Deep Understanding of Urban Dynamics from Imprint Urban Toponymic Data Using a Spatial-Temporal-Semantic Analysis Approach

Yurong Chen, Feng Zhang, Xinba Li, Chuanrong Zhang, Ninghua Chen, Zhenhong Du, Renyi Liu, Bo WangIsprs Int. J. geo inf 2021 · DOI

TL;DR

This paper introduces a novel spatial-temporal-semantic framework to analyze urban dynamics using street names as "imprint data." By combining text classification of street names with geographical analysis, the study reveals how urban development, cultural shifts, and economic focus have evolved in Hangzhou over the past century, offering a deeper understanding than traditional methods.

What Problem Does This Paper Address?

Understanding urban dynamics—the complex, long-term evolution of cities driven by human activity and land use changes—is crucial for sustainable urban planning. Existing research often relies on two-dimensional spatiotemporal analysis using data like satellite imagery or human mobility patterns, which can be limited in temporal scope and lack contextual depth. This paper addresses the gap in understanding long-term urban development by proposing a method to leverage the rich, often overlooked, semantic and historical information embedded within urban toponyms, specifically street names. The challenge lies in effectively processing this textual data and integrating it with spatial and temporal information to extract meaningful insights into a city's historical trajectory and characteristics.

Methods

The core of this research is a spatial-temporal-semantic analytical framework.

  1. Street Name Classification: The authors employed the FastText model, a shallow neural network known for its efficiency in short text classification, to categorize street names. They experimented with various Chinese text segmentation methods (single char, exact, full mode, search engine) and n-gram configurations to optimize classification accuracy. The chosen approach involved single-character segmentation and n-gram analysis, achieving a classification accuracy (F-measure) of 0.699. This classification was then used to group street names into 31 types, broadly categorized as "nature-driven" and "society-driven."
  2. Spatial Visualization and Analysis: Kernel Density Estimation (KDE) was used to visualize the spatial distribution and density of street names across different time periods. This method helps identify areas of high street name concentration, indicating significant development or change.
  3. Featured Functional Area Aggregation: A three-step process was developed to identify key functional areas within the city. This involved:
  4. Extracting center points from street line data.
  5. Aggregating these points using a specific threshold (3500 meters) to form initial areas.
  6. Adjusting these areas using a buffer based on street length to better reflect the spatial aggregation of street names. This process allowed for the identification of areas characterized by specific types of street names, such as administrative, scenic, or industrial zones.

  7. Temporal Analysis: Street names were categorized into eight distinct time periods, spanning from the Qing dynasty to the new millennium, based on historical events. By analyzing the proportion of different street name types within each period, the study tracked changes in urban focus and development over time.

Datasets

  • Hangzhou Gazetteer: This dataset, compiled by the Hangzhou Civil Affairs Bureau and Hangzhou Toponymy Committee, provided historical street name information, including naming reasons, dates, and historical changes. It served as the source for semantic and temporal data.
  • OpenStreetMap: This dataset provided current road network data, including spatial geometry, street names, and types, which were used to match with the historical semantic data.

The combination of these datasets resulted in a spatial dataset of 903 streets with historical name information, spanning from the Song dynasty to 2010.

Key Insights and Ideas

  • Street names as historical archives: The study powerfully demonstrates that street names are not just labels but "imprint data" that record a city's historical evolution, cultural shifts, and development priorities.
  • Spatial-Temporal-Semantic synergy: The true power lies in integrating these three dimensions. Analyzing street names across time, space, and meaning reveals a richer narrative of urban development than any single dimension alone.
  • Evolution of urban focus: Hangzhou's street names show a clear shift from an early emphasis on natural scenery and historical sites to a later focus on economic development, industry, and technology, reflecting broader societal changes.
  • Functional area identification: The aggregation method effectively highlights distinct functional zones within the city, linking them to specific types of street names and providing insights into urban planning and development patterns.
  • Toponyms as indicators of urban planning: The study suggests that the types and distribution of street names can reflect planned urban development, such as the designation of administrative centers or industrial zones.

Key Results

  • Classification Accuracy: The FastText model achieved a classification accuracy (F-measure) of 0.699 for street names, with 18 types showing accuracy above 60%, accounting for 87% of the data. Certain types like "digit" and "political motivations" achieved precision and recall rates of 1.000.
  • Temporal Trends:
    • "Historic site" was consistently the most common street name type across all periods.
  • "Commemorative personage" was prominent before 1990 but declined, while "wish" and "industry" increased in later periods, indicating a shift towards economic aspirations.
  • Time periods G and H showed increased street naming related to "wish" and "industry," reflecting rapid urban expansion and a focus on economic development.
  • Spatial Distribution:
    • Changed street names expanded from old urban zones to suburban areas, with density peaks shifting to suburban zones in later periods.
  • The latest period (H) showed four main density peak areas: the old urban zone near the lake, the canal zone, the Binjiang high-tech zone, and the Xiasha university zone, highlighting the coexistence of historical and new development areas.
  • Functional Areas: The study successfully identified and aggregated featured functional areas, such as administrative, scenic, industry, culture/sport, and wish areas, providing visual representations and word clouds of dominant themes within each. For instance, industry areas were concentrated in Binjiang, Xiaoshan, and Yuhang districts, with keywords like "industry," "park," and "high-tech."

Limitations and Future Work

  • Classification Standard Limitations: The authors acknowledge that the classification of street names into "nature-driven" and "society-driven" could be more detailed. While they identified 31 subcategories, the lack of universal standards for street name classification means these types might not be directly applicable to Western countries.
  • Accuracy of Urban Space Recognition: While the method effectively identifies interesting areas and development directions, it's not intended to precisely map every industrial layout or function.
  • Future Work:
    • Web-based Interactive Maps: The authors suggest developing interactive web maps for public use to visualize street name data, enhancing accessibility and understanding.
  • Multiscale Analysis: The framework's temporal and spatial scales could be adapted for multiscale analysis to provide a more comprehensive understanding of urban dynamics.
  • Semantic Information for Functional Zones: Further exploration of semantic information is recommended for identifying urban functional zones and highlighting regional differences beyond temporal and spatial activity patterns.

Summary updated

BibTeX
@article{chen2021deep,
  author = {Yurong Chen and Feng Zhang and Xinba Li and Chuanrong Zhang and Ninghua Chen and Zhenhong Du and Renyi Liu and Bo Wang},
  title = {{Deep Understanding of Urban Dynamics from Imprint Urban Toponymic Data Using a Spatial-Temporal-Semantic Analysis Approach}},
  journal = {Isprs Int. J. geo inf},
  year = {2021},
  doi = {10.3390/ijgi10050278},
}
Figure