APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature with Large Dynamic Range and Bandwidth

Transcription

1 Bulletin of the Seismological Society of America, Vol. 109, No. 1, pp , February 2019, doi: / APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature with Large Dynamic Range and Bandwidth by Earl E. Davis, Martin Heesemann, Joseph J. Farrugia, Greg Johnson, and Jerome Paros Abstract A simple tool has been developed to facilitate the study of interrelated geodetic, geodynamic, seismic, and oceanographic phenomena in marine settings. It incorporates quartz pressure and triaxial acceleration sensors and a low-power, high-precision frequency counter. The sensors are housed in a 6-cm outside-diameter, 1-m-long pressure case that is pushed vertically into the seabed with a submersible or remotely operated vehicle, with no profile remaining above the seafloor to cause current-induced noise. The mass of the tool is designed to match that of the sediment it displaces to optimize coupling. Intrinsic measurement precision of the order of 10 8 of full scale (in this instance, a pressure range equivalent to 4000 m of water depth and an acceleration range of 3g) allows observations of pressure, acceleration, and tilt variations of 0.4 Pa, 0:6 μms 2, and 0:06 μrad, respectively. Temperature variations measured near the top and at the bottom of the instrument are resolved to better than 0.1 mk. With the large dynamic ranges, high sensitivities and broad bandwidth (10- Hz Nyquist to drift-limited zero-frequency DC), ground motion associated with microseisms, strong and weak seismic ground motion, tidal loading, and slow and rapid geodynamic deformation all normally studied using disparate instruments can be observed with this single tool. Examples of data are provided from four deployments with connections to the Ocean Networks Canada Northeast Pacific telemetred undersea networked experiment (NEPTUNE) observatory cable. Introduction Instrument Description Many seafloor instruments have been developed over the past several decades to observe ground motion and pressure signals from seismic, oceanographic, and geodynamic sources, the most common being ocean-bottom seismometers and pressure recorders. These have been extremely useful for a broad range of studies in seismically and tectonically active regions. In this article, we describe a new tool (referred to here as APT, for acceleration, pressure, and temperature) for ground-motion monitoring. It expands seafloor observational capabilities beyond those of cable-connected instruments with similar sensors, such as one underway off Japan (e.g., Mochizuki et al., 2017), by way of its low-power consumption for autonomous operations. It incorporates a triaxial accelerometer designed by Quartz Seismic Sensors, Inc., and a pressure sensor built by Paroscientific, Inc. Both are housed in a slim (6.03-cm outside diameter), 0.53-cm wall titanium pressure case that is pushed 1 m below the seafloor (Fig. 1a). These sensors use quartz crystal oscillators loaded by a Bourdon tube for measuring pressure and coupled to masses for measuring acceleration (Paroscientific Technical Notes, a, b). Both also contain independent unloaded crystals that allow the temperature sensitivity of the pressure and acceleration crystals to be accounted for and that provide observations of temperature near the top and at the bottom of the tool for scientific applications. The quartz sensors have extremely broad bandwidth by nature and, with precise frequency counting technology, they provide measurements over a large dynamic range. Because of their small size, low-power requirements, and robustness, the sensors are well suited for autonomous (battery-powered) deployments. The acceleration sensor is mounted rigidly to the lower pressure-case end cap, and the pressure sensor is mounted below the upper cap with a pressure feed-through leading to an external port protected against biofouling by a Cu/Ni screen (Fig. 1b). Having no profile above the seafloor, the tool is unaffected by bottom currents, and the length of the tool buffers the acceleration sensor from the effects of bottom-water temperature variations. The mass of the tool matches closely that of the sediments displaced, providing good coupling (e.g., see Sutton et al., 1981). A prototype of the tool without a pressure sensor was deployed on the outer Cascadia accretionary prism in September 2015 (Fig. 2) and connected to the Ocean Networks Canada (ONC) Northeast Pacific telemetred undersea networked experiment (NEPTUNE) offshore fiber-optic cable system to allow realtime data transmission and long-term operations without 448

