Volume 45, Issue 12 p. 6008-6017
Research Letter
Free Access

Augmenting Onshore GNSS Displacements With Offshore Observations to Improve Slip Characterization for Cascadia Subduction Zone Earthquakes

Jessie K. Saunders

Corresponding Author

Jessie K. Saunders

Cecil H. and Ida M. Green Institute of Geophysics and Planetary Physics, Scripps Institution of Oceanography, University of California, San Diego, San Diego, CA, USA

Correspondence to: J. K. Saunders,

[email protected]

Search for more papers by this author
Jennifer S. Haase

Jennifer S. Haase

Cecil H. and Ida M. Green Institute of Geophysics and Planetary Physics, Scripps Institution of Oceanography, University of California, San Diego, San Diego, CA, USA

Search for more papers by this author
First published: 08 June 2018
Citations: 5

Abstract

For the Cascadia subduction zone, Mw~8 megathrust earthquake hazard is of particular interest because uncertainties in the predicted tsunami size affect evacuation alerts. To reduce these uncertainties, we examine how augmenting the current Global Navigation Satellite Systems (GNSS) network in Cascadia with offshore stations improves static slip inversions for Mw~8 megathrust earthquakes at different rupture depths. We test two offshore coseismic data types: vertical-only bottom pressure sensors and pressure sensors combined with GNSS-Acoustic aided horizontal positions. We find that amphibious networks best constrain slip for a shallow earthquake compared to onshore-only networks when offshore stations are located above the rupture. However, inversions using vertical-only offshore data underestimate shallow slip and tsunami impact. Including offshore horizontal observations improves slip estimates, particularly maximum slip. This suggests that while real-time GNSS-Acoustic sensors may have a long development timeline, they will have more impact for static inversion-based tsunami early warning systems than bottom pressure sensors.

Key Points

  • Amphibious station configurations better constrain shallow megathrust static slip characteristics compared to onshore-only networks
  • Three component offshore coseismic data are needed to retrieve >80% of the maximum slip for shallow ruptures with static slip inversions
  • Onshore displacements from first arrival to peak ground displacement can be used to rapidly evaluate the quality of static slip estimates

Plain Language Summary

The Cascadia subduction zone is the region of highest tsunami hazard within the contiguous United States. This region has experienced many tsunamis over the last 10,000 years that were generated by earthquakes of magnitude 8 to 9. Magnitude 8 earthquakes in the subduction zone can be tricky for tsunami early warning systems because it is difficult to determine the depth of the earthquake rupture, which strongly affects the anticipated tsunami size. This can make the difference between an evacuation order being issued or not. This study tests how estimating total slip on the earthquake fault during rupture and the resulting tsunami wave height for magnitude 8 earthquakes can be improved when combining the current land-based Global Navigation Satellite Systems monitoring network in the Pacific Northwest with offshore seafloor networks. We test hypothetical arrangements of offshore stations that measure the vertical seafloor motion with ocean bottom pressure sensors. We also test networks that measure motion in all three directions by including Global Navigation Satellite Systems measurements at the sea surface linked by acoustic communication to measurement points on the seafloor. This work can help plan where best to put new offshore instruments as they are developed for future tsunami early warning systems.

1 Introduction

The Cascadia subduction zone, where the Juan de Fuca plate is subducting under the North America plate, is a region of high earthquake and tsunami hazard for the west coast of the United States. While no great earthquakes of Mw ≥ 8 have occurred in the Cascadia subduction zone during its documented history beginning around 1790 (Atwater et al., 1995), paleoseismic studies using turbidites, shoreline subsidence, and tsunami deposits confirm that the region is capable of great earthquakes (Goldfinger et al., 2003, 2012; Nelson et al., 2006). In addition to Mw~9 events, the paleoseismic record shows that the southern portion of Cascadia offshore Oregon experiences Mw~8 tsunamigenic earthquakes (Goldfinger et al., 2012).

