QUALIS-CAPES

B1

2021-2024
quadriênio

Language

Revista Brasileira de Sensoriamento Remoto

e-ISSN: 2675-5491 | ISSN: 2675-5491


Abstract

This study seeks to investigate the different uses of fire in territorial management by analyzing prescribed, controlled, and suppression burns, combined with data from the Fire Panel of the Amazon Protection System Operational and Management Center (CENSIPAM) for the year 2023. The research was carried out in the Brazilian Legal Amazon, following the official administrative boundaries, covering the states of Acre, Amapá, Amazonas, Pará, Rondônia, Roraima, Tocantins, Mato Grosso, and part of Maranhão. Fire events recorded in 2023 were retrieved from the Fire Panel platform, which compiles a database of environmental and socio-spatial variables, thereby enabling a more comprehensive characterization of fire occurrences. These data were integrated with records of prescribed, controlled, and suppression burns conducted by ICMBio brigades during the same period. The methodological framework was based on machine learning techniques to generate predictive models and provide insights into the impacts of fire use across the Brazilian Federative Units (states). The Random Forest model achieved the highest overall accuracy (76%). Suppression and prescribed burns showed the highest precision values (0.90 and 0.69, respectively), while the ROC curves and AUC scores demonstrated good generalization performance for these classes (0.84 and 0.79). Conversely, controlled burns exhibited lower predictive performance. SHAP (Shapley Additive Explanations) analyses revealed that variables such as biomass and unmonitored areas play a key role in modeling the classification of fire occurrences.

License

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

Copyright (c) 2026 Jack Endrick Pastrana Mojica, Tássia Fraga Belloli , Pamela Boelter Herrmann, Deyvis Cano , Camila Souza Silva