2 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 449 (a) (b) this design resulted in greater precision and logging capabilities (130 million samples), higher sampling frequencies (programmable from 1 sample per hour to 20 samples per second vs. a maximum of 1 sample per second in the original PPCs), less power consumption for autonomous battery powered deployments (160 mw when operating continuously, and less when operating intermittently, e.g., 4.5 mw average at 1 sample per minute), and greater flexibility for external interfacing, including options for serial (RS232 or 485) or Ethernet communications and for network time protocol (NTP) time synchronization for cable-connected deployments (see Data and Resources for details). Figure 1. Photo of the first acceleration, pressure, and temperature (APT) tool being tested in a salt water tank. (a) An ROV manipulator handle and a cable to a shore-based Ocean Networks Canada (ONC) junction box can be seen at the top of the tool. (b) Construction details of the current-generation APT (deployed in 2017 and 2018) are shown in the cut-away schematic. See Data and Resources for further details. The color version of this figure is available only in the electronic edition. battery power. This instrument included six lithium thionyl D cells to power the instrument for a period of several months in the event of an extended cable connection failure. In the second version of the instrument, a pressure sensor replaced the space occupied by the D cells; smaller batteries were mounted as part of the electronics chassis. During the first deployment, seafloor pressure observations were provided by a separate bottom pressure recorder (BPR) located 70 m away (Fig. 2). This instrument is one among many pressure recorders deployed for long-term autonomous and cabled seafloor and borehole pressure monitoring. The BPRs use Paroscientific quartz sensors and the original precise-period counter (PPC) system designed and built by John Bennest of Bennest Enterprises, Ltd., originally for use in borehole hydrologic experiments (e.g., Davis et al., 2010). These counters use a stable frequency reference oscillator to count variable frequency sensor cycles over a userspecified counting interval (typically 800 ms), and they determine fractional cycles at the beginning and end of each counting interval with a capacitor charging circuit keyed by the leading edges of the reference and sensor square-wave outputs. Averaging multiple fractional count determinations allows measurement of sensor frequency with a precision of 1 ppb. The full dynamic range of the Paroscientific sensors spans 10% of the sensor frequency; thus, with the PPCs, pressure variations (and now accelerations) are determined to 10 ppb of full scale. Signal period counters that operate on the same principle are now being built by RBR Ltd., and they were used for the APT instruments described here. Refinements achieved in Data Reduction Sensor signal frequencies are converted to engineering units using calibrations carried out by Paroscientific, Inc., for pressure sensors, and by Quartz Seismic Sensors, Inc., for acceleration sensors. Because of the imperfect orientation of the tools (Table 1), data have been rotationally corrected, first by applying a matrix transform with factory calibration values to orthogonalize the sensor s axial components (see Paroscientific Technical Note, b), then by minimizing horizontal-axis values through rotations about the x and y sensor axes in sequence, and finally by azimuthally rotating to optimize the fit of iteratively rotated horizontal APT waveforms of microseisms and earthquake surface waves to those observed by nearby buried broadband seismometers, where the orientation is known to better than 1 (see Table 1). Spurious values (spikes), originating with the period counters but of as-yet unknown cause, occur in the data records with a frequency of occurrence defined in the period of operation of the first APT of 10 6 (20 occurrences over a 9-month period of mostly 1 sample per second operation). In the current version of the APT, these are searched for in real time with a 20-point running window. Isolated values outside a predefined threshold (relative to previous and subsequent values) are replaced by linear interpolation and flagged. Relative time is determined in all instruments with an onboard oscillator, with accuracy limited by the offset from its nominal frequency ( 1:5 ppm from 0 to 40 C) and by changes with age ( 0:5 ppm=yr after 1 yr of operation). In the case of autonomous deployments, offsets must be determined by comparison with time checks made at times of deployment, recovery, and/or submersible visits. Improvements to internal clock accuracy for autonomous seismic applications are being explored. In initial cable-connected use, clock offsets were defined by comparison to shore-based time stamps. Although uncertainties arise from cable transmission latencies,

3 450 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros The NTP client and the transmission control protocol/ Internet protocol (TCP/IP) port (used only in cabled-mode operations) are by far the most power consumptive components; they bring the total power consumption up to 1.5 W. Both are located near the top of the instrument just beneath the pressure sensor. The associated temperature perturbation at the pressure sensor is large (1.2 K), but comparison of the records of this sensor and of the nearby BPR and a conductivity/temperature/depth (CTD) instrument shows that the offset is constant. Thus, temperature variations recorded by the APT pressure sensor can be considered reliable. Temperatures measured by the accelerometer at the bottom of the tool also may be offset (although far less than the offset seen at the top of the tool), and temporal variations can be accepted as reliable. Preliminary Testing Figure 2. Regional (inset) and detailed maps showing the locations of APT and broadband seismometer instruments connected to the ONC/Northeast Pacific telemetred undersea networked experiment (NEPTUNE) cable system (see Table 1 for details and site naming), the latter showing locations of the first and second APT installations at the Clayoquot slope site (labeled here and in Table 1 as APT1 and APT2), the buried Güralp broadband seismometer NC89, the Nanometrics accelerometer, and the bottom pressure recorder (BPR) used for comparisons in this article. Locations of gas-charged groundwater vents at Clayoquot are shown for context (Römer et al., 2016; M. Scherwath, personal comm., 2018). The location of the earthquake discussed later in the article is shown in the inset as a star. The color version of this figure is available only in the electronic edition. over long periods of time and with redundant checks, their effects could be made small. After construction of the first prototype tool (referred to as APT1), timing accuracy was improved by adding an NTP client. Earlier time checks allowed the drift of the onboard clock to be verified as steady (6: ) and to be corrected for with an uncertainty of < 50 ms over the early APT2 deployment period. Regular NTP time checks are now operating in all instruments at a refresh rate of once per minute. Frequency Counter Evaluation Before construction of the first APT instrument, comparative tests of different types of frequency counters and sensors were performed in the laboratory and in the seismic vault at the Pacific Geoscience Centre (PGC). First, three different frequency counter technologies were evaluated: (1) a Bennest PPC, (2) an RBR counter, and (3) a Paroscientific nano-resolution frequency counter. Initial tests of the Bennest PPC and the RBR counter, done by measuring the signal frequency of a stable oscillator, showed peak-to-peak noise levels of 3 ppb of the nominally 35-kHz output frequency and a root mean square deviation of less than 1 ppb when making measurements at 1 sample per second. Later, a comparison of all three counters was done with input from a pressure sensor insulated to reduce temperature variations in the laboratory. Results of these tests also demonstrated frequency counting resolution in the neighborhood of 1 ppb (better by a factor of three in amplitude in the case of the Paroscientific counter) when sampling at 1 sample per second. (Fig. 3). Thus, all were considered appropriate candidates for application to the APT instrument. Although the Paroscientific counter was capable of higher sampling rates than the former two and displayed lower noise (as would Table 1 Nomenclature, Locations, and Other Details of Instruments Used in or Relevant to This Article ONC Site/Subsite/Instrument Type/Station Code Latitude ( ) Longitude ( ) X Tilt Y Tilt Azimuth Depth Date Deployed Clayoquot slope/bullseye/prototype APT 1/NC89.Z true 1255 m 15 September 2015 Clayoquot slope/bullseye/apt 2/NC89.Z true 1258 m 14 June 2017 Clayoquot slope/bullseye/bpr N/A 1258 m 6 September 2009 Clayoquot slope/bullseye/güralp CMG-1T/NC true 1256 m 17 September 2009 Clayoquot slope/bullseye Nanometrics TitanEA/ m 8 June 2017 NC89.W1 Barkley Canyon/Node/APT/BACND.Z true 643 m 22 June 2018 Barkley Canyon/Upper slope Güralp CMG-1T/NCBC true 396 m 7 September 2009 Cascadia Basin/EastAPTCBC27.Z true 2656 m 24 June 2018 Cascadia Basin/West/Güralp CMG-1T/NC true 2656 m 22 May 2014 ONC, Ocean Networks Canada.

