CREATOR: Km-scale climate downscaling through machine learning emulators

• Grant number: 25/05182-7
• Support Opportunities: Regular Research Grants
• Start date: October 01, 2025
• End date: September 30, 2029
• Field of knowledge: Physical Sciences and Mathematics - Geosciences - Meteorology
• Principal Investigator: Rosmeri Porfírio da Rocha
• Grantee: Rosmeri Porfírio da Rocha
• Principal researcher abroad: Maria Laura Bettolli
• Institution abroad: Centre National de la Recherche Scientifique, France
• Host Institution: Instituto de Astronomia, Geofísica e Ciências Atmosféricas (IAG). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated researchers: Amanda Rehbein

Abstract

Relying on observational data, empirical statistical downscaling models (ESD) relate large-scale predictors and regional climate variables. In turn, emulators are hybrid downscaling models based on machine learning techniques that are trained using regional climate model (RCM) simulations to learn the physical relationships between low-resolution predictors and high-resolution predictand as much as is done in the classical dynamical downscaling. The varied nature of ESD and the relatively new emulation approach based on more sophisticated machine learning techniques, requires the improvement of our understanding of how well new machine learning techniques are able to represent local climate and produce reliable and cost-effective future projections. CREATOR tackles these challenges fostering inter- institutional collaboration and further networking, integrating not only South American research communities but also French downscaling communities. The main aim of this initiative is to comprehensively assess machine learning-based ESD models and RCM emulators at km-scale to obtain high-resolution climate projections over Subtropical South America (SSA) where the understanding of changes in extreme climate events in a climate change context is still an open question. This project (i) builds up on taking full advantage of the available convection permitting RCM evaluation simulations that cover the complete SSA sector; (ii) designates SSA as a core region for coordinated RCM simulations at km-scale for future climate change scenarios; and (iii) proposes the development of an evaluation framework of machine learning-based ESD and RCM emulators at km-scale. The ultimate goal of this research is to provide the basis for leveraging the merits of dynamical and statistical downscaling to increase the Global Climate Model/RCM matrix in the region, hence producing comprehensive ensembles to address the uncertainty assessment in climate projections. (AU)