Mapping soil organic matter contents at field level with Cubist, Random Forest and kriging

Abstract

The soil organic matter (SOM) content is strongly related to soil fertility and greenhouse gas emissions. Knowledge of the soil SOM content is therefore necessary for efficient and sustainable management practices. In this study, we compare the performance of five machine learning techniques for the prediction of SOM contents using remote sensing, proximal soil sensors and topographic data as environmental predictor. The methods used were kriging, Cubist, Random Forest and regression-kriging (Cubist- kriging and Random Forest kriging). 285 soil samples were collected from a 10 ha field in Denmark. 75% of the data were used for training and 25% for validation. NDVI (Normalized Difference Vegetation Index), RVI (Ratio Vegetation Index), DVI (Difference Vegetation Index) and SAVI (Soil-Adjusted Vegetation Index) were derived from Sentinel 2 images with 10 m resolution as auxiliary variables. Topographic variables were derived from a digital elevation model (DEM) with 1.6 m resolution. The performance of the methods was compared based on the coefficient of determination (R2) and root mean square error (RMSE). The results showed that kriging achieved the best performance, followed by regression-kriging. The models using only Cubist and Random forest had the poorest performance. The results, therefore, demonstrate that kriging can predict SOM contents without the need of auxiliary variables for fields with high sampling densities.