An automatic atmospheric correction algorithm for visible/NIR imagery
Abstract:The automatic correction of atmospheric effects currently requires visible to short‐wave spectral bands (400–2500 nm) to derive high accuracy surface reflectance data. Common techniques employ spectral correlations of dark targets in the short‐wave infrared (SWIR, around 2.2 µm), blue (480 nm) and red (660 nm) regions to derive the aerosol optical depth. A large number of current Earth‐observing satellite sensors have only three or four spectral channels in the visible and near‐infrared (VNIR) region (400–1000 nm), making an automatic image‐based atmospheric correction very difficult. This contribution presents a new algorithm and first results with VNIR imagery. The method starts with the assumption of average clear atmospheric conditions (aerosol optical depth AOD = 0.27, corresponding to a visibility of 23 km) and calculates the surface reflectance in the red and near‐infrared (NIR) bands. The second step derives a mask of dark vegetation pixels. It is calculated using multiple thresholds of vegetation index combined with red and NIR surface reflectance values. Then the red band surface reflectance for the dark pixels is estimated from the NIR reflectance as ρ red = 0.1 ρ nir , from which the aerosol optical depth (or visibility) can be calculated. The core of the VNIR algorithm consists of two subsequent iteration loops (visibility and ρ red ) to improve the visibility estimate. Results of the VNIR method are presented for Landsat‐5 Thematic Mapper (TM) and Landsat‐7 Enhanced Thematic Mapper Plus (ETM+) imagery using only the first four bands. The performance of the method is compared to the established dark pixel technique where the SWIR bands are included. Results show that the deviation between both methods is usually less than 0.005 reflectance units if measured in terms of the scene‐average reflectance, indicating a useful potential for this approach.
Document Type: Research Article
Affiliations: 1: DLR, German Aerospace Center, Remote Sensing Data Center, D‐82234 Wessling, Germany 2: University of Zurich, Remote Sensing Laboratories, Winterthurerstr. 190, CH‐8057‐Zurich, Switzerland
Publication date: May 1, 2006