Machine learning‐based statistical downscaling of wind resource maps using multi‐resolution topographical data

The need for high-resolution wind resource maps is increasing with the increase in supply and development of power. Many physical downscaling models have been developed applied to make these maps. However, as existing require extensive computations time, statistical higher efficiency are being studied. Statistical such regression machine learning can quickly calculate maps, but they a problem low accuracy. This study proposes model new topography-derived variables interpret characteristics wind. As shape topography, which was unable be interpreted previous studies, considered derived variables, significant performance improvement identified. analysis conducted using 1 km Weather Research Forecasting (WRF) results ERA5 reanalysis data from South Korea. Two Weibull distribution parameter were calculated used input output data. Three collections devised compared. Therefore, multi-resolution topography showed highest improvements approximately 15% reduction root mean square error (RMSE) both linear models. In particular, land area decrease 20%. best proposed an RMSE 7% 8% two parameters. expected serve reference continuing research utilization