Abstract:Geospatial data has been transformative for the monitoring of the Earth, yet, as in the case of (geo)physical monitoring, the measurements can have variable spatial and temporal sampling and may be associated with a significant level of perturbations degrading the signal quality. Denoising geospatial data is, therefore, essential, yet often challenging because the observations may comprise noise coming from different origins, including both environmental signals and instrumental artifacts, which are spatially and temporally correlated, thus hard to disentangle. This study addresses the denoising of multivariate time series acquired by irregularly distributed networks of sensors, requiring specific methods to handle the spatiotemporal correlation of the noise and the signal of interest. Specifically, our method focuses on the denoising of geodetic position time series, used to monitor ground displacement worldwide with centimeter- to-millimeter precision. Among the signals affecting GNSS data, slow slip events (SSEs) are of interest to seismologists. These are transients of deformation that are weakly emerging compared to other signals. Here, we design SSEdenoiser, a multi-station spatiotemporal graph-based attentive denoiser that learns latent characteristics of GNSS noise to reveal SSE-related displacement with sub-millimeter precision. It is based on the key combination of graph recurrent networks and spatiotemporal Transformers. The proposed method is applied to the Cascadia subduction zone, where SSEs occur along with bursts of tectonic tremors, a seismic rumbling identified from independent seismic recordings. The extracted events match the spatiotemporal evolution of tremors. This good space-time correlation of the denoised GNSS signals with the tremors validates the proposed denoising procedure.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to effectively remove the noise in GNSS (Global Navigation Satellite System) position - time series in order to extract weak slow - slip events (SSEs). Specifically, the research aims to develop a method capable of processing multi - station, multi - variable time - series data. This method can capture spatially and temporally correlated noise and distinguish it from the signals of interest (such as slow - slip events). ### Specific background of the problem GNSS data are widely used to monitor crustal movement with an accuracy reaching the centimeter - to - millimeter level. However, these data are affected by noise from multiple sources, including environmental signals and instrument errors. These noises are spatially and temporally correlated and difficult to separate. Especially when extracting slow - slip events, since the signals of these events are very weak and are easily masked by other stronger signals, denoising becomes more difficult. ### Main contributions of the paper 1. **Multi - station spatio - temporal graph neural network (STGNN)**: The paper proposes a deep - learning model based on graph neural networks and spatio - temporal attention mechanisms - SSEdenoiser. This model can process spatial and temporal information simultaneously, thereby more effectively removing noise and extracting slow - slip events. 2. **Adaptive graph structure**: By learning the graph structure between GNSS stations, the model can discover hidden spatial relationships, rather than simply relying on predefined distances or geometric structures. 3. **Spatio - temporal attention mechanism**: Combining time - attention and space - attention mechanisms enables the model to focus on important features in both the time and space dimensions. 4. **Application verification**: This method is applied to the Cascadia subduction zone and successfully extracts slow - slip events that match the low - frequency tremors of earthquakes, verifying its effectiveness. ### Mathematical formula representation - **Displacement model**: \[ \xi_i(t)=n_i(t)+d_i(t) \] where \(\xi_i(t)\) is the position - time series of the \(i\) - th station, \(n_i(t)\) is the noise term, and \(d_i(t)\) is the synthesized slow - slip signal. - **Temporal evolution of the slow - slip signal**: \[ d_i(t)=\frac{D}{1 + e^{-\beta(t - t_0)}} \] where \(D\) is the static displacement, \(\beta\) is the growth rate of the curve, and \(t_0\) is the inflection point time of the logistic function. - **Loss function** (mean - squared error): \[ \text{MSE}(d(t),\hat{d}(t))=\frac{1}{n\cdot N_s\cdot N_t\cdot N_d}\sum_{i = 1}^{n}\sum_{j = 1}^{N_s}\sum_{t = 1}^{N_t}\sum_{k = 1}^{N_d}(d_{i,k}^j(t)-\hat{d}_{i,k}^j(t))^2 \] Through these methods and techniques, the paper provides a powerful tool for more accurately extracting weak but important geological signals from GNSS data.

Denoising of Geodetic Time Series Using Spatiotemporal Graph Neural Networks: Application to Slow Slip Event Extraction

Denoising of Geodetic Time Series Using Spatiotemporal Graph Neural Networks: Application to Slow Slip Event Extraction

Slow slip detection with deep learning in multi-station raw geodetic time series validated against tremors in Cascadia

Detecting slow slip events in the Cascadia subduction zone from GNSS time series using deep learning

Wavelet-like denoising of GNSS data through machine learning. Application to the time series of the Campi Flegrei volcanic area (Southern Italy)

Adaptive Denoising Approach for High-Rate GNSS Seismic Waveform Preservation: Application to the 2010 EI Mayor-Cucapah Earthquake and 2012 Brawley Seismic Swarm

Tremor Waveform Denoising and Automatic Location with Neural Network Interpretation

Magnetotelluric Noise Attenuation Using a Deep Residual Shrinkage Network

A Deep-Learning Neural Network for Postseismic Deformation Reconstruction From InSAR Time Series

Low-Frequency Magnetotelluric Data Denoising Using Improved Denoising Convolutional Neural Network and Gated Recurrent Unit

Time-Varying GPS Displacement Network Modeling by Sequential Monte Carlo

Autonomous extraction of millimeter-scale deformation in InSAR time series using deep learning

Subduction zone fault slip from seismic noise and GPS data

Deep Learning Denoising Applied to Regional Distance Seismic Data in Utah

Least-squares decomposition with time-space constraint for denoising microseismic data

Unsupervised Deep Learning for Single-Channel Earthquake Data Denoising and Its Applications in Event Detection and Fully Automatic Location

SEDENOSS: SEparating and DENOising Seismic Signals With Dual‐Path Recurrent Neural Network Architecture

Fast Window-Based Event Denoising with Spatiotemporal Correlation Enhancement

Denoising Deep Learning Network Based on Singular Spectrum Analysis—DAS Seismic Data Denoising With Multichannel SVDDCNN