New process- or knowledge-guided machine learning (ML) techniques can predict flow and temperature in river networks more accurately even when data is scarce, according to researchers at the University of Minnesota, the University of Pittsburgh (Pitt), and the U.S. Geological Survey.
The work involved an algorithm that was taught physical rules to generate more accurate forecasts and identify physically significant relationships between inputs and outputs.
The method was designed to avoid common traps in ML-based prediction by informing the model through correlation across time, spatial links between streams, and energy budget equations.
Pitt's Xiaowei Jia said, "Accurate prediction of water temperature and streamflow [can assist in] decision making for resource managers, for example helping them to determine when and how much water to release from reservoirs to downstream rivers."
From University of Minnesota News and Events
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA