Abstract:Earthquakes are commonly estimated using physical seismic stations, however, due to the installation requirements and costs of these stations, global coverage quickly becomes impractical. An efficient and lower-cost alternative is to develop machine learning models to globally monitor earth observation data to pinpoint regions impacted by these natural disasters. However, due to the small amount of historically recorded earthquakes, this becomes a low-data regime problem requiring algorithmic improvements to achieve peak performance when learning to regress earthquake magnitude. In this paper, we propose to pose the estimation of earthquake magnitudes as a metric-learning problem, training models to not only estimate earthquake magnitude from Sentinel-1 satellite imagery but to additionally rank pairwise samples. Our experiments show at max a 30%+ improvement in MAE over prior regression-only based methods, particularly transformer-based architectures.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to use Sentinel - 1 satellite image data to estimate the magnitude of an earthquake. Traditionally, the detection of earthquakes and the estimation of their magnitudes rely on physical seismic stations, but the installation and maintenance of these stations are costly and their global coverage is limited. Therefore, researchers have proposed a new method to process Earth - observation data, especially Sentinel - 1 satellite images, through machine - learning models, in order to achieve earthquake monitoring on a global scale. Specifically, the paper proposes to re - define the problem of earthquake magnitude estimation as a metric - learning problem, training the model not only to predict magnitudes but also to rank samples. This method aims to overcome the problem of limited historical earthquake data and improve the generalization ability of the model under low - data conditions. ### Main contributions: 1. **Re - define the earthquake magnitude estimation task as a metric - learning problem**: Combine regression and ranking objectives, enabling the model to learn better from limited data. 2. **Apply this method to various advanced convolutional neural networks (CNN) and Transformer - based models**: Experimental results show that in all architectures, using Margin Ranking Loss (MRL) significantly improves the mean absolute error (MAE) by more than 30%. ### Method overview: - **Dataset**: Use the QuakeSet dataset, which contains 3,327 pairs of dual - temporal Sentinel - 1 images (VV and VH polarizations), with each image being 512×512 in size. - **Loss function**: Combine the mean - squared - error (MSE) loss and the margin ranking loss (Margin Ranking Loss), with the expression as follows: \[ L = L_{\text{MSE}}+L_{\text{MR}} \] where the margin ranking loss \( L_{\text{MR}} \) is expressed as: \[ L_{\text{MR}}(x_1, x_2, y)=\max(0, - y\cdot(x_1 - x_2)+m) \] where \( x_1 \) and \( x_2 \) are the predicted magnitudes of two sets of images respectively, \( y \) is the label vector (taking the value of 1 or - 1 according to the relative size between samples), and \( m \) is the margin parameter. ### Experimental results: - Experiments were carried out using different models (ConvNeXt v2, MobileNet, MobileViT v2, ViT - tiny), and the results show that the introduction of margin ranking loss significantly improves performance, especially in the ConvNeXt v2 and ViT - tiny models. Through this method, the researchers have successfully improved the accuracy of earthquake magnitude estimation and provided a low - cost and efficient solution for global earthquake monitoring.

Estimating Earthquake Magnitude in Sentinel-1 Imagery via Ranking

Rapid Estimation of Single-Station Earthquake Magnitudes with Machine Learning on a Global Scale

Rapid Earthquake Magnitude Estimation Using Deep Learning.

QuakeSet: A Dataset and Low-Resource Models to Monitor Earthquakes through Sentinel-1

Earthquake magnitude and location estimation from real time seismic waveforms with a transformer network

Application of machine learning to magnitude estimation in earthquake emergency prediction system

Multiclass Post-Earthquake Building Assessment Integrating Optical and SAR Satellite Imagery, Ground Motion, and Soil Data with Transformers

SeisT: A Foundational Deep-Learning Model for Earthquake Monitoring Tasks

Exploring a CNN Model for Earthquake Magnitude Estimation using HR-GNSS data

A Deep Attention Model for Onsite Estimation of Earthquake Epicenter Distance and Magnitude

A Deep Learning-Based Electromagnetic Signal for Earthquake Magnitude Prediction

MEANet: Magnitude estimation via physics-based features time series, an attention mechanism, and neural networks

Data-Driven Earthquake Multi-impact Modeling: A Comparison of Models

SeisT: A foundational deep learning model for earthquake monitoring tasks

Employing Machine Learning for Seismic Intensity Estimation Using a Single Station for Earthquake Early Warning

Earthquake magnitude estimation using a two-step convolutional neural network

Real‐Time Earthquake Detection and Magnitude Estimation Using Vision Transformer

Towards advancing the earthquake forecasting by machine learning of satellite data

An integrated approach for prediction of magnitude using deep learning techniques

Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking