Arvin Samadi Koucheksaraee (2024)

Project Title: A Novel Hybrid Machine Learning Method for Streamflow Forecasting in Cold Regions

Project ID: Fellow: Arvin Samadi Koucheksaraee

Adviser(s): Xuefeng Chu

Project summary: Machine learning (ML) has been widely applied for hydrologic forecasting, but its effectiveness in snow-dominated watersheds is limited by data quality, predictor selection, and generalization across scales. This study developed a hybrid ML framework (KRR-EN-EINFO) for daily streamflow prediction in cold regions, combining kernel ridge regression, elastic net, and the enhanced weighted mean of vectors, alongside a multi-step preprocessing method using MVMD and Boruta-SHAP for non-stationarity handling and feature selection. Applied to the Forest River watershed in North Dakota with MODIS snow cover and hydroclimatic data, the model outperformed existing ML approaches, especially for 3- and 7-day forecasts.