A Study of the Non-Linear Relationship Between Urban Morphology and Vitality in Heritage Areas Based on Multi-Source Data and Machine Learning: A Case Study of Dalian

He Li, Li MiaoISPRS Int. J. Geo Inf 2025 · DOI

TL;DR

This study uses multi-source geospatial data and machine learning (XGBoost and SHAP) to analyze the complex, non-linear relationship between urban morphology and vitality in Dalian's heritage and urban areas. It reveals distinct factors influencing vitality during the day and night, offering crucial insights for targeted urban planning and heritage preservation.

What Problem Does This Paper Address?

Urban vitality, a key indicator of a city's dynamism and attractiveness, is crucial for sustainable development and community well-being. While urban morphology (the physical form of cities) is known to influence vitality, existing research often focuses on general urban areas and tends to use linear models, potentially missing complex, non-linear relationships and threshold effects. Furthermore, the specific influence of urban morphology on the vitality of heritage areas—which have unique characteristics and preservation constraints—remains less understood. This paper aims to bridge this gap by investigating the non-linear impacts of various urban form elements on both daytime and nighttime vitality in both heritage and general urban areas of Dalian, China.

Methods

The study employs a quantitative research framework that combines big geospatial data with advanced machine learning techniques.

  • Data Collection and Indicator Extraction: A comprehensive set of urban morphology indicators was extracted from multi-source data, covering ground plan (street system, block pattern), building form, and land use patterns. Vitality was measured using daytime indicators (Points of Interest, specifically small food facilities) and nighttime indicators (nighttime lighting data).
  • Machine Learning Models:
    • XGBoost (Extreme Gradient Boosting): This algorithm was chosen for its high accuracy in analyzing complex, non-linear relationships and identifying threshold effects, outperforming traditional linear regression and other machine learning models like Random Forest.
  • SHAP (SHapley Additive exPlanations): To address the "black-box" nature of machine learning models, SHAP values were used to interpret the model's predictions, quantifying the contribution of each urban form element to vitality and visualizing these non-linear relationships.
  • Comparative Analysis: The study explicitly compares the findings between heritage areas and general urban areas, as well as between daytime and nighttime vitality, to highlight specific influencing factors and their spatial and temporal variations.

Datasets

The research utilizes a variety of publicly available geospatial datasets for Dalian, China:

  • Points of Interest (POI) Data: Collected from Gaode Map and Baidu Map (2022-2024) to identify fundamental urban functions (food, hotel, shopping, etc.).
  • Building Data: Crawled from Gaode Map (post-2022) for spatial coordinates, base contours, and floors.
  • Street Data: Sourced from OpenStreetMap (OSM) for road network analysis.
  • Nighttime Lighting Data: Obtained from the High Resolution Earth Observation System of Systems (HBEOS) for measuring nighttime vitality.
  • Small Catering Business Data: Collected from the Volkswagen website (2022) as a proxy for daytime vitality.
  • Housing Price Data: Sourced from Beike.
  • Population Data: Acquired from East View Cartographic.

No new benchmark datasets were introduced; rather, existing open data sources were integrated and processed.

Key Insights and Ideas

  • Diurnal Variation is Key: The factors influencing urban vitality differ significantly between daytime and nighttime. This highlights the need for distinct planning strategies for different times of the day.
  • Heritage vs. Urban Areas Diverge: Urban morphology elements that drive vitality are not uniform across heritage and general urban areas. Land use patterns are paramount for heritage areas, while ground plan characteristics are more influential in urban areas.
  • Non-Linearity Matters: Simple linear relationships are insufficient. The study reveals critical thresholds for various urban form elements (e.g., block area, public transport proximity) where their impact on vitality changes dramatically.
  • Diversity Breeds Vitality (Especially in Heritage): In heritage areas, a higher density of commercial and service facilities (RPOI) and a mix of land uses (EPOI) are crucial for attracting people and fostering vibrancy, aligning with Jane Jacobs' "diversity" theory.
  • Street System's Dual Role: While a well-connected street system (RIQ) is generally beneficial, an excessive number of intersections can negatively impact vitality by reducing continuity and safety.

Key Results

  • Model Performance: The XGBoost model demonstrated superior performance across all vitality metrics (R², MAE, RMSE) compared to OLS and RF models for both heritage and urban areas, both day and night. Cross-validation results showed high R² values (e.g., 0.778 for heritage daytime vitality).
  • Spatial Distribution: Vitality in Dalian generally follows a "core-edge" distribution, decreasing from the city center outwards. Heritage areas, however, show more evenly distributed and relatively high nighttime vitality.
  • Factor Importance:
    • Daytime Vitality: In heritage areas, RPOI (commercial/service density) and EPOI (land use diversity) are highly influential. In urban areas, land use functions like commercial (COM), residential (RES), and recreational (REC) are dominant.
  • Nighttime Vitality: Building form elements, particularly building density (BD) and floor area ratio (FAR), have a notable positive impact on nighttime vitality in heritage areas. For urban areas, building form and ground plan elements are significant.
  • Threshold Effects: For instance, urban vitality shows a sharp increase with block area (A) up to 4 hm², after which the effect stabilizes. Similarly, housing prices have a critical threshold above which they negatively impact vitality.

Limitations and Future Work

  • Vitality Measurement: The authors acknowledge that vitality is a multi-dimensional and abstract concept, and the chosen indicators may lead to some bias or incomplete results.
  • Methodology Expansion: While machine learning was effective, the study suggests exploring deep learning models (CNN, GNN) for more accurate feature extraction from street view images and spatial data.
  • Real-time Data Integration: Incorporating real-time data streams (social media, mobile signals) could capture dynamic urban changes more effectively.
  • Predictive Modeling: Future work could focus on time-series modeling (LSTM, Transformer) for long-term trend prediction and agent-based modeling to simulate policy impacts.
  • Data Availability: The authors note that they do not have permission to share the data used in the study.

Summary updated

BibTeX
@article{li2025study,
  author = {He Li and Li Miao},
  title = {{A Study of the Non-Linear Relationship Between Urban Morphology and Vitality in Heritage Areas Based on Multi-Source Data and Machine Learning: A Case Study of Dalian}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2025},
  doi = {10.3390/ijgi14040177},
}
Figure