The local tsunami impact from Mw~8 megathrust earthquakes is strongly dependent on earthquake depth. Rupture of the shallowest portion of the subduction interface, from ~15 km depth to the trench, can produce long-duration earthquakes that have anomalously low seismic moment and short-period energy release accompanied by large coseismic offsets (Lay et al., 2012; Lay & Bilek, 2007). Such events are called “tsunami earthquakes” due to their disproportionately large tsunamis compared to their seismically determined magnitude (Kanamori, 1972; Polet & Kanamori, 2000). In Cascadia, it is difficult to determine shallow rupture potential as the pattern of subduction interface coupling is not well constrained near the deformation front due to the lack of present-day seismicity and long-term offshore geodetic data (Pollitz & Evans, 2017; Schmalzle et al., 2014). However, local tsunami warning systems should be prepared for a shallow megathrust rupture.

Standard warning level thresholds for tsunami early warning used by the National Oceanic and Atmospheric Administration Tsunami Warning Centers are tsunami wave heights of 30 cm, 1 m, and 3 m (Whitmore et al., 2009). Mw 8 subduction zone earthquakes can generate tsunamis that fall within the 1–3 m range and exceed 3 m in the case of shallow rupture. It is therefore important to correctly estimate tsunami wave heights in order to issue appropriate warnings. Early warning systems that rely only on hypocenter and magnitude estimates using seismic sensors may miscalculate the potential impact of tsunami earthquakes because the lack of high-frequency energy may underestimate earthquake magnitude (Kanamori, 1972). Magnitude estimates using long-period W-phase are less susceptible to this miscalculation (Kanamori & Rivera, 2008; Zhao et al., 2017), but near-field Global Navigation Satellite Systems (GNSS) data provide observations that can allow for detailed tsunami warning models in faster time frames (Melgar, Allen, et al., 2016).

Near-source GNSS displacement data measure low-frequency motions and coseismic offsets helpful for characterizing tsunamigenic earthquakes (Melgar et al., 2012). Magnitude can be rapidly estimated through peak ground displacement (PGD) scaling (Crowell et al., 2013; Melgar et al., 2015), and the pattern of onshore coseismic displacement can be used to estimate downdip rupture extent (Singh et al., 2012). The Pacific Northwest has a dense, real-time GNSS network that is being incorporated into early warning systems that use rapid coseismic offsets estimations to compute finite fault slip distributions (Crowell et al., 2016; Grapenthin et al., 2014; Minson et al., 2014). These fault slip distributions can be input to tsunami simulations to provide tsunami wave height estimates for early warnings. However, the one-sided distribution of data from land-based networks may be unable to uniquely determine the slip distribution of a Mw 8 rupture, especially for a shallow tsunami earthquake. Measurements of offshore coseismic displacement could help reduce this nonuniqueness, if strategically placed.

In this project, we examine how augmenting the current real-time GNSS stations in the Pacific Northwest with offshore sensors can improve the static slip estimates of shallow ruptures for tsunami early warning. At the 2017 Offshore Geophysical Monitoring in Cascadia Workshop (University of Washington, 2017; report available at www.cascadiaoffshore.org), various network configurations were discussed along with different offshore instrument types, including preliminary analysis of this project (Saunders & Haase, 2017). In addition to GNSS-Acoustic (GNSS-A) sensors, fiber-optic strainmeters (Blum et al., 2008; Zumberge, 1997) were considered as potential sources of horizontal deformation information. Ocean bottom strong motion accelerometers were also discussed. These could be useful for kinematic slip inversions but are excluded in this study because of the difficulty estimating coseismic displacements due to baseline offset integration errors (Boore & Bommer, 2005). Linking these sensors on a cabled array would provide real-time data streaming, which adds constraints on network design and additional motivation for this sensitivity study.

