Improving Precipitation Accuracy: A Rescaling Method for PERSIANN Using NDVI, LST, and DEM Data

This study is based on the rain gauge data from Torbat-e Jam over a 23-year period (2001–2023). PERSIANN satellite rainfall data with a spatial resolution of 27 kilometers were enhanced to a 1-kilometer resolution using NDVI, land surface temperature (LST), and digital elevation model (DEM) data, aided by the random forest (RF) algorithm. To evaluate the accuracy of satellite rainfall downscaling compared to ground station data, statistical metrics such as correlation coefficient (CC), root mean square error (RMSE), and mean absolute error (MAE) were utilized. Additionally, a residual correction method was implemented to refine model predictions further. Results demonstrated that integrating spatial datasets with the RF algorithm significantly improved rainfall modeling accuracy. Applying the residual correction method led to substantial improvements in forecasting accuracy across all studied stations on both monthly and annual timescales. On the monthly scale, the correlation coefficient increased by 22-29%, while RMSE and MAE decreased by 61-64% and 60-68%, respectively. On an annual scale, the correlation coefficient showed an increase of 7-35%, with RMSE and MAE reductions of 69-74% and 69-76%, respectively. This study underscores the effectiveness of the applied method in enhancing prediction accuracy across various temporal scales within the studied region. Additionally, the practical implications of this research provide valuable insights for hydrological modeling and water resource management, especially in regions with limited ground station data. The findings of this research can significantly aid in better water resource management and climatic planning, particularly in arid and semi-arid areas.