| Popis: |
In riverine ecosystems the species distribution, determined primarily by their environment often shows zonation patterns that are also typical in the case of net-spinning caddisfly larvae (Trichoptera: Hydropsychidae). In the present research, we aimed to build an ensemble of base learner machine learning (ML) models based on the most important environmental parameters shaping the sequential distribution of ten Central European species of the genus Hydropsyche in the North Hungarian catchment area of Tisza, one of the major rivers of Central and Eastern Europe. The model could explain and effectively predict the occurrence of species and/or groups of them with similar niche preferences. Variable selection revealed the importance of predictors, measured at various spatial scales and with gradient-like characteristics, such as elevation, annual means of discharge, water temperature or the composition of habitat substrates as well as those related to the ecological quality of water or anthropogenic impacts, like annual means of dissolved oxygen and orthophosphate-phosphorous content. Trained on the predictions of different base learner models a final ensemble model predicted the presence and absence of three individual species and three species-groups with significantly improved overall accuracy. High group-wise balanced accuracies of the final model shows that longitudinal, catchment-scale distribution models in stream ecosystems are best built on predictors with variable spatial scales, several of which are routinely measured or recorded in environmental monitoring programmes. Accurate species distribution models (SDMs), capable of adequately predicting presence and absence of bio-indicator taxa, such as Hydropsyche species, can be applied to support environmental management or conservation measures regarding streams and rivers, that are among the most vulnerable of anthropogenic pollution, hydrologic alteration, climate change and biodiversity loss. |
Full Text from ScienceDirect