Modelling the effects of multiple stressors on respiration and microbial biomass in the hyporheic zone using decision...

Integrity of freshwater surface- and groundwater ecosystems and their ecological and qualitative status greatly depends on ecological processes taking place in streambed sediments overgrown by biofilm, in the hyporheic zone (HZ). Little is known about the interactions and effects of multiple stressors on biologically driven processes in the HZ. In this study, machine learning (ML) tools were used to provide evidence-based information on how stressors and ecologically important environmental factors interact and drive ecological processes and microbial biomass. The ML technique of decision trees using the J48 algorithm was applied to build models from a data set of 342 samples collected over three seasons at 24 sites within the catchments of five gravel-bed rivers in north-central Slovenia. Catchment-scale land use data and reach-scale environmental features indicating the HZ morphology and physical and chemical characteristics of water were used as predictive variables, while respiration (R) and microbial respiratory electron transport system activity (ETSA) were used as response variables indicating ecological processes and total protein content (TPC) indicating microbial biomass. Separate models were built for two HZ depths: 5-15 cm and 20-40 cm. The models with R as a response variable have the highest predictive performance (67-89%) showing that R is a good indicator of complex environmental gradients. The ETSA and TPC models were less accurate (42-67%) but still provide valuable ecological information. The best model show that temperature when combined with selected water quality elements is an important predictor of R at depth of 5-15 cm. The ETSA and TPC models show the combined effects of temperature, catchment land use and selected water quality elements on both response variables. Overall, this study provides new knowledge on how ecological processes occurring in the HZ respond to catchment and reach-scale variables, and provides evidence-based information about complex interactions between temperature, catchment land use and water quality. These interactions are highly dependent on the selection of the response variable, i.e., each response variable is influenced by a specific combination of predictive environmental variables.