Building Earth Observation Based Systems for Grain forecasting and Agricultural Outlooks in Eastern and Southern Africa

Abstract:

We report on the status of building an end-to-end system for supporting earth-observation based grain yield forecasts and agricultural outlooks in Eastern and Southern Africa. We outline strategies for data ingest, modeling, validation, and reporting, as well as challenges to implementation. We frame our example in the context of responding to a simulated weather shock and the impacts of that shock on sub-national grain-production. Focusing first on data requirements we provide an overview of optimal earth observation products and rationale for which products were chosen for this effort. We then compare modelling approaches: examining traditional regression, machine learning, and deterministic modelling approaches as well as different approaches to model validation. Finally we examine different approaches for reporting and tool development as they pertain to different end users.