A quantitative method to evaluate the performance of topographic correction models used to improve land cover identification

Abstract

Topographic correction methods have been widely used prior to land cover identification in sloping terrain because the topographic variation on the Earth's surface can interfere with the classifications. The topographic correction involves the normalization of brightness or surface reflectance values from the slanted to the horizontal plane. Several topographic correction models have been proposed, and a quantitative evaluation method is needed for these models because the performance can vary according to the surface cover types and spectral bands. In this study, we proposed an efficient method to evaluate the performance of topographic correction models through measuring the histogram structural similarity (HSSIM) index estimated from the sunlit and sun-shaded slope areas before and after the correction. We tested the HSSIM index by using three different land cover types derived from Landsat-8 Operational Land Imager (OLI) images and eight commonly used topographic correction models. When the proposed HSSIM index was compared with the visual analysis technique, the results matched exactly. Using the HSSIM index, the best correction methods were then determined, and the best ones included the statistical-empirical or SCS+C methods (where SCS+C refers to the sun-canopy-sensor plus C-correction) for the R, G, and B bands and the Minnaert+SCS method for the NIR, SWIR-1, and SWIR-2 bands. These results indicate that (i) the HSSIM index enables quantitative performance evaluations of topographic correction models and (ii) the HSSIM index can be used to determine the best topographic correction method for particular land cover identification applications.