4 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 451 Figure 3. Raw data determined with three types of period counters connected to a single Paroscientific pressure sensor. Time is referenced to the beginning of this bench test on 29 August The several-minute-long variations that are coherent among the counters reflect variations in laboratory pressures. Incoherent high-frequency variations reflect intrinsic counter noise at levels indicated. Period values are given as a fraction of the sensor signal period of 28 μs. Full-scale pressure variations range over 10% of this value; hence, the noise levels reflect pressure variations ranging from 4 to 13 ppb of the full range of the sensor. Absolute period values vary among instruments and among channels of a single instrument by typically 1%. The color version of this figure is available only in the electronic edition. have been the case for a counting system developed about this time by Webb and Nooner, 2016), it required several times more power, too much for extended autonomous operations. Thus, continued developments adopted the RBR counter technology along with its integrated logging system. Sensor Evaluation Two different types of sensors both built by Quartz Seismic Sensors, Inc. were evaluated during tests in the PGC seismic vault: (1) a triaxial accelerometer similar to the one used in the APT tool described here and (2) a prototype biaxial tilt sensor. The triaxial accelerometer has a dynamic range of 3g, and if 1-ppb frequency counting is achieved, acceleration can be resolved to 0:6 μms 2 and tilt to 0:06 μrad. Although less robust, the tilt sensor is designed to apply its sensitivity over a range limited to 10, or 0:2g acceleration, thus providing greater precision for measurements of tilt (i.e., 4 nrad with 1-ppb counting resolution) and horizontal acceleration. Data from both sensors were compared to measurements from a Güralp CMG-1T broadband seismometer, which was being tested in parallel before its deployment on the ONC NEPTUNE cabled observatory. The output of this feedback-controlled inductively balanced mass seismometer includes three-axis velocity (with linear response over a bandwidth from 40 Hz to 0:03 Hz) and mass position data, which, at periods longer than the seismometer velocity bandwidth, are linearly related to acceleration. At very long periods (e.g., tidal), the horizontal axes readings are geometrically related to tilt (Davis et al., 2017). All test data were synchronized using Global Positioning System and TCP time servers, with the exception Figure 4. Tidal tilt variations (offset for plotting convenience) observed with the mass position data from a Güralp broadband seismometer, with a quartz tilt sensor, and with a quartz accelerometer situated on a stable pier at the Pacific Geoscience Centre (PGC), along with ocean tides observed in Patricia Bay with a BPR connected to the ONC/VENUS near-shore cable system. Sensors and instruments are described in the Introduction. Signals roughly four hours after the beginning of the record seen by the tilt and acceleration sensors are from an earthquake on the Charlie Gibbs fracture zone in the North Atlantic Ocean (illustrated in Fig. 5). The color version of this figure is available only in the electronic edition. of tests with the prototype RBR period counter that ran on its own independent clock. Signals observed during tests carried out in the PGC vault with the accelerometer and tilt sensor included tidal deformation, seismic waves, and ubiquitous microseisms. Tidal tilts of 1 μrad were observed by all sensors (Fig. 4), although the accelerometer signal appears to have been influenced by the temperature variations present in the vault (true by varying amounts for all sensors). Tidal signals of greatest amplitude were aligned in a north south direction, perpendicular to the local shoreline which lies 50 m north of the vault. A direct relationship to local ocean tide suggests that the origin lies with local ocean loading. Arrivals from an earthquake on the Gibbs fracture zone provided a fortuitous test of the sensitivity of the various sensors at seismic frequencies (Fig. 5). A good match among all sensors was seen up to the frequency limit imposed by the tilt sensor sampling interval (using an RBR counter running at only 1 sample per second at this time) and down to 0.01 Hz (spectra for all sensors are indistinguishable in Fig. 5c). Below 0.01 Hz, the seismometer and quartz sensor spectra diverge, possibly as a result of noise introduced into the quartz sensors by large thermal noise at this time. At higher frequencies, signals from microseisms and local site noise recorded by the quartz accelerometer and the broadband seismometer continue to agree well up to near the 20-Hz Nyquist frequency. A comparison of pre-earthquake signals recorded at night with no cultural sources (lower curves of Fig. 5c) revealed more fully the practical resolution of the quartz accelerometer and tilt sensor. Signals recorded by the tilt sensor agreed well with those recorded by the seismometer over the full frequency range from near-nyquist (5 Hz, defined at this time by a sampling rate of 10 samples per second) down