Here we test multiple offshore station configurations and different offshore coseismic data types, using ocean bottom pressure and GNSS-A sensors as our hypothetical offshore instrumentation. As these instruments have different expected data uncertainties, we use synthetic slip models for our analysis. This also allows us to assess the influence of these data types and their uncertainties on the resulting rapid tsunami forecast for these slip models. The station configurations examined here do not form an exhaustive list but capture the main differences between proposed network concepts for static slip inversion.

2 Methods

2.1 Slip Models and Tsunami Prediction

We consider two synthetic ruptures of Mw8.0 with dimensions chosen using the empirical scaling relationships of subduction zone earthquakes by Blaser et al. (2010). The fault model is a 10-km uniform grid approximation of the McCrory et al. (2012) 3-D Cascadia slab geometry. Both slip models have uniform 5-m slip with linear slip taper at the rupture edges (Figure 1). The first slip model is between 20 and 30 km depth, which produces coastal subsidence similar to paleoearthquake observations (Atwater et al., 1995). Coseismic offsets are computed using Green's functions from the frequency-wave number approach of Zhu and Rivera (2002) at the low-frequency limit with a 1-D multilayered Earth structure model for Cascadia (Gregor et al., 2002). The second model places slip in the shallow subduction interface above 15 km, the depth range of observed tsunami earthquakes (Lay et al., 2012). The shallow rupture produces little coseismic displacement onshore, which will make it difficult for land-based instrumentation alone to characterize the rupture region.

Details are in the caption following the image
Input slip models for the (a) deeper rupture and (b) shallow rupture. (c) Maximum tsunami amplitude at the coastline for the input ruptures. (d) Eastward and vertical coseismic offsets for the input slip models along a transect that runs through the center of the ruptures. The region where onshore GNSS stations are available is shaded in purple. (e) Location of the modeling region.

We then compute wave amplitudes at the coastline from tsunamis generated by the input ruptures using the GeoClaw open-source software, which solves the 2-D shallow water equations using a finite volume approach (Berger et al., 2011; LeVeque et al., 2011). We use the Shuttle Radar Topography Mission 15+ data set (Sandwell et al., 2014) as the input bathymetry in our tsunami simulations because input bathymetry with larger grid spacing can produce spurious coastal wave height estimates (Figure S1 in the supporting information). We forward compute vertical coseismic offsets for a grid on the seafloor and interpolate to the SRTM15 grid spacing before input to the GeoClaw simulations, where we assume instantaneous seafloor uplift. We also include the vertical displacement of water caused by horizontal motion of steeply-sloping bathymetry, which can contribute significantly to the tsunami if slip occurs near the deformation front (Tanioka & Satake, 1996). The deeper and shallow slip models generate tsunamis with maximum wave heights of 2 and 3 m at the coastline, respectively (Figure 1), which would trigger different evacuation levels.

2.2 Station Configurations and Offshore Data Types Tested

We test five station configurations in this study (Figure 2). The first is the existing real-time GNSS network in the Pacific Northwest. This network is composed of the Plate Boundary Observatory (Herring et al., 2016) and the Pacific Northwest Geodetic Array (Popovici et al., 2015). The second configuration is onshore GNSS augmented with a dense offshore network with 0.4° station spacing similar to the Japanese S-net system (Uehira et al., 2012). The third and fourth networks are the GNSS network with offshore trench-perpendicular transect profiles with 2° and 1° latitude profile spacing, respectively. And the fifth network is the GNSS network with an offshore trench-parallel line of stations above the deformation front.

Details are in the caption following the image
Networks considered in this study. The black dots show existing real-time Global Navigation Satellite Systems (GNSS) station locations and magenta dots show hypothetical offshore instrument locations. (a) GNSS-only configuration. The other networks are the GNSS network augmented with the following offshore networks: (b) a configuration similar to the S-net cabled array offshore Japan, (c) a sparse transect profile configuration, (d) a dense transect profile configuration, and (e) a trench parallel profile configuration.

