Tracking Seamless All-Hour PM_2.5 in China Using a Gridded Surface Visibility-Based Transformer Model.

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Accurate spatiotemporal monitoring of PM2.5 is crucial for environmental and health impact assessments.
• Existing satellite methods for PM2.5 estimation are limited by cloud cover and lack of nighttime aerosol optical depth (AOD) data, causing fragmented retrievals.

Purpose of the Study:

• To develop a novel method for seamless, all-hour PM2.5 monitoring.
• To address the limitations of current satellite-based PM2.5 products, particularly the nighttime data gap.

Main Methods:

• Introduction of surface visibility (SV) as an alternative to AOD.
• Development of a gridded SV-based transformer model (GSVTM) integrating multisource meteorological and environmental data.
• Utilizing multihead attention mechanisms and residual networks to model the relationship between SV and PM2.5.

Main Results:

• The GSVTM enables seamless all-hour PM2.5 tracking at 6.25 km and hourly resolution across China.
• Hourly scale: R² of 0.80 and RMSE of 15.14 μg m⁻³.
• Daily scale: R² of 0.89 and RMSE of 9.73 μg m⁻³, comparable to existing satellite products.
• Successfully captured PM2.5 transport dynamics during a large-scale pollution event.

Conclusions:

• The GSVTM represents a significant advancement in continuous PM2.5 monitoring.
• The model effectively overcomes the nighttime data gap in satellite-based PM2.5 products.
• Provides reliable, real-time data for understanding diurnal patterns and impacts of PM2.5 at various scales.