High frequency monitoring

Abstract

RTI International built upon its crop modeling work in Rwanda (Chew et al., 2020; Hegarty-Craver et al., 2020) to apply the use of drones (also known as unmanned aerial vehicles or UAVs) to generate crop type labels in the Konni region of Niger and use them to develop machine learning models that predict crop types. Drone data offer the ability to see and label crops at high resolution (~2 cm) at multiple intervals during a given growing season, providing a potential alternative to in-person field data collection. RTI captured and used the drone data for the area of investigation to create large crop label datasets containing multiple crop examples at specific locations and times. These labeled data in turn provided the training data for satellite-based crop type models that can be applied over large areas. These models can be run multiple times during a compact as part of a project’s monitoring and evaluation strategy so that a shift in crop types and extents can be detected more quickly than with traditional survey-based methods, allowing for program intervention and recalibration if needed.