This study presents a prolonged artificial nighttime-light dataset of China (PANDA-China) spanning from 1984 to 2020, generated using a Night-Time Light Convolutional Long Short-Term Memory (NTLSTM) network. The dataset is produced by integrating historical DMSP-OLS data with deep learning techniques to enhance temporal consistency and extend the time span of available nighttime light data. The NTLSTM model is trained to predict nighttime light patterns, and the results are adjusted using a modified RLOWESS method (MODEST) to ensure temporal consistency. The PANDA-China dataset demonstrates high accuracy, with an RMSE of 0.73, R² of 0.95, and a linear slope of 0.99 at the pixel level. It outperforms existing nighttime light datasets in terms of temporal consistency, longer time span, and correlation with socioeconomic indicators such as built-up areas, GDP, and population. The dataset provides valuable insights into urbanization dynamics and socio-economic changes over the past four decades. PANDA-China is freely accessible and offers unprecedented opportunities for studying economic and energy-related topics since 1984. The dataset is stored in TIF format and is compatible with software such as ArcGIS. The study also includes detailed comparisons with other datasets, highlighting the strengths and limitations of different approaches in capturing nighttime light dynamics and their correlations with socioeconomic factors.This study presents a prolonged artificial nighttime-light dataset of China (PANDA-China) spanning from 1984 to 2020, generated using a Night-Time Light Convolutional Long Short-Term Memory (NTLSTM) network. The dataset is produced by integrating historical DMSP-OLS data with deep learning techniques to enhance temporal consistency and extend the time span of available nighttime light data. The NTLSTM model is trained to predict nighttime light patterns, and the results are adjusted using a modified RLOWESS method (MODEST) to ensure temporal consistency. The PANDA-China dataset demonstrates high accuracy, with an RMSE of 0.73, R² of 0.95, and a linear slope of 0.99 at the pixel level. It outperforms existing nighttime light datasets in terms of temporal consistency, longer time span, and correlation with socioeconomic indicators such as built-up areas, GDP, and population. The dataset provides valuable insights into urbanization dynamics and socio-economic changes over the past four decades. PANDA-China is freely accessible and offers unprecedented opportunities for studying economic and energy-related topics since 1984. The dataset is stored in TIF format and is compatible with software such as ArcGIS. The study also includes detailed comparisons with other datasets, highlighting the strengths and limitations of different approaches in capturing nighttime light dynamics and their correlations with socioeconomic factors.