Field-based crop cut experiment data is a gold standard of ground truth data for crop analytics, yet it is time-consuming and challenging to scale over large areas. Satellite remote sensing-based estimates can cover a large area, yet their accuracy is highly subject to the availability and quality of ground truth data. To address these two interlinked challenges, IFPRI, University of Twente/ITC, ICRISAT, and aWhere developed a two-stage pilot project and applied it in Odisha, India.
This webinar will present a new two-stage crop production analytics approach piloted in this project that leverages cutting-edge satellite remote sensing and geo-statistical techniques to address the dual issues of inefficient ground-truth sampling design and inaccurate in-field crop yield measurement methods.
The project team piloted spatially detailed weather information to cue field data collection over areas of high in-season variability, analyzed using long-duration temporal NDVI profiles. At the field level, a smartphone-based 3D imaging technique was used to expedite the collection of crop yield measurements without cutting crops. These data will be used along with photos of the crop to train a deep-learning model to estimate yield, which can then be bootstrapped for use in smartphones.
Link to the project website and reports: [ Ссылка ]
