Machine Learning Models in Estimation and Mapping Soil Phosphorus and Potassium

Abstract The objective of this research is to compare utilizing remote sensing techniques to field data at the Gonbad Kavous area. To quantify of two nutrients, namely, available phosphorus (Pav) and exchangeable potassium (Kex), soil samples were obtained from the soil surface (0–15 cm depth). Four machine learning techniques, including Random Forest (RF), support vector machine, Boosted Regression Trees (BRT) and Generalized linear model (GLM) were utilized to predict the soil organic carbon map using Sentinel 2. Comparison of different predictive models showed that Random Forest (RF) and Support Vector Machine (SVM) models are efficient machine learning models in estimating and digital maps of two nutrients, namely, (Pav) and (Kex). While Random Forest showed the lowest error, SVM also performed well in capturing the variability of in the study area. The findings of this study can help decision-makers and policy makers to better understand the spatial distribution of in the study area and to develop appropriate strategies for soil management and conservation.