Mapping cropland types in heterogeneous arid environments using machine learning algorithms and dataset variations on Google Earth Engine

Monitoring crops ensures global, regional, and national food security. It entails gathering information such as the health and growth of crops and environmental conditions. This aids farmers in making well-informed decisions, enhancing productivity, and reducing environmental impact. This, in turn, leads to improved economic outcomes and long-term agricultural sustainability. In arid and semi-arid lands, effective crop monitoring is particularly critical due to the limiting factor of water availability. This study used Sentinel-2 (S2) image collection in Google Earth Engine (GEE) to mapping crop types in North Darfur State, Sudan during the 2022 growing season (July 1 to September 30, 2022), using support vector machine (SVM) and random forest (RF) classifiers. Eight vegetation indices (VIs), i.e., normalised difference vegetation index (NDVI), enhanced vegetation index (EVI), soil-adjusted vegetation index (SAVI), green-normalised difference vegetation index (GNDVI, weighted Difference vegetation index (WDVI), red edge NDVI (NDVIre), ratio-vegetation index (RVI), and normalised difference infrared index (NDII) were used as additional bands. The results show that the RF models produced an overall accuracy (OA) of 90-97% with a kappa coefficient (κ) of 0.87-0.96. The SVM models reported OA and κ values in the range of 84-95% and 0.81-0.94, respectively. Producer’s (PA) and user’s accuracies (UA) were in the range 83-97% and 81-98%, respectively. Highest F1 scores for both classifiers were 0.98. The findings of this study, along with the derived classification maps, would enable farmers, policymakers, and other stakeholders to make well-informed decisions regarding agricultural production, land use planning, and resource management in North Darfur and arid environments.