dispersal—on the accuracy of species distribution models. Distribution models were developed for 129 phytoplankton taxa, 164 littoral invertebrate taxa and 44 profundal invertebrate taxa sampled in 105 Swedish lakes, using six different modeling techniques (generalized linear models (GLM), multivariate adaptive regression splines (MARS), classification tree analysis (CTA), mixture discriminant analysis (MDA), generalized boosting models (GBM) and random forests (RF)). Model accuracy was not affected by dispersal ability (i.e., invertebrate flight ability), but the accuracy of phytoplankton assemblage predictions and, to a lesser extent, littoral invertebrate assemblages were related to ecosystem size and connectivity. Although no general pattern across species or spatial configuration was evident from our study, we recommend that dispersal and spatial configuration of ecosystems should be considered when developing species distribution models.

Highlights

Species distribution models (SDMs) have become important tools in applied ecology [1]
Model accuracy was low for lakes presumed to be more isolated, e.g., the accuracy of assemblage prediction (TPR) was significantly and negatively correlated with altitude for littoral invertebrates (Figure 2 and Table 4)
Both littoral invertebrate and phytoplankton assemblages were more accurately predicted when distances to other lakes were greater, and littoral assemblages were more accurately predicted when fewer lakes were present in the 10,000 m buffer zone

Summary

Introduction

Species distribution models (SDMs) have become important tools in applied ecology [1]. Species with narrow niches were more accurately modeled by climatic envelope models than species with broader niches [12], possibly because the species studied did not occupy their entire climate tolerance range due to other environmental and historical factors. Related to niche width, species range is another factor affecting model accuracy. Newbold [13] found that models for species with a small range size are more accurate than models for species with larger range size Dispersal ability is another factor thought to affect model accuracy. Increased model accuracy was attributed to dispersal limitation, mass effects and spatial autocorrelation in environmental factors, all of which were included in the combined model [16]

Methods

Results

Discussion

Conclusion