5 452 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros Figure 5. (a) Details of teleseismic surface waves and (b) relatively high-frequency site noise observed in the PGC vault with the quartz accelerometer (similar to one used later in the APT instrument) and the collocated Güralp broadband seismometer (horizontal components are shown; see Fig. 4 for context). (c) Spectra of 1-hr-long records including the seismic arrivals and of nighttime background noise (avoiding cultural site noise), with background and earthquake spectra for the quartz accelerometer labeled 1 and 2, for the broadband seismometer labeled 3 and 4, and for the quartz tilt sensor labeled 5 and 6, respectively. The characteristic lower and upper background spectra of Peterson (1993), labeled 7 and 8, are shown for context. In this and all other figures, acceleration values from the Güralp seismometer are computed from velocity channel data (unless the mass position channel or the strong-motion acceleration channel is specified). Spectra shown here and in figures that follow were computed with a SigmaPlot macro that applies a Hanning taper. The color version of this figure is available only in the electronic edition. to the limit of the seismometer (360-s period). Resolution with the quartz accelerometer was less, although its noise ( 130 db in power relative to 1 ms 2 ) was understandable in light of the limits imposed by its greater dynamic range (six times that of the tilt sensor). Although the tilt sensor was seen to be clearly superior for seismic recording, the limits to resolution of the accelerometer were thought to be acceptable for many applications, particularly in light of typically high noise levels on the seafloor (as seen in examples that follow). Advantages of the accelerometer include its robustness and large dynamic range, properties that make it serviceable as a seismic strong-motion instrument, insensitive to deployment orientation, and calibratable along all axes using the Earth s gravity field. Overall, the results provided strong justification for continuing the design, construction, and deployment of the first APT instrument using the RBR counter and logger and the Quartz seismic systems triaxial accelerometer. For instruments devoted primarily to seismic observations and when leveling can be assured, a tilt sensor could easily be used or even added for the horizontal axes. For cable-powered deployments when external power is available, further small gains in resolution could be realized through use of a Paroscientific or other higher precision counter. Initial Results of Seafloor Deployments After lab and vault testing and evaluation, the first APT instrument, built at the PGC (referred to in Fig. 2 and Table 1 as APT1), was deployed in September Installation was done with the remotely operated vehicle (ROV) Jason at the Clayoquot slope NEPTUNE site on the Cascadia subduction zone accretionary prism (labeled NC89 in Fig. 2). The BPR and seismometer at this site are located 70 and 140 m west southwest of the APT1 position, respectively. A 1-m-long push core was taken with the aid of an omnidirectional level to evaluate the sediment bulk density and to create a pilot hole for the APT installation. After initial installation in the hole created, it was believed that the coupling of the tool to the sediment was poor, so the APT was removed and pushed in directly (with no pilot hole) a few meters away. Azimuthal orientation relative to the ROV s heading was determined visually, and inclination was determined from the values of each of the three components relative to 1g (Table 1). The seafloor at the site is flat and underlain by accreted sediments, superimposed slope deposits, and a ubiquitous gas hydrate bottom-simulating reflector

6 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 453 Figure 6. Tidal-period acceleration and derived tilt (sin 1 a=g ) observed with the first APT deployed at the ONC Clayoquot site and with the nearby buried broadband NC89 seismometer as derived from the mass position data (see Fig. 2 and Table 1 for locations). Total tilts resolved from the horizontal axes data are in a direction parallel to the dip of the subduction prism structure; the cause of the difference in phase is unknown (see Davis et al., 2017 for details). The color version of this figure is available only in the electronic edition. (Riedel et al., 2002, 2006). Numerous seafloor gas vents are present nearby (Römer et al., 2016; see Fig. 2). The instrument was connected to a NEPTUNE scientific junction box at the site and began transmitting data to shore shortly thereafter (15 September 2015) at 1 sample per second. In March 2016, the sample rate was increased to 6 samples per second. Sampling at highest rate possible with this prototype tool, 12 samples per second, was not done to avoid conflict with military regulations. In June 2016, the first APT instrument was disconnected when its junction box was moved, and operation of the tool was terminated. Over the following year, the technology was transferred to RBR Ltd. who designed and produced tools that had enhanced capabilities, including a 20-Hz sampling rate, and that would be made commercially available. One of these new tools was deployed 70 m west of the original APT site closer to the NC89 seismometer and less than 20 m from the Clayoquot site BPR (labeled NC89 in Fig. 2) and connected in June 2017 (referred to in Fig. 2 and Table 1 as APT2). After evaluation of data from this tool, additional instruments were acquired and two were deployed and connected in June 2018, one at the ONC/NEPTUNE Barkley Canyon site and the other at the Cascadia Basin site (labeled NCBC and NC27 in Fig. 2, Table 1), both within a few kilometers of buried broadband seismometers identical to the one at the Clayoquot site. Tidal Deformation An early comparison of signals from oceanographic and earthquake sources recorded by a Güralp broadband seismometer and by the Quartz Seismic Sensors accelerometer after being packaged in the prototype APT1 tool and installed at the ONC/NEPTUNE Clayoquot slope site near seismometer NC89 is summarized in Figures 6 and 7. Without the perturbing contributions of varying temperature and groundwater levels that contaminated the vault test results, tidal period signals are more clearly documented by the offshore accelerometer data (Fig. 6) than they were in the land-based data. Signal amplitudes match those determined using seismometer mass positions, although there is a shift in phase of unknown origin. As discussed by Davis et al. (2017), the vertical signal at this site matches expected gravity variations arising from body and ocean attraction terms, but the horizontal acceleration is believed to reflect formation tilt resulting from ocean pressure loading. The amplitude of tilt is surprisingly large, given that the very long wavelength of open-ocean tides produces a spatially uniform load (unlike the situation at the coastal location of the PGC vault) and that the local topographic relief is small. The direction of maximum tilt is oriented in a direction across the strike of the subduction zone accretionary prism, suggesting a subseafloor structural origin (Davis et al., 2017). The cause of the tidal signals is not important for this article, however; what is important is that the observations demonstrate that changes in tilt can be resolved with the APT instrument at a level of a fraction of a microradian and more generally that oceanographic loading produces perturbations that must be taken into consideration in any search for geodynamic signals. Earthquake Seismic Waves Since the deployment of the first APT in 2015, a variety of local earthquake and teleseismic signals have been recorded. An example spanning a broad frequency band and dynamic range provided by ocean background signals and teleseisms from an M w 7.1 earthquake off Honduras is shown in Figure 7. An excellent match between the

