Model averaging with logistic autoregressive conditional peak over threshold models for regional smog.

Area of Science:

• Environmental Science
• Statistics
• Time Series Analysis

Background:

• Extreme air pollution events, particularly PM2.5, pose significant public health risks.
• Existing static models fail to capture the time-dependent nature of air pollutant concentrations.
• Accurate modeling of extreme smog events is crucial for effective environmental policy and public health interventions.

Purpose of the Study:

• To propose a novel dynamic generalized Pareto distribution (GPD) framework for modeling time-dependent peak over threshold (POT) events in PM2.5 time series.
• To introduce three autoregressive conditional Pareto (ACP) models with time-dependent parameters.
• To enhance model flexibility and predictive performance through logistic function autoregression and model averaging.

Main Methods:

• Developed three dynamic ACP models using GPD with time-varying scale and shape parameters.
• Incorporated past PM2.5 levels, other air quality factors (SO2, NO2, CO), and weather variables (temperature, humidity, wind speed) as predictors.
• Applied a logistic function for autoregressive structures on scale and shape parameters.
• Utilized model averaging with AIC and BIC criteria for optimal weight selection.
• Employed eight automatic threshold selection procedures and Maximum Likelihood Estimation (MLE) for parameter estimation.

Main Results:

• The proposed dynamic ACP models demonstrate superior performance in modeling extreme PM2.5 time series compared to static approaches.
• The logistic function autoregressive structure provides flexible and computationally efficient modeling of parameter dynamics.
• Model averaging effectively improves predictive accuracy.
• Automatic threshold selection procedures ensure objective model setup.
• MLE parameter estimation proved stable and reliable across simulations and real-world data.

Conclusions:

• The novel dynamic GPD framework, specifically the ACP models, offers a robust and accurate approach for analyzing and forecasting extreme air pollution events.
• The integration of time-varying parameters and external factors enhances the understanding of smog dynamics.
• The proposed methodology provides a valuable tool for environmental monitoring and risk assessment.