A Prolonged Artificial Nighttime-light Dataset of China (1984-2020)

Area of Science:

• Earth Observation
• Remote Sensing
• Geospatial Analysis

Background:

• Nighttime light remote sensing is crucial for monitoring human activities.
• Limited availability of long-term nighttime light datasets hinders comprehensive analysis.
• Synthesizing long-term products requires advanced methodologies.

Purpose of the Study:

• To develop a long-term, high-quality nighttime light dataset for China.
• To create a novel dataset covering the period from 1984 to 2020.
• To enable detailed analysis of China's socioeconomic development and human activity patterns.

Main Methods:

• Development of a Night-Time Light convolutional Long Short-Term Memory (LSTM) network.
• Application of the network to generate a 1-km annual nighttime light dataset (PANDA-China).
• Rigorous validation using pixel-level assessments (RMSE, R², slope) and comparisons with existing datasets.

Main Results:

• The PANDA-China dataset demonstrates high accuracy with an average RMSE of 0.73 and R² of 0.95.
• PANDA-China exhibits superior temporal consistency and longer nighttime light dynamics compared to other datasets.
• The dataset shows strong correlations with socioeconomic indicators like built-up areas, GDP, and population.

Conclusions:

• The PANDA-China dataset provides a reliable and unprecedented resource for studying nighttime light dynamics over four decades.
• This product enhances the understanding of China's socioeconomic evolution and human activity.
• The developed methodology offers a robust approach for creating similar long-term remote sensing products globally.