Automatic intercalibration of night-time light imagery using robust regression

In remote-sensing community, radiometric calibration of night-time light images has long been a problem, hindering change detection of images in different dates. Currently, an intercalibration model is regarded as the unique solution for the problem, but prior knowledge is needed to extract reference pixels with stable lights, which are hard to obtain in most of the applications. This study proposed an automatic algorithm to extract the reference pixels for convenient use of the intercalibration model, with an assumption that there are sufficient pixels with stable night-time lights in the multi-temporal images. To automatically extract the stable pixels from images in two dates, all pixels in the two dates were entered into a linear regression model, and the outliers viewed as suspected changed pixels were discarded iteratively. Consequently, some stable pixels were extracted and the intercalibration model was implemented. Annual night-time light composites in Beijing, China, from 1992 to 2010 were taken as the study material, and the results show that the multi-temporal calibrated night-time light data have higher correlation with gross domestic production (GDP) (R 2 = 0.8734) and urban population (UP) (R 2 = 0.9269) than those of the uncalibrated images (with the R 2 values 0.7963 and 0.8575, respectively). Furthermore, the data inconsistency from different night-time light satellites in the same year was reduced with the proposed algorithm. The results demonstrate that the proposed algorithm is effective in intercalibrating the Defense Meteorological Satellite Program's Operational Linescan System (DMSP-OLS) images automatically.