![Sample our Environment & Agriculture journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5934/TOC-Subject-Sample-2021-Environment-Agriculture.jpg"})
Abstract
Environmental agencies frequently require tools for quick assessments of areas affected by large fires. Remote sensing techniques have been reported as efficient tools to evaluate the effects of fire. However, there exist few quantitative comparisons about the performance of the diverse methods. This study quantitatively evaluated the accuracy of five different techniques, a field survey and four satellite‐based techniques, in order to quickly classify a large forest fire that occurred in 1998 in Solsonès (north‐east Spain) by means of an IRS LISS‐III image. Three pure classes were determined: burned area, unburned vegetation, and bare soil; along with a non‐pure class that we called mixed area. These selected techniques were included into a tree classifier to investigate their partial contribution to the final classification. The most accurate methods when focusing on pure classes were those directly related to the spectral characteristics of the pixel: Reflectance Data and Spectral Unmixing (82% of overall accuracy), versus the poorer performances of Vegetation Indices (70%), Textural measures (72%) and the field survey (68.6%). Since no image processing technique was applied to the Raw Reflectance Data, it can be considered the most cost‐effective method, and the tree classifier reinforces its importance. The results of this study reveal that time consuming and expensive methods are not necessarily the most accurate, especially when potentially easily distinguishable classes are involved.
Acknowledgements
This research was partly supported by a Governmental grant (BEAI grants‐2000. BEAI200111). We thank personnel in the Geography Department in the University of California, Santa Barbara for their support, Jose Luis Ordóñez for helping with the figures, Emilio Chuvieco for the program of matrix normalization and variance elaboration, Izayas Numata for his help with the final tree, and two anonymous referees for helpful comments.