Transferability of stream benthic macroinvertebrate distribution models to drought-related conditions

Abstract

Freshwater ecosystems increasingly face pressures from climate change induced extreme events, like droughts, posing significant threats to biodiversity. While Species Distribution Models (SDMs) serve as vital tools for predicting species responses to environmental shifts, their transferability to novel environmental conditions, especially during and after drought remains poorly understood. In this study, we delve into the transferability of SDMs for freshwater macroinvertebrates from drought-free to drought-influenced conditions. We examine how sensitive the transferability is to traits such as tolerance scores according to their distribution along longitudinal gradients, as well as the used modelling method. We constructed and validated SDMs for freshwater macroinvertebrates in a central German catchment under drought-free conditions using four different algorithms (Generalised linear models; GLM, Spatial Stream Networks; SSN, Random Forests; RF, and Maximum Entropy; MaxEnt). We then projected these models to environmental conditions influenced by drought, and obtained their transferability by computing the difference in accuracy when predicting under drought-free and drought-influenced conditions (AUCgap). Our findings reveal a marked reduction in SDM accuracy under drought conditions, illustrating the challenges of accurately predicting species distributions into novel environmental conditions. Our results show a slightly better transferability when using SSN and RF. In addition, we observed that SDM transferability can be influenced by species tolerance, with sensitive and tolerant species presenting higher AUCgap (i.e. lower transferability). Furthermore, we found that when sensitive species were modelled using SSNs, the AUCgap was reduced. Our study underscores the limitations of SDMs in capturing species responses to drought and advocates for integrating ecologically relevant predictors and modelling methods that account for the stream connectivity to allow for robust predictions. These considerations could enhance the ability of SDMs to effectively estimate the impacts of extreme events on freshwater biodiversity.