Urban Road Collapse Risk Assessment Study Based on InSAR Spatiotemporal Data

Juncai Jiang, Wenfeng Bai, Yizhao Wang, Fei Wang, Qinglun He, Long Chen, Yuming Qiao, Zhi Wang, Haitao Luo2024 · DOI

TL;DR

This paper introduces a novel approach to urban road collapse risk assessment by integrating InSAR-derived ground subsidence data with Long Short-Term Memory (LSTM) networks. This method effectively captures temporal deformation patterns, enabling a more dynamic and predictive risk assessment that overcomes the limitations of traditional, static methods.

What Problem Does This Paper Address?

Urban road collapses pose a significant threat to public safety and infrastructure, causing fatalities, injuries, and substantial economic losses. Traditional risk assessment methods often rely on numerous, static indicators, making them less effective at capturing the dynamic, evolving nature of ground deformation that precedes a collapse. There's a clear need for a more predictive and dynamic approach that can leverage continuous monitoring data to identify high-risk areas before a catastrophic event occurs.

Methods

The core of this study is the application of a Long Short-Term Memory (LSTM) network, a type of recurrent neural network well-suited for processing sequential data and identifying temporal dependencies. The researchers used InSAR (Interferometric Synthetic Aperture Radar) data to generate ground subsidence maps, which represent the subtle, ongoing ground movements that can indicate underlying instability. These temporal subsidence patterns were then fed into the LSTM model. The LSTM was trained to correlate these deformation trends with historical road collapse incidents. For comparison and validation, other machine learning models like GRU, SVM, KNN, and RF were also employed.

Datasets

The study utilized two primary datasets:

  • Historical Road Collapse Accident Data: Compiled from Guangzhou's historical road collapse accident reports and national geological disaster bulletins. This data includes the time, geographical coordinates, and causes of collapses in Guangzhou from 2014 to 2021, totaling 210 accident points.
  • InSAR Ground Subsidence Data: Derived from Sentinel-1 satellite data spanning from 2014 to 2020. This data was processed using the Small Baseline Subset InSAR (SBAS-InSAR) technique to create subsidence maps with a 50m spatial resolution.

Key Insights and Ideas

  • Temporal Dynamics are Key: The study strongly emphasizes that the rate and pattern of ground subsidence over time are more indicative of collapse risk than static measurements. LSTM's ability to model these temporal sequences is its major advantage.
  • InSAR as a Powerful Data Source: InSAR provides a cost-effective and spatially comprehensive way to monitor ground deformation, offering a continuous stream of data that traditional surveys cannot match.
  • Simplifying Risk Assessment: By focusing on InSAR-derived temporal data, the model bypasses the need for a multitude of complex, often hard-to-obtain assessment indicators, making the risk assessment process more streamlined.
  • Spatial Heterogeneity of Risk: The generated risk map highlights that road collapse risk is not uniform across the urban landscape, with distinct high-risk zones and generally low-risk areas, suggesting the need for localized risk management strategies.

Key Results

The LSTM model demonstrated superior performance compared to other evaluated models (GRU, SVM, KNN, RF) in predicting road collapse risk.

  • Accuracy: The LSTM model achieved an accuracy of 0.764.
  • AUC: It also achieved a high Area Under the Curve (AUC) of 0.766, indicating its strong ability to differentiate between collapse and non-collapse scenarios.
  • Risk Map: A risk map generated for Guangzhou in late 2022 visually identified areas with high, medium, and low road collapse risk, revealing significant spatial variations and clustering of high-risk zones, particularly in western and southern regions.

Limitations and Future Work

  • Data Availability: While the study used historical data, the authors mention that raw data can be made available upon request, implying that broader access could facilitate further research.
  • Model Generalization: The study focused on Guangzhou. Future work could involve testing the model's generalizability to other urban environments with different geological and urban development characteristics.
  • Integration of Other Data: Although the paper highlights the advantage of reducing indicators, future research could explore integrating other relevant dynamic data (e.g., rainfall, traffic load) with the InSAR-LSTM framework to further enhance predictive capabilities.
  • Real-time Monitoring: The current study is based on historical data. Implementing this model for real-time risk assessment and early warning systems would be a crucial next step.

Summary updated

Abstract

Urban road collapse is a common disaster in modern cities, posing a severe threat to the safety of life and property of urban residents. Effective risk assessment is crucial for preventing collapses. This study proposes a novel method for assessing the risk of urban road collapse by integrating Interferometric Synthetic Aperture Radar (InSAR) technology with Long Short-Term Memory (LSTM) networks. We conduct experimental analysis using collapse incidents and corresponding subsidence maps from five areas in Guangzhou. The results demonstrate that the proposed model successfully establishes a mapping relationship between the temporal features of InSAR data and the risk of road collapse, enabling a risk assessment of specific areas solely based on InSAR data. This model overcomes the difficulty of obtaining numerous assessment indicators in traditional risk assessment methods, providing strong support for the prevention and control of regional road collapse risks.
BibTeX
@misc{jiang2024urban,
  author = {Juncai Jiang and Wenfeng Bai and Yizhao Wang and Fei Wang and Qinglun He and Long Chen and Yuming Qiao and Zhi Wang and Haitao Luo},
  title = {{Urban Road Collapse Risk Assessment Study Based on InSAR Spatiotemporal Data}},
  year = {2024},
  doi = {10.3390/proceedings2024110024},
}
Figure