Estimate crop type distribution in South Africa using Google Earth Engine cloud computing

The accurate location-specific information of the crop area, yield and production is critical for food security planning especially in developing countries where the living standards of the majority population largely rely on agriculture practices. Meanwhile, this type of information collected from sample or household surveys lacks spatial components and is often sparse, costly and dated. Alternatively, spatiotemporal remotely sensed data especially vegetation index have been widely applied in land cover and crop type mapping in U.S. and Europe where land parcels are relatively large. The crop type mapping in the context of Africa remains a challenge due to the small plot size and mixed farming practice. In this research, we develop a machine learning application to map crop types of smallholders in South Africa using Google Earth Engine (GEE) cloud computing with spatial-temporal remotely sensed datasets. The model can successfully map major crop types and crop combinations in the region. The validation shows that R-squared is over 0.7 when characterizing crop types in the region.