Streamflow Prediction in Ungauged Streams Using Random Forest

Abstract:

Streamflow gauges are commonly installed in major streams covering a relatively small fraction of the watershed. As a result, most, especially lower order, streams, are not gauged. This study aims to predict streamflow at ungauged streams at three different Hydrologic Unit Code (HUC) levels, including Basin (HUC-6), Subbasin (HUC-8), and Watershed (HUC-10), using Random Forest. Random Forest provides a reliable way to estimate streamflow in ungauged streams by including time feature and HUC level in the nodes of decision trees. The application of this approach is tested in the Wabash River Basin and Patoka-White Basin in Indiana and Illinois, respectively. Prediction of streamflow in ungauged basin at HUC-10 level using Random Forest is found to be acceptable, and shows potential to predict streamflow in HUC-6 level ungauged basins.