A global continuous 500 m nighttime light dataset (1992–2024) via NDVI-guided DMSP-OLS correction and U-TransNet cross-sensor harmonization

• Proposed a novel combination method to mitigate DMSP-OLS saturation effects. • Developed a new deep learning framework to generate cross-sensor VIIRS-DNB. • Produced a global, long-term annual VIIRS-DNB at 500 m resolution (1992–2024). Nighttime light (NTL) remote sensing is a widely used proxy for quantifying human activities, urbanization and economic dynamics. The two most commonly used NTL archives are the Defense Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) and the Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS-DNB). However, substantial differences in radiometric sensitivity, spatial resolution and measurement units between DMSP-OLS and VIIRS-DNB hinder the construction of continuous long-term NTL time series. To address this issue, we proposed and implemented a novel method for improving spatial resolution which combined a Landsat-based NDVI correction index (NDVI-corrected Nighttime Light Index, NWLI) to mitigate DMSP-OLS saturation and overflow effects, with a U-Net architecture augmented by a Transformer bottleneck (U-TransNet) for pixel-level cross-sensor mapping. The U-TransNet was trained on paired NWLI–VIIRS samples from 2013 to learn the nonlinear mapping from NWLI to VIIRS radiance, the trained model was then applied to predict NWLI for 1992–2012 on an annual basis to generate simulated VIIRS images. Finally, the simulated series were concatenated with observed VIIRS-DNB (2012–2024) to produce a continuous, VIIRS-compatible NTL dataset for 1992–2024 at 500 m resolution. Multi-scale validation indicated strong agreement with reference data, the pixel-level R 2 was 0.92, provincial/state-level and national-level R 2was 0.97 and 0.99, respectively. Comparisons with several radiometrically calibrated DMSP products (RNTL) also yield provincial R 2 values were greater than 0.90. The NTL time series products from 1992 to 2024 constitute a high-quality, spatio-temporally consistent observational baseline for long-term studies of urbanization, population dynamics and economic activity.