7 454 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros Figure 7. (a) Waveforms and (b) spectra of seismic arrivals from an earthquake in the Caribbean Sea off Honduras recorded by the second APT instrument and the NC89 broadband seismometer. Seismic-wave arrivals are split into early high-frequency body waves and later lowfrequency surface waves. Spectra, including those for APT pressure, are computed for 1-hr periods during and before the earthquake. Upper and lower background spectra of Peterson (1993) are shown as dashed lines. The color version of this figure is available only in the electronic edition. waveforms (Fig. 7a) and spectral content (Fig. 7b) of signals seen by the APT and of those seen by the nearby NC89 broadband seismometer provided confidence that the efforts to couple the instruments to the formation by way of pushed insertion (in the case of the APT probe), burial (in the case of the seismometer), and instrument sediment mass matching (both instruments) had succeeded. As seen in the vault test results (Fig. 5), the limit imposed on resolution by the APT accelerometer counter, about 130 db relative to 1 ms 2,is apparent in the background oceanographic record before the earthquake at submicroseismic frequencies (< 0:05 Hz), although the contrast between the Güralp and APT noise limits is not as great. As frequency declines, the signal level climbs in a manner similar to that seen in Figure 5, but the site signal levels recorded by the APT are typically only 10 db or less above those recorded by the broadband seismometer. This limit to resolution is seen to be insignificant relative to the earthquake teleseismic signal levels. The primary factor that limits detection of seismic signals at higher frequencies (e.g., generated by small local earthquakes) is seen to arise from oceanographic noise. This is clear both from the intrainstrument comparison of signal levels of each of the components and from the comparison of pressure and vertical acceleration, which shows a relationship following that expected for the acceleration of the overlying water mass at the Clayoquot site under the influence of the ground motion ( 1:3 Pa=μms 2 ) over much of the seismic and microseismic frequency band. Microseisms and Infragravity Waves Representative waveforms and spectra of signals recorded during a local storm by the second Clayoquot APT are shown in Figures 8 and 9. As in the case of teleseismic arrivals, a good match is found between the oceanographically sourced waveforms recorded by the APT and seismometer (Fig. 8a) despite the separation of the instruments and possible differences in local site characteristics, and in this

8 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 455 Figure 8. Waveforms of microseisms and infragravity waves observed during a storm on 1 December Vertical acceleration seen by the second APT and the broadband NC89 seismometer (as the time derivative of velocity) are compared in (a), pressure observed by the APT and the nearby BPR are compared in (b), APT vertical acceleration and pressure are compared to one another with infragravity signals removed in (c), and with microseismic signals removed in (d). The color version of this figure is available only in the electronic edition. instance, there is little difference between signal levels anywhere above a frequency of about 0.01 Hz (Fig. 9a). No differences are seen in the comparison of pressures recorded by the APT and the BPR (Figs. 8b, 9b). Where coherence between acceleration and pressure is high within the microseismic band (periods centered at 8 s; Fig. 9c), the relationship between acceleration and pressure yields information about the spatial scale of coherence of double-frequency ocean-wave loading at the seafloor, with the longest period microseisms behaving the same as long-period seismic signals (i.e., 1D loading at 1:3 Pa=μms 2 ). In the infragravity waveband (Fig. 8d), coherence between acceleration and pressure is also seen (at periods in this case centered at 1 min; Fig. 9c), showing that during this and other times of strong ocean wave energy, the resolution of the APT may be adequate for compliance determinations. A more comprehensive view of the practical resolution of acceleration measured with the APT instruments is provided by a comparison of sepectra, expressed as probability density functions, of data from the APTs and nearby Guralp broadband seismometers at the three ONC deployment sites (Fig. 10; see Fig. 2 and Table 1 for locations). The data span intervals ranging from about two weeks (late June to early July 2018 for the most recently deployed Barkley Canyon and Cascadia basin instruments, Fig. 10a,c, respectively) to nearly 1 yr (for Clayoquot, Fig. 10b). Several things are apparent in this comparison. Despite the data window available for the Cascadia and Barkley Canyon sites being in an oceanographically quiet season (e.g., Thomson et al., 2014) and the nearly full year of recording at Clayoquot being dominated by quiet intervals, signal levels fall near the upper bound of background signals as defined by the standard characteristic curves of Peterson (1993). This noisy condition is well known to be characteristic of the northeastern Pacific Ocean (e.g., Webb, 1998). As seen in time-specific spectra presented in other figures of this article, background signals are strongest at microseismic frequencies. Much less energetic infragravity waves are present in the vertical-component seismometer data, but these signals generally fall below the inherent resolution of the APTs as defined by the frequency counting accuracy and full-scale range, that is, 125 db in power relative to 1 ms 2. Infragravity wave signals climb above this only occasionally (e.g., Figs. 8 and 9). Another comparison that might best illustrate the inherent resolution of the APTs as they are currently configured is made in Figure 11. These spectra represent a brief (12-hr) interval (within the longer one used in Fig. 10b) when oceanographic background signals were at a particularly low amplitude. Choice of this period allowed not only an evaluation to be done when site noise was minimized but also a comparison with a new instrument, a Nanometrics TitanEA accelerometer, one of several now buried much like the Güralp broadband instruments and commissioned by ONC/ NEPTUNE at each of the locations listed in Table 1. To complete the comparison, results from the strong-motion sensor of the Güralp seismometer are also shown. At frequencies above 0.1 Hz, data from the Titan, APT, and Güralp (velocity

