Fatores naturais e antrópicos na ocorrência de queimadas no Pantanal: uma abordagem preditiva

Abstract

The Pantanal is recognized as a National Heritage by the Brazilian Constitution and hosts internationally significant sites under the Ramsar Convention on Wetlands. This biome is rapidly losing its status as a natural frontier due to changes in its vegetation cover, with fires being one of its main environmental problems. Many significant impacts, such as the loss of fauna and flora habitats, the reduction of biodiversity, the emission of greenhouse gases, and direct risks to human life are caused by fires. In 2020, INPE recorded historical records of burned areas in the biome. This research aimed to: (1) predict areas of fires using machine learning algorithms; (2) determine important natural and anthropogenic covariates for prediction; and (3) map areas of greatest occurrence in the last 30 years. The methodology was divided into three phases: creation of the database, trend modeling, and training/spatialization. The Random Forest (RF), Weighted k-Nearest Neighbor Classifier (KKNN) and Neural Networks Using Model Averaging (avNNet) algorithms were used. The results showed better performance of RF, although all models overestimated the burned areas. The predictions were concentrated in the south of the biome, a region with the highest historical frequency of fires. Vegetation moisture was the most important variable in RF, while anthropogenic factors such as proximity to roads, urban centers, buildings and waterways showed significance in all models. The spatial distribution of the frequency of fires highlights the vulnerability of the southern region of the biome, especially in the border areas.