Modeling urban air pollution with optimized hierarchical fuzzy inference system

Area of Science:

• Environmental Science
• Computer Science
• Public Health

Background:

• Urban air pollution modeling is essential for environmental exposure assessments and epidemiological studies.
• Existing models face limitations due to data uncertainty and inflexibility, particularly in developing countries.
• There is a need for advanced models with appropriate spatial and temporal resolutions.

Purpose of the Study:

• To develop a novel hierarchical fuzzy inference system (HFIS) for urban air pollution modeling.
• To address challenges of data uncertainty and model inflexibility in air pollution assessment.
• To improve the accuracy and applicability of air pollution models in diverse conditions.

Main Methods:

• A three-step approach was employed: geospatial information system (GIS) and probabilistic data preprocessing, hierarchical structure generation, and multi-objective particle swarm optimization (MOPSO) for simultaneous accuracy and complexity optimization.
• The model was designed to handle large-scale, high-dimensional air pollution data.
• Fivefold cross-validation was used to evaluate model performance.

Main Results:

• The developed HFIS demonstrated strong predictive capabilities for daily and annual mean PM2.5 and NO2 concentrations.
• Key performance metrics included RMSEs of (8.13, 0.78) for daily PM2.5, (4.96, 0.80) for annual PM2.5, (5.63, 0.79) for daily NO2, and (2.89, 0.83) for annual NO2.
• Comparative analysis confirmed the benefits of probabilistic preprocessing, multi-objective optimization, and hierarchical structure.

Conclusions:

• The developed hierarchical fuzzy inference system (HFIS) offers a robust and accurate solution for urban air pollution modeling.
• The model's features, including advanced preprocessing and optimization, enhance its utility for environmental exposure assessments and epidemiological studies.
• This approach shows significant potential for application in regions with data limitations.