Forecasting agricultural drought using VCI and VHI across Africa

Droughts are complex and a major threat globally as they can cause substantial damage to society, especially in regions that depend on rain-fed agriculture. It is understood that acting early based on alerts provided by early warning systems (EWS) can potentially provide substantial mitigation, reducing the financial and human cost of such hazards. Several satellite-based indicators such as the Vegetation Condition Index (VCI) or the Vegetation Health Index (VHI) are included in these EWS to monitor the agricultural and ecological droughts. In this presentation, we first present a suite a machine-learning techniques that we developed to forecast up to 12 weeks ahead these indicators at the second administrative boundaries across Kenya. Our approaches (Gaussian Process, auto-regressive distributed lag model, Hierarchical Bayesian Model) all provided skilful forecasts at various lead times. Finally, we show our Africa-wide forecasts of VCI and VHI using Gaussian Processes where we analyse whether the performance of the forecasts is influenced by season, land cover, or agro-ecological zone. Providing highly skilful forecast on vegetation condition will allow disaster risk managers act early to support vulnerable communities and limit the impact of a drought hazard.