9 456 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros Figure 9. Comparison of (a) spectra of acceleration and (b) pressure variations calculated for a 1-hr segment of data spanning the records shown in Figure 8. Upper and lower background spectra of Peterson (1993) are shown as dashed lines. Correlation between APT acceleration and pressure seen in Figure 8 is confirmed by coherence between acceleration and pressure near unity within microseismic and infragravity bands (c). The color version of this figure is available only in the electronic edition. channel) agree well (with a small exception of the vertical component of the Titan at 0.12 Hz), but noise levels exhibited by the Güralp acceleration sensor (used for strongmotion detection) are understandably high. Below Hz, instrument noise of the Titan and APT causes their signals to fall above the site noise level defined by the Güralp velocity channel. Low-amplitude signals in the infragravity waveband that are seen during this calm period by the vertical component of the Güralp seismometer are not resolved with the APT, which is limited by its frequency counter resolution to about 125 db, or with the Nanometrics instrument, which is characterized by an inherent noise level typically 5 db greater than that of the APT. Noise characteristics are simple, with APT, Nanometrics, and Güralp accelerometer powers rising with falling frequency at rates of 1.5, 2.5, and 4.5 db/octave, respectively. In summary, the resolution of the APT in the seismic frequency band is generally adequate in this environment, with seismic detection thresholds set primarily by oceanographic noise that is to a certain extent exaggerated by site characteristics (e.g., high-signal levels at frequencies within and above the microseismic band associated with sedimentary shear modes). Resolution of vertical motion associated with infragravity waves will be limited to oceanographically energetic periods (e.g., Fig. 9). Data from a Recent Local Earthquake On 22 October 2018, a series of earthquakes on the Sovanco transform fault that bounds the Pacific and Explorer plates provided one more useful illustration of the utility of the APT instrument. The first earthquake of the series was an M w 6.1 event 200 km distant, and although recording has not yet been made routine, 20 samples per second data were acquired from the second APT deployed near the NC89 broadband seismometer. A comparison of records at the time of this event is shown in Figure 12. As expected from Figure 11, high-frequency instrument noise can be seen riding on top of approximately 7-s period background microseisms in the case the APT and more so in the case of the Güralp accelerometer, which displays long-period noise as well (Fig. 12a). The high-frequency noise partly masks the small Pn arrival (attenuated by the thick accretionary prism sediments) in the APT data and fully precludes its detection by the Güralp accelerometer (Fig. 12b). Later compressional and shear waves seen by all sensors track one another well, but 40 s into the seismic wavetrain, the large amplitude surface-wave signals recorded by the Güralp seismometer are clipped (Fig. 12c) and distorted by adjustments activated by the mass-centering electronics. This behavior can be dealt with by use of the Güralp strong-motion sensor data, but the lack of distortion and clipping by the high-dynamic range quartz sensor of the APT makes such corrections unnecessary. Temperature Variations An example of temperatures measured at the seafloor by the nearby BPR and at the bottom of the tool by the accelerometer temperature compensation crystal is shown in Figure 13. Seafloor temperatures at this site vary over tidal to month-long periods, and these variations diffuse into the sediment section. The longer term variations observed at the bottom of the APT are attenuated by roughly a factor of 5 and

10 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 457 (a) (b) Figure 10. Comparison between north- (upper panels) and vertical-component (lower panels) probabilistic power spectral density functions for APTs and nearby Güralp CMG-1T broadband seismometers (right and left panels, respectively, with the seismometer velocities differentiated to accelerations) at (a) Barkley Canyon, (b) Clayoquot slope, and (c) Cascadia basin sites. Standard high- and low-noise models of Peterson (1993) are shown as thick gray lines. Mean power spectral densities are shown as thick solid black lines. Probabilistic power spectral density functions were computed following the work of McNamara and Buland (2004). The probability density functions were constructed for the APT and broadband seismometer over the period 22 June July 2018 using (a) 731 and hr segments, respectively and (b) 528 and hr segments, respectively. Probability density functions in (c) were constructed for the APT with data over the period 2 August July 2018 using hr segments and for the broadband seismometer with data over the period 20 October July 2018 using 11,478 1-hr segments. Low-rate broadband seismometer data were used to prevent data availability conflicts with military data-diversion schedules. (Continued)

