ML-Enhanced Vegetation Condition Forecasting — Kenya's National Drought Management Authority Early-Warning Extension

Kenya's National Drought Management Authority (NDMA) classifies drought severity across 23 arid and semi-arid counties using the satellite-derived Vegetation Condition Index (VCI3M), issuing an alert whenever VCI3M falls below 35 in monthly county bulletins.

RCMRD, which supplies NDMA's satellite data, partnered with University of Sussex researchers to build machine-learning models forecasting VCI3M several weeks before it can be observed directly from satellite imagery, extending NDMA's rainfall-forecast lead time of roughly 15 days.

Linear autoregression and Gaussian Process regression models were trained and validated against Kenya's historical VCI3M record and operational alert threshold.

The models achieved approximately 89% hit rate with a 4% false-alarm rate at a 4-week lead time, and approximately 81% hit rate with a 6% false-alarm rate at a 6-week lead time. Results were published in a peer-reviewed paper, also posted on arXiv, independently of any single agency's press materials.

The intent, per the University of Sussex, is for RCMRD to feed these forecasts into the same interagency technical working group (including NDMA and the Kenya Red Cross) that already coordinates drought response. No source dated after the initial publication confirms the forecasts are now running inside NDMA's official monthly bulletins, so this is best read as a near-operational research-to-practice pilot rather than a finished, audited production system.