For each rupture, synthetic coseismic offsets are computed at all stations, and Gaussian noise is added before inversion. We assume GNSS errors with standard deviation 1.5 cm in the horizontal and 5 cm in the vertical for the onshore coseismic displacements assuming a high-rate, real-time precise point positioning solution (Geng et al., 2013). For offshore coseismic data, we assume errors with standard deviation 5 cm in the horizontal based on GNSS-A observations (Chadwell, 2016) and 1.5 cm in the vertical based on seafloor pressure sensor accuracy (Tsushima et al., 2009).

We test two types of offshore coseismic data configurations in the static slip inversions: vertical-only and three component. Vertical coseismic offsets have been measured accurately offshore using existing seafloor pressure sensors (Bürgmann & Chadwell, 2014; Saito & Tsushima, 2016). Such data can be obtained in near real-time through cabled systems like the S-net in the Japan subduction zone, DO-NET in the Nankai subduction zone (Kawaguchi et al., 2008), and the OOI and NEPTUNE cabled arrays for the Cascadia subduction zone (Kelley et al., 2014). Horizontal coseismic data could be acquired by a GNSS-A system composed of acoustic transponders on the seafloor linked to a GNSS receiver on the sea-surface directly above the transponders (Bürgmann & Chadwell, 2014). GNSS-A receivers deployed along the Japan Trench recorded ~50 m of horizontal displacement near the trench after the 2011 Mw9.0 Tohoku-oki earthquake (Fujiwara et al., 2011). These observations were key in confirming that the Tohoku-oki earthquake ruptured into the shallow megathrust. Real-time observations of these large horizontal displacements would be very useful in determining the rupture location during a megathrust earthquake. GNSS-A networks such as the one in the Nankai subduction zone are surveyed in annual campaigns (Tadokoro et al., 2012; Yokota et al., 2015), but the addition of buoys or remotely operated vehicles could help develop real-time capabilities in the future (Chadwell, 2016).

2.3 Static Slip Inversion Methodology

In the forward problem, coseismic displacements can be computed given an earthquake slip model and Green's functions that relate slip on a buried fault to surface deformation over a layered half space. These Green's functions can also be used to invert for earthquake slip if the coseismic deformation is known. We perform static slip inversions using the open-source MudPy inversion code (Melgar & Bock, 2015) with the assumptions that slip is oriented within 45° of the dip-slip direction and occurs on a known fault geometry, the subduction interface. This allows the static slip inversion problem to be solved using a nonnegative least squares approach (Trifunac, 1974). We employ Laplacian spatial smoothing, where the optimal smoothing weight is objectively chosen using the minimization of Akaike's Bayesian information Criterion given a range of inversion results with different smoothing parameter weights (Akaike, 1998; Ide et al., 1996). We perform slip inversions for every combination of rupture, station configuration, and offshore data type using 15 realizations of observational noise. The resulting fault slip estimates are used to compute the vertical seafloor deformation and input to the tsunami calculation. We compare the slip and tsunami wave height estimates at the coastline from the slip inversion results to that produced by the input slip models to determine which offshore network and offshore data type provide the most valuable additions to the GNSS network.

A comparison of model resolution matrices for these configurations is included in Text S1 and Figures S2S6 in the supporting information. While these are informative, they do not consider the impact of data uncertainty. The optimal solution is determined objectively based on the ABIC, which considers the relative size of the data uncertainties with the deformation amplitudes when selecting the regularization weighting. An important point we consider in this study is that the data types have different levels of uncertainty that would affect the slip inversion results and affect the choice of offshore network that is implemented.

3 Results

The quality of the retrieved slip inversion and tsunami results are compared among offshore data types (vertical-only and three-component) for each rupture. Figure 3 shows one realization of slip inversion results using GNSS-only and the trench parallel profile configurations. Example static slip inversion results for all station configurations and both offshore data types are shown in Figures S7 and S8 for the shallow and deeper ruptures, respectively. The tsunami wave height estimates along the coastline for these inversion solutions are shown in Figure S9. For the deeper rupture, all station configurations recover the slip distribution and maximum slip. Some smearing of slip is present in the updip direction for the GNSS-only and vertical-only offshore inversion results that is reduced by having three-component offshore data directly updip of the rupture (Figures 3 and S8).

