Explainable AI and Earth Observation for seasonal drought prediction
Drought prediction is essential to managing some of the most impactful hazards for human life today -wildfires, agricultural food loss or water availability. This research project aims to improve seasonal climate predictions of drought through the broad understanding of Earth Observation data and climate variables, thanks to powerful artificial intelligence techniques.
A crucial need:
Drought management in a changing climate
Under a global warming scenario where a drier climate is projected for the coming decades in numerous regions, including most of Europe, the capability to anticipate the occurrence, intensity, and duration of drought emerges as a crucial need to adapt to its multifaceted impacts effectively.
Seasonal climate prediction systems (SPS) represent the most suitable tool to inform drought-related decision making at appropriate temporal and spatial scales. Yet, despite two decades of considerable improvements, the provision of skilful seasonal precipitation predictions, especially in the extra-tropics, is notoriously challenging.
Our research
Improving seasonal drought prediction
We present a comprehensive approach combining Artificial Intelligence, dynamical seasonal climate prediction systems, and multiple Earth Observation products. The study aims to improve drought prediction capabilities and enrich our understanding of droughts’ causes, evolution, and consequences at seasonal time scales.
The AI for Drought project (2022-2024) is implemented by the European Space Agency (ESA). It focuses on Extreme Events, Multi-Hazards and Compound Events, and contributes to the ESA Extremes and Natural Disasters Science Cluster.
Case study
Central Europe and Iberia
The limited predictive skill over the European continent, both for temperature and precipitation is a well-known problem for the research community.
In this project, we consider seasonal climate predictions in the Central Europe and the Iberian peninsula, where a drier climate is very likely to be established, and where the understanding of the causal relationships within the cascade effects of droughts will represent a milestone in generating actionable knowledge. All in all, a solid case study to validate and demonstrate the results of our research and its potential impact.
Methodology
Why is Earth Observation important?
Earth observation products like soil moisture, vegetation indices, lake levels, snow cover, or burned area can contribute to specifying the initial climate state of models and, at the same time, properly characterize the multifaceted nature of droughts. Beyond the physical drivers provided by seasonal prediction systems, Earth Observation records directly depict existing environmental impacts. These two intimately related but different representations of the climate system can significantly benefit one another with recent breakthroughs in eXplainable Artificial Intelligence.
