Enhancing County-Level GDP Estimation Accuracy With Downscaled NPP-VIIRS Nighttime Light Data

Nighttime light (NTL) data have provided invaluable support for estimating gross domestic product (GDP). However, commonly used global-scale NTL data acquired by the Visible Infrared Imaging Radiometer Suite (VIIRS) aboard the Suomi National Polar-orbiting Partnership (Suomi NPP) satellite suffers from relatively coarse spatial resolution (15 arcsec), limiting its potential for fine-scale applications. In this study, we employed a deep learning-based NTL Conditional Multi-Scale Downscaling Model (NTL-CMDM), incorporating multi-source scale factors as conditional constraints, to downscale NPP-VIIRS NTL data (500m) to a finer spatial scale of 130m. Furthermore, the effectiveness of downscaled NTL data for county-level GDP estimation was evaluated through comparison with NPP-VIIRS and Luojia1-01 NTL data in 205 Chinese county-level cities with varying economic development levels in the Beijing, Shanghai, and Guangzhou regions. The results show that regressions between GDP and both Total Nighttime Light (TNL) and Nighttime Light Area (NLA) using the downscaled NTL data (R > 0.782 and R > 0.634) achieve higher fitting accuracy than those using NPP-VIIRS NTL data (R > 0.716 and R > 0.110), and approach the performance of Luojia1-01 NTL data (R > 0.796 and R > 0.267). Additionally, the downscaled NTL data improve the accuracy of GDP estimates by reducing the relative error (RE) between estimated and statistical GDP compared to NPP-VIIRS NTL data. More importantly, the spatial distribution of GDP estimates derived from the downscaled NTL data more closely aligns with statistical GDP data, reflecting a more realistic geographic pattern. This study demonstrates that the downscaled NTL data generated by NTL-CMDM offer a promising data source for more accurate and fine-scale socioeconomic analysis.