Modeling high-resolution precipitation by means of a regional climate model coupled with a machine learning algorithms: Application to Dong Nai-Sai Gon river watershed.

Abstract:

The modelling of large rainfall events play an important role in water resources and floodplain management. Rainfall is a result of complex interactions between climate factors as air moisture, temperature, wind speed, and land surface including topography, soil, and land cover conditions. Therefore, deriving accurate areal rainfall is not only relied on the atmospheric boundary conditions but also reliability and availability of soils, topography, and vegetation. Consequently, the uncertainties of both atmospheric and land surface conditions also result in rainfall model errors. In this study, a blended technique combining dynamical and statistical downscaling has been explored. The proposed downscaling technology uses input provided from three different global reanalysis data including ERA-Interim, ERA20C, and CFSR. These reanalysis atmospheric data are hybrid downscaled by means of the Weather Research and Forecasting (WRF) and followed by the application of an artificial neural network (ANN) model to further downscale the WRF output to a finer resolution over studied watershed. The results of this study suggest that the proposed approach can improve the accuracy of simulated data, since it merges model simulations with observations over the modeled region. Another highlight of this technology is inexpensive computational demand, both in computation times and output storage.