Prediction and Typification of Fire Use in the Legal Amazon Region of Brazil: A Study Based on Events Detected by the Fire Panel
DOI:
https://doi.org/10.1590/SciELOPreprints.14897Keywords:
brazilian amazon, Fire ecology, machine learningAbstract
This research aims to understand the different uses of fire in management, through prescribed, controlled, and combat fires, cross-referenced with data from the CENSIPAM Fire Panel in 2023. Machine learning techniques were applied to make predictions. It was possible to generate insights into the impacts of fire use in the Federal Units (UFs). The Random Forest model showed the highest overall accuracy (76%). Combat and prescribed fires had the highest precision (0.90% and 0.69%, respectively). The ROC curves and AUC values indicated good discrimination capacity between these classes (0.84 and 0.79), while controlled burning showed lower generalization ability. Shapley Additive Explanations (SHAP) values highlighted biomass and areas without oversight as factors contributing to the modeling of fire occurrence classification.
Downloads
Posted
How to Cite
Section
Copyright (c) 2026 Jack Endrick Pastrana Mojica, Tássia Fraga Belloli, Pamela Boelter Herrmann, Deyvis Cano, Camila Souza Silva

This work is licensed under a Creative Commons Attribution 4.0 International License.
Funding data
-
Instituto Nacional de Ciência e Tecnologia da Criosfera
Grant numbers 380182/2024-6, Categoria: DTI-B – B
Plaudit
Data statement
-
The research data is available on demand, condition justified in the manuscript


