A Multiscale Land Use Regression Approach for Estimating Intraurban Spatial Variability of PM2.5 Concentration by Integrating Multisource Datasets

Yuan Shi, Alexis K.h. Lau, Edward Ng, Hung Chak Ho, Muhammad BilalInternational Journal of Environmental Research and Public Health 2021 · DOI

TL;DR

This study introduces a novel multiscale Land Use Regression (LUR) approach to map fine-scale PM2.5 pollution in urban environments, specifically Hong Kong. By independently modeling at city and neighborhood levels and integrating diverse data sources, they achieve improved spatial estimation of PM2.5, crucial for understanding and mitigating urban air quality issues.

What Problem Does This Paper Address?

Air pollution, particularly PM2.5, is a significant environmental and health concern in large, high-density cities, especially in Asia. The complex interplay of urban structures and atmospheric processes leads to highly variable PM2.5 concentrations across different spatial scales, from city-wide patterns to street-level variations. Existing modeling approaches often struggle to capture this multiscale complexity effectively. Traditional deterministic models can be computationally intensive and limited in spatial coverage, while statistical models, like standard LUR, may not adequately differentiate between broad urban influences and localized micro-environmental factors. This paper aims to bridge this gap by developing a multiscale LUR framework that can provide both broad spatial coverage and capture fine-grained intraurban variability in PM2.5 concentrations.

Methods

The core of this research is a multiscale Land Use Regression (LUR) modeling approach. Unlike many previous studies that integrate predictors at various scales into a single model, this study adopts a two-stage strategy:

  1. City-Scale LUR: This stage uses long-term monitoring data and broad-scale predictors (e.g., AOD, land use, meteorological data) to model seasonal average PM2.5 concentrations across the entire city. The authors employed Geographically and Temporally Weighted Regression (GTWR), a method that accounts for spatial and temporal variations in predictor-response relationships, offering better performance than standard regression.
  2. Neighborhood-Scale LUR: This stage focuses on capturing fine-scale spatial variability. It utilizes in-situ PM2.5 sampling data collected at specific sites and incorporates detailed, micro-scale predictors. Here, the model predicts the variation in PM2.5 (APM2.5) relative to the site average, rather than the absolute concentration. Multiple Linear Regression (MLR) was used for this stage, as the microscale effects are considered more geolocation-independent.

The final PM2.5 map is generated by combining the outputs from these two scales, leveraging the city-scale model for overall coverage and the neighborhood-scale model to downscale and refine the spatial estimates.

Datasets

The study integrates several types of datasets:

  • Air Quality Data:
    • Long-term PM2.5 monitoring data: Hourly data from 16 EPD (Environmental Protection Department) air quality monitoring stations across Hong Kong (2018-2019).
  • In-situ PM2.5 sampling data: High-resolution (1-second interval) PM2.5 measurements collected via a portable backpack unit at ten selected neighborhood-scale sites.
  • Satellite Data:
    • MODIS Multi-Angle Implementation of Atmospheric Correction (MAIAC) AOD: 1-km resolution Aerosol Optical Depth data, used as a proxy for PM2.5 at the city scale.
  • Meteorological Data:
    • Near-surface data (air temperature, relative humidity, wind speed, rainfall, mean sea level pressure) from the Hong Kong Observatory (HKO).
  • Upper-air sounding indices.
  • Spatial Data:
    • Land Use Data: Five-class reclassified land use map (residential, commercial, industrial, governmental/facilities, greening/open space) at 10m resolution.
  • Population Density Data: WorldPop Global Project data at 100m resolution.
  • Road Network Data: Extracted from OpenStreetMap (OSM), including road density metrics and road segment centrality/accessibility metrics.
  • Building Morphological Data: * City-scale: Frontal Area Index (FAI) calculated within circular buffers.
  • Neighborhood-scale: Point-based FAI and Sky View Factor (SVF) calculated at sampling locations.

