Abstract

Soil mapping is critical for precision agriculture, environmental monitoring, and land management, but traditional methods are labor-intensive and limited in scale. This study explores the integration of hyperspectral, Light Detection and Ranging (LiDAR), and Synthetic Aperture Radar (SAR) data using artificial intelligence (AI) to enhance soil mapping accuracy in a 500-hectare agricultural site in Iowa, USA. A Multimodal Transformer (MMT) model was employed to fuse data from Sentinel-2 hyperspectral imagery, LiDAR-derived topographic metrics, and Sentinel-1 SAR data, predicting soil properties such as organic carbon (SOC), pH, clay content, and available nitrogen (N). The model achieved an R² of 0.92 for SOC and 0.87 for pH, outperforming single-sensor models by 15–20%. Attention weight analysis revealed key contributions from Short-Wave Infrared (SWIR) bands and topographic wetness index (TWI). The results demonstrate that AI-driven multimodal data fusion significantly improves soil mapping precision, offering scalable solutions for sustainable land use.