Benchmarking Vegetation Forecasts for Drought Early Warning in Eastern Africa

By August 2022, drought in the Greater Horn of Africa had resulted in 3.6 million livestock deaths and left 28 million people highly food insecure, urgently requiring humanitarian assistance. Pastoralist communities, whose livelihoods depend on the availability of pasturelands, are particularly vulnerable to the impacts of drought.

Operational drought Early Warning Systems and Early Action Protocols in the region predominantly rely on real time observations and precipitation forecasts. However, vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) and the Vegetation Condition Index (VCI), provide a more direct measure of pasture conditions. Incorporating vegetation forecasts into these systems could shift the focus toward impact-based forecasting, offering a more accurate basis for early action.

Numerous statistical and machine learning approaches have been developed to forecast vegetation conditions using satellite-derived vegetation indicators, often in combination with hydroclimatic and land surface variables. Despite this, a gap remains between academic research and the methods currently applied in operational settings.

Here, we conduct a benchmarking analysis of existing statistical and machine learning models that forecast vegetation indices (NDVI and VCI) to provide decision-makers with an informed overview of the range of available solutions.

We evaluate four types of models: autoregressive models, Gaussian processes, convolutional long short-term memory neural networks, and transformers, assessing their ability to forecast vegetation indices across different spatial resolutions: VIIRS (500 m) and Sentinel-2 (20 m). We also examine model performance during documented extreme drought events in cross-border arid and semi-arid pastoralist regions of the Greater Horn of Africa. Our analysis highlights the relative strengths and limitations of these models, providing guidance for integrating vegetation-based forecasts into operational early warning systems to better support drought-affected pastoralist communities.