Towards Real-Time Mapping of Smallholder Maize Yields Using Multiple New Sensors

Remote sensing is an invaluable tool for efficiently tracking crop production and understanding causes of yield losses in farmers' fields. Previous work has used MODIS and Landsat imagery to estimate crop yield, but the limited resolution of these satellites (250m and 30m per pixel, respectively) precluded the exposure of between- and within-field patterns in smallholder-dominated landscapes. In this project we use a satellite-based technique named Scalable Crop Yield Mapper (SCYM) with data collected from various finer resolution sensors, including commercial "cubesats", in Morogoro, Tanzania. Results derived from four sensors - Sentinel-2 MSI (10m), RapidEye (5m), PlanetScope Dove (3m), and Terra Bella SkySat (2m) - were compared with ground data collected from whole-field harvests in 33 maize fields. The results help to demonstrate the ability to map yields in smallholder systems and the relative merits of the different data sources by themselves and in combination.