Details are in the caption following the image
Comparison of input ruptures, slip inversion results using GNSS-only and the trench parallel profile offshore configuration, and their resulting tsunamis. (top row) shallow rupture and (bottom row) deeper rupture.

For the shallow rupture, the GNSS-only inversion correctly places slip in the shallow region but significantly underestimates the amount of slip due to the lack of stations near the rupture (Figure 1d). This is similar for the sparse transect profiles. When there are stations located directly above the rupture, maximum slip improves. However, slip is consistently underestimated when using vertical-only offshore data compared to three component offshore data. The along-strike extent is not recovered unless there are multiple offshore stations above the rupture in the along strike direction.

We provide a quantitative measure of the differences among network configurations in Figure 4. The quality of the solution is measured in four ways: (1) rms difference between recovered and input fault slip (Figure 4a), (2) difference between recovered and input maximum fault slip (Figure 4b), (3) difference between maximum tsunami amplitude at the coast for the recovered and input slip models (Figure 4c), and (4) percentage of coastline hit by high amplitude waves for recovered relative to input fault slip tsunami simulations (Figure 4d). These measures are plotted with respect to the number of additional offshore stations, where the error bars are determined from the standard deviation of the parameters from all noise realizations. These figures show that there is significant improvement in all measures for the shallow rupture by incorporating three component data and can be accomplished with only a moderate number of stations.

Details are in the caption following the image
Comparison of slip inversion results using the two offshore coseismic data types. Configurations (left) for the shallow rupture and (right) for the deeper rupture. Inversion solutions using vertical-only offshore data are in lighter-colored circles, while those using three-component offshore data are in darker-colored squares. The rows show the comparison of (a) rms difference in slip, (b) maximum slip, (c) maximum tsunami wave heights, and (d) percentage of high amplitude waves >1 m at the coastline recovered in the tsunami estimates. All comparisons are plotted against the number of additional offshore stations in the networks. The error bars indicate the standard deviation of these values from all noise realizations.

Utilizing offshore data brings maximum wave height estimates to within 0.5 m (70%) for vertical-only data, and three-component offshore networks with stations located above or directly updip of the rupture estimate maximum wave height to within 0.25 m (90%) for the deeper rupture. For the shallow rupture, the tsunami simulations based on GNSS-only and offshore data that lack sites above the source (the sparse transect profiles) severely underestimate tsunami impact, estimating maximum wave height at <10% of the input model, because maximum slip is not recovered. Tsunami simulations improve when using the other offshore configurations but still tend to underestimate maximum tsunami wave height. Adding three-component offshore data improves maximum wave height estimates by at least 20% compared to estimates using vertical-only offshore data.

In addition to the maximum tsunami wave height at the coastline, the coastal extent of high-amplitude waves is a useful metric for determining the potential tsunami evacuation region. Figure 4c shows that offshore networks with multiple stations located above the rupture in the along-strike direction (the S-net and trench parallel profile configurations) are able to estimate the extent along the coastline hit by the highest amplitude tsunami waves (>1 m) when using three-component offshore data. This is most obvious for the shallow rupture. The ability of the network geometry to constrain the percentage of coastline affected with >1 m waves is quantified in Figure 4d.

4 Discussion

Among the networks investigated, increasing the number of offshore stations generally improves slip inversion results in terms of matching maximum slip and rms difference as well as improving tsunami wave height estimates along the coast. For the deeper rupture, adding additional offshore stations does not significantly improve slip results and tsunami simulations with either offshore data type due to the rupture's proximity to onshore stations. However, three-component offshore data directly above the rupture region are required to obtain the most improvement in slip and tsunami estimates for the shallow rupture.

