An Algorithm for Inter-calibration of Time-Series DMSP/OLS Night-Time Light Images

The night-time lights (NTL) captured by the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP/OLS) are widely used as proxy for studying human activities. Lack of on-board calibration necessitates inter-calibration of time-series images acquired by different satellites before using them in long-term studies. Pseudo-invariant features (PIFs) i.e., stable night-light features over time, are often used as references for calibrating the temporal images. This study presents an algorithm for inter-calibration of DMSP/OLS time-series images using a semi-automatic PIFs identification procedure through the combined use of Getis statistic (Gi*) and coefficient of variation (CV). Annual stable NTL images over India from 1992 to 2006 are used in this process. While Gi* helps finding lit clusters of spatial homogeneity, CV is useful in finding areas of low spatial variability. The combined use of both these statistic across the time-series images thus helps identifying the PIFs. The identified PIFs are then used to develop calibration models for all the uncalibrated images taking one year image as a reference. Among the three parameter estimation methods evaluated in this study, least trimmed squares is most suitable for inter-calibration. The quality of calibrated images is evaluated using three metrics, sum of lights, gross domestic product and urban population. The calibration outputs are also compared with that of Elvidge et al. (Energies, 2009. https://doi.org/10.3390/en20300595). We find that the algorithm developed in this study for inter-calibration of time-series NTL images provides good results. The PIFs identification procedure proposed here is primarily data dependent and thus minimizes the human errors introducible by manual selection method.