Abstract

Accurate soil property mapping is crucial for precision agriculture, environmental monitoring, and sustainable land management. This study presents an innovative approach combining artificial intelligence (AI) techniques with multi-source remote sensing data to predict soil properties at high spatial resolution. We integrated Sentinel-2 multispectral imagery, Synthetic Aperture Radar (SAR) data, and digital elevation models (DEM) with machine learning algorithms including Random Forest (RF), Support Vector Machines (SVM), and deep learning models. The methodology was tested across 500 km² of agricultural land, targeting key soil properties: organic carbon content (SOC), clay content, pH, and moisture levels. Results demonstrated that the ensemble deep learning approach achieved R² values of 0.89 for SOC, 0.85 for clay content, 0.83 for pH, and 0.91 for moisture prediction. The integration of multi-temporal remote sensing data improved prediction accuracy by 23% compared to single-date imagery. This research provides a scalable framework for high-resolution soil mapping, supporting precision agriculture applications and sustainable land management decisions.