Atmospheric corrector in Landsat images

Abstract

Atmospheric effects in satellite imagery distort the available information and generate errors in the estimation of biophysical variables from spectral data. An atmospheric correction algorithm was developed based on the correlation between the reflectance of band 7 (medium infrared, 2.2 μm) and band 1 (Blue, 0.485 μm) of vegetation, for TM and ETM+ sensors. The intercept of the regression between band 7 and band 1 of Landsat imagery is an estimator of path reflectance in band 1. With this path reflectance and a def ined atmosphere and aerosol model, it is possible to estimate the optical thickness of aerosols centered at 0.55 μm. The algorithm was based on coupling of several simulation models and spectral data libraries to represent the soil-vegetation-atmosphere optical system. Vegetation in an image is identif ied by a generic object classif ier in four vegetation variants: dark dense, high, medium and low coverage. The algorithm was evaluated in two phases, the f irst was based on empirical analysis of the results of the simulation of the soil-vegetation-atmosphere optical system, in which controlled conditions were maintained. In the second phase, the algorithm was validated with 7 ETM + images whose scenes contained a site of the aerosol robotic network (AERONET), which accurately measures the optical thickness of aerosols in different wavelengths. The analysis of results showed that the corrector estimates the optical thickness of the scene with good adjustment: R2 = 0.97, Root Mean Squared Error of 0.059 (20.3%) and represents well the spatial variability of aerosol load in Landsat imagery.