Air quality forecasting using artificial neural networks with real time dynamic error correction in highly polluted
Shivang Agarwal1, Sumit Sharma1, Suresh R1
1TERI, The Energy and Resources Institute, IHC Complex, Lodi Road, New Delhi 110003, India.
The Science of the Total Environment
|June 3, 2020
Summary
This study developed an Artificial Neural Network (ANN) model for forecasting air pollutants like PM2.5 and ozone in Delhi. Real-time correction significantly improved forecast accuracy for short-term air quality prediction.
Area of Science:
- Environmental Science
- Atmospheric Chemistry
- Data Science
Background:
- Air pollution poses significant risks in global megacities.
- Accurate short-term air quality forecasting is crucial for public health and exposure reduction.
Purpose of the Study:
- To develop and validate an Artificial Neural Network (ANN) model for forecasting key air pollutants (PM10, PM2.5, NO2, O3) in Delhi.
- To enhance forecast accuracy using a Real-Time Correction (RTC) mechanism.
Main Methods:
- An ANN model was trained using hourly pollution data and meteorological parameters from 2018.
- The model was implemented for real-time forecasting and validated using 2018 and 2019 data.
- A Real-Time Correction (RTC) module was integrated to dynamically adjust forecasts.
Main Results:
- The ANN model demonstrated good performance across all tested pollutants.
- Forecast accuracy improved significantly with the integration of RTC.
- Correlation coefficients ranged from 0.79-0.88 (Day0) to 0.49-0.68 (Day4), with ozone showing the least performance degradation.
Conclusions:
- The developed ANN model with RTC is effective for real-time air quality forecasting in highly polluted urban areas.
- RTC significantly enhances the reliability and accuracy of short-term air pollution predictions.
- The model provides valuable tools for managing air quality and mitigating public exposure in megacities.
