Project in collaboration with Dr. Zhou Zhang in the Department of Biological Systems Engineering.
Future land use and climate changes will intensify demands for and challenges on water quality simulation and prediction. The effects of land use on nutrient inputs to surface waters are highly variable, showing the complex interactions among watershed characteristics and water quality, with uncertainties in model predictions. Statistical models built decades ago without consideration of climate change might not be able to predict water quality today or in the near future because they were highly reliant on limited data. Physics-guided models will be more applicable for forecasting compared to statistical models because they can be self-updated with more recent or projected meteorological data. However, physical models take a longer time to calibrate and validate with reliable estimation of parameters and the corresponding results might not be accurately simulated or predicted. Many studies have been conducted on the use of machine-learning models to increase prediction accuracy, but a potential weakness is the interpretation or causality of relationships between inputs and the system variables of interests.
We aim to develop a coupled physics-guided and machine-learning nutrient loading model that has high accuracy and will be able to inform land managers and water-related stakeholders to sustainably plan future land use and protect water quality. We expect this approach to be advantageous over a single statistical/machine-learning model or physical model which are limited in their transfer ability across time and regions and their time-consuming low-accuracy model-building process, respectively.