Methodology for the discrimination of areas affected by forest fires using satellite images and spatial statistics

Abstract

Abstract We evaluated the effectiveness of statistical operators Moran's I, Geary's C and Getis and Ord Gi in Landsat ETM satellite images to measure and analyze the spatial dependence of the spectral characteristics of the affected areas, comparing these results with those obtained in the use of spectral indices for discrimination of burned areas. The study was carried out using a multitemporal analysis (before and after the fire) in some areas located in the Tuparro National Park in the department of Vichada, in the northeastern part of the Orinoco region in Colombia, which were affected by the presence of this phenomenon. To perform the analysis used tools provided by the software R version 2.11.1, packages spdep and rgdal for handling raster data and calculation of spatial autocorrelation tests. The use of statistical operators provide better results with the band of near infrared (NIR) and allow better identification of burned areas, which facilitates the detection of spatial and spectral changes. This methodology provides information that can improve the monitoring of the effects of fire in time and helps to prevent possible disruption and the inventory of locations where this phenomenon has been presented.