Using a deep convolutional neural network to predict 2017 ozone concentrations, 24 hours in advance

Area of Science:

• Environmental science
• Atmospheric chemistry
• Artificial intelligence in environmental monitoring

Background:

• Ozone pollution poses significant risks to public health and ecosystems.
• Accurate forecasting of ozone concentrations is crucial for implementing timely mitigation strategies.
• Existing forecasting models often face limitations in accuracy and computational efficiency.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) model for predicting 24-hour ozone concentrations.
• To assess the model's performance across various ambient monitoring stations in Texas.
• To explore the model's robustness with varying input data availability.

Main Methods:

• Utilized a deep convolutional neural network (CNN) architecture.
• Input data included historical meteorological data (wind, temperature) and air pollution concentrations (NOx, ozone).
• Trained and validated the model using data from 21 Continuous Ambient Monitoring Stations (CAMS) in Texas (2014-2017).

Main Results:

• The CNN model achieved a yearly index of agreement (IOA) above 0.85 for 19 out of 21 stations, indicating acceptable accuracy.
• The model demonstrated robust performance even with limited meteorological input variables and for stations with diverse monthly ozone trends.
• A consistent underprediction of daily maximum ozone concentrations was observed, highlighting an area for future model refinement.

Conclusions:

• The developed CNN model provides accurate and computationally efficient 24-hour ozone concentration predictions.
• The model's performance across diverse Texas CAMS stations validates its applicability.
• Future research should focus on improving the prediction of daily maximum ozone levels to enhance the early warning system's effectiveness.