Graduate Thesis Or Dissertation


Understanding Feature Importance in Streamflow Prediction Using Machine Learning in Snow-Dominated Basins Public Deposited
  • Reliable prediction of streamflow is vital to ensuring a community’s access to water. Traditional prediction methods often rely on regression models and both station measured snow water equivalent (SWE) and precipitation accumulation. However, climate change is expected to negatively affect SWE’s ability to accurately predict streamflow. With an increased emphasis on the western United States water availability, the goal of this study was to explore input data, or features, and machine learning models that could potentially aid streamflow prediction in the future. For five case study sites, multiple linear regression (MLR), multivariate adaptive regression splines (MARS), and random forest (RF) machine learning models are used in conjunction with station-based, meteorological, and climate input data to examine the make-up of high-performing models. Every possible combination of the features is used to build individual models for each model type (MLR, MARS, and RF). The performance of each model is then judged using the Nash and Sutcliffe coefficient of efficiency (NSE) and ranked against one another for each basin. Results suggest that having a diverse selection of features arising from physical station-based data, large-scale metrological data, and climate indices will produce better models than relying solely on station-based features alone. Rather than individual features standing out as powerful predictors, the representation of all three feature categories (station-based, meteorological, and climate) appeared most important to high-performing models. Each basin’s top performing models had differing feature sets and model types. Overall, this study presents an approach to feature selection in hydrological machine learning modeling that provides improved understanding and assists in the streamflow prediction, benefiting water managers as well as assisting in the foundation of future studies attempting to improve the results of streamflow prediction.
Date Issued
  • 2022-06-29
Academic Affiliation
Committee Member
Degree Grantor
Commencement Year
Last Modified
  • 2022-09-14
Resource Type
Rights Statement