Comparison of methods for topographic correction in Landsat-8/OLI image over a topographically complex region
ISBN 978-85-88783-11-9
Authors
1Neves de Souza Lima, R.; 2Damiati Ferreira, R.
1INSTITUTO BRASILEIRO DE GEOGRAFIA E ESTATÍSTICA - IBGE Email: ricardo.s.lima@ibge.gov.br
2INSTITUTO BRASILEIRO DE GEOGRAFIA E ESTATÍSTICA - IBGE Email: rafael.damiati@ibge.gov.br
Abstract
The information of the Earth's surface acquired by satellite imagery enables identification of environmental processes in space and their temporal dynamics. However, the geometric and radiometric distortions must be minimized in order to ensure better efficiency on obtaining and analysis of these data. The topographic shading is one of the main factors that hamper information extraction in satellite imagery, especially in mountainous areas, and results from variations in the spectral radiance induced by inclination and orientation angles of the terrain, combined with the zenith and azimuth angles of the sun. With the recent availability of orbital imagery such as Landsat series, as well as Digital Elevation Models (DEM), several methods have been developed to reduce topographic effects in satellite imagery. In most cases, the topographic correction is performed by calculating the cosine of the solar incident angle (cosi) relative to the normal of each pixel, estimated from topographic and solar angles. Based on this parameter, the reflectance on a sloped surface can be normalized with respect to a flat surface using lambertians and non-lambertians topographic correction methods. The objective of this study was to compare the efficiency of three non-lambertians topographic correction algorithms on an OLI/Landsat-8 image over a topographically complex region in the southeastern of Minas Gerais state, Brazil. The corrections were applied to seven spectral bands of one scene of OLI/Landsat-8 sensor, acquired from the US Geological Survey database, corresponding to the path/row 217/75, and pass date from August 02, 2013. Initially, to obtain the surface reflectance of the scene, the atmospheric correction was performed using the Quick Atmospheric Correction - QUAC module in ENVI software, version 5.1. To obtain the cosi, was used the DEM from the TOPODATA project. The topographic correction of the scene was performed by using the Minnaert, C-correction and Empirical Rotation methods, and the results were evaluated based on qualitative (visual inspection) and quantitative (statistical variables) parameters. The implementation of the operational procedures for obtaining cosi and topographic correction was performed in GIS (Geographic Information System) environment, through the development of geoprocessing tools using the modelbuilder module of ArcGIS software, version 10.1. The atmospheric correction applied to the OLI/Landsat-8 image produced significantly decrease of reflectance factor in the visible spectral region (bands 1, 2, 3 and 4), by minimizing the effect of Rayleigh scattering, while for the infrared region (bands 5, 6 and 7) there was an increase in reflectance factor, due to the reduction of the effect of spectral absorption, resulting from atmospheric moisture. Regarding the topographic correction, it was observed that the three algorithms significantly reduced the presence of shadows caused by topography, as well as the brightness in facing the sun areas, normalizing the reflectance factor between shadowed and illuminated slopes. However, the corrected images by the C-correction and Minnaert methods produced the overcorrection effect in intense shadowed areas, whereas in the Empirical Rotation method this effect was less noticeable. Statistically the three methods presented similar performance. The correlation coefficient between cosi and the corrected reflectance showed a mean reduction from 108.9 to 115% compared to the original scene, indicating that the dependence of the reflectance to the topography has been eliminated. In a similar way there was a mean reduction in standard deviation of reflectance, ranging from 8.9 to 12% between corrected and non-corrected scenes. These results corroborate numerous researches aiming to reduce topographic effects in satellite images over mountainous areas, highlighting the importance of these techniques for improve information extraction from remote sensing with Landsat-8 data.