11 458 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros (c) Figure 10. Continued. delayed by 1 week relative to those in the bottom water (Fig. 13a). The full frequency-dependent phase and amplitude relationship between the seafloor and sediment temperatures constrains the thermal diffusivity of the intervening sediment section (Davis and Villinger, 2018). Tidal-period variations are also present at depth (Fig. 13b); although small ( 0:2 mk), they are much too large to be a consequence of thermal diffusion from the seafloor. Their highly regular character and close correspondence with seafloor pressure (Fig. 13b) suggest that they are a consequence of adiabatic heating and cooling under the influence of tidal loading (Davis and Villinger, 2018). Secular Acceleration Variations Resolution of secular geodetic signals is severely limited by sensor drift. Drift exhibited by loaded quartz crystals probably arises from crystal aging, outgassing, and creep under load and is difficult to avoid (e.g., Paros and Kobayashi, Tech. Note G8101). Drift of the acceleration sensors over the full history of the first APT record is illustrated in Figure 14, along with the computed total gravity. Several first-order aspects of this long-term record include (1) Initial drifts of the x and z sensors decreased rapidly, reversed direction, and then became generally linear after the first few months after deployment. The initial drifts may have been a consequence of creep under load because of the change in the orientation of the tool, which had been kept horizontal for several months before deployment or to the change in ambient temperature from the laboratory and ship to the seafloor. Physical equilibration of the tool in the sediment may also have contributed to this early transient, although this contribution is probably small, given that a similar transient is seen in the total gravity. (2) The sign of the long-term drift exhibited by the y axis is opposite that exhibited by the x and z axes. This may be noteworthy, but the cause is unknown. Drifts of all three channels of the second APT deployed at the Clayoquot site were positive. (3) The magnitudes of drift displayed by each channel after 1 2 months is large but roughly linear ( 2:5 mm s 2 =yr). Drifts exhibited by each channel of the second tool were lower (0:1 0:8 mm s 2 =yr) for the horizontal channels and 0:45 mm s 2 =yr for the vertical. Longterm characteristics of the third and fourth APTs deployed most recently at the Barkley Canyon and Cascadia basin sites are not yet determined. As in the case of the early transients, the long-term secular trends cannot be caused by settling of the tool. Whereas real changes in tilt would be seen primarily by the horizontal axes, secular trends observed are similar among all three axes. Furthermore, the total gravity value computed from the three components displays a secular trend that is nearly identical to the trend of the vertical channel the expected behavior of an instrument with a sensor drift that is small compared with g and that is installed with little inclination. This confirms that the cause of the drift must arise from the sensors or the period counters. Laboratory tests at RBR using a rubidium clock as a source have shown that the counter reference crystals provide counting stability to within

12 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 459 Figure 11. Signal levels observed during an oceanographically quiet period with four independent instruments located at the ONC/ NEPTUNE Clayoquot slope observatory site, including the APT and buried Güralp CMG-1T broadband seismometer (differentiated velocity output) discussed extensively in this article, the strongmotion acceleration sensor in the Güralp seismometer, and a buried Nanometrics TitanEA strong-motion accelerometer (see Fig. 2 and Table 1 for locations). Power spectral densities in this figure were generated using the Welch method. APT data were demeaned and filtered using a high-pass harmonic reduction filter at Hz. Other data from Incorporated Research Institutions for Seismology (IRIS) were decimated to 5 samples per second and bandpass filtered similarly. Upper and lower background spectra of Peterson (1993) are shown as dashed lines. The color version of this figure is available only in the electronic edition. 0.5 ppm per year or 0:3 mm s 2 per year when converted to acceleration. This is generally smaller than the annual drift exhibited by the APT accelerometers. Thus, much of the observed drift must arise from the sensor crystals themselves. These rates are comparable to drift rates exhibited by pressure sensors studied by Polster et al. (2009) and of sensors used in ONC NEPTUNE BPRs early in their operating histories. However, after 2 3 yrs of NEPTUNE operations, annual rates of BPR drifts stabilized at ppm (again, stated as a fraction of the crystal frequency). Hence, the accelerometer sensor drifts seem large, and they will limit the detection of secular geodetic changes to Figure 12. Comparison of observations of acceleration recorded by the APT and the buried Güralp broadband seismometer (velocity channels converted to acceleration = HHZ and HHN, acceleration channels = HNZ and HNN) at the ONC/NEPTUNE Clayoquot slope site (a) before and (b,c) during seismic-wave arrivals from an M w 6.1 earthquake on the Sovanco transform fault 200 km to the northwest. Minor ticks are shown at intervals of 10 s in the upper and lower panels and 1 s in the middle panels. The time of the Pn body-wave arrival is indicated with an arrow. Clipping of the Güralp surface-wave velocity signal in (c) results in zero computed acceleration values. The color version of this figure is available only in the electronic edition. 0:5 mm s 2 =yr in gravity and 0:1 mrad=yr in tilt (in the case of the second APT). With trends removed, however, transient signals of amplitudes greater than those associated with oceanographic loading should be easily resolved. A technique for determining the composite drift of the three axes (the total g value) involves periodically rotating the tool without changing its gravitational position, with the goal of exposing each axis to 1g (J. M. Paros, Triaxial Acceleration Assembly and In-situ Calibration Method for Improved Geodetic and Seismic Measurements, U.S. Patent 9,645,267 B2, 9 May 2017). A singular example of such an operation carried out at the time of the installation of the second APT instrument is shown in Figure 15. Doing this

