Speaker
Matteo Giuliani, assistant professor, Politecnico di Milano
Moderator
Andrea Toreti, senior scientist, European Commission, Joint Research Centre (TBC)
Additional info will be soon available.
Abstract
Drought is a slowly developing natural phenomenon that can occur in all climatic zones and
propagates through the entire hydrological cycle with long-term socio-economic and
environmental impacts. Intensified by anthropogenic climate change, drought has become one of
the most significant natural hazards in Europe. Different definitions of drought exist, i.e.
meteorological, hydrological, and agricultural droughts, which vary according to the time horizon
and the variables considered. Just as there is no single definition of drought, there is no single
index that accounts for all types of droughts. Capturing the evolution of drought dynamics and
associated impacts across different temporal and spatial scales still remains a critical challenge.
In this talk, we discuss the role of Machine Learning for advancing impact-based drought
detection. Our main goal is the identification of relevant drivers of observed drought impacts (e.g.,
water deficits or crop stress) from a pool of candidate hydro-meteorological predictors. The
selected predictors are then combined into an index representing a surrogate of the drought
impacts in the considered area. To support this task, we developed a ML pipeline that integrates
(1) a novel dimensionality reduction method that allows an interpretable aggregation of spatially
distributed drivers, (2) feature extraction techniques including both filters and wrappers to select
the most informative and non-redundant information, and (3) existing and new causal inference
algorithms for verifying the causal links between the selected drivers and the target impacts.
Our new indexes advance state-of-the-art drought monitoring practices, which often rely on
standardized drought indexes that are poorly correlated with drought impacts, and provide
reliable projections of drought impacts’ trends under different climate change scenarios. Several
real-world examples will be used to provide a synthesis of recent applications of our methodology
in case studies featuring diverse hydroclimatic conditions, variable levels of data availability, and
increasing spatial domain from single river basins to a pan-European analysis.
