Statistics and Data Science Seminar: Mapping the spatio-temporal distribution of burned areas in the Amazon from 2001 to 2020: an ensemble modeling approach (Dr Rodriguez de Rivera Ortega)

The Amazon region is the world’s largest tropical rainforest and plays an important role in global climate regulation, biodiversity conservation, and sustainable development. Forest fires in the Amazon have been a growing concern as they can rapidly damage the region. This article studies wildfires’ geographic and temporal distribution and risk factors in the Amazon region from 2001 to 2020. Our focus lies on examining the burned area as a crucial variable that signifies the occurrence of small-scale fire incidents. To capture the wide-ranging variability in the burned area, we propose an ensemble approach that combines the predictions of a number of models that effectively incorporate the influence of various high-resolution factors, including land cover information, temperature, precipitation, humidity, wind speed, and other environmental elements that significantly impact fire occurrences. Given the substantial amount of data involved, we employ a downsampling strategy to reduce the computational burden, accounting for the imbalance between the number of burned and unburned areas. Furthermore, we integrate individual models into a comprehensive ensemble model, capitalizing on their collective strengths and leveraging their combined insights simultaneously. By employing this comprehensive approach, we aim to better understand the diverse factors contributing to the spatial variability in burned areas within the Amazon region. This research sheds light on the complex dynamics of Amazon fires, providing valuable insights that can inform future fire management and prevention strategies. This is the link to the paper: https://link.springer.com/article/10.1007/s10651-025-00661-x