A geospatial framework for soil characterization: Integrating multitemporal remote sensing with advanced statistical methods

The possibility to create accurate and functional soil maps through Digital Soil Mapping (DSM) techniques are increasingly widespread, thanks to the recent availability of adequate remotely sensed imagery. In this study, multispectral data from Sentinel-2 satellite images and chemical-physical soil analyses were integrated to characterize the soil properties of two experimental areas located in Emilia Romagna (3,622 ha, Northern Italy) and Toscana (1,082 ha, Center Italy). The methodology included topsoil sampling at 384 locations, multi-temporal image processing to generate a Synthetic Soil Image (SYSI) via the GEOS3 technique and applying spectral indices to highlight bare soil. Principal Component Analysis (PCA) was used to reduce data dimensionality, followed by cluster optimization to delineate homogeneous zones based on soil texture, Soil Organic Carbon (SOC) content, and macro-elements (NPK). The results revealed significant spatial variability, with a decreasing trend in clay percentage and SOC and an increase in sand fraction moving from the first to the third zone in both study areas. Geographically Weighted Regression (GWR) models showed good performance for clay (R2= 0.77 and 0.88; RPD = 1.59 and 2.72 for Area 1 and Area 2, respectively), SOC (R2= 0.71 and 0.64; RPD = 1.59 and 1.50), and nitrogen (R2 = 0.72 and 0.66; RPD = 1.61 and 1.54). At the same time, predictive power for phosphorus and potassium was less reliable. In conclusion, this study highlights the potential of integrating satellite data with advanced statistical techniques to generate high-resolution soil maps, thereby supporting precision farming practices.