[Predictive Model for O₃ in Shanghai Based on the KZ Filtering Technique and LSTM].

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Context:

• Ozone (O3) pollution poses significant environmental and health risks, necessitating accurate long-term prediction models.
• Traditional O3 prediction methods often struggle with complex temporal dynamics and meteorological influences.
• Identifying key meteorological drivers is crucial for enhancing O3 forecasting accuracy.

Purpose:

• To develop an improved method for predicting long-term ozone (O3) concentration series.
• To enhance feature selection for meteorological variables influencing O3 levels using enhanced maximal minimal redundancy (mRMR) and support vector regression (SVR).
• To assess the predictive performance of a long short-term memory (LSTM) network for high O3 concentration periods.

Summary:

• A Kolmogorov-Zurbenko (KZ) filter was employed to decompose O3 sequences, separating baseline and short-term components.
• The enhanced mRMR technique combined with SVR identified pressure, temperature, humidity, boundary layer height, and wind direction as optimal meteorological features.
• An LSTM model was applied to predict high O3 concentration periods in Shanghai, demonstrating strong performance with R2 values ranging from 0.83 to 0.86.

Impact:

• Decomposing the O3 sequence significantly improved prediction accuracy compared to standard methods.
• Feature selection effectively identified crucial meteorological variables, maintaining model predictive performance.
• The study provides a robust framework for enhancing O3 forecasting, aiding in pollution control and public health strategies.