Diagnosing drivers of PM 2.5 simulation biases from meteorology, chemical composition, and emission sources using an efficient machine learning method

Shuai Wang, Mengyuan Zhang, Yueqi Gao, Peng Wang, Qingyan Fu, Hongliang Zhang2023 · DOI

Abstract

Chemical transport models (CTMs) are widely used for air pollution modeling but suffer from significant biases due to uncertainties in simplified parameterization, meteorological fields, and emission inventories. Accurate diagnosis of simulation biases is critical for model improvement, especially for fine particulate matter ($PM_{2.5}$). This study designed an efficient machine learning (ML) method to diagnose the drivers of Community Multiscale Air Quality (CMAQ) model biases from three perspectives: meteorology, chemical composition, and emission sources. Using source-oriented CMAQ data for 2019, the ML models demonstrated strong fitting ability. Results show that CMAQ underestimates $PM_{2.5}$ by -19.25 to -2.66 μg/m³, particularly in winter and during high-pollution events. Secondary organic aerosols (SOA) and residential emission sources were identified as the largest contributors to simulation bias, highlighting significant uncertainties in these areas. This ML-based approach provides a valuable, efficient complement to traditional mechanism-based methods for model refinement.

Introduction

Chemical Transport Models (CTMs), such as the Community Multiscale Air Quality (CMAQ) model, are fundamental tools for air quality management and policy-making. However, these models frequently exhibit biases when compared to real-world observations. These inaccuracies stem from simplified chemical mechanisms, errors in meteorological inputs, and uncertainties in emission inventories.

Traditionally, identifying the specific drivers of these biases required exhaustive sensitivity testing, such as Monte Carlo simulations, which are computationally demanding and often rely on restrictive prior assumptions. This research addresses the need for a more efficient, data-driven diagnostic framework to pinpoint the origins of model errors, enabling targeted improvements in air quality forecasting.

Methodology

The researchers developed a diagnostic framework utilizing LightGBM (Light Gradient Boosting Machine), a high-performance gradient boosting algorithm.

Simulation and Data

The study utilized a source-oriented version of the CMAQ model to simulate air quality across China for the year 2019. This version "tags" $PM_{2.5}$ by its source sector (e.g., industry, energy, residential, transportation). Observational data was obtained from the China National Environmental Monitoring Centre (CNEMC), while meteorological inputs were generated using the Weather Research & Forecasting (WRF) model.

Machine Learning Framework

Rather than predicting $PM_{2.5}$ concentrations directly, the ML models were trained to predict the simulation bias—the difference between CMAQ outputs and ground-station observations. Three distinct models were constructed to analyze bias drivers:

  1. Meteorological Factors: Including wind speed, relative humidity, and surface pressure.
  2. Chemical Composition: Analyzing components like sulfate, nitrate, ammonium, and secondary organic aerosols (SOA).
  3. Emission Sources: Evaluating contributions from various anthropogenic and biogenic sectors.

By calculating the relative importance of these features, the study identified the primary factors responsible for model inaccuracies.

Key Results and Insights

The Dominance of Secondary Organic Aerosols (SOA)

A critical finding was that SOA represented the largest chemical contributor to $PM_{2.5}$ simulation bias, ranging from 13.8% to 22.6% across different regions. This suggests that current mathematical representations of organic gas-to-particle conversion remain overly simplified in CTMs.

Residential Sector Uncertainties

Among emission sources, the residential sector (primarily heating and cooking) contributed most significantly to the bias (approximately 20.2%). This highlights the difficulty in accurately tracking small-scale coal and biofuel combustion compared to large, regulated industrial sources.

Regional Meteorological Drivers

The impact of meteorology on model bias varied geographically:

  • North China Plain: Relative humidity was the primary driver.
  • Pearl River Delta: Cloud cover dominated the bias.
  • Northwest China: Soil moisture was the key factor, primarily affecting dust simulations.

The CMAQ model generally underestimated $PM_{2.5}$ by -19.25 to -2.66 μg/m³. These underestimations were most severe during winter and spring, as well as during "heavy haze" events. The ML models achieved high diagnostic accuracy, with cross-validation $R^2$ values reaching up to 0.80 for meteorological bias.

Discussion and Future Directions

While the ML-based approach is highly efficient, it faces challenges such as feature collinearity (e.g., the correlation between temperature and humidity) and a current focus on local surface data rather than vertical atmospheric transport.

The researchers emphasize that ML should be viewed as a "compass" to guide traditional mechanism-based model improvements. Future work will aim to extend this framework to other pollutants like ozone ($O_3$) and incorporate more complex vertical transport dynamics to further refine air quality simulations.

Summary updated

Abstract

Chemical transport models (CTMs) are widely used for air pollution modeling, which suffer from significant biases due to uncertainties in simplified parameterization, meteorological fields, and emission inventories. Accurate diagnosis of simulation biases is critical for improvement of models, interpretation of results, and efficient air quality management, especially for the simulation of fine particulate matter (PM2.5). In this study, an efficient method based on machine learning (ML) was designed to diagnose the drivers of the Community Multiscale Air Quality (CMAQ) model biases in simulating PM2.5 concentrations from three perspectives of meteorology, chemical composition, and emission sources. The source-oriented CMAQ were used to diagnose influences of different emission sources on PM2.5 biases. The ML models showed good fitting ability with small performance gap between training and validation. The CMAQ model underestimates PM2.5 by -19.25 to -2.66 μg/m3 in 2019, especially in winter and spring and high PM2.5 events. Secondary organic components showed the largest contribution to PM2.5 simulation bias for different regions and seasons (13.8–22.6 %) among components. Relative humidity, cloud cover, and soil surface moisture were the main meteorological factors contributing to PM2.5 bias in the North China Plain, Pearl River Delta, and northwestern, respectively. Both primary and secondary inorganic components from residential sources showed the largest contribution (12.05 % and 12.78 %), implying large uncertainties in this sector. The ML-based methods provide valuable complements to traditional mechanism-based methods for model improvement, with high efficiency and low reliance on prior information.
BibTeX
@misc{wang2023diagnosing,
  author = {Shuai Wang and Mengyuan Zhang and Yueqi Gao and Peng Wang and Qingyan Fu and Hongliang Zhang},
  title = {{Diagnosing drivers of PM <sub>2.5</sub> simulation biases from meteorology, chemical composition, and emission sources using an efficient machine learning method}},
  year = {2023},
  doi = {10.5194/egusphere-2023-1531},
}
Figure