Band-Based Best Model Selection for Topographic Normalization of Normalized Difference Vegetation Index Map

Topographic effect in remote sensing images is severe in high mountainous areas. Efficiently to reduce the effects, several topographic normalization models have been proposed. Since the performance of the models is largely dependent on the spectral band and land surface type, the best performance model can vary from image to image in an area as well as from band to band in an image. The normalized difference vegetation index (NDVI) map has been widely used for the vegetation monitoring and assessment. An efficient reduction of the topographic effect in the NDVI map must be required for the spatial analysis of the vegetation monitoring and assessment. In this paper, we propose an efficient method to select the best topographic normalization model in each band to reduce the topographic effect of NDVI maps. The histogram structural similarity (HSSIM) index was used for the model selection because the index allows to select the best model in each band of an image. Five topographic normalization models were used for the test, which include the sun-canopy-sensor (SCS), statistical-empirical, C-correction, Minnaert, and Minnaert + SCS. The performance of the proposed method was validated by using two different season Landsat-8 OLI images including the forest area of northern Malaysia. The standard deviations of the two NDVI maps generated from the test images were reduced by about 53.1% and 28.6% after correction in profile analysis. The coefficient of determination ( $R^{2}$ ) between the two different NDVI maps increased from 0.626 to 0.759. It indicates that the proposed method effectively reduced the topographic effect of the NDVI maps. This result implies that the proposed method can work well in the topographic normalization. Furthermore, the proposed method would be successfully applied to index maps including the normalized difference snow index (NDSI), normalized difference water index (NDWI), etc.