Delineating Urban High-Risk Zones of Disease Transmission: Applying Tensor Decomposition to Trajectory Big Data

Tianhua Lu, Wenjia ZhangIsprs Int. J. geo inf 2025 · DOI

TL;DR

This paper introduces a novel method for identifying urban high-risk zones for disease transmission by applying CP tensor decomposition to large-scale mobile phone trajectory data. By analyzing spatiotemporal patterns and community overlaps, it offers a more dynamic and precise approach to pandemic containment compared to traditional methods.

What Problem Does This Paper Address?

Effectively containing large-scale pandemics requires accurately identifying "high-risk zones" for targeted interventions. Traditional methods often rely on static demographic data or simplistic spatial analysis, failing to capture the dynamic interplay of human mobility and social contact that drives disease spread. This leads to either overly broad, economically damaging lockdowns or insufficient measures that allow for covert transmission. The challenge lies in moving beyond static administrative boundaries to dynamically delineate risk zones that reflect the complex, multi-dimensional nature of human movement and interaction.

Methods

The core of this research is the application of CP (CANDECOMP/PARAFAC) tensor decomposition. This technique is used to analyze high-dimensional, spatiotemporal trajectory data.

Here's a breakdown of the approach:

  1. Data Aggregation: Individual mobility trajectories are aggregated into a grid-based origin-destination (OD) matrix. This matrix captures daily mobility patterns.
  2. Tensor Construction: These OD matrices are stacked over time (152 days in this study) to form a three-dimensional tensor. This tensor represents the spatiotemporal interactions within the city, with dimensions for origin grid, destination grid, and time.
  3. CP Tensor Decomposition: The constructed tensor is decomposed into a sum of rank-one sub-tensors. Each sub-tensor represents a distinct "community" with a unique spatiotemporal activity pattern. This is the key innovation, allowing for the identification of overlapping community structures.
  4. Community Analysis: The spatial distribution and temporal patterns of these identified communities are analyzed.
  5. Entropy Calculation: To identify high-risk zones, an entropy-based measurement is used. This quantifies the degree of "mixing" or co-presence of individuals from different communities in specific geographic locations. Higher entropy indicates a greater potential for inter-community transmission.

Datasets

The study utilizes anonymized mobile phone signaling data from China Unicom for Shenzhen residents. This dataset comprises over 500 million travel records from approximately 180,000 individuals, spanning from January 1 to May 31 in both 2019 and 2020.

Key Insights and Ideas

  • Dynamic Risk Zones: Traditional administrative boundaries are poor proxies for disease transmission risk. High-risk zones are defined by the convergence of human activity patterns, not by arbitrary lines on a map.
  • Community Fragmentation: The COVID-19 pandemic significantly fragmented urban community structures. In 2020, the number of detected communities doubled compared to 2019, reflecting a shift from large-scale, city-wide mobility to more localized activity patterns due to restrictions.
  • Temporal Rhythms Matter: Even with spatial overlap, distinct temporal activity patterns can reduce contact probability. Communities with synchronized temporal rhythms are more likely to create sustained transmission pathways.
  • Entropy as a Risk Metric: The degree of community mixing (quantified by entropy) in a location is a strong indicator of its potential as a disease transmission hotspot. "Melting pot" areas with diverse community presence pose a higher risk.
  • Urban Villages as Hotspots: Areas like urban villages (e.g., Baishizhou) and industrial zones, characterized by high population density and mixed land use, are identified as significant high-risk zones due to intense spatiotemporal co-location.

Key Results

  • The CP tensor decomposition successfully identified 12 distinct communities in Shenzhen during the COVID-19 impact phase (2020), compared to 6 communities in the pre-pandemic year (2019).
  • The analysis revealed significant spatial overlaps between communities, particularly in southern districts and industrial areas, highlighting these as critical zones for inter-community transmission.
  • The entropy-based analysis identified specific high-risk zones with high degrees of community mixing, correlating with reported outbreak incidents in areas like Baishizhou.
  • The study demonstrated that communities with synchronized temporal rhythms and stable spatial distributions (Communities 1-4 in 2020) were more likely to generate sustained transmission dynamics.

Limitations and Future Work

  • Causality: While tensor decomposition identifies patterns, it cannot elucidate the causal mechanisms behind them. Aggregated data limits the ability to model relationships with individual or environmental factors.
  • Data Granularity: The use of aggregated geographic units prevents the direct illustration of comparative advantages against studies using precise individual trajectory data.
  • Validation: Direct quantitative validation against official, geolocated epidemiological data is limited by the unavailability of fine-grained historical case data.
  • Single-Source Bias: Data from a single mobile network operator may introduce sampling bias. Future work could integrate data from multiple carriers for a more comprehensive model.

Future research directions include:

  • Incorporating additional data sources to construct multi-dimensional tensors and analyze interaction mechanisms.
  • Constructing tensors using non-aggregated individual trajectories for more direct methodological comparisons.
  • Conducting quantitative validation if fine-grained epidemiological data becomes accessible.
  • Fusing data from multiple carriers to create a more representative urban population model.

Summary updated

BibTeX
@article{lu2025delineating,
  author = {Tianhua Lu and Wenjia Zhang},
  title = {{Delineating Urban High-Risk Zones of Disease Transmission: Applying Tensor Decomposition to Trajectory Big Data}},
  journal = {Isprs Int. J. geo inf},
  year = {2025},
  doi = {10.3390/ijgi14080285},
}
Figure