Related Experiment Video
Updated: Jun 24, 2026
16:14
Trajectory Data Analyses for Pedestrian Space-time Activity Study
Published on: February 25, 2013
13.4K
Deep transfer learning for spatiotemporal mapping of PM2.5 nitrate across China: Addressing small data challenges in
Xi Zheng1, Haiyan Meng1, Zixiang Zhao2
1College of Carbon Neutrality Future Technology, Sichuan University, Chengdu, Sichuan 610065, China; Department of Environmental Science and Engineering, Sichuan University, Chengdu, Sichuan 610065, China.
Journal of Hazardous Materials
|April 9, 2025
Summary
Transfer learning effectively reconstructed PM2.5 nitrate concentrations in China despite limited data. This approach improved air quality modeling and identified pollution hotspots, aiding environmental management.
Area of Science:
- Environmental Science
- Atmospheric Chemistry
- Data Science
Background:
- Increasing particulate matter (PM2.5) nitrate concentrations in China pose significant health risks.
- Limited nationwide monitoring data hinders accurate spatiotemporal distribution research of PM2.5 nitrate.
- Data scarcity presents a major challenge for environmental modeling and policy development in China.
Purpose of the Study:
- To develop a novel transfer learning method to overcome data scarcity in reconstructing PM2.5 nitrate concentrations across China.
- To generate a high-resolution daily nitrate dataset for China from 2005-2020.
- To analyze the spatiotemporal trends of PM2.5 nitrate pollution and their correlation with policy interventions.
Main Methods:
- A deep neural network (Source model) was pre-trained using extensive US nitrate data.
- The Source model was fine-tuned with limited Chinese nitrate observations to create a Transfer model.
- Validation using cell-based cross-validation (R²=0.72, RMSE=7.5 μg/m³) demonstrated the Transfer model's improved generalizability.
Main Results:
- The Transfer model significantly improved the reconstruction of PM2.5 nitrate concentrations compared to traditional machine learning methods.
- Generated a 10 km gridded daily nitrate dataset for China (2005-2020), revealing severe pollution in the North China Plain and economic zones.
- Nitrate concentrations showed a fluctuating upward trend (2005-2014) followed by a reduction (2015-2020), aligning with air pollution control policies.
Conclusions:
- Transfer learning is an effective strategy for addressing small data challenges and data disparity in environmental research.
- The developed dataset and model provide valuable insights for understanding and managing air pollution in China.
- This approach is particularly beneficial for resource-limited regions and developing countries in environmental management.
