Beyond Backscatter: Leveraging Diverse Sentinel-1 Radar Features for Improved Soil pH Prediction

While the application of Synthetic Aperture Radar (SAR) in soil pH prediction has been explored, its predictive capacity has been underestimated because the radar indicators employed are mostly mathematical combinations of backscatter observations. We combined soil pH observations from the LUCAS 2018 topsoil dataset across the Iberian Peninsula with Sentinel-1 backscatter, texture, and interferometric coherence features to develop random forest models and compared their performance with models based on Sentinel-2 optical data. Coherence values decreased and spatial details diminished as the temporal baseline increased, while shorter baselines provided better modeling performance. Models incorporating multiple temporal baselines outperformed those using a single baseline. Among models using a single type of radar information, coherence-based models incorporating all temporal baselines achieved the highest accuracy (R2 = 0.54–0.55), followed by texture-based models, while backscatter-based models performed the poorest (R2 = 0.46–0.48). Notably, most radar-based models achieved higher accuracy than those using Sentinel-2 optical data alone (R2 = 0.44–0.46). Integrating diverse Sentinel-1 radar features further improved prediction accuracy, achieving R2 = 0.59 for pH in H2O and R2 = 0.58 for pH in CaCl2, while adding Sentinel-2 data only slightly enhanced accuracy. Coherence was the most influential variable group, particularly at short temporal baselines. These results highlight the substantial potential of diverse Sentinel-1 radar information for soil pH mapping.