Bi-temporal Radiometric Normalization of Landsat 8 Images Using Pseudo-Invariant Features

Relative radiometric normalization (RRN) is one of the radiometric corrections for satellite imagery besides absolute radiometric normalization (ARN). In contrast to the absolute method that corrects various components such as atmospheric condition, earth-sun distance, illumination and viewing angle of satellite to find true reflectance, relative method does not find true reflectance but do the transformation of digital number to fit with reference image digital number or try to find common scale of digital number both of reference and target images instead. Several studies have conducted relative radiometric normalization to solve radiometric inconsistency issues by using pseudo-invariant features (PIFs). PIFs are reference objects that has an insignificant or near stable reflectance value change over time. This study is aimed to evaluate radiometric normalization result for Landsat 8 surface reflectance product that utilized Google Earth Engine platform for the computations. Normalization in this paper applied Multivariate Alteration Detection for PIFs selections. The selection of PIFs is based on data distribution of MAD result, the threshold values for selection are 10%, 15%, 20% and 25% of data distribution. Finally, the normalization used selected PIFs as sample data for calculate the slope and aspect of linear regression. On this study show the normalization result have the highest Pearson correlation value on 10% PIFs blue band which achieve 97.6% then the lowest Pearson correlation on 25% PIFs SWIR1 band which achieve 91.4%. The results suggest that developed approach have a potential solution to deal with inconsistency issues.