Abstract:
Food security is an issue of global concern; this has mandated research on the development of systems for monitoring of agriculture using cost effective techniques such as remote sensing. Smallholder maize farms are dominant in Africa; they produce 80% of the maize in the region. The majority of the African population lives in rural areas and their livelihoods are dependent on smallholder agriculture particularly maize production. Thus, smallholder maize production plays a vital role in combating food insecurity in rural areas. Targeting food insecurity in developing countries is one of the important objectives of the Sustainable Development Goals (SDGs). However, local planning agencies and governments do not have adequate spatial information on smallholder farmers, and this affects the monitoring of the SDGs. Additionally, these farmers are faced with economic and environmental constraints that limit their productivity. Furthermore, the estimates of total planted area are unknown in most developing countries. Techniques for undertaking such estimates are either absent or very unreliable. This study explores the use of Sentinel-1 and Sentinel-2 data products for mapping and monitoring smallholder farms with machine learning. Findings suggest that the multi-temporal approach with the application of support vector machine and extreme gradient boosting is the recommended method for mapping smallholder maize farms in comparison to single date imagery based on lower standard deviation errors. The random forest model was suitable for estimating soil nitrogen. Furthermore, the findings suggest that maize yields can be accurately predicted from two months before harvest. The frameworks developed in this study can be used to generate spatial agricultural information in areas where agricultural survey data are limited. We recommend the use of Sentinel-1 and Sentinel-2 in conjunction with machine learning algorithms to map smallholder maize farms to support the SDGs.
