Quantifying Individual PM2.5 Exposure with Human Mobility Inferred from Mobile Phone Data

Zhaoping Hu, Le Huang, Xi Zhai, Tao Yang, Yaohui Jin, Yanyan XuSustainability 2023 · DOI

TL;DR

This paper introduces a novel big data framework to precisely quantify individual PM2.5 exposure by integrating mobile phone mobility data with high-resolution PM2.5 concentration. This approach moves beyond traditional aggregate methods to provide hourly, individual-level exposure estimates, revealing significant spatiotemporal variations in exposure across Shanghai.

What Problem Does This Paper Address?

Accurately assessing individual exposure to air pollutants like PM2.5 is crucial for understanding public health risks and developing effective environmental policies. Traditional methods often rely on aggregated population data or focus solely on residential locations, failing to capture the dynamic nature of human mobility and its impact on exposure. This "Neighborhood Effect Averaging Problem" (NEAP) can lead to imprecise conclusions. The paper aims to bridge this gap by developing a method that accounts for where individuals spend their time throughout the day, not just where they sleep.

Methods

The core of this research is a big data analytics framework that combines two key data sources:

  • Mobile Phone Mobility Data (Call Detail Records - CDRs): The researchers used CDR data from one million anonymous users in Shanghai, representing approximately 60 billion records. This data, collected from January to April 2014, captures user interactions with base stations, allowing for the inference of user locations and mobility patterns.
  • High-Resolution PM2.5 Concentration Data: To achieve fine-grained spatial and temporal resolution, the study fused two PM2.5 datasets:
  • Station Monitoring Data: Hourly data from fourteen fixed monitoring stations in Shanghai provided high temporal resolution but low spatial coverage.
  • China High Air Pollutants (CHAP) Data: This dataset offered daily average PM2.5 concentrations at a 1-km grid level across China, providing high spatial resolution.

A novel aspect of their methodology involves:

  1. Stay Location Recognition: A clustering method was employed to process CDR data, distinguishing between temporal and spatial proximity to identify "stay points" and infer users' daily trajectories, including their home and work locations.
  2. High-Resolution PM2.5 Data Fusion: They developed a "correction factor" to adjust the low-temporal-resolution CHAP data using the high-temporal-resolution station data. This process generated hourly PM2.5 concentration data for each 1-km grid in Shanghai.
  3. Individual Exposure Calculation: Two methods were used to calculate individual exposure:
  4. ER (Residence-based): Exposure calculated solely based on the user's home location.
  5. ES (Stay Behavior-based): Exposure calculated by integrating the user's mobility behavior (time spent at different locations) with the corresponding PM2.5 concentrations.

Datasets

  • Call Detail Records (CDRs): Data from one million anonymous users in Shanghai, collected from January to April 2014, comprising approximately 60 billion records. This data was provided by a communication operator and anonymized.
  • PM2.5 Concentration Data:
    • Hourly data from fourteen fixed monitoring stations in Shanghai (China National Environmental Monitoring Station).
  • Daily average PM2.5 surface concentration data at a 1-km grid level for Shanghai (China High Air Pollutants - CHAP dataset).

Key Insights and Ideas

  • Mobility Matters: The study strongly emphasizes that considering individual mobility patterns is crucial for accurate PM2.5 exposure assessment. The "Stay behavior based" exposure (ES) was found to be slightly higher than the "Home location based" exposure (ER), highlighting that time spent away from home in areas with higher pollution significantly impacts overall exposure.
  • Temporal Exposure Peaks: PM2.5 exposure tends to be higher during morning and evening rush hours, likely due to increased traffic-related emissions. This suggests that targeted interventions during these periods could be effective.
  • Seasonal Variations: January exhibited significantly higher PM2.5 exposure levels compared to February, March, and April, underscoring the severity of winter pollution.
  • Spatial Disparities: A clear west-to-east gradient in PM2.5 exposure was observed in Shanghai, mirroring the PM2.5 concentration distribution. This suggests that geographical factors and industrial activity play a significant role.
  • Residential Environment Dominance: Despite accounting for mobility, the study found that the residential environment still strongly correlates with individual exposure. Higher pollution around a user's residence generally leads to higher overall exposure.

Key Results

  • The average PM2.5 exposure for residences in Shanghai was calculated to be 60.37 ug·h·m⁻³ during the study period.
  • The "Stay behavior based" exposure was slightly higher than the "Home location based" exposure, indicating the importance of mobility.
  • Exposure levels were notably higher during morning (9-11 am) and evening (6-8 pm) rush hours.
  • January showed significantly higher average exposure than other months.
  • Spatial analysis revealed a trend of higher exposure in western Shanghai and lower exposure in eastern Shanghai. Residents in the Pudong new area had the lowest average exposure, while those in the Jinshan district had the highest.

Limitations and Future Work

  • Station Monitoring Imbalance: The authors acknowledge a limitation in the spatial distribution of PM2.5 monitoring stations, which are concentrated in urban centers, potentially affecting the accuracy of suburban data. They suggest improving the balance of station placement and exploring low-cost sensors.
  • Data Privacy: The CDR data used was from 2014 and anonymized to protect user privacy. While considered representative of general mobility patterns, it's not the most recent data.
  • Future Work: The researchers propose several avenues for future research:
  • Differentiating exposure for indoor and outdoor spaces.
  • Incorporating the effect of wearing masks on exposure estimation.
  • Further refining individual exposure estimation by analyzing detailed transition-based travel behavior.
  • Exploring finer-grained user behavioral data and trajectory data.

Summary updated

Abstract

Treatment of air pollution and health impacts are both crucial components of long-term sustainability. Measuring individual exposure to air pollution is significant to evaluating public health risks. In this paper, we introduce a big data analytics framework to quantify individual PM2.5 exposure by combining residents’ mobility traces and PM2.5 concentration at a 1-km grid level. Diverging from traditional approaches reliant on population data, our methodology can accurately estimate the hourly PM2.5 exposure at the individual level. Taking Shanghai as an example, we model one million anonymous users’ mobility behavior based on 60 billion Call Detail Records (CDRs) data. By integrating users’ stay locations and high-resolution PM2.5 concentration, we quantify individual PM2.5 exposure and find that the average PM2.5 exposure of residences in Shanghai is 60.37 ug·h·m−3 during the studied period. Further analysis reveals the unbalance of the spatiotemporal distribution of PM2.5 exposure in Shanghai. Our PM2.5 exposure estimation method provides a reliable evaluation of environmental hazards and public health predicaments confronted by residents, facilitating the formulation of scientific policies for environmental control, and thus advancing the realization of sustainable development.
BibTeX
@inproceedings{hu2023quantifying,
  author = {Zhaoping Hu and Le Huang and Xi Zhai and Tao Yang and Yaohui Jin and Yanyan Xu},
  title = {{Quantifying Individual PM2.5 Exposure with Human Mobility Inferred from Mobile Phone Data}},
  booktitle = {Sustainability},
  year = {2023},
  doi = {10.3390/su16010184},
}
Figure