In the automatic GNSS-A data acquisition system, the WG transits to a seafloor station, initiates the GNSS-A observation, and remains at the station for several hours (up to ∼20 h) to record the GNSS-A observation data. These actions should be performed automatically and as autonomously as possible. It is also desirable that the WG remains offshore and continues recording observations for as long as possible. Thus, the required functions of the system are as follows:

Autonomous transit according to a pre-configured course or target waypoints while avoiding other ships

The SV3 WG (manufactured by Liquid Robotics, Inc., CA, United States) has the capability for functions 1 and 4. We then constructed a system that would accomplish function 2 and considered power needs (function 3). Note that the observation includes waking-up the seafloor transponders, transmitting the acoustic ranging signals, receiving/recording the replies from the transponders, and storing GNSS and gyroscope data.

Figure 1A shows the block diagram of the GNSS-A data acquisition system developed for this study. It consists of two GNSS antennas, a GNSS receiver, an acoustic transducer, an acoustic unit, an MEMS gyroscope, a controlling and recording single board computer (SBC), a wired/wireless router, band filters to filter out Iridium satellite signals, and power supply units (a DC/DC converter, a supercapacitor, and a charge–discharge relay board to the supercapacitor). The supercapacitor is necessary to supply the acoustic unit with instantaneous high current to transmit a wake-up signal to the seafloor transponder. The GNSS antennas and the acoustic transducer are mounted on the float, while the other instruments are installed inside a waterproof container (Figures 1B,C). The router has a wireless port that allows for access after the waterproof container is sealed. We can modify the controlling script installed in the SBC, configure the GNSS receiver setting, and check the recording status of acoustic ranging and gyroscope data via the wireless port on the deck before deployment.

The power supply to the payload can be remotely turned on and off (i.e., the S1 and S3 ports on the WG (Figure 1A) from land via communication from an Iridium satellite. The power supply for the acoustic unit is started when the S3 port is turned on. Once the S1 port is turned on, the GNSS receiver and the gyroscope are activated, and the SBC is booted up. The GNSS receiver itself records the raw data. The startup script in the SBC checks the current geographic location through the GNSS output. If the location is found sufficiently close to a certain GNSS-A observation site, it starts recording GNSS National Marine Electronics Association (NMEA) 0183 sentence output and raw gyroscope data, as well as starts sending a sequence of acoustic commands of wake-up and delay setting to the corresponding seafloor transponders at the site. After confirming that all the transponders have woken-up based on their reply messages, the acoustic ranging session is initiated. The acoustic ranging signal is transmitted every minute, and the replies from the seafloor transponders are recorded by the acoustic unit. When the WG leaves the site at a threshold distance, acoustic ranging and data recording of GNSS NMEA and gyroscope are automatically stopped, and the SBC halts and is ready to turn off the S1 port. In this experiment, data processing to estimate the position of the site, which is defined as the center of the array of seafloor transponders, is performed after the cruise ends. However, it can be estimated onboard in semi-real time and the position of the site can be transmitted to shore via satellite communication. Effecting such semi-real time observations is the next step in the development of this system.

We conducted GNSS-A observations at a seafloor station off the coast of Aomori Prefecture in July 2019 during a survey cruise (KS-19-12) using the research vessel Shinsei-maru (Joint Usage/Research Center for Atmosphere and Ocean Science and Japan Agency for Marine-Earth Science and Technology, 2019). Figure 2 shows the tracks of the Shinsei-maru and the WG throughout the cruise. Table 1 shows the extracted time schedule with respect to the GNSS-A observations made by the WG during the survey cruise. Shinsei-maru deployed the WG at site G02 on July 4. After conducting the GNSS-A observation by means of its own system parallel with the WG for about 3 h, Shinsei-maru moved toward northeast to install observation instruments at other sites. On July 7, it returned to the retrieval point ER4 to recover the WG. Note that the activation of the seafloor transponders at site G02 was performed using the GNSS-A observation equipment mounted on the Shinsei-maru before the deployment of the WG, and the GNSS-A observation by Shinsei-maru was suspended and resumed at 23:36 on July 3 and 00:05 on July 4 in order to confirm that the WG was transducing the acoustic ranging signal as shown in Table 1.

The items tested during the trial observations were as follows:

Data acquisition from the sensors, namely, the GNSS carrier phase, attitude of the float (heading, roll, and pitch), and acoustic waveforms from the transponders

To examine these items, in the initial plan, the WG was deployed at a site (G02), performed GNSS-A observations with tracing a figure-eight course that was centered at the center of the array of seafloor transponders (Figure 2), and then was ordered to transit to another site (G03), which is located just south of site G02 (item 3), and GNSS-A observations were initiated at site G03 (item 2). However, as the weather forecast stated that the state of the sea will become worse from the south, we altered the recovering point of the WG (G03) to site ER4, where a long-term ocean bottom seismograph was installed during the same survey cruise. Therefore, item 2 was examined at site G02 by suspending GNSS-A observations for a period of time. One hour after the final acoustic shot is received, the seafloor transponders go into a “sleep mode.” Thus, we suspended and resumed the GNSS-A observations at 08:17 and 23:00 by turning off and on the S1 port, respectively, to test whether the autonomous activation of the seafloor transponders was successful. We could ascertain whether the data acquisition was successful (item 1) only after we recovered the WG on deck.

