Locally weighted total least-squares variance component estimation for modeling urban air pollution

Area of Science:

• Environmental Science
• Geospatial Analysis
• Statistical Modeling

Background:

• Land use regression (LUR) models are standard for urban air pollution estimation but often lack accuracy.
• Inaccurate stochastic models and measurement errors in variables limit the precision of traditional LUR models.
• There is a need for advanced methods to address non-stationary effects and variable errors in air pollution modeling.

Purpose of the Study:

• To propose and evaluate a novel method, locally weighted total least-squares variance component estimation (LW-TLS-VCE), for improved urban air pollution modeling.
• To address limitations of existing LUR models by incorporating non-stationary effects and measurement errors of dependent and independent variables.
• To enhance the accuracy of PM2.5 concentration estimation in urban environments.

Main Methods:

• Developed a locally weighted total least-squares (LW-TLS) regression to handle non-stationary effects and errors in variables simultaneously.
• Implemented variance component estimation for the stochastic model to achieve best linear unbiased estimation.
• Applied the LW-TLS-VCE method to model PM2.5 concentrations using meteorological, land use, and traffic data in Isfahan, Iran.

Main Results:

• The proposed LW-TLS-VCE method demonstrated increased efficiency in modeling PM2.5 concentrations compared to four other methods.
• Accounting for non-stationary effects and random errors in all variables significantly improved model performance.
• Accurate estimation of observation variance contributed to the enhanced accuracy of the proposed approach.

Conclusions:

• The LW-TLS-VCE method offers a more accurate and robust approach for urban air pollution modeling, particularly for PM2.5.
• Employing appropriate stochastic and functional models is crucial for maximizing the efficiency of air quality estimation.
• This study highlights the importance of addressing variable errors and spatial heterogeneity in environmental modeling.