Comparative analysis of the atmospheric correction results for inter- and cross-sensor application in LUCC studies

Kurzfassung

Remote sensing enables continuous Earth observations and change detection. On the one hand satellite data archives provide great records for multi-temporal analysis, on the other an increasing number of new remote sensing datasets offer opportunities for multi-scale and cross-sensor applications. Atmospheric correction and radiometric normalization of satellite imagery is a prerequisite in order to achieve reliable and comparable results in land change studies. Atmospheric correction reduces effects of scattering and absorption by gases and aerosols in the atmosphere between the Earth’s surface and the sensor, and minimizes the influence of solar illumination and topography on the registered signal. Well performing atmospheric correction algorithms should provide identical results for individual images acquired on the same date and over the same area. Moreover, pseudo-invariant features such as for example dark water, asphalt or sand should return the same spectral signature in corrected imagery from different acquisition dates. This study presents a comparison of atmospheric correction results obtained from correcting Landsat TM and RapidEye imagery using the ATCOR software version 8.2.1. First, we tested intra-sensor differences between correction results for the along- and across-track overlap areas of two scenes. Subsequently, the cross-sensor variation of surface reflectance between Landsat and RapidEye imagery was investigated. Finally, we compared ground-truth measurements of aerosol optical thickness (AOT) obtained simultaneously to the satellite acquisitions with that ATCOR derives from the imagery. Overall, our results indicate high consistency in reflectance within the overlap areas of the separately corrected scenes. The mean difference in reflectance for the overlap area of two successive scenes is less than 0.01 . The highest differences occurred in near-infrared spectral range. Nevertheless, larger disparities above 0.04 were observed. Such differences in surface reflectance cause uncertainties in the retrieval of biophysical parameters (leaf area index, aboveground biomass, etc.) or spectral indices (e.g. the normalized differenced vegetation index) across image frames and thus hinder their application. Regarding aerosol optical thickness, the ATCOR-based AOT values were mostly overestimated when compared to the ground-truth measurements. Differences in AOT in the overlap areas were up to 0.05. The inconsistency of AOT for the overlapping area as well as uncertainties in AOT retrieval confirm the limitations of atmospheric correction found for the reflectance retrieval. To overcome retrieval limitations our results underline the need for relative radiometric normalization performed additionally to atmospheric correction of imagery and conducted prior to the atmospheric correction, particularly for intra- and cross-sensor data integration.