After the recovery of the WG, we analyzed the collected GNSS-A data by applying the method devised by Kido et al. (2006). We estimated the translational horizontal displacements of the seafloor transponder array relative to the pre-defined position and net sound velocity change for each shot by combining the absolute position of the surface transducer and its distances to the seafloor transponders under the condition of a stratified sound velocity structure. The pre-defined array geometry and its position were estimated by Honsho et al. (2019) in the past using vessel-based moving observations. The positions of the surface transducer at timings of transmission and reception in a global coordinate system are derived based on the position of GNSS antenna and the WG’s attitude. The GNSS NMEA and MEMS gyroscope data provide information about the attitude of the WG, which is defined by the heading, pitch, and roll angles. The heading angle was estimated based on the dual antenna GNSS, while the pitch and roll angles were observed directly by the MEMS gyroscope. The relative position between the GNSS antenna and surface transducer in the body-fixed frame is also regarded as an unknown that can be solved by minimizing travel time residuals throughout the campaign. Once the relative position is solved, we can finally estimate the position of the array and temporal change in the average sound velocity for each acoustic shot to evaluate the temporal stability of the estimated array positions.

The data acquisition from the sensors performed well at site G02. Figure 3A shows the time series of the pseudo ranges between the acoustic transducer mounted on the WG and the seafloor acoustic transponders. Note that the pseudo ranges include apparent distances due to artificial response delays of seafloor transponders. Owing to a bug in the controlling script of the acoustic unit, the wake-up sequence, in which the acoustic commands to wake-up the seafloor transponder and configure the artificial delay were sent for each transponder, and an acoustic ranging session that was composed of 20 acoustic signal transmissions at the rate of 1 shot/min were repeated. The wake-up sequence and acoustic ranging session took ∼35 min and 20 min, respectively. Therefore, the acoustic ranging data is discontinuous. Although the total amount of the acoustic ranging data was reduced by this script error, the acquired data are sufficient to estimate the position of the seafloor transponder array. The wake-up sequence succeeded in activating the seafloor stations when the power supply to the payload was turned on from the shore via satellite communication after the observations were suspended. This proved that the WG can autonomously perform observations from a site on another site.

The solar panels generated sufficient power to perform the GNSS-A observations although the weather was mostly cloudy during the deployment. Figure 3B shows the time series of the battery power. During the periods when GNSS-A observations were performed, the total operation consumed approximately 24 Wh per hour at night. However, the total consumption rate was almost balanced or was exceeded by the power generation rate during the day.

The GNSS data were processed by IT (interferometric translocation) software (Colombo et al., 2000). The quality of the GNSS data has no significant problem considering the temporal stability of the vertical positions which was evaluated using the technique provided by Fujita and Yabuki (2003).

An example of the recorded acoustic raw waveform and the time series of cross-correlation between the transmitted and received signals is shown in Figure 3C. The raw waveform shows that the background noise is high, and the signal-to-noize ratio is low, while the cross-correlation time series exhibits clear correlation peaks that correspond to the arrival of reply signals from the four seafloor transponders. The correlation waveforms are similar to those obtained using a research vessel (e.g., Azuma et al., 2016), and the differences in the correlation waveforms of individual seafloor transponders are not significant. Thus, the quality of the acoustic data is equivalent to that obtained using a research vessel. We obtained the travel time of the acoustic wave between the acoustic transducer mounted on the WG and the seafloor transponders by applying the detection method devised by Azuma et al. (2016) to the cross-correlation time series for the estimation of the position of the seafloor transponder array. Note that an ingenious method to obtain the travel time is necessary because the correlation peak sometimes does not correspond with the direct arrival of the reply signal from the seafloor transponder in the correlation waveform (Azuma et al., 2016).

Figure 3D shows the position of the center of the array estimated for each acoustic shot. The scatter in the WG-derived positions (red points) is not larger than the positions estimated from observations made using a research vessel (blue points). The precision of the estimated position of the center of the array for this campaign is less than 10 cm; that is, the standard deviations of the scatter for each acoustic shot are 7.0 cm and 6.1 cm for both N–S and E–W components, respectively.

To estimate the seafloor displacement rate with high precision, both precision and accuracy (i.e., repeatability between observation campaigns) are required. Figure 3E shows the time series of the average array position of site G02. The array position estimated from the WG data is consistent with past observations. This indicates that negligible systematic bias is included in the estimated position. Therefore, we can directly compare the results from the WG data with past observations that were performed using research vessels and expect to estimate displacement rates as large as the plate convergence rate (i.e., ∼8 cm/year) along the Japan Trench with the observations at least once a year for a couple of years. For instance, we can estimate the displacement rate with the estimation error of less than 5 cm/year based on three observations with the observation error at 7.0 cm during 3 years at even intervals.