Abstract

Soil mapping is a cornerstone of precision agriculture, environmental management, and sustainable land use. Traditional methods, reliant on physical sampling, are labor-intensive and struggle to capture the three-dimensional variability of soil properties. This article explores the use of 3D Convolutional Neural Networks (3D-CNNs) to generate high-resolution 3D soil maps for properties like texture, organic matter, and moisture content. Using geophysical data from ground-penetrating radar (GPR) and electrical resistivity tomography (ERT), combined with soil samples from a 100-hectare agricultural field, we trained a 3D-CNN model to predict subsurface soil characteristics. The model achieved a predictive accuracy of 92% for soil texture and 88% for organic matter content. These results underscore the potential of 3D-CNNs to provide scalable, non-invasive soil mapping solutions. Challenges such as computational demands and data quality are discussed, alongside future prospects for integrating 3D-CNNs with advanced sensing technologies.