An Open-Source Framework of Generating Network-Based Transit Catchment Areas by Walking

Jian Yang, Diao Lin, Ruoxin Zhu, Jian Yang 0034, Liqiu MengIsprs Int. J. geo inf 2020 · DOI

TL;DR

This paper introduces an open-source framework for generating network-based transit catchment areas (TCAs) by walking. It addresses the need for accurate and efficient TCA generation, crucial for public transport planning, by proposing a novel methodology and releasing the implementation for public use.

What Problem Does This Paper Address?

Transit catchment areas (TCAs) are fundamental for public transport planning, helping to understand passenger origins, design urban infrastructure, and model ridership. However, accurately defining these areas, especially for pedestrian access, has been challenging. Existing methods often rely on simplified buffer zones (Euclidean distance) that overestimate coverage or use complex models that are computationally expensive or proprietary. There's a clear need for a method that accurately reflects real-world pedestrian travel along road networks, is computationally efficient, and is accessible to researchers and planners. This paper aims to fill this gap by providing an open-source, network-based approach.

Methods

The core of the proposed framework involves three main components:

  • Subgraph Construction: To speed up calculations, a localized subgraph is created around each transit facility. This involves building an R-tree for efficient spatial querying, projecting facility points to the nearest road segments, and extracting relevant road segments (sub-edges) that intersect a searching box around the projected point.
  • Extended Shortest Path (SP) Tree Construction: This is a key innovation. Instead of a standard SP tree, an "extended" version is built. This involves identifying "non-SP tree edges" (edges not part of the shortest path from the root) and inserting "break points" on them. This ensures that distances are accurately measured even when the shortest path deviates from a direct edge. This extended tree is crucial for accurate distance interpolation.
  • Contour Generation: Using the extended SP tree, edges are segmented, and a constrained Delaunay triangulation is built. This triangulation, which respects the road network structure, is then used to generate contour lines at the specified cut-off distance, defining the boundaries of the catchment area.

The framework is also generalized for directed road networks and non-point facilities (like multiple station entrances). For non-point facilities, a "virtual node-based method" is proposed as a more efficient alternative to simply dissolving individual catchment areas.

Datasets

The study uses two primary datasets for its case study and evaluation:

  • Metro Station Dataset: 1223 metro station entrances from 301 metro stations in Shanghai, collected via the Gaode map API.
  • Road Network Dataset: The road network of Shanghai, downloaded using the OSMNX Python package. This was modeled as an undirected graph for initial experiments.

The paper does not introduce a new benchmark dataset but uses existing real-world data to validate its method against established approaches.

Key Insights and Ideas

  • The Power of "Extended" Shortest Paths: The most significant insight is the development of the "extended shortest path tree." By explicitly accounting for non-shortest path edges and inserting break points, the method achieves much more accurate distance measurements along the road network, which is critical for defining realistic pedestrian access.
  • Constrained Triangulation is Key: Using a constrained Delaunay triangulation, which incorporates the road network edges as constraints, prevents inaccuracies that can arise from standard Delaunay triangulations intersecting road segments. This ensures that distance interpolations accurately follow the road network.
  • Open-Source for Accessibility: The decision to release the framework as an open-source C++ program is a major contribution, democratizing access to sophisticated TCA generation tools for researchers and planners who may not have access to commercial GIS software.
  • Efficiency Gains for Non-Point Facilities: The "virtual node-based method" for non-point facilities offers a substantial speedup compared to traditional dissolving methods, making it practical for large-scale analyses.

Key Results

  • High Accuracy: The proposed method achieves precision and recall scores above 90%, with an average F1 score comparable to or better than the ArcGIS service area tool, indicating high geospatial accuracy.
  • Computational Efficiency: The framework is significantly faster than the ArcGIS method, being at least twice as fast for point-based facilities. For non-point facilities, the virtual node-based method shows a dramatic reduction in running time, especially with an increasing number of entrances.
  • Visual Comparison: Visualizations show that the proposed method (and ArcGIS) correctly includes accessible edges and excludes inaccessible ones, outperforming simpler methods like convex hull or standard SP tree-based approaches.
  • Directed Graph Handling: The framework successfully differentiates between "to-facility" and "from-facility" catchment areas in directed road networks.

Limitations and Future Work

  • Road Density Impact: While not a strict limitation, the study notes that higher road density in urban areas generally correlates with better F1 scores, suggesting that interpolation accuracy might be lower in sparser areas.
  • Off-Network Area Interpolation: The triangulation-based interpolation, while accurate for on-network travel, might not guarantee perfect accuracy for off-network areas.
  • Future Work:
    • Non-Overlapping Catchment Areas: Developing methods to generate mutually exclusive catchment areas, potentially using network-based Voronoi polygons or grid-based assignment.
  • Other Transport Modes: Adapting the framework for modes other than walking, which might require different impedance models and handling larger cut-off distances.
  • User-Friendly Interface: Creating a more accessible interface, possibly as a QGIS plugin, to further enhance usability.
  • Incorporating Environmental Factors: Extending the framework to include factors like road quality or travel time (instead of just distance) by modifying the edge impedance.

Summary updated

BibTeX
@article{yang2020open,
  author = {Jian Yang and Diao Lin and Ruoxin Zhu and Jian Yang and Liqiu Meng},
  title = {{An Open-Source Framework of Generating Network-Based Transit Catchment Areas by Walking}},
  journal = {Isprs Int. J. geo inf},
  year = {2020},
  doi = {10.3390/ijgi9080467},
}
Figure