Reliable Real-time Seismic Signal/Noise Discrimination with Machine Learning

Abstract: In Earthquake Early Warning (EEW), every sufficiently impulsive signal is potentially the first evidence for an unfolding large earthquake. More often than not, however, impulsive signals are mere nuisance signals. One of the most fundamental - and difficult - tasks in EEW is to rapidly and reliably discriminate real local earthquake signals from all other signals. This discrimination is necessarily based on very little information, typically a few seconds worth of seismic waveforms from a small number of stations. As a result, current EEW systems struggle to avoid discrimination errors, and suffer from false and missed alerts. In this study we show how modern machine learning classifiers can strongly improve real-time signal/noise discrimination. We develop and compare a series of non-linear classifiers with variable architecture depths, including fully connected, convolutional (CNN) and recurrent neural networks, and a model that combines a generative adversarial network with a random forest (GAN+RF). We train all classifiers on the same data set, which includes 374k local earthquake records (M3.0-9.1) and 946k impulsive noise signals. We find that all classifiers outperform existing simple linear classifiers, and that complex models trained directly on the raw signals yield the greatest degree of improvement. Using 3s long waveform snippets, the CNN and the GAN+RF classifiers both reach 99.5% precision and 99.3% recall on an independent validation data set. Most misclassifications stem from impulsive teleseismic records, and from incorrectly labeled records in the data set. Our results suggest that machine learning classifiers can strongly improve the reliability and speed of EEW alerts.

• The paper demonstrates that advanced ML models, notably CNN and GAN+RF, achieve 99.5% precision and 99.3% recall in seismic signal classification.
• It leverages a robust dataset of 1.32 million records and extracts key seismological features using a 2nd order Butterworth high-pass filter for model training.
• The study highlights the potential for faster, more reliable earthquake alerts in EEW systems while addressing challenges from teleseismic events and noise.

Reliable Real-time Seismic Signal/Noise Discrimination with Machine Learning

Introduction

The study by Meier et al. titled "Reliable Real-time Seismic Signal/Noise Discrimination with Machine Learning" addresses a critical challenge in earthquake early warning (EEW) systems: the reliable discrimination between legitimate earthquake signals and various forms of noise. With the objective of enhancing the performance of EEW systems, this paper investigates the application of ML classifiers to improve the real-time discrimination of seismic signals from noise. The authors explore several ML models, ranging from fully connected networks to convolutional and recurrent neural networks, as well as a hybrid generative adversarial network with a random forest classifier (GAN+RF).

Data Set and Model Design

The authors utilize a robust dataset comprising 374,000 local earthquake records and 946,000 noise records, allowing for extensive training and evaluation of different ML models. They employ three-component waveform data and extract seismologically relevant features, processed through a 2nd order causal Butterworth high-pass filter. The classification models evaluated include a Fully Connected Neural Network (FCNN), Recurrent Neural Network (RNN), RNN with Attention (RNNa), Convolutional Neural Network (CNN), and the GAN+RF. These models were trained to predict the likelihood of a signal being from a local earthquake.

Results and Analysis

The study's findings demonstrate that complex deep learning models can significantly outperform traditional linear classifiers, such as the one employed by the OnSite algorithm, in seismic signal classification tasks. The CNN and GAN+RF classifiers, trained directly on raw waveform data, achieved the highest classification accuracy with 99.5% precision and 99.3% recall on an independent validation set. This performance underscores the capability of deep learning models to discern signal characteristics in scenarios where standard models fall short.

A detailed analysis indicates that most misclassifications arise from impulsive teleseismic records and incorrectly labeled data, yet the CNN model in particular shows robustness in its predictions, attributing near-certain probabilities to both quake and noise records. The paper also highlights the potential for EEW systems to leverage these ML models to issue reliable alerts based on fewer station triggers, reducing alerting delays in critical regions like southern California.

Potential and Limitations

Deep learning classifiers exhibit profound implications for EEW systems by providing more rapid and reliable alerts. The ability to operate effectively even during high noise scenarios, such as aftershock sequences, represents a significant advancement over traditional detection methods that rely on amplitude-based metrics. However, challenges remain in managing teleseismic events and adapting models to stations with unique noise characteristics. The authors suggest that larger data sets and region-specific adaptations may mitigate these issues.

Conclusion

Meier et al.'s study convincingly demonstrates the utility of ML-based approaches in seismic signal classification, presenting a promising avenue for enhancing EEW systems. By harnessing advanced ML algorithms, specifically deep learning models, the reliability and speed of earthquake alerts can be significantly improved. These developments offer the potential to mitigate risks in densely populated areas, providing timely warnings and minimizing false alarms. Future investigations may focus on expanding training datasets, refining model architectures, and assessing the applicability of these methods across diverse geological regions.