Spatial modelling of soil mechanical-hydraulic behavior for precision agriculture

Within-field soil management in Precision Agriculture is still largely based on scalar observations or on geostatistical modelling of individual curve parameters, despite the inherently functional nature of many soil properties. This study proposes a geostatistical framework for the spatial modelling and zoning of function-valued soil descriptors, explicitly accounting for data heterogeneity, spatial non-stationarity, and multiscale variability. Two agronomically relevant functional responses are considered: soil penetration resistance as a function of soil moisture and soil water retention as a function of matric suction, represented through the Stock-Downes and van Genuchten models, respectively. The proposed workflow combines polygon-based estimation of spatially varying local means with multivariate geostatistical modelling of functional parameters and factorial cokriging, allowing joint mechanical-hydraulic behavior to be analyzed across spatial scales. Applied to a 200-ha agricultural field with 100 sampling locations, the approach enabled the reconstruction of complete functional responses and the assessment of their uncertainty at unsampled points as well as the extraction of scale-dependent regionalized factors synthesizing the joint soil hydraulic and mechanical behavior. These factors supported a multiscale partition of the field into zones with contrasting soil conditions, revealing spatial patterns that are not directly accessible through scalar-based or single-scale approaches. Rather than competing with simpler methods on scalar prediction accuracy, the proposed framework addresses a complementary decision-support problem: the interpretation and zoning of functional soil behavior under non-stationary conditions.