13 460 E. E. Davis, M. Heesemann, J. J. Farrugia, G. Johnson, and J. Paros Figure 13. Temperatures recorded with the temperature compensating crystals of the APT acceleration sensor at the bottom of the tool 1 m below the seafloor along with seafloor temperature and pressure recorded with the nearby BPR (see Fig. 2). (a) Long-term sediment temperature variations are attenuated relative to and lag those in the bottom water. (b) Short-term sediment temperature variations (with 10-min averaged values superimposed on raw data) exhibit a clear tidal component that is in phase with seafloor pressure (smooth curve). The color version of this figure is available only in the electronic edition. repeatedly with the existing design cannot be done without physically removing the instrument from the seabed. Efforts to do this by automated rotation of the sensor within an instrument are being made by W. Wilcock (personal corr., 2018), but application to a tool having the form factor of the APT would not be realistic. Deformation Transients During the first APT deployment, a small number of acceleration anomalies stood well above oceanographic and seismic signals and instrument drift (Fig. 14). Some of these include a change in the total g value; their origin is unknown but likely to be instrumental. The abrupt change in early April provides one example; this change in g originates exclusively from the vertical axis. Other anomalies are not accompanied by a change in total g and are likely to reflect formation tilt. Arguments supporting this supposition, made on the basis of aspects of the example shown in Figure 16, include (1) anomalous values of acceleration are seen simultaneously on all three channels but dominantly along the horizontal axes (as would be expected from tilt); (2) total tilt determined from the combination of the horizontal axes follows tilt determined from the vertical channel, and as a corollary to this, the total gravity value does not change (see Fig. 14); (3) no temperature anomaly is seen at the time of the event ; one would be expected if the event were related to physical disturbance of the tool; and (4) acceleration (tilt) returns to a value identical to that preceding the event. It is possible that this and other transients (e.g., the one spanning early April May, also prominent in the horizontal channel data) are associated with hydrologic activity at or below the fluid and gas vents in the area (Fig. 2), although no coordinated signals were seen at the seismometer located 140 m away, suggesting that the signal is very local in origin. Whatever the cause, it seems clear that with the redundancy provided by the triaxial observations, signals associated with deformation can be confidently resolved when they rise above the 1 μrad level of oceanographically induced tilts. Summary A new instrument has been developed that is capable of resolving a broad range of seismic, oceanographic, and geodynamic signals. Its salient characteristics include robustness, ease of deployment, low-power consumption, small size, high precision, large dynamic range, and broad bandwidth. It houses a triaxial quartz accelerometer with each axis having a range of 3g and a pressure sensor having a range of 40 MPa, mounted inside a slim 1-m-long pressure case designed to be pushed into seafloor sediment. It can be adapted easily for other deployments, for example, in boreholes, and in low-profile monuments on hard substrates. The tool uses high-precision (1 ppb) period counters to determine the output frequency of the pressure and acceleration sensors. Each sensor also includes a temperature sensitive crystal. Acceleration, pressure, and temperature variations are resolved at levels of 0:6 μms 2, 0.4 Pa, and 0.08 mk, respectively. With sampling rates up to 20 samples per second, the tool functions well as a seismometer, with noise levels being in the range of 120 to 130 db (power relative to 1 ms 2 ) over a frequency band ranging from 0.01 to 5 Hz, higher than those characteristic of broadband seismometers but only slightly higher than characteristic seafloor site noise and significantly lower than noise characteristic of other types of sensors (Fig. 11). The total frequency range, which spans from the Nyquist frequency to drift-limited DC and the large

14 APT: An Instrument for Monitoring Seafloor Acceleration, Pressure, and Temperature 461 Figure 14. Full histories of acceleration and computed total gravity data for the first APT deployment. Raw data plotted in the background show periods of enhanced infragravity and microseismic signals associated with local storms and swell from distant sources, short-lived signals from earthquakes, and occasional spurious spikes; 10- min average data (dark lines) show tidal components (see expanded plot in Fig. 6), secular trends related to sensor drift, and transient anomalies. The color version of this figure is available only in the electronic edition. and slow slip on faults. The high sensitivity of the temperature compensation crystal in the accelerometer has permitted observations of tidal adiabatic heating and hence determination of sediment thermodynamic properties. Planned improvements to the tool include addition of (1) a thermistor at the top of the tool to allow bottom-water temperature monitoring, (2) an auxiliary module for in situ pressure calibration, and (3) an exchangeable memory and battery module for extended autonomous operations. Use of a quartz tilt sensor for horizontal channels would provide increased resolution for seismic studies but with some loss of dynamic range, ruggedness, and deployment simplicity. Data and Resources Pressure and temperature data from Ocean Networks Canada (ONC/ NEPTUNE) bottom pressure recorder (BPR) data are available at the ONC data portal Velocity and mass position data from the seismometers are available at the Incorporated Research Institutions for Seismology (IRIS) data portal service.iris.edu/irisws/timeseries. Pressure, temperature, and acceleration data from the APT instruments will soon be available also at the IRIS data portal. Data for this investigation were last accessed in October Details about the Güralp CMG-1T seismometer beyond those described in Davis et al. (2017) canbefoundathttp:// Details about the current generation APT instrument can be found at docs.rbr-global.com/apt. All websites were last accessed on October Figure 15. Pressure and acceleration records acquired at the time of installation of the second APT instrument. The instrument was held with x then y axes oriented vertically up ( and min, respectively) before the tool was inserted partially (40 45 min) and then fully into its final position (post-50 min). The computed total gravity amplitude shown in the lower panel probably contains early transient reaction of the sensor from its stored to deployed thermal state and orientation. The color version of this figure is available only in the electronic edition. dynamic range also allows the tool to serve in monitoring strong ground motion, tidally induced deformation, and geodynamic deformation associated with seismic, tsunamigenic, Acknowledgments The authors thank the National Science Foundation for supporting the development of the prototype instrument as part of a Grant (OCE ) to Laura Wallace for tilt and pressure monitoring at the Hikurangi subduction zone, New Zealand. The authors also thank Robert Macdonald of Biologica Ltd., Chris Foreman of Foreman CNC, and Robert Meldrum of Pacific Geoscience Centre (PGC) for responding quickly to the design, fabrication, and assembly of mechanical and electronic components of the first acceleration, pressure, and temperature (APT) instrument. John Bennest developed the original 1-ppb precise-period counters (PPCs) for the PGC that have been used in seafloor and borehole pressure monitoring