TL;DR
This study introduces a novel "Travel-related Heat Exposure Index" (THEI) that combines ride-hailing data with satellite-derived land surface temperature to dynamically assess urban heat exposure. By analyzing this index in Chengdu, China, the research identifies high-risk zones and reveals that urban form and transportation infrastructure are key drivers city-wide, while demographic and socioeconomic factors become more influential in these high-risk areas.
What Problem Does This Paper Address?
Cities are increasingly facing the threat of extreme heat events, exacerbated by the Urban Heat Island (UHI) effect. Traditional methods for assessing heat exposure often rely on static land surface temperature (LST) data, which fail to capture the dynamic nature of human activity and its interaction with thermal environments. This overlooks how people move through cities and the varying levels of heat they encounter during their daily commutes and travels. Understanding this "travel-related" heat exposure is crucial for developing effective, spatially targeted urban planning strategies to build climate resilience and ensure public health.
Methods
The study employs a multi-stage methodological framework:
- Travel-related Heat Exposure Index (THEI) Construction: This is the core innovation. THEI is calculated by normalizing ride-hailing trip intensity (as a proxy for human activity/exposure frequency) and LST (as a proxy for thermal hazard intensity). This follows the "hazard intensity × exposure frequency" model.
- Spatial Identification of High-Exposure Risk Zones (HERZ): Local Indicators of Spatial Association (LISA) are used to identify spatial clusters of high THEI values, effectively delineating areas with significantly elevated travel-related heat exposure.
- Nonlinear Impact Analysis: An XGBoost model is utilized to explore the complex, nonlinear relationships between various built environment factors and THEI. This is performed for both the entire study area and the identified HERZ.
- Interpretability Techniques: To understand the XGBoost model's findings, Shapley Additive Explanations (SHAP) values are used for variable importance and directionality, while Partial Dependence Plots (PDP) and Two-Dimensional Partial Dependence Plots (2D PDP) are employed to visualize marginal and interaction effects of built environment variables on THEI.
Datasets
- Ride-hailing Trajectories: Anonymized data from DiDi's GAIA Open Data Initiative, covering July 2021 in Chengdu, China. This dataset provides detailed spatial and temporal information on human mobility.
- Land Surface Temperature (LST): MODIS MOD11A1 product data for July 2021, resampled to a 600m grid resolution.
- Built Environment Data: A comprehensive set of variables derived from various sources, including:
- Spatial network/POI data from Gaode Maps (for intersection density, road density, public transport stops, etc.).
- Population density data (WorldPop 2021).
- Housing price data (web scraped from Anjuke).
- Remote sensing data (for NDVI).
Key Insights and Ideas
- Dynamic Exposure is Key: Static LST is insufficient; understanding how people move through hot environments is critical for accurate heat exposure assessment.
- Scale Matters: The drivers of heat exposure differ significantly between the city-wide scale and concentrated high-risk zones.
- Vegetation is a Universal Coolant: NDVI consistently shows a negative association with THEI, highlighting its crucial role in mitigating heat risks across all scales. Its effect is even more pronounced in high-risk areas.
- Infrastructure vs. Demographics: City-wide, urban form (like intersection density and FAR) and transportation accessibility are dominant. In HERZ, human activity (population density, housing price) and localized infrastructure (road density) become more influential.
- Nonlinear Relationships: The impact of built environment factors isn't linear. For example, the cooling effect of vegetation plateaus, and the risk from increased intersections might level off after a certain point.
Key Results
- THEI Distribution: High THEI values are concentrated in central urban districts and transit hubs, showing significant spatial heterogeneity.
- HERZ Identification: Approximately 14% of the study area was identified as HERZ, but these zones contain over 28% of the population, indicating a disproportionate concentration of heat risk.
- XGBoost Model Performance: The models achieved strong explanatory power, with R² values of 0.72 for the entire study area and 0.63 for the HERZ subset.
- Dominant City-Wide Drivers: Intersection density (InS), subway/bus station density (SBS), and floor-area ratio (FAR) were key drivers of THEI city-wide.
- Dominant HERZ Drivers: Population density (Pop), housing price (HP), and road density (RD) became more influential in HERZ.
- NDVI's Amplified Role: NDVI showed stronger negative SHAP values in HERZ, suggesting its cooling effect is more critical under intensified human activity and environmental stress.
Limitations and Future Work
- Ride-hailing Bias: The reliance on ride-hailing data may introduce socioeconomic bias, potentially underrepresenting vulnerable populations (elderly, low-income, non-ride-hailing users) who might face higher risks. Future work could incorporate more inclusive mobility data (e.g., public transit cards, shared bikes).
- Potential Overfitting: Despite robust modeling techniques, potential overfitting risks remain, especially in the smaller HERZ subset. Future research could explore spatial extensions of XGBoost or geospatial weighting.
- Temporal Scope: The study focused on July. Future research should expand to include seasonal variations and finer temporal resolutions (e.g., hourly).
- Data Sources: Incorporating additional data sources like pedestrian counts or wearable sensors could provide a more comprehensive picture of heat exposure dynamics.
- Socioeconomic Dimensions: Further exploration of socioeconomic dimensions of heat vulnerability, including age and income, is needed for more equitable adaptation strategies.
Summary updated