Detecting slow slip events in Alaska using seafloor pressure data from the AACSE array

Abstract:

Detecting and measuring shallow slow slip events (SSEs) offshore is important for understanding the mechanics of these events as well as assessing tsunami hazard. However, this remains a challenge due to the high cost of collecting seafloor deformation data and high noise level from the water column. Here we applied machine learning to detect SSEs using seafloor pressure data between the summers of 2018-2019 from the Alaska Amphibious Community Seismic Experiment (AACSE). 13 shallow-water (<300m) and 9 deep-water (>1000m) stations were considered in this study. For shallow-water stations, we used an ocean circulation model to reduce the noise of the ocean on seafloor pressure. The variance reduction is up to 70%. For deep-water stations, we used a two-step method: subtract a 2-day low-pass filtered ocean model and then subtract the pressure residual from a reference site. The total variance reduction is up to 80% considering these two steps. A machine learning based method is then implemented to detect SSEs in the ocean-noise-corrected data. In addition to the previous developed method (He et al., 2020), we added an up-ramp shaped synthetic SSE to train the model, which allows both uplift and subsidence to be detected. We found two possible events in the shallow stations and three possible events in the deep stations. Independent data is required to distinguish between SSEs and ocean events. GPS data near the shallow stations also had signals during the second event around days 20-50 in 2019. Seismic data from the AACSE ocean bottom seismometers will be used to determine if any tremors occur or if the small earthquake rate increases during the possible SSEs.