A new study in Remote Sensing of Environment (ELSEVIER), led by VITO in collaboration with ECMWF, both partners in the CENTAUR Horizon Europe project, shows how Earth Observation (EO) data combined with machine learning models allow to forecast the impacts of agricultural droughts on vegetation up to three months in advance.
This publication adds directly to CENTAUR’s mission of building operational capabilities to better anticipate the impacts of extreme weather events, such as droughts, which threaten food and water security.
From monitoring to forecasting
Most drought monitoring systems today rely on near-real-time assessment of vegetation conditions. While valuable, this approach signals drought only after its impacts have begun, allowing reactive rather than proactive drought management.
In contrast, this new study proposes a forecast system which integrates EO data, meteorological forecast and soil moisture into a machine learning framework. By predicting below average anomalies of Normalized Difference Vegetation Index (NDVI), a key indicator of vegetation health or “greenness”, the model can anticipate vegetation stress up to three months before it becomes visible.
This approach is highly relevant for the CENTAUR Horizon Europe project, as it contributes to the development of early warning systems. By integrating such frameworks with environmental, societal, and economic risk factors, the project aims to improve anticipatory action and enable more coordinated crisis response in the face of drought.
Tested in drought-prone regions
The study focused on Mali, Mozambique, and Somalia, countries which are highly vulnerable to agricultural drought. Across all three cases, the model consistently outperformed traditional near-real-time monitoring approaches.
The figure below compares UN’s Food and Agriculture Organisation (FAO) Vegetation Condition Index (VCI) (visualised by bars) with the Agricultural Drought Risk (ADR) (visualised by lines) derived from the NDVI predictions. It illustrates how the model’s forecasts align with existing agricultural drought monitoring products.
Qualitative comparison between FAO’s Vegetation Condition Index (VCI) and the modelled agricultural drought risk (ADR)
Towards proactive drought management
Although the model still has shortcomings in predicting NDVI anomalies during periods of seasonal transition, the findings highlight the strong potential of combining EO data and machine learning to move from reactive responses to anticipatory action.
By forecasting where and when agricultural drought will have the most severe impacts, the system sets a new benchmark for further developments aimed at transforming drought management, across Africa’s most vulnerable regions.
This work marks an important step in advancing CENTAUR’s ambition to deliver anticipatory insights increasing food and water security. It also exemplifies CENTAUR’s broader mission: to strengthen the Copernicus Emergency Management and Copernicus Security Services with timely, actionable, and policy-relevant information.
The full article is freely available here.
For the latest updates on CENTAUR activities and upcoming events subscribe to our newsletter and follow us on LinkedIn, X, and BlueSky.