Mohammed Moayad, Qahtan Ahmed, M. Shareef
2026.1.31Iraqi Geological Journal
tlooto Summary
This framework provides a foundation for soil monitoring and precision agriculture in arid and semi-arid regions, with spectrum datasets producing the greatest results.
Abstract
This study evaluates Sentinel-2A multispectral imaging and machine learning techniques to predict 17 semi-arid soil physicochemical characteristics. A thorough framework was created to assess ANN, RF, and Linear Regression models using 21 sample sites in Baqubah, Iraq. To improve prediction accuracy and eliminate data redundancy, 13 Sentinel-2 spectral bands and derived indices were studied across five feature combinations. The ANN models performed well, achieving perfect R² scores (1.000) for five chemical parameters (gypsum concentration, sulfate content, total dissolved solids, organic matter, and pH using the 13-band spectral dataset. Although physical properties showed some reduction, they still performed well, with R² values between 0.973 and 1.000. Forest models achieved intermediate performance, with an average R² of 0.851 for physical characteristics. Linear regression often performed poorly, confirming the non-linear nature of soil–spectral interactions. The Bare Soil Index (BSI) was the most discriminative feature for salt-related properties (importance scores 0.368-0.391), while the coastal aerosol band (B1) was sensitive to total dissolved solids (0.656) and specific gravity (0.374). The Iron Oxide Ratio showed the highest predictive importance (0.293). Chemical properties outperformed physical properties in all modeling methods, with spectrum datasets producing the greatest results. This framework provides a foundation for soil monitoring and precision agriculture in arid and semi-arid regions.
Citation format
MOAYAD, Mohammed; AHMED, Qahtan; SHAREEF, M. A comparative analysis of ANN, random forest, and linear models for predicting soil physicochemical properties using sentinel-2a spectral indices. Iraqi Geological Journal, 2026: 255–273.