Air Quality Analysis

Overview

This project analyzes Beijing's air quality using multi-source time-series data, following the methodology of Zhang et al. (2017). The main goal is to disentangle the effects of meteorology from emission trends, providing a "weather-adjusted" view of PM2.5 pollution.

Approach

1. Data Integration & Preprocessing

• Combined hourly air quality and meteorological data from multiple monitoring stations (2013–2017).
• Created a unified DataFrame with a datetime index and station labels.
• Handled missing values and engineered features (e.g., cyclical time, wind direction encoding).

2. Exploratory Data Analysis (EDA)

• Explored spatial, temporal, and meteorological patterns in PM2.5.
• Visualized trends by station, hour, month, and wind direction.

3. Predictive Modeling

• Trained three models to classify PM2.5 air quality categories based on meteorological features:
  • Random Forest
  • XGBoost
  • LSTM (PyTorch)
• Evaluated models using accuracy, F1-score, and confusion matrices.

4. Weather-Adjusted PM2.5 Analysis

• For each year, replaced actual weather with long-term average ("baseline") weather conditions.
• Used trained models to predict what PM2.5 categories would have been under average weather.
• Compared observed and weather-adjusted trends to reveal the true impact of emission controls, independent of meteorological variability.

Key Findings

• Weather adjustment reveals greater air quality improvement than raw data suggests.
• In years with unfavorable weather (e.g., 2017), observed PM2.5 can appear worse than it truly is.
• Meteorological factors, especially wind speed and direction, are major drivers of short-term PM2.5 variation.

How to Run

1. Install dependencies:

   pip install -r requirements.txt

2. Train models:

   python -m src.train_rf
   python -m src.train_xgb
   python -m src.train_lstm

3. Run weather-adjusted analysis:

   python -m src.weather_adjusted_pm25
   python -m src.weather_adjusted_pm25_xgb
   python -m src.weather_adjusted_pm25_lstm

4. Check the output/ directory for visualizations and results.

References

• Zhang, Q., et al. (2017). "Cautionary tales on air-quality improvement in Beijing." PNAS.

---

This project demonstrates the importance of meteorological adjustment in air quality trend analysis and provides a reproducible workflow for similar studies.