Modeling the spatial distribution of soil nitrogen content at smallholder maize farms using machine learning regression and sentinel-2 data

Abstract

Nitrogen is one of the key nutrients that indicate soil quality and an important component for plant development. Accurate knowledge and management of soil nitrogen is crucial for food security in rural communities, especially for smallholder maize farms. However, less research has been done on generating digital soil nitrogen maps for these farmers. This study examines the utility of Sentinel-2 satellite data and environmental variables to map soil nitrogen at smallholder maize farms. Three machine learning algorithms—random forest (RF), gradient boosting (GB), and extreme gradient boosting (XG) were investigated for this purpose. The findings indicate that the RF (R2 = 0.90, RMSE = 0.0076%) model performs slightly better than the GB (R2 = 0.88, RMSE = 0.0083%) and XG (R2 = 0.89, RMSE = 0.0077%) models. Furthermore, the variable importance measure showed that the Sentinel-2 bands, particularly the red and red-edge bands, have a superior performance in comparison to the environmental variables and soil indices. The digital maps generated in this study show the high capability of Sentinel-2 satellite data to generate accurate nitrogen content maps with the application of machine learning. The developed framework can be implemented to map the spatial pattern of soil nitrogen. This will also contribute to soil fertility interventions and nitrogen fertilization management to improve food security in rural communities. This application contributes to Sustainable Development Goal number 2.