Evaluation of Topographic Correction on Subpixel Impervious Cover Mapping With CBERS-2B Data

This study compared the effectiveness of six commonly used topographic correction methods for subpixel impervious surface mapping in selected mountainous areas of Southwest Virginia. One 2008 China-Brazil Earth Remote Sensing 2B (CBERS-2B) image was processed using selected topographic algorithms and then used as input for subpixel impervious cover mapping. High-resolution National Agriculture Imagery Program (1-m resolution) images were used to build proportional subpixel impervious cover as training/validation data. We then applied a classification and regression tree algorithm to establish relationships between CBERS signals and impervious surfaces. Accuracy assessment showed that both R-2 (0.644-0.767) and RMSE (0.118-0.150, reported as proportion of impervious surface) values vary across different topographic correction algorithms. The accuracy differences (R-2: 0.448-0.771; RMSE: 0.118-0.247) were most pronounced for areas facing away from the sun azimuth angle, suggesting aspect-sun azimuth-dependent map accuracy. For terrain shadowing areas, the Minnaert method, the minslope method, and the C-correction substantially outperformed the cosine and improved cosine correction. These findings indicate that users should apply caution in using topographic correction algorithms and that aspect-stratified accuracy assessment needs to be conducted for detailed comparisons. We also repeated the analyses using Landsat TM and obtained better overall results compared to the CBERS-2B data. The differences in R-2 (or RMSE) for two data sources were not substantial, suggesting the high potential of CBERS data for subpixel impervious surface mapping.