Key Insights and Ideas

  • Multiscale Modeling is Crucial: The study strongly supports the idea that a single modeling scale is insufficient for capturing urban air pollution variability. Separating city-wide influences from localized micro-environmental factors leads to more accurate and nuanced estimations.
  • Building Morphology Matters at the Microscale: At the neighborhood level, detailed building characteristics like the Frontal Area Index (FAI) and Sky View Factor (SVF) emerged as significant predictors of PM2.5. This highlights how the physical structure of the urban environment directly impacts air pollutant dispersion at the pedestrian level.
  • Road Network Centrality as a Proxy for Traffic: Road network metrics (centrality and accessibility) proved to be effective proxies for traffic distribution and, consequently, PM2.5 levels in street canyons, even without direct traffic volume data.
  • MAIAC AOD Outperforms Previous Products: The study found that the MAIAC AOD product, with its finer resolution, offered better prediction performance for city-scale PM2.5 estimation compared to coarser resolution products.
  • Transferability of Methods: The authors emphasize that the data requirements for their approach are generally accessible globally, suggesting the methodology's broad applicability beyond Hong Kong.

Key Results

  • City-Scale Model: The GTWR model for city-scale PM2.5 estimation achieved a high R² of 0.72 to 0.80 for the annual average, outperforming previous studies using AOD.
  • Neighborhood-Scale Model: The MLR model for estimating PM2.5 variation at the neighborhood scale achieved an R² of 0.508. This model successfully identified building morphological indices and road network centrality as key predictors.
  • PM2.5 Reduction: The study noted a significant reduction in annual average PM2.5 in Hong Kong from 45.8 µg/m³ (2003-2015) to 24.5 µg/m³ (2018-2019), suggesting the effectiveness of regional air pollution control efforts.

Limitations and Future Work

The authors acknowledge several limitations:

  • AOD Data Filtering: The filtering scheme used for MAIAC AOD data, while necessary to extend coverage, might have introduced some unreliable data. Future work should focus on evaluating these uncertainties.
  • Seasonal Models: The current GTWR models are developed separately for each season. Future research could explore developing models that account for different weather types more dynamically.
  • Traffic Proxy: While road network topology metrics were used as a proxy for traffic, using actual traffic volume data from simulations could further reduce uncertainties.
  • Extrapolation: The neighborhood-scale regression model is only valid within the numerical range of input predictor variables, meaning it does not have full spatial coverage of the city.

Future work suggested includes:

  • External Validation: Validating the models in other testbed cities.
  • Dynamic Weather Modeling: Incorporating more dynamic weather classifications into the models.
  • Improved Traffic Data: Utilizing computational traffic simulation data for more accurate traffic impact assessment.
  • Integrating Other Pollutants: Extending the multiscale approach to other air pollutants.

Summary updated

Abstract

Poor air quality has been a major urban environmental issue in large high-density cities all over the world, and particularly in Asia, where the multiscale complex of pollution dispersal creates a high-level spatial variability of exposure level. Investigating such multiscale complexity and fine-scale spatial variability is challenging. In this study, we aim to tackle the challenge by focusing on PM2.5 (particulate matter with an aerodynamic diameter less than 2.5 µm,) which is one of the most concerning air pollutants. We use the widely adopted land use regression (LUR) modeling technique as the fundamental method to integrate air quality data, satellite data, meteorological data, and spatial data from multiple sources. Unlike most LUR and Aerosol Optical Depth (AOD)-PM2.5 studies, the modeling process was conducted independently at city and neighborhood scales. Correspondingly, predictor variables at the two scales were treated separately. At the city scale, the model developed in the present study obtains better prediction performance in the AOD-PM2.5 relationship when compared with previous studies (R2¯ from 0.72 to 0.80). At the neighborhood scale, point-based building morphological indices and road network centrality metrics were found to be fit-for-purpose indicators of PM2.5 spatial estimation. The resultant PM2.5 map was produced by combining the models from the two scales, which offers a geospatial estimation of small-scale intraurban variability.
BibTeX
@article{shi2021multiscale,
  author = {Yuan Shi and Alexis K.H. Lau and Edward Ng and Hung Chak Ho and Muhammad Bilal},
  title = {{A Multiscale Land Use Regression Approach for Estimating Intraurban Spatial Variability of PM2.5 Concentration by Integrating Multisource Datasets}},
  journal = {International Journal of Environmental Research and Public Health},
  year = {2021},
  doi = {10.3390/ijerph19010321},
}
Figure