Atmospheric and topographic correction for the detection of forest transitions in mountain areas

Differential illumination effects complicate the interpretation and classification of satellite imagery in mountain areas. Different topographic correction methods ranging from simple empirical to complex physically-based models have been developed to quantify and remove differential illumination effects from the recorded reflectance values. However, the preprocessing of satellite imagery requires additional input data and complicates the design of automatic classification process chains for large scale mapping. It is at present not clear what the relative added value of atmospheric and topographic correction techniques is. This study compares the influence of 2 atmospheric and 4 topographic correction methods on two Landsat-TM images in the Romanian Carpathian mountains. The results show that atmospheric corrections improve the images substantial in flat regions. However, in mountain areas, the influence of topographic corrections is more important than atmospheric corrections. The empirical topographic correction methods based on band ratioing and cosine corrections do not significantly improve the results (Figure). Topographic correction methods based on a pixel-based Minnaert procedures and physically-based methods such as the integrated radiometric correction (IRC) do improve the results significantly but their parameterization impedes automation in classification process chains. The combined application of atmospheric and topographic correction techniques improves the accuracy of land cover mapping significantly and allows to distinguish more land cover classes with an acceptable class purity. This is especially important for the analysis of forest transitions that typically result in patch landscapes with a mixture of various stages of forest degradation and forest regeneration.status: publishe