CHROMATIC

Extreme meteoceanic conditions can have devastating consequences on coastal areas, and for this reason it is important to identify the drivers of coastal risks and to quantify their impacts. Classical approaches to evaluate climate impact can impose strong assumptions between the drivers and the risk index variable. Instead, machine learning methods have the advantage of discovering complex relations in the data without assuming strong structural assumptions, and thus in many scenarios, they can reach outstanding performances in predicting the effect of drivers on a target variable. However, so far they are not tailored for predicting rare scenarios in the data including extreme events. Our proposal to the PEPR TRACCS: “Transforming climate modelisation for climate services" aims to investigate and develop advanced learning methods for correctly predicting the effects of extreme climate events to conduct reliable impact studies on the consequences of offshore forcing conditions like wave, sea-level, wind on coastal risks and human infrastructures.

Regarding the methodology of our project,
-First, we propose new algorithms for distributional regression relying on extrapolation techniques that provide satisfactory approximations of the effects of extreme predictors on the response variable. Developing and evaluating their statistical guarantees will be the launching part of our project.
-In a second part, these new developments will be applied to predict the extreme meteoceanic conditions in a data-driven way from climate variables (or any variables that are typical outputs of large-scale climate models, e.g., CFSR, ECMWF) that are of interest for impact assessment (e.g. extreme sea states at the coast) and impact indicators of interest for coastal risk management (e.g. maximum spatial flood extent, maximum water height at vulnerable assets).
-Ultimately, the trained prediction models can be evaluated on different climate projections from numerical models (e.g. CMIP6 projections via Copernicus platform) to forecast the probable impacts of climate change.