Impacts of Catchments Derived from Fine-Grained Mobility Data on Spatial Accessibility (Short Paper)

Alexander Michels, Jinwoo Park, Bo Li, Jeon-Young Kang, Shaowen WangGIScience 2023 · DOI

TL;DR

This paper investigates how different methods for defining "catchment areas" (regions accessible within a certain travel time) from fine-grained mobility data impact spatial accessibility analyses. They found that the choice of method, specifically between convex hulls and alpha shapes, significantly alters accessibility measures and that these differences disproportionately affect socially vulnerable populations, potentially leading to policy misinterpretations.

What Problem Does This Paper Address?

Spatial accessibility is crucial for understanding how people can reach essential services like healthcare. Traditional methods often rely on broad administrative zones, which can mask local disparities. More recent approaches use fine-grained mobility data (like GPS traces) to define more precise "catchment areas" around service locations. However, there's a lack of consensus on the best way to derive these catchments from point data. This paper addresses the question of how different geometric algorithms (convex hull vs. alpha shapes) used to create these catchments influence the resulting spatial accessibility measures, and importantly, how these differences might relate to social vulnerability.

Methods

The authors employ the Enhanced Two-Step Floating Catchment Area (E2SFCA) method to measure spatial accessibility. This is a standard technique that calculates the ratio of supply (e.g., ICU beds) to demand (e.g., population) within defined travel-time thresholds.

The core novelty lies in how they define the "catchment areas" for these travel-time zones:

  • Convex Hull: This method creates the smallest convex polygon that encloses a set of points. It's a simple approach but can oversimplify the shape of accessible areas.
  • Alpha Shapes: This algorithm, based on Delaunay triangulation, creates more complex, non-convex shapes. It allows for more nuanced representations of accessible regions by filtering triangles based on a parameter 'α' (alpha). A lower 'α' value results in a shape closer to the convex hull, while higher values create more intricate shapes that better fit the point distribution.

They compare these two methods by calculating catchments for various travel-time thresholds (10, 20, and 30 minutes) around hospitals and then applying the E2SFCA method.

Datasets

  • Population and Social Vulnerability Data: Centers for Disease Control (CDC) Social Vulnerability Index (SVI) at the census tract level (American Community Survey 5-Year, 2014-2018).
  • Hospital and ICU Bed Data: Homeland Infrastructure Foundation-Level Data Geoplatform.
  • Road Network Data: OpenStreetMap, processed using the osmnx Python package.

No new benchmarks were introduced; the focus was on evaluating existing methods with real-world data.

Key Insights and Ideas

  • Catchment Shape Matters Significantly: The choice between a convex hull and an alpha shape (even with a low alpha value) can lead to substantial differences in measured spatial accessibility. This highlights that seemingly minor methodological choices can have a large impact on research outcomes.
  • Alpha Shapes Offer Nuance: Alpha shapes, by their nature, can capture more complex and realistic shapes of accessible areas compared to the often overly broad convex hull.
  • Disproportionate Impact on Vulnerable Populations: The study found a statistically significant correlation between the "alpha-access effect" (how accessibility changes with alpha) and social vulnerability. Specifically, socially vulnerable communities tend to show a decrease in measured accessibility as the alpha value increases.
  • Convex Hulls May Over-Report Access: This correlation suggests that convex hulls and low-alpha shapes might over-report access for socially vulnerable populations. This could lead to policy decisions that underestimate the need for support in these areas.
  • Spatial Autocorrelation is Present: The differences in accessibility measures are not random but exhibit spatial patterns, indicating that the choice of catchment method has localized and correlated effects.

Key Results

  • The study found that the choice of catchment construction (convex hull vs. alpha shape) significantly impacts spatial accessibility metrics.
  • A weak but statistically significant negative correlation (Kendall's τ: -5.45e-02, p-value: 3.06e-03) was found between the "alpha-access effect" and the CDC Social Vulnerability Index (SVI).
  • This implies that census tracts with higher social vulnerability tend to experience a greater decrease in measured accessibility as the alpha value increases, meaning convex hulls and low-alpha shapes may over-report access in these areas.

Limitations and Future Work

  • Generalizability: The study was conducted in Cook County, Illinois. The authors suggest future work should apply this methodology to a variety of cities to assess the generalizability of their findings.
  • Data Variety: While OpenStreetMap data was used, the authors note that other mobility datasets (like Floating Car Data or temporally dynamic mobility data) could offer different insights.
  • Individual-Level Mobility: The current methods capture general patterns, but real-world mobility is highly individual. Future research could explore how to better incorporate individual-level characteristics into catchment definitions.
  • Policy Implications: The authors emphasize the need for careful consideration of methodological choices, as they can have unintended policy implications, particularly for socially vulnerable populations. They suggest that alpha shapes with appropriate values are more likely to accurately represent mobility data.

Summary updated

BibTeX
@inproceedings{michels2023impacts,
  author = {Alexander Michels and Jinwoo Park and Bo Li and Jeon-Young Kang and Shaowen Wang},
  title = {{Impacts of Catchments Derived from Fine-Grained Mobility Data on Spatial Accessibility (Short Paper)}},
  booktitle = {GIScience},
  year = {2023},
  doi = {10.4230/lipics.giscience.2023.52},
}
Figure