Main content area

Study on radiometric intercalibration methods for DMSP-OLS night-time light imagery

Li, Chang, Ye, Jia, Li, Shice, Chen, Guangping, Xiong, Hao
International journal of remote sensing 2016 v.37 no.16 pp. 3675-3695
algorithms, least squares, models, radiometry, remote sensing, satellites, urbanization, variance
The Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) night-time light image sets has been successfully used in various fields to monitor temporal and spatial distributions, such as urbanization and socioeconomic activities. However, the radiometric calibration of night-time light images has long been a problem that limits the application of night-time light image sets in multi-temporal analyses. The key to an intercalibration model is to automatically extract the reference pixels with stable lights and to simultaneously remove unstable (or variant) light pixels viewed as outliers (or gross error). This paper systemically compares five weighted least squares regression (WLSR) algorithms for radiometric intercalibration to determine the method with the highest accuracy, including (1) classic methods: empirical rule (ER), Danish method (DM), and posterior variance estimation (PVE); and (2) state-of-the-art methods: random sample consensus (RANSAC) and least median of squares (LMedS). Moreover, a more objective adjusted root mean square error (RMSE) method is proposed to evaluate accuracy. Through the experiments, we systemically analyse the performance of different estimators and propose recommendations for optimizing intercalibration for specific applications. Moreover, the study reveals that gross error approximatively obeys the stochastic or expansion model in the radiometric intercalibration of DMSP-OLS image sets. Overall, LMedS works best and is proposed to intercalibrate the radiometric values of night-time light image sets.