Regional dynamics of VIIRS-like nighttime lights 1992-2023

🌐 A Global Annual Simulated VIIRS Nighttime Light Dataset (1992-2023)

• Authors: Xiuxiu Chen, Zeyu Wang, Feng Zhang, Guoqiang Shen, Qiuxiao Chen
• Published in: Scientific Data (2024)
• DOI: https://doi.org/10.1038/s41597-024-04228-6

🔬 Background & Summary

• Nighttime light (NTL) data is widely used to measure human activity, urbanization, and socioeconomic trends.
• Existing NTL datasets (DMSP-OLS & NPP-VIIRS) have limited temporal coverage and inconsistencies.
• The study presents a new dataset, SVNL (Simulated VIIRS NTL), using deep learning to provide a continuous, high-resolution (500m) dataset from 1992-2023.
• SVNL allows for long-term monitoring of human activity and urbanization trends.

📚 Data Collection

• DMSP-OLS Stable NTL (1992-2013): Oldest available nighttime light dataset.
• NPP-VIIRS Annual VNL V2 (2012-2023): Higher resolution and more accurate than DMSP.
• Landsat NDVI (1992-2013): Used to improve calibration and reduce saturation.
• Other datasets: Extended NTL datasets (ChenVNL, LiDNL), GDP data, and administrative boundaries.

🎯 Research Framework

• Step 1: Preprocess and calibrate DMSP-OLS NTL data for consistency.
• Step 2: Develop and train a U-Net super-resolution network (NTLSRU-Net) for cross-sensor calibration.
• Step 3: Apply the trained model to convert DMSP NTL into VIIRS-like data (1992-2011).
• Step 4: Merge simulated VIIRS data (1992-2011) with real VIIRS data (2012-2023) to create SVNL dataset.

🤖 U-Net Super-Resolution Model

• The model enhances spatial resolution and corrects inconsistencies between DMSP & VIIRS.
• Modifications:
  • Removed pooling layers to preserve spatial details.
  • Used transposed convolutions for up-sampling.
  • Integrated Landsat NDVI data to correct for saturation.
• Model trained using DMSP & VIIRS data from 2012-2013 and then applied for historical reconstruction.

🌍 Evaluation & Validation

• Accuracy Assessment:
  • Histogram and scatter plot comparisons between SVNL & real VIIRS data (2012-2013).
  • High correlation observed at pixel, city, province, and national levels.
• Spatial Pattern Validation:
  • SVNL data closely matches real VIIRS data, avoiding saturation issues in urban areas.
• Temporal Trend Validation:
  • SVNL aligns well with economic indicators (GDP growth) and urban expansion patterns.

🔄 Key Findings

• SVNL dataset provides a high-resolution, long-term global record of nighttime lights.
• Outperforms previous datasets by maintaining spatial and temporal consistency.
• Enables more accurate studies on urbanization, socioeconomic trends, and environmental monitoring.
• Publicly accessible for researchers and policymakers.

💡 Conclusion

• The SVNL dataset fills a crucial gap in long-term nighttime light data.
• Facilitates detailed analysis of human activities from 1992-2023.
• Future work includes further refinements using additional remote sensing data.
• Dataset Access: Original data repository
• GEE dataset Access: Awesomme GEE community catalog
• Exploratory Tool: GEE web app by Carlos Mendez

See web app in full screen HERE