UAV-mounted electrochemical soil probing system for geospatial phosphorus mapping

• UAV-mounted system enables autonomous in situ soil phosphorus sensing. • Electrochemical probing overcomes limitations of imaging-based nutrient estimation. • Spatially informed machine learning converts sparse samples into coherent P maps. • Field validation across Canada and Tanzania achieved F1-scores up to 0.91. • Framework supports scalable, data-driven fertilizer management systems. Accurate spatial mapping of soil phosphorus remains challenging due to the high cost and limited spatial density of laboratory soil analyses and the poor performance of remote-sensing approaches for phosphorus estimation. This study presents a UAV-mounted soil sensing system that integrates autonomous mechanical probing with in-situ electrochemical measurement and spatial machine learning to enable high-resolution phosphorus mapping under field conditions. The platform combines a lightweight drone-deployable probing mechanism, an ion-selective electrochemical probe, onboard control electronics, and wireless telemetry to collect GPS-referenced soil nutrient measurements during automated flight operations. Field experiments were conducted at two contrasting agricultural sites: a vineyard in British Columbia, Canada, and an intensively managed flower farm in Arusha, Tanzania. Three complementary datasets were generated to evaluate spatial nutrient structure at both location and within-location scales. Spatial feature engineering incorporated positional geometry and neighborhood statistics, and spatial autocorrelation was assessed using Moran’s I and variogram analysis. Random Forest classifiers were trained using spatially grouped cross-validation to prevent performance inflation due to spatial dependence. Across study sites, the proposed approach achieved weighted F1-scores ranging from 0.82 to 0.91, representing improvements of 25–40% over baseline classifiers. The resulting phosphorus maps were spatially coherent and agronomically plausible, successfully identifying low-phosphorus zones in the Canadian vineyard despite limited nutrient variability, and distinct low- and high-phosphorus regions in the Tanzanian field where fertility gradients were stronger. Analysis of feature importance highlighted the dominant role of local spatial contrast and neighborhood information, underscoring the importance of explicitly modeling spatial context when using in situ electrochemical sensors. The results demonstrate that coupling UAV-based automated soil probing with spatially informed machine learning provides an effective pathway for transforming sparse physical measurements into continuous phosphorus distribution maps. This integrated sensing-to-mapping framework offers a scalable foundation for site-specific nutrient management and can be extended to additional soil properties and precision agriculture applications.