A Hybrid Approach for Soil Nutrient Estimation using Multispectral Data in Belgium and Luxembourg

Planning for agricultural production and sustainable land management depend on accurately measuring soil nutrient levels. Leveraging machine learning to estimate soil nutrients simplifies farming and land management decision-making. This paper presents a novel hybrid model to forecast soil nutrient concentrations in the Belgium and Luxembourg region by combining Partial least squares regression (PLSR) with Gaussian process regression (GPR). An integrated approach is provided for predicting various soil nutrients, utilizing the Lucas topsoil dataset and multispectral Landsat 8 satellite images, extending the model’s applicability to numerous agricultural and environmental applications. The hybrid model has undergone evaluation, producing RMSE and RPD values within acceptable ranges for various soil parameters, such as Soil Organic Carbon (SOC), Electrical Conductivity (EC), Nitrogen (N), Potassium (K), Phosphorus (P), pH in Water (H2O), pH in Calcium Chloride (CaCl2), and pH in Calcium Carbonate (CaCO3).