Multi-model and Multi-watershed Integration in Flood Forecasting System

In operational flood forecasting, it is essential to have accurate detections of the flood risk, reliable forecasts of upcoming flood hazards, timely early warnings for assuring proper responses to protect lives and properties. Despite the advances and potentials of emerging techniques and the computational power available, the traditional rainfall-runoff models have yet played an important role in the operational system. Multi-model integration in flood forecasting system is able to address the gaps between hydrologic models and therefore provide more reliable forecasts. It can also support the various needs of the model selection by operating authorities. In addition, managers of forecasting systems typically cover multiple watersheds such that the multi-watershed configuration of the flood forecasting system is preferred. This study integrates three hydrologic models using the Delft-FEWS platform which is configured for two rural and one urban watersheds in southern Ontario, Canada. In specific, SAC-SMA combined by SNOW17, RAVEN, and WATFLOOD models are adapted to the DELFT-FEWS platform by their respective adapters. Canadian Surface Prediction Archive (CaSPAr) has stored various deterministic and ensemble forecasts since 2017. This study explores forecast/hindcast performance of the integrated forecasting system using High-Resolution Deterministic Prediction System (HRDPS), of which spatial resolution is about 2.5km and temporal resolution is four predictions per day. Deterministic hindcast applications of the three hydrologic models focused on the recent flood events will be compared and discussed. The individual model performance and the potential of the model ensemble and the multi-model ensemble forecast will also be discussed.