Regional PM2.5 pollution forecasting using a hybrid model based on multi-scales feature fusion and deep learning...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Regional haze pollution, particularly PM2.5, is a growing environmental concern.
• Existing early warning models for regional haze pollution are insufficient.
• Accurate prediction of PM2.5 is crucial for pollution management and public health.

Purpose of the Study:

• To develop and validate a novel deep learning model for accurate regional PM2.5 pollution prediction.
• To forecast PM2.5 concentrations for all cities within the Chengdu-Chongqing urban agglomeration simultaneously.
• To assess the model's performance against baseline models and real-world pollution data.

Main Methods:

• Utilized hourly PM2.5 concentration and meteorological data from January 1, 2021, to December 31, 2023.
• Developed a multi-input-multi-output deep learning framework named the multi-scales feature fusion regional pollution prediction network (MSFRPM).
• The MSFRPM model integrates temporal dependencies, inter-city pollutant transport, and meteorological factors.

Main Results:

• The MSFRPM model demonstrated significantly superior prediction performance compared to baseline models.
• The model effectively captured complex spatiotemporal dependencies in PM2.5 pollution.
• In 2023, moderate pollution was dominant, with pollution concentrated in winter; model predictions aligned with observations.

Conclusions:

• The MSFRPM model offers a robust solution for accurate regional PM2.5 pollution forecasting.
• Effective prediction capabilities are vital for coordinated regional pollution management and early warning systems.
• The study highlights the potential of deep learning in addressing complex environmental challenges like regional haze.