Improved estimation of surface soil moisture based on soil properties and dual-satellite spectral fusion

Soil moisture (SM) deficiency presents a significant challenge, highlighting the need for improved estimation methods. This study aims to enhance the accuracy of SM content prediction by integrating remote sensing-derived spectral indices into pedotransfer functions (PTFs). Surface soil samples were collected from 100 sites across three regions in China, and key soil physical properties were measured. Multispectral satellite images from Landsat 8 and Sentinel-2 were fused using the Gram-Schmidt (G-S) algorithm to generate enhanced composite datasets. From these, various spectral indices were derived, including soil spectral indices (SSIs), vegetation spectral indices (VSIs), and moisture spectral indices (MSIs). PTFs were developed in four stages: (1) using only readily measurable soil properties (RM-SPs) (PTF1), (2) combining RM-SPs with SSIs (PTF2-PTF6), (3) combining RM-SPs with VSIs (PTF7-PTF11), and (4) combining RM-SPs with MSIs (PTF12-PTF14). The baseline model (PTF1), based on bulk density (BD), clay content, and the silt-to-sand ratio, showed limited predictive accuracy (R2 = 0.22, RMSE = 8.57 cm3. cm⁻3, MAE = 7.38 cm3. cm⁻3). Including spectral indices significantly improved model performance. The model combining RM-SPs and MSIs (PTF13) achieved the highest accuracy (R2 = 0.89, RMSE = 3.28 cm3. cm⁻3, MAE = 2.29 cm3. cm⁻3), outperforming the models based on RM-SPs with SSIs (PTF6; R2 = 0.88) and VSIs (PTF11; R2 = 0.69). These findings underscore the effectiveness of integrating RM-SPs with spectral indices derived from G-S fused multispectral datasets, substantially improving the precision of SM estimation.