Machine-Learning-Based Near-Surface Ozone Forecasting Model with Planetary Boundary Layer Information.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Machine Learning Applications

Background:

• Surface ozone is a harmful air pollutant impacting health and the environment, necessitating accurate forecasting.
• Conventional and data-driven methods exist for ozone forecasting, with machine learning showing promise.
• Planetary Boundary Layer (PBL) height influences surface ozone levels but is often underutilized in forecasting models.

Purpose of the Study:

• To investigate the effectiveness of incorporating Planetary Boundary Layer (PBL) height as a feature in data-driven surface ozone forecasting models.
• To evaluate the performance of multilayer perceptron (MLP) and bidirectional long short-term memory (LSTM) models using PBL height data.

Main Methods:

• Developed and compared MLP and bidirectional LSTM models for 24-hour surface ozone forecasting.
• Utilized hourly measurement data and predicted PBL height from the Weather Research and Forecasting (WRF) model as input features.
• Evaluated model performance using Index of Agreement (IOA), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE).

Main Results:

• Both MLP and bidirectional LSTM models showed reduced MAE and RMSE when incorporating forecasted PBL height.
• No significant change in IOA was observed when comparing models with and without forecasted PBL height data.
• The inclusion of PBL height data demonstrably improved the accuracy of surface ozone concentration predictions.

Conclusions:

• Utilizing forecasted PBL height is an effective strategy to enhance the performance of data-driven surface ozone forecasting models.
• MLP and bidirectional LSTM models benefit from the inclusion of PBL height, leading to more accurate air quality predictions.
• This approach offers a pathway to improved public health advisories regarding dangerous ozone concentrations.