Slow slip events were discovered in many subduction zones during the last two decades thanks to recordings of the displacement of Earth’s surface by dense GNSS networks. Slow slip can last from a few days to several years and have a relatively short recurrence time (months to years), compared to the recurrence time of regular earthquakes (up to several hundreds of years), allowing scientists to observe and study many complete event cycles. In many places, tectonic tremor is also observed in relation to slow slip and can be used as a proxy to study slow slip events of moderate magnitude where surface deformation is hidden in GNSS noise. However, in subduction zones where no clear relationship between tremor and slow slip occurrence is observed, these methods cannot be applied, and we need other methods to be able to better detect and quantify slow slip. Wavelets methods such as the Discrete Wavelet Transform (DWT) and the Maximal Overlap Discrete Wavelet Transform (MODWT) are mathematical tools for analyzing time series simultaneously in the time and the frequency domain by observing how weighted averages of a time series vary from one averaging period to the next. In this study, we use wavelet methods to analyze GPS time series and seismic recordings of slow slip events in Cascadia. We use detrended GPS data, apply the MODWT transform and stack the wavelet details over several nearby GPS stations. As an independent check on the timing of slow slip events, we also compute the cumulative number of tremors in the vicinity of the GPS stations, detrend this signal, and apply the MODWT transform. In both time series, we can then see simultaneous waveforms whose timing corresponds to the timing of slow slip events. We assume that there is a slow slip event whenever there is a peak in the wavelet signal. We verify that there is a good correlation between slow slip events detected with only GPS data, and slow slip events detected with only seismic data. The wavelet-based detection method detects all events of magnitude higher than 6 as determined by independent event catalogs (e.g. Michel et al., 2019, Pure Appli. Geophys.).

Slow-slip events were discovered in many subduction zones during the last two decades thanks to recordings of the displacement of Earth’s surface by dense GPS networks. They can last from a few days to several years, and have a relatively short recurrence time (months to years), compared to the recurrence time of regular earthquakes (up to several hundreds of years), allowing scientists to observe and study many complete event cycles. Moreover, whereas regular earthquakes occur along the shallow part of the dipping plate boundary (in the seismogenic / locked zone), slow-slip events often occur on the plate boundary downdip of the locked zone. Slow-slip events could potentially trigger large earthquakes. This phenomenon provides a potential opportunity to further our understanding of subduction zone processes, and evaluate the time-varying seismic hazard. Wavelets methods such as the Discrete Wavelet Transform (DWT) and the Maximal Overlap Discrete Wavelet Transform (MODWT) are mathematical tools for analyzing time series simultaneously in the time and the frequency domain by observing how weighted averages of a time series vary from one averaging period to the next. In this study, we use wavelet methods to analyze GPS recordings of slow-slip events in New Zealand. An important application of the DWT and the MODWT is the estimation of a signal hidden by noise within an observed time series. We used synthetic time series with slow slip events of different durations, to which a Gaussian noise has been added, and denoised the signal using a wavelet-based method, and a low-pass filter. Although the signal was barely visible behind the noise, we could see a unique ramp-like signal in the data denoised with the wavelet-based method, whereas the low-pass filtered signal showed several ramp-like features which were not present in the original synthetic data. Eventually, we aim to be able to detect possible smaller (magnitude 5) slow-slip events that may be currently undetected with standard methods, detect longer (months to years) slow-slip events that are more difficult to detect than slow-slip events with a short duration (days to weeks), and determine the vertical displacement of the ground surface during a slow-slip event.