Soil Organic Matter Prediction with Sentinel-2 and Geostatistical Methods

Abstract

Predicting soil organic matter (SOM) using advanced tools like Sentinel-2 satellite imagery and geostatistical methods can help achieve Sustainable Development Goals related to food security and climate change. Spectral capabilities of Sentinel-2 allows the characterization of soil properties from its reflectance patterns, enabling indirect SOM estimation. This is imperative for countries like South Africa with limited SOM data as some areas are reported to have low SOM levels, and Okhahlamba Local Municipality was chosen as a testing site in this study. Here, we used Sentinel-2 data with geostatistical methods Ordinary Kriging (OK) and Simple Kriging (SK) and hybrid geostatistical methods Regression Ordinary Kriging (ROK) and Regression Simple Kriging (RSK) to predict SOM. Hybrid methods performed better than ordinary methods. Specifically, ROK and RSK using spectral bands had the highest R² and lowest errors, followed by these methods using principal components. OK and SK showed the lowest R² and highest errors. Auxiliary information boosted predictive performance. Overall, ROK and RSK accounted for 63% of predicted SOM variability using spectral bands. SOM plays a crucial soil role and should be sustainably managed for to improve food security and for future generations. The combined use of Sentinel-2 data and geostatistical methods has great potential for predicting SOM to achieve the sustainable development agenda.