Predicting Crop Loss Remotely Through Satellite Imagery and Drones in Tanzania

Advances in satellite imagery have brought with them increased expectations for their use in early warning systems that could inform policy responses. Use in this way would be dependent upon the ability to predict crop failure and loss and as a result, threats to food security. Despite the value these predictions could have, their development faces a number of obstacles including the heterogeneity of crops in developing countries as well as cloud cover that can obfuscate satellite imagery. In an effort to address these challenges, the authors worked with local drone operators to map farm boundaries and crops in the Chemba district of Dodoma region, Tanzania. The study collected information from 300 producers and 600 plots of maize and beans using two drone flights (just after sprouting and just before harvesting) as well as producer-reported yield and the Green Chlorophyll Vegetation Index (GCVI) derived from harmonized Landsat and Sentinel-2 datasets. Our results demonstrate the utility of medium resolution satellite data in differentiating between low and high yielding fields and the challenges of accurately forecasting in-season yields for smallholder farmers.