Comparison of spatio-temporal low-flow models for predicting remobilization of water pollutants

Droughts are significant hydrological and environmental hazards that threaten the ecological functioning of water bodies. Low flow with increased water temperature leads to a cascade of hydrochemical processes. This is a particular cause of concern for regions like eastern Austria, where agricultural land use and the projected risk of low flows and increased water temperatures due to climate change are particularly high. Under these scenarios, nutrient release from river sediments may become the dominant factor for the water quality of Iotic ecosystems. The role of this remobilization-potential for water quality is assessed in the project DIRT based on a combination of laboratory experiments with at-site water quality monitoring and regionalized streamflow observations.

Here we focus on space-time models of low flow and stream temperature, which are crucial for upscaling the remobilization potential along the river network. We present a study that compares different models for spatio-temporal low flow regionalization at the monthly scale in eastern Austria.

We evaluate three different statistical models: (i) a tree-based boosting model, (ii) a simple linear regression model with 3-way interactions, and (iii) a combination of a non-linear boosting approach and Top-kriging. Our results show a very high performance for all models, with an overall R² of 0.88 and a median R² of 0.70. The best performance is reached by the combination of Top-kriging with a non-linear boosting approach. However, accuracy of the model is somewhat lower in headwater gauges, whereas non-headwater catchments are even better modeled by a simple spatio-temporal Top-kriging approach. In a next step, the model shall be integrated with laboratory experiments and water-quality monitoring to develop space-time models that can predict the remobilization of pollutants from river sediments.