Improving Digital Mapping of Soil Organic Carbon and Total Nitrogen Through Integration of Sentinel‐1 Backscatter, <scp>InSAR</scp> Coherence, and Sentinel‐3 <scp>OLCI</scp> / <scp>SLSTR</scp> Observations

ABSTRACT The potential of optical satellites for the spatial prediction of soil organic carbon (SOC) and total nitrogen (STN) has long been recognized. Owing to their balanced spatial–temporal resolution, multisource observational capabilities, and unprecedented availability of free data, Sentinel satellites offer unique opportunities for mapping soil properties. Despite these opportunities, there is still a lack of studies systematically evaluating the complementary contributions of Sentinel‐1 backscatter/coherence and Sentinel‐3 optical/thermal data for SOC and STN mapping. This work investigated the synergistic use of Sentinel‐1 backscatter and coherence, as well as Sentinel‐3 optical (OLCI) and thermal (SLSTR) data, as tools for the spatial prediction of SOC and STN in Spain. Various scenarios constructed from diverse satellite‐derived information from Sentinel‐1 and Sentinel‐3 data were employed to construct random forest‐based soil prediction models using 3867 LUCAS 2018 soil samples. The results indicated that both coherence and backscatter observations from Sentinel‐1 present high potential for SOC and STN prediction, with coherence exhibiting greater predictive capability than backscatter. Their joint use significantly increased the predictive accuracy ( R 2 = 0.44 for SOC and R 2 = 0.42 for STN). Strategies for utilizing backscatter observations substantially influence modeling performance, with cross‐polarization data outperforming copolarization data, whereas backscatter observations from ascending and descending orbits exhibit similar predictive performance. In terms of independent modeling accuracy, the data were ranked as follows: OLCI &gt; coherence &gt; backscatter &gt; SLSTR. Variable importance analysis revealed that coherence/backscatter observations exerted comparable or even greater influence than OLCI/SLSTR data did. The complementary information provided by Sentinel‐1 and Sentinel‐3 observations substantially increased the prediction accuracy, achieving the best results, with R 2 values of 0.52 for SOC and 0.47 for STN. The digital soil maps produced by independently or synergistically applying various remote sensing products from Sentinel‐1 and Sentinel‐3 displayed spatial patterns that were consistent with those generated from extensive soil‐forming factor data in previous studies. Our findings emphasize the value of the diverse information from Sentinel‐1/3, particularly coherence, which has been minimally documented in previous studies.