Denoising Daily Displacement GNSS-Time series using Deep Neural Networks In a Near Real-Time Framing: a Single-Station Method

Recent ground observations from Global Navigation Satellite Systems (GNSS) displacement time series have provided compelling evidence that the motion of tectonic plates is ubiquitously non-steady-state. In some cases, these anomalous transient motions have been identified as potential precursors occurring months, days, or hours before large-magnitude earthquakes. However, effectively detecting these signals in daily geodetic time-series at the earliest opportunity remains challenging due to the levels of high-frequency noise. Currently, there is a lack of established methodologies to reduce this noise in near-real-time thereby hindering our ability to timely monitor tectonic transient motions. Precursors are typically modeled retrospectively, and the use of geodetic data for seismic hazard surveillance remains limited. To address this limitation, this study demonstrates an approach to model high-frequency noise in daily GNSS displacement time-series, with the removal of this modeled noise allowing for tectonic transients to be potentially more clearly identified. Using Deep Neural Networks (DNNs), we develop a denoising approach that removes noise from GNSS displacement time-series on a station-by-station basis. To more effectively train our DNN models, we generate a comprehensive and diverse dataset by combining synthetic trajectories with synthetic noise time-series created using Generative Adversarial Networks (GAN). To train the GAN, we use noise time series extracted from ~5000 GNSS displacement time series distributed globally. Validating our approach with real data confirms its capability to significantly reduce the high-frequency noise that characterizes GNSS time-series. The flexibility of the method allows for near-real-time noise removal (with a latency of a few days), opening up the possibility of detecting and modeling small tectonic transients in a timely fashion. By introducing this novel approach, we present exciting opportunities to advance the geodetic surveillance of tectonic motions and usher in a new era of improved monitoring of seismic activity.