Problem: DMSP/OLS nighttime light data has been widely used to monitor human activities from space. However, data acquired from different sensors are not directly comparable (Fig. 1), due to a number of factors. Thus, change analysis with this data source is limited.

Research goals: 1) designing a robust method to calibrate DMSP/OLS nighttime light time series; 2) generating a new time series with this method; 3) evaluating its performance.

Methodology: The densest part of the scatterplot between the reference image and a target image forms a ridgeline that represents the majority of lit pixels (Fig.2a). Data points selected along the ridgeline are then used to build a linear regression model for calibrating the target image (Fig.2b).

Results and conclusions:
1. Visual (Fig. 3) and quantitative assessment results show that the RSR method can successfully generate a consistent nighttime light time series from DMSP/OLS data.
2. The RSR method avoids subjective selection of pseudo-invariant features on ground and used all lit pixels to develop calibrating models, thus it is more robust than existing methods. It further reduces computation burden to build models.
3. The newly generated nighttime light time series has the potential to study human activities at night over time. 

Fig. 1. DMSP/OLS nighttime light time series of Beijing, China. Large gaps can be easily observed among data segments from different sensors.

Fig. 2. The proposed Ridgeline Sampling and Regression (RSR) method.

Fig. 3. Raw and calibrated DMSP/OLS nighttime light time series of selected regions worldwide.