A novel statistical method for detecting short-term slow slip events in GNSS time series

Slow slip events (SSEs), a type of slow earthquakes, generally recorded by the Global Navigation Satellite System (GNSS), play an important role in releasing strain in subduction zones. Understanding the relationship between SSEs and damaging earthquakes in nearby velocity-weakening portions of the plate interface could provide a valuable tool for forecasting large earthquakes, thus aiding in hazard mitigation. Detecting accurately the occurrence times of SSEs is one prerequisite to illuminate their interactions with large earthquakes. However, robust detection methods remain limited. Most undetected SSEs in GNSS data are short-term SSEs, i.e. SSEs with short durations ranging from days to weeks, since the amplitude changes in the GNSS data trend from short-term SSEs are somewhat small, close to (or even lower than) the background noise. Therefore, more urgent efforts should be devoted to developing an automated detection method for short-term SSEs in GNSS data.

Both observed and simulated GNSS data containing SSEs exhibit a typical piecewise nonlinear trend. In periods without SSEs, the data generally follow a noisy linear process. When an SSE occurs, the trend shifts to a different trajectory and returns to its original state once the event concludes. In this context, the start and end of an SSE correspond to change points in statistics, which refer to the times when the underlying dynamics of the signal transition between regimes. Thus, detecting SSEs in GNSS data can be formulated as a change point detection problem for piecewise nonlinear signals. However, developing a nonparametric change point detection method specifically for SSEs is challenging because constructing a suitable contrast function requires knowledge of the exact piecewise structure, which is currently unknown. This limitation also prevents existing change point detection methods from being directly applied to detect SSEs.

In this study, we propose Singular Spectrum Analysis Isolate-Detect (SSAID), a novel change-point detection method for automatically estimating the start and end times of short-term SSEs in GNSS data. A key advantage of SSAID is that it does not require prior knowledge of the specific form of the underlying SSE signal. The core idea of SSAID is to obscure the differences between the nonlinear SSE signal and a piecewise-linear model, allowing existing change-point detection techniques for piecewise-linear signals to be directly applied for SSE detection. We evaluate SSAID through extensive simulations on both synthetic and observed SSE data, demonstrating its robustness across varying noise levels and its superior performance compared to two existing approaches: linear regression with AIC and the L-1 trend filtering method. Finally, we confirm the effectiveness of our detections in observed GNSS data via the co-occurrence of non-volcanic tremors, hypothesis tests, and fault estimation.