Abstract
Short-term monitoring campaigns are commonly used for assessing intra-urban air pollution (AP), capturing spatial variations at non-reference sites while measuring temporal variations at reference sites. This study aims to predict continuous AP data at non-reference sites, aligning their temporal patterns with those of reference sites. Using data from the New York City Community Air Survey (NYCCAS), which collected bi-weekly AP measurements at reference sites and seasonal measurements at non-reference sites, we examine four methods for estimating bi-weekly PM2.5, NO2, and EC at non-reference sites. These methods include two model-based approaches—mixed and fixed-effect generalized linear models—and two non-model-based data-driven approaches. The mixed and fixed-effect models outperformed the data-driven approaches, showing R2 values above 0.80 for PM2.5 in all years except 2019 (0.76), while data-driven approaches had slightly lower R2 values and higher RMSE. The performance of both model-based approaches was consistent across the years, with the lowest RMSE values observed for PM2.5, NO2, and EC. Additionally, our methodology successfully applied the Principal Component Analysis (PCA) to reduce climate variable dimensionality, contributing to model improvement. These findings provide robust bi-weekly AP estimates at non-reference sites, offering a valuable tool for spatiotemporal air quality modeling and health risk assessments in urban settings.
Original language | English |
---|---|
Article number | 102294 |
Journal | Urban Climate |
Volume | 59 |
DOIs | |
State | Published - 1 Feb 2025 |
Externally published | Yes |
Keywords
- Air pollutants
- Generalized linear models
- NYCCAS
- Predictive modeling
- Spatio-temporal inverse distance weighting
ASJC Scopus subject areas
- Geography, Planning and Development
- Environmental Science (miscellaneous)
- Urban Studies
- Atmospheric Science