Seasonal precipitation forecasting in Southeastern Brazil remains a significant scientific challenge, particularly in anticipating extreme events such as droughts and heavy rainfall. These events have substantial impacts on water security, energy systems, and socio-economic conditions in the region, reinforcing the need for more efficient and decision-oriented predictive tools. 

This project proposes the development and evaluation of a machine learning-based approach for seasonal forecasting of precipitation extremes, with an emphasis on interpretability, low computational cost, and operational applicability. The methodology includes model development, validation using both deterministic and probabilistic metrics, and a targeted assessment of predictive skill for extreme events. 

The expected outcomes are to advance seasonal forecasting capabilities in Brazil, providing more accessible and actionable tools to strengthen resilience to climate extremes and related crises.

To understand the impact of climate change on mesoscale convective systems (MCSs), we need to understand how the environment in which they are generated influences them. That is, it is essential to understand the dynamic and thermodynamic atmospheric patterns associated with the occurrence of MCSs in the current climate to advance our understanding of future climate conditions using climate change scenarios.

In the Amazon basin, this type of investigation presents a high degree of difficulty, mainly due to the lack of homogeneity and scarcity of observational data and the high variability of larger scale systems and circulations. Furthermore, the Amazon basin exhibits relatively constant convective instability and abundant convection while various MCSs occur throughout the territory. In addition to these factors, the interaction of atmospheric variables with physical territorial elements, such as topography, river intermittency, vegetation, constant and accelerated changes in land use, among others, can be added.