Towards an Accurate and Reliable Downscaling Scheme for High-Spatial-Resolution Precipitation Data

Accurate high-spatial-resolution precipitation is significantly important in hydrological and meteorological modelling, especially rain-gauge-sparse areas. Some methods strategies have been applied for satellite-based downscaling, residual correction calibration. However, which downscaling scheme can provide reliable high-resolution efficiently remains unanswered. To address this issue, study aimed to present a framework combining the machine learning algorithm post-process procedures. Firstly, four ML-based models, namely support vector regression, random forest, spatial forest (SRF) eXtreme gradient boosting (XGBoost), were tested compared with conventional methods. Then, effectiveness of process using ordinary Kriging calibration geographical difference analysis (GDA) method was investigated. The results showed that had better performance than regression interpolation approaches. SRF XGBoost outperformed others generating accurate estimation high resolution. GDA improved downscaled results. decreased models. Combining or could be promising data. used generate precipitation, areas urgently requiring data, would benefit regional water resource management disaster prevention.