Machine Learning-Based Evaluation of Seasonal and Spatial Variations in Soil Quality Across Agroecological Zones of Lagos State, Nigeria

This study assesses seasonal and spatial variations in soil quality across four agroecological zones (Badagry, Epe, Ikorodu, and Ojo) in Lagos State, Nigeria, using physico-chemical soil properties and machine learning techniques. Samples of soil were gathered in both the rainy and dry seasons at 0–15 cm and 15.1–30 cm depths, were analyzed for physico-chemical parameters, including pH, electrical conductivity (EC), organic carbon (OC), nitrogen (N), cation exchange capacity (CEC), and base saturation (BS). The Soil Quality Index (SQI) was computed employing a weighted scoring method based on fertility indicators. EC was higher in the rainy season, while pH decreased, indicating leaching effects. Organic matter and nutrient levels showed moderate seasonal and depth-related variations, with Ojo exhibiting the highest OC content. A t-test found no significant difference in SQI between seasons (t = 0.76, p = 0.448), suggesting temporal stability in soil quality. Among machine learning models, Random Forest achieved the highest predictive accuracy for SQI (R² = 0.94, MSE = 0.0003), outperforming Partial Least Squares Regression and Cubist models. Cross-validation (Root Mean Square Error (RMSE) = 0.015–0.017) supported targeted soil management strategies, identifying Ojo as a high-fertility zone (mean SQI = 0.45). These findings highlight the efficacy of integrating machine learning approaches for soil quality assessment, offering insightful information on soil management and sustainable agricultural engagement in Lagos State’s diverse agroecological zone.