Field-Scale Yield Estimation Using Remote Sensing Time-Aggregate Variables

The need for reliable ground data increases as frameworks leveraging Earth observations and machine learning, dedicated to monitoring crops and analyzing agricultural risks, advance - especially across Sub-Saharan Africa’s changing farming landscape. Through the Enabling Crops Analytics At Scale (ECAAS) project, we developed and implemented a framework to estimate yield at the end of the 2023 main growing season in Kenya with Ministry of Agriculture Livestock and Fisheries (MoALF) partners. We utilize 502 data samples collected from Trans-Nzoia, Kakamega, Bungoma and Uasin-Gishu counties of Kenya to train machine learning models for crop yield prediction across time and geography in Kenya. A Random Forest model generated yield predictions with an RMSE (Root mean squared error) value of 0.203 mg/ha. Preliminary analysis focused on reliable model estimation through feature engineering and shows potential of EO / ML to scale yield estimation across Kenya, leading to timelier and more appropriate interventions.