Causal inference and machine learning on complex climatic systems

• Grant number: 24/08278-2
• Support Opportunities: Scholarships in Brazil - Master
• Start date: November 01, 2024
• End date: October 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Francisco Aparecido Rodrigues
• Grantee: Matheus Victal Cerqueira
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

As an effect of climate change, we are increasingly witnessing the occurrence of extreme weather events such as hydrological anomalies, fierce cyclones and heat waves. The socioeconomic impact of such phenomena is immense and there is great scientific interest in understanding their behavior, characterization and forecast, exceptionally challenging tasks considering their extreme behavior in relation to the historic distribution. Among these types of events there are the extreme hydrological events comprised of droughts and extreme precipitation. Their occurrence can lead to several impacts such as destruction of agricultural capacity, intensification of food insecurity, economic and people fluxes disruption, public and private propriety destruction, among many others, and their forecast is extremely important in mitigating these effects. As in many other fields, correlation and regression methods are still the most common statistical tools used in Earth Sciences and are still largely used applied for the study of extreme weather phenomena. These tools, although practical, obtain limited knowledge of the causal dynamics of a given system. At the same time, we are seeing the evolution of the use of deep learning methods for extreme event forecasting but without the knowledge of the system s causal structure, these methods often lead to black-box models of difficult interpretation. Hence, this project proposes the study and integration of causal inference, complex network science and deep learning modern techniques aiming to create early warning systems for extreme hydrological events based on the case study of drought in the Amazon Basin. The main objective is to develop a general methodology for the creation of these systems based on the aforementioned techniques and the study case of drought in the Amazon s hydrological cradle.