First, the data stored in the recorders were copied to hard disks on computers for data processing and archiving. Timing of data from each OBS was adjusted to GPS time using time differences between the GPS time and an internal clock in each OBS, which were measured just before deployment and after recovery. Finally, data from each OBS were combined into a single file in 1-hour segments.
All LT-OBSs for the first-term observation frequently recorded characteristic events from the beginning of the records. The early part of the events dominates waves with a high frequency of a few-Hz and large-amplitude waves with a period of a few seconds follows (Fig. 2). Characteristic events were recorded throughout the records of the first-term observation, the early part of the second-term observation, and the late part of the fourth-term observation. In contrast, few ordinary earthquakes were recorded by the LT-OBSs. Characteristic events were also reported by Okada et al. (2016). These characteristic events have good S/N ratio in a frequency range from 4 to 8 Hz. The event durations were < 1 min in a frequency band of 4–8 Hz.
The record section of the characteristic event shows that the arrival time of the event on the furthest OBS from Nishinoshima (NI11) is the latest and amplitude is smallest (Fig. 2). In addition, arrival times and amplitudes are similar for other OBSs with almost identical distances from Nishinoshima. We picked up arrival times of the events on February 28, 2015 for each OBS on computer display (Urabe and Tsukada 1991), and located the events (Hirata and Matsu’ura 1987). A homogeneous velocity structure with P-wave velocity of 4 km/s was adopted for the location. Although it is difficult to pick arrival times precisely, especially S-wave arrivals, the epicenters of the events were concentrated near the volcano crater on Nishinoshima (Fig. 2).
On February 27, 2015, infrasound and video observations were performed on the R/V Kairei during the deployment of the LT-OBSs. Infrasound was observed during the plume release from the crater. We selected visual data just above the crater from the video and determined the luminance based on their collection times (Fig. 3). When Nishinoshima volcano emitted plumes, a black region traveled toward the upper right. During the infrasound and video observations, one OBS (NI11) was deployed and began the observations. We compared the events recorded by NI11 with infrasound data (1–7 Hz) and visual data of the volcano (Fig. 3). For comparison, infrasound and seismic data times were shifted in consideration of the sonic and seismic speeds in seawater. The wave packets in the LT-OBS records with bandpass filter of 4–8 Hz seem to correlate with the plume emissions (Fig. 3).
We estimated the number of characteristic events that were considered to be related to plume release using the Short-Term Average (STA)/Long-Term Average (LTA) trigger method. The method was applied to records from more than three LT-OBSs. First, a bandpass filter of 4–8 Hz was applied to all records and we adopted the parameters of the STA/LTA method as follows: 2-s STA window, 40-s LTA window, 1.5 STA/LTA ratio, 3-s trigger duration, 4-s re-trigger time. Events were identified by more than three triggers at the same time, and a continuous trigger was interpreted as one event. We changed the STA/LTA ratio parameter for each observation period due to a change in configuration of the LT-OBS network; all other parameters were identical for all terms. For the first-term observation, STA/LTA ratio was set to 1.5. We detected 363,367 events from February 27, 2015 to October 3, 2015 (the first-term observation). During the second-term observation (from October 4, 2015 to May 5, 2016), 27,544 events were identified. For the second term, we changed the STA/LTA ratio to 2.0. For the third-term observation (from May 5, 2016 to October 20, 2016), 8192 events were detected with an STA/LTA ratio of 1.6. We identified 39,618 events for the fourth-term (from October 19, 2016 to May 26, 2017) using an STA/LTA ratio of 1.6. Figure 4 shows examples of detecting the events using the STA/LTA method. We also calculated the trigger length for each event and measured the maximum amplitude for 1 h on the vertical component. In addition, visual inspection using the waveform of the OBS records was useful for recognizing seismic activities. We calculated the averaged absolute values of vertical component in the seismograms for time windows of 1 s and plotted all processed data for one LT-OBS for each term. The resulting graphs were examined to confirm events detected using the STA/LTA method.