Development of Hydrological Assessment Model (HAM) Based on A Hybrid Machine Learning Approach

Since various hydrological models have been developed and used for various purposes such as flood forecasting, drought analysis and water resources management, it is essential to identify and improve the model performance. This study addresses development of the Hydrological Assessment Model (HAM) based on a Hybrid Machine Learning Technique (HMLT) to evaluate the performance of a hydrologic model for flood forecasting. The HMLT in this study consists of a clustering module to group four label ratings (very good, good, satisfactory, and unsatisfactory) and a classification module to rate the simulated streamflow as one of the ratings against to the observed. The HAM is able to assess three phases (rising limb, recession limb, and total hydrograph) of an independent hydrograph. This study applied the HAM to the NOAA National Water Model (NWM) in the San Francisco Bay area consisting of 9 counties. The NWM which is a hyper resolution distributed hydrologic model produces a variety of hydrologic analysis and forecast products, including gridded fields of soil moisture, evapotranspiration, snowpack as well as streamflow and velocity for about 2.7 million river reaches in the United States. The simulated and observed streamflow from Oct. 2013 to Jan. 2017 for training and a month, Feb. 2017, for testing are used. The HAM represented the reasonable rating performance for three phase of a hydrograph. This accomplishment in this study is expected to be effective for many water-related agencies have used the NWM outputs for various purposes.