Using new satellites to assess maize productivity in Tanzania

Principle Investigator
David Lobell, Stanford University
Inbal Becker-Reshef, University of Maryland
Geographic area of interest
Maize is the staple crop in Tanzania, where yields remain well below the yield potential estimated for the predominant soil and weather conditions. Sustainable intensification of these systems will be important for future food security in the region, and will require better knowledge of how yields respond to different management and genetic changes. At the same time, the scope for yield gains is not well understood, because basic information on both the current and potential productivity of the system is lacking. Generating knowledge about current levels of intensification and the pathways and prospects for improvement will require ways to rapidly measure productivity in farmers’ fields. We propose to test a remote sensing based approach to measuring maize yields in Tanzania, building off recent work in Kenya and Uganda. The approach relies on microsatellite data provided by Skybox and Planet Labs sensors, along with algorithms designed to leverage crop simulation models and thus minimize the need for ground calibration. An important component is detailed field work to map field boundaries and obtain ground-based measures of crop type and production as validation.The goal of this project is to test methods to map maize areas and yields within Tanzania, using a variety of sensors that are capable of resolving smallholder fields. This will allow better understanding of yield constraints in the region and the scope for intensification.
  • Develop classification models for crop type, and apply to the available imagery, both within and outside the area of intensive field collection
  • Develop maps of maize yields for areas with available microsatellite imagery. Compare these estimates of productivity with estimates from traditional governmental sources
Key Achievements (last update: Sep 2016)
  • Field campaign successfully completed to conduct whole-field harvests and farmer interviews on 30+ fields
  • Images from various sources obtained and ingested into Google Earth Engine
  • PlanetLab and Skysat images geoereferenced and radiometrically corrected; Cloud masking and shadow removal performed
  • Yield estimation obtained from crop model simulation
  • Whole-field harvests compared to image based vegetation indices to determine the importance of each image