A novel Improved Geographically Weighted Random Forest (IGWRF) model for low-resolution soil moisture data downscaling in Africa

Soil moisture (SM) products such as SMAP have a coarse spatial resolution, which limits their applicability in agricultural management and drought monitoring. Downscaling techniques can overcome this limitation by enhancing the spatial resolution of SM data. However, mainstream methods rarely address both spatial heterogeneity and the high-dimensional nonlinear coupling between SM and environmental factors. To address this gap, this study proposes an Improved Geographically Weighted Random Forest (IGWRF) model that integrates local adaptation and global generalization for high-accuracy SM downscaling. This study focuses on Kenya in East Africa, where we downscale the 9 km SMAP SM product to 1 km and evaluate IGWRF against traditional RF and GWRF using in-situ measurements. The results show that: (1) all methods produced results strongly correlated with the original SMAP SM (R > 0.9) and significantly enriched spatial detail and texture; (2) in complex terrain, GWRF and IGWRF achieved higher accuracy than RF, with IGWRF showing the greatest consistency with in-situ measurements (R = 0.771); (3) relative to RF, IGWRF increased R by 4.5 % and reduced RMSE and ubRMSE by 7 % and 5.8 %, respectively, demonstrating the superior performance of IGWRF. The study confirms that the IGWRF effectively captures the spatial heterogeneity of SM and addresses the challenge of jointly modeling spatial heterogeneity and nonlinear relationships in SM downscaling, significantly improving the accuracy of downscaled results. This research provides high-resolution (1 km) SM data to support agricultural decision-making and water resource management in drought-prone regions of Africa, filling a critical gap in fine-scale SM products across the continent.