| Summary: | Nighttime light data has been widely used in socioeconomic research and in studies related to light pollution. The most commonly used nighttime light global datasets include the Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) and Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS-DNB). However, the quantization level and spatial resolution of DMSP-OLS data are poor. A combined DMSP-OLS and VIIRS-DNB time series would be valuable, and we propose that the challenges associated with DMSP-OLS data can be overcome by integrating multi-resolution and multi-temporal data. This research introduces a new remote sensing index based on multi-source data fusion, which we use to improve the resolution and quantization range of DMSP-OLS nighttime light data through deep learning methods. The main results of the study are as follows: First, we propose a nighttime light enhancement index (NLEI) that can be used to develop synthetic DMSP-OLS data series with improved spatial resolution and quantization level. Second, features from VIIRS-DNB data and NLEI in 2013 are extracted and trained using a variational autoencoder framework based on the principle of zonal and blocking processing to obtain a simulated VIIRS-DNB dataset from 1992 to 2012. Finally, consistency verification with VIIRS-DNB data from 2012 shows a pixel level R2 validation of 0.94, and the R2 of accuracy verification at the provincial level is 0.98. The correlation coefficient values between the simulated VIIRS-DNB data from 1996, 2006, 2000, and 2010 and the corresponding DMSP-OLS radiance calibrated nighttime light (RNTL) product are 0.92, 0.94, 0.95, and 0.96, respectively. The simulated VIIRS-DNB dataset obtained in this study from 1992 to 2011 has a resolution of 500 m and higher quantization levels, which can be combined with VIIRS-DNB data to support economic, social, urban change and light pollution research from 1992 to 2022.
|
|---|
