Rapid Learning of Earthquake Felt Area and Intensity Distribution with Real-time Search Engine Queries

Hengshu Zhu, Ying Sun, Wenjia Zhao, Fuzhen Zhuang, Baoshan Wang, Hui XiongScientific 2020 · DOI

Abstract

Immediately after a destructive earthquake, the real-time seismological community has a major focus on rapidly estimating the felt area and the extent of ground shaking. This estimate provides critical guidance for government emergency response teams to conduct orderly rescue and recovery operations in the damaged areas. While considerable efforts have been made in this direction, it still remains a realistic challenge for gathering macro-seismic data in a timely, accurate and cost-effective manner. To this end, we introduce a new direction to improve the information acquisition through monitoring the real-time information-seeking behaviors in the search engine queries, which are submitted by tens of millions of users after earthquakes. Specifically, we provide a very efficient, robust and machine-learning-assisted method for mapping the user-reported ground shaking distribution through the large-scale analysis of real-time search queries from a dominant search engine in China. In our approach, each query is regarded as a "crowd sensor" with a certain weight of confidence to proactively report the shaking location and extent. By fitting the epicenters of earthquakes occurred in mainland China from 2014 to 2018 with well-designed machine learning models, we can efficiently learn the realistic weight of confidence for each search query and sketch the felt areas and intensity distributions for most of the earthquakes. Indeed, this approach paves the way for using real-time search engine queries to efficiently map earthquake felt area in the regions with a relatively large population of search engine users.

Introduction

When a destructive earthquake strikes, the first few minutes are critical for emergency response. Responders need immediate data on where shaking was most intense to prioritize life-saving resources. Traditionally, this information relies on dense seismic observation networks or post-event field investigations. However, high-quality seismic networks are prohibitively expensive to maintain across vast regions like mainland China, leading to sparse coverage. Field investigations, while highly accurate, often take days or weeks to produce actionable maps.

While social media platforms like Twitter have been explored for rapid assessment, they are frequently saturated with "noise"—posts from users who did not feel the quake but are merely discussing it. This research introduces the Q-Felt Map system, which leverages real-time search engine queries as "crowd sensors" to generate accurate impact maps in near real-time, bypassing the limitations of traditional hardware and social media noise.

The Q-Felt Map Framework

The Q-Felt Map system processes search queries (specifically the keyword "Earthquake" on Baidu) through a three-stage pipeline designed for speed and robustness.

Overview of the Q-Felt Map generation process, illustrating the accumulation of search queries over time and the subsequent three-stage processing pipeline to generate isoseismal maps.
  1. Query Screening (Deep Learning): The system distinguishes between "signal" and "noise." A query from a high-rise in a distant city is less indicative of local intensity than one from a rural area near the epicenter. A 3-layer Deep Neural Network (DNN) assigns a "confidence weight" to each query based on submission time, location, and local population density.
  2. Direction Detection (Weighted PCA): Shaking rarely spreads in a perfect circle; it is influenced by fault lines and local geology. By applying Weighted Principal Component Analysis (PCA) to the screened queries, the system identifies the primary axis of propagation.
  3. Isoseismal Line Segmentation: The final stage uses density-based algorithms to draw concentric ellipses representing perceived intensity levels. These boundaries are determined by identifying sharp gradients in the cumulative weight of queries as distance from the epicenter increases.

Machine Learning-Assisted Calibration

To ensure the reliability of these "crowd sensors," the researchers developed a training framework that uses historical earthquake data to calibrate the DNN.

The machine learning workflow for query screening: individual queries are processed through a neural network to determine weights, which are then optimized using a distance-based loss function against known epicenters.

By analyzing 554 earthquakes in mainland China from 2014 to 2018, the model learns to associate specific query features with the true epicenter. The loss function minimizes the distance between the estimated epicenter (derived from weighted query locations) and the actual seismic record. This allows the system to proactively "weight" users' reports based on their historical accuracy and context.

Key Results and Performance

The system's performance highlights its potential for real-world emergency management:

  • Speed: The entire mapping process takes approximately 1.22 seconds on standard consumer hardware.
  • The Five-Minute Window: Analysis shows that search volume peaks rapidly. The model's error rate stabilizes after just five minutes of data collection, providing a "sweet spot" for rapid response where waiting for more data offers diminishing returns.
  • Accuracy: In tests on 58 earthquakes of magnitude $M_s \ge 5.0$, the system successfully generated maps for 51 events. These maps showed high regional overlap with official intensity maps produced by seismologists days later.
  • Efficiency: The system is robust even with sparse data. For the 2017 Tashkurgan earthquake, it accurately sketched the maximum shaking area using only 161 high-confidence queries.

Discussion and Limitations

The Q-Felt Map provides a unique "human-centric" viewpoint of earthquake impact, reflecting perceived shaking rather than just structural damage. However, it faces two primary limitations. First, it is highly dependent on population density; sparsely populated regions like Tibet may lack sufficient "sensors" for a reliable map. Second, it cannot detect deep earthquakes (e.g., the 2016 Linkou event at 580 km depth) that do not produce surface shaking felt by humans.

Future work will focus on correlating these felt intensity maps with physical ground motion parameters, such as peak ground acceleration, to fully integrate search-based sensing into standard seismological workflows.

Summary updated

Abstract

Immediately after a destructive earthquake, the real-time seismological community has a major focus on rapidly estimating the felt area and the extent of ground shaking. This estimate provides critical guidance for government emergency response teams to conduct orderly rescue and recovery operations in the damaged areas. While considerable efforts have been made in this direction, it still remains a realistic challenge for gathering macro-seismic data in a timely, accurate and cost-effective manner. To this end, we introduce a new direction to improve the information acquisition through monitoring the real-time information-seeking behaviors in the search engine queries, which are submitted by tens of millions of users after earthquakes. Specifically, we provide a very efficient, robust and machine-learning-assisted method for mapping the user-reported ground shaking distribution through the large-scale analysis of real-time search queries from a dominant search engine in China. In our approach, each query is regarded as a "crowd sensor" with a certain weight of confidence to proactively report the shaking location and extent. By fitting the epicenters of earthquakes occurred in mainland China from 2014 to 2018 with well-designed machine learning models, we can efficiently learn the realistic weight of confidence for each search query and sketch the felt areas and intensity distributions for most of the earthquakes. Indeed, this approach paves the way for using real-time search engine queries to efficiently map earthquake felt area in the regions with a relatively large population of search engine users.
BibTeX
@inproceedings{zhu2020rapid,
  author = {Hengshu Zhu and Ying Sun and Wenjia Zhao and Fuzhen Zhuang and Baoshan Wang and Hui Xiong},
  title = {{Rapid Learning of Earthquake Felt Area and Intensity Distribution with Real-time Search Engine Queries}},
  booktitle = {Scientific},
  year = {2020},
  doi = {10.1038/s41598-020-62114-8},
}
Figure