Authors: Jacob Anderson, Daniel P. Ames – Brigham Young University
Title: Streamlining Streamflow: Post-Processing National Water Model Long-Range Forecasts with Machine Learning
Abstract: Correcting bias in streamflow forecasts is crucial for better flood mitigation. Many hydrologic models exist for runoff simulation, yet errors persist, leading to uncertainty in flood warning systems. Bias correction is essential to enhance model predictive performance. We used machine learning techniques, specifically random forest regression, to correct bias in streamflow predictions from the National Water Model (NWM) against U.S. Geological Survey (USGS) gauge station observed measurements of actual streamflow. We assessed the effectiveness of the bias correction technique through error metrics including RMSE to gauge improvement. Post-processing NWM forecasts through bias correction will elevate accuracy and trust in the NWM, amplifying its efficacy in flood management efforts.