Assessment of post fire vegetation cover using spectral mixture analysis. Application and comparison of different endmember characterization methods

Abstract

The analysis of satellite images allows one to monitor the regeneration of vegetation after a fire. In this work, a methodology for quantifying post fire vegetation cover was developed. The proposed methodology is based on the examination of Landsat 7 ETM+ images by using Spectral Mixture Analysis (SMA) and involves the following steps: a) pre-processing, b) inherent dimensionality image determination c) endmember characterization following two methods that thus lead to different models: traditional method based on the knowledge of the area worked as well as Minimum Noise Fraction and Pixel Purity Index method, d) model inversion and combination, e) comparison between the vegetation cover estimated by each model and that measured in field, and f) selection of the most accurate model and mapping of the vegetation cover for the study area. The cover estimated provided by the different models exhibited a high correlation (Spearman’s correlation coefficient >0.89). The average absolute difference between the estimated and field-measured vegetation cover obtained with the most accurate model for each fire never exceeded 6%.