Forecasting crop yields through the use of Machine Learning algorithms and Earth Observation data in Kenya

Keywords: Agriculture forecasts, Kenya, Machine learning, Agriculture yields, Food security
Session Type: Virtual Poster Abstract
Day: Friday
Session Start / End Time: 2/25/2022 03:40 PM (Eastern Time (US & Canada)) - 2/25/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 39

Abstract

Low crop productivity and weather variability contribute to food insecurity in Eastern African countries. Having knowledge on what end of season crop yields will be earlier on would provide decision-makers with more time to prepare. Recent improvements to earth observation satellites and machine learning technology have provided new data products that can be used to develop better methods of forecasting crop yields to create this solution. The purpose of this study is to develop a machine learning model to forecast crop yields for the maize crop in Kenya and other Eastern African countries. Various agrometeorological variables derived from satellite data, including evaporative stress index, precipitation totals, soil moisture, and Normalized Difference Vegetative Index, for the years between 2002 and 2016 are used as inputs for various model types. The models being compared in this study include a ridge regression, random forest model, and mixed effect random forest model. Testing of the models is focused on optimizing performance for years with lower yields. Initial values for correlation coefficients, RMSE and feature importances suggest that these models hold promise for providing early warning predictions that can preempt actions being taken when crop yields are impacted negatively by adverse weather conditions.