Original paper

Are species distribution models based on broad-scale environmental variables transferable across adjacent watersheds? A case study with eleven macroinvertebrate species

Gies, Maria; Sondermann, Martin; Hering, Daniel; Feld, Christian K.

Fundamental and Applied Limnology Volume 186 Nr. 1-2 (2015), p. 63 - 97

published: Feb 1, 2015

DOI: 10.1127/fal/2014/0600

BibTeX file

ArtNo. ESP141018601004, Price: 29.00 €

Download preview PDF Buy as PDF


Species distribution models (SDMs) allow the prediction of the spatially explicit presence and absence of species based on environmental predictors that reflect the species' habitat requirements. In river macroinvertebrates, the habitat is often defined at very fine scales spanning one to several tens of square meters (e.g., substrate preferences). Such habitat information, however, is usually not available for entire river networks at the large scale, which limits the application of SDMs in conservation ecology. In this study, we present SDMs of eleven lotic macroinvertebrate species based on two broad-scale environmental variable groups: land use (derived from ATKIS high resolution land cover map) and physical habitat structure (derived from regional surveys in Germany). The actual species distributions were scanned through a field survey at 225 sites in two adjacent watersheds in a mountainous region (river Ruhr and Lenne, Federal State of North Rhine-Westphalia, Germany). The aim of this study was, first, to test the usefulness of broad-scale variables in SDMs using measures of model goodness-of-fit and predictive power. Second, local habitat variables (physico-chemistry and meso-scale substrates) were included in SDMs to examine model improvement. Third, we tested the transferability of models of the same species between the two watersheds. Due to the similar environmental characteristics of both watersheds, we hypothesized concordant SDMs for both watersheds. Overall, we found a reliable performance and predictive power of models of Dinocras cephalotes in both watersheds. Models of several other species performed fair in the Ruhr (Leuctra geniculata, Silo piceus, Siphlonurus lacustris) but not in the Lenne system or vice versa (Hydropsyche instabilis). Broad-scale SDMs included predictors on physical habitat quality as well as riparian land use at a similar extent. Only for five models the SDMs including fine-scale predictors (e.g., physico-chemistry, microhabitat distribution) outperformed those models with broad-scale predictors only (AUC > 0.70). We suggest that species specifically distributed in upstream reaches explicitly respond to fine-scale variables due to stronger dependency of their occurrences on local conditions. Model transferability from one watershed to another was low (transferability index < 0.6), thus revealing SDMs not only to be species-specific but also variable across adjacent watersheds. We suggest that the transferability was limited not only by actual environmental differences between both watersheds, but also by legacy land use effects that may continue to affect the recent distribution of macroinvertebrates.


physical habitat qualitymodel transferabilityland usenon-parametric multiplicative regression (nmpr)mountainous rivers