A spatially adaptive fast atmospheric correction algorithm

Abstract

An atmospheric correction algorithm for high spatial resolution satellite sensors like Landsat Thematic Mapper (TM) has been developed. The algorithm works with a catalogue of atmospheric correction functions stored in look-up tables. The catalogue consists of a broad range of atmospheric conditions (different altitude profiles of pressure, air temperature, and humidity; several aerosol types; ground elevations from 0-1 km above sea level; solar zenith angles ranging from 0°-70°). The catalogue covers visibilities (surface meteorological range) from 5-40 km, values can be extrapolated down to 4 km and up to 80km. The 1994 edition of the catalogue was compiled using the MODTRAN-2 and the SENSAT-5 codes. The algorithm consists of an interactive and an automatic part. The interactive phase serves for the definition of a reference target (dense dark vegetation or water) as well as haze and cloud. The reflectance of the reference target in a single spectral band (dark vegetation: TM band 3; water: TM band 4) has to be specified. Additionally, the image can be partitioned into sub-images, called sectors. This phase also selects one of the atmospheres available in the catalogue, i.e. the altitude profile of pressure, temperature and humidity as well as the aerosol type (e.g. rural) are fixed. The automatic phase first calculates the visibility of the reference areas for the selected atmosphere. The visibility is obtained by matching the measured signal (i.e the digital number (DN) converted to a radiance using the sensor calibration) to the model-derived signal in the spectral channel of known target reflectance. The sector-average visibility of the reference pixels is assigned to the non-reference pixels. The second step is the haze removal performed by histogram matching the statistics of the haze regions to the statistics of the clear part of the scene for each sector and each channel. The last step is the calculation of the ground reflectance image including the adjacency correction, and the computation of the ground brightness temperature image (TM band 6)