The S-net configuration produces the most accurate slip inversion solutions and tsunami simulations, but maximum tsunami wave height is still underestimated for the shallow rupture by 15% for three-component offshore data and 45% for vertical-only offshore. Transect profile spacing of 1° latitude improves the likelihood that sites will be located over the rupture plane for a Mw8.0 earthquake compared to transects with greater spacing. While this configuration will estimate maximum slip for fault patches underneath the offshore stations using three-component coseismic data, the profiles are too sparse to be able to accurately determine along-strike rupture extent and underestimate the coastal extent that will experience tsunami amplitudes >1 m by 45%.

The preferred offshore station configuration is the trench parallel profile because it provides the most improvement in the shallow slip solution while also limiting the number of offshore stations. The trench parallel profile configuration captures the along-strike rupture extent, and when using three-component offshore data, the resulting tsunami computation accurately estimates the coastal impact of tsunami waves with >1.0 m amplitude and produces comparable tsunami amplitudes to the S-net configuration. Additionally, such GNSS-A stations in combination with onshore stations will provide robust long-term plate boundary deformation measurements above the shallow subduction interface. These offshore observations are necessary for determining the locking pattern of the Cascadia subduction interface (Pollitz & Evans, 2017; Schmalzle et al., 2014).

An additional verification step that could be implemented on a practical time scale for early warning is to confirm shallow megathrust slip using high-rate onshore GNSS displacement waveforms. To explore this, we examine 1-Hz synthetic displacement waveforms for the four ruptures shown in Figure 3: the input model of shallow slip, the slip inversion results using GNSS-only, and the two trench parallel profile inversion results (vertical and three component; Figure 5). We develop the static slip model into a kinematic rupture following the approach of Melgar, LeVeque, et al. (2016) and Graves and Pitarka (2010) (Text S3). As expected, all ruptures produce similar coseismic offsets, but there are deterministic differences in the early portion of the displacement waveforms that could be used for quality verification for the static slip solution, notably in differences in the amplitude and timing of the PGD of the waveforms in each component. Underestimation of the PGD by the GNSS-only and vertical-only offshore slip models indicates that larger slip offshore is required to match the observed waveforms. For example, the east component at station P367 shows a 5-cm difference between the PGD of the input model and the GNSS only slip model. The timing differences of the PGD are likely due to underestimation of along-strike rupture extent. Such comparisons of the first part of the waveform can be useful for determining if the static slip inversion underestimates the slip and rapidly checking solution quality.

Details are in the caption following the image
Comparison of displacement waveforms for the shallow rupture model and three static inversion results shown in Figure 3 at three stations highlighted by the pink diamonds in Figure 1. The dashed lines indicate sensitivity levels of real-time GNSS data.

5 Conclusions

Offshore data near the Cascadia subduction zone improve maximum tsunami height predictions by >20% for a deep (>15 km) Mw8 megathrust earthquake assuming known fault geometry. For a shallow megathrust rupture, maximum tsunami height estimates are improved by >2.5 m (>75%) when using strategically placed offshore stations. An error of 2.5 m has significant consequences where the 1 to 3-m wave height estimate determines whether an evacuation order is issued. Vertical-only offshore data (for example, derived from ocean-bottom pressure sensors) improve shallow slip estimates when combined with onshore data, but slip is consistently underestimated, which in turn underestimates the tsunami impact even with a dense offshore station configuration with 30–40 km station spacing. Offshore estimates of horizontal coseismic displacement improve retrieval of maximum slip, and fewer stations are required to compute a tsunami prediction with necessary accuracy sufficient for warning. Among the networks tested, an offshore station configuration with sites located parallel to the deformation front on the hanging wall satisfactorily recovers the along-strike rupture extent and the length of coastline that experienced high-amplitude tsunami waves. The onshore high-rate displacement waveforms from first motions to PGD can be used to check the quality of the static slip inversions.

Acknowledgments

This work was funded by NASA grant NNX16AO32H through the NASA Earth and Space Science Fellowship. We thank Diego