A data-driven approach to predicting the impacts of streamflow disturbance on water quality in river corridors

Abstract:

Floods and droughts are projected to increase over the next few decades due to climate change. These streamflow disturbances can worsen water quality by increasing salt, nutrient, and contaminant concentrations, which will have direct consequences for human and ecosystem health. Hence, it is important to understand and predict how water quality in streams and rivers will respond to new flow- disturbance regimes.

Here we describe results from a U.S. Department of Energy (DOE) early career project that utilizes data-driven approaches to understand how changes in streamflow-disturbance events, ranging from floods to low-flow conditions, impact water quality over time. Our initial study is focused on water temperature and conductivity predictions in the Delaware river, using a combination of deep learning methods with data-driven analyses to infer physical information that can inform model selection and architecture as well as input features. We also describe the application of surrogate models to optimize the hyperparameter choice for the machine-learning algorithms. Finally, we present the use of a data brokering tool BASIN-3D that can be used to reproducibly synthesize diverse data from distributed sources, and assimilate new data into the analytical framework.

The outcome of this work will yield new predictions of watershed water quality characteristics, in response to extreme perturbations. The same framework will also serve as a hub to enable the generation and analysis of integrated interagency watershed datasets in real-time that are transferable to many stakeholders.