Simulation Based Inference for climate models

• Grant number: 25/06168-8
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: September 30, 2025
• End date: September 29, 2026
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Rafael Izbicki
• Grantee: Luben Miguel Cruz Cabezas
• Supervisor: Pedro Luiz Coelho Rodrigues
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil
• Institution abroad: Centre Inria De L'Université Grenoble Alpes, France
• Associated to the scholarship: 22/08579-7 - Validation and calibration of prediction models, BP.DD

Abstract

This project aims to develop machine learning methods to enhance the understanding and modeling of physical phenomena, with a particular focus on climate science and related inference problems. Simulation-Based Inference (SBI) provides a powerful framework for estimating nonlinear model parameters from observational data, which is crucial for tasks such as tuning parameterizations in atmospheric and oceanic models. However, existing SBI methods are largely restricted to small-scale models, limiting their applicability to complex climate simulations.Our goal is to extend the SBI framework to handle large-scale models while also addressing a key challenge: the lack of statistical calibration in posterior estimation, particularly in high-dimensional settings. Poorly calibrated posteriors can lead to unreliable uncertainty quantification and flawed inferences. To tackle this, we propose conformal-based statistical calibration techniques specifically designed for large-scale simulators.Thus, this project focuses on two main challenges: (1) developing new SBI algorithms that account for the computational costs of climate simulations, and (2) designing scalable statistical calibration strategies to improve posterior estimation reliability. (AU)