Spatiotemporal assessment of forest fires occurred in the Serra da Canastra National Park between 1991 and 2011

Abstract

This study involved a spatiotemporal analysis of forest fires in the Serra da Canastra National Park (SCNP) located in the State of Minas Gerais, Brazil. The analysis was based on remotely sensed data acquired between 1991 and 2011. Landsat5 TM imagery was geometric and radiometric corrected and them transformed by applying Principal Component Analysis (PCA), which removed correlation between bands, increasing the discrimination of burned areas. The derived PC3, PC4, and PC5 were used as input for a decision tree classification technique, which showed better performance to detect forest fire scares.This classification technique showed overall accuracy of approximately 93%, Kappa’s index of 92% and yet 0% and 18% of omission and commission accuracies, respectively. The results show that forest fires mostly occurred in 1991, 1997, 2001, 2007, and 2010 within the Serra da Canastra National Park, which burned approximately 80% of it. A total of 157000 and 140000 hectares were burned in 2010 and 2000, respectively. Forest fires events and recurrent fires increased during the period of analysis. This is a crucial indicator that institutional and political measures must be enforced in the coming years to avoid further environmental issues in this protected area.