🤖 AI Summary
Detecting abrupt change points in short-term slow slip events (SSEs) within GPS time series from southwestern Japan remains challenging due to noise corruption, lack of prior models, and poor robustness of existing methods to piecewise nonlinear structures. To address this, we propose a novel change-point detection framework integrating Singular Spectrum Analysis (SSA) with Isolate-Detect. SSA is employed—first in this context—for joint noise suppression and trend decomposition, while Isolate-Detect enables model-free, adaptive, multiscale change-point localization, specifically tailored for nonlinear time series exhibiting continuous piecewise behavior. Experiments on both synthetic and real GPS data demonstrate that our method significantly improves accuracy in identifying SSE onset and termination times and enhances noise robustness: the average F1-score increases by 12.6% over conventional piecewise linear models and outperforms standalone decomposition or detection approaches.
📝 Abstract
Detecting change-points in noisy data sequences with an underlying continuous piecewise structure is a challenging problem, especially when prior knowledge of the exact nature of the structural changes is unknown. One important application is the automatic detection of slow slip events (SSEs), a type of slow earthquakes, in GPS measurements of ground deformation. We propose a new method based on Singular Spectrum Analysis to obscure the deviation from the piecewise-linear structure, allowing us to apply Isolate-Detect to detect change-points in SSE data with piecewise-non-linear structures. We demonstrate its effectiveness in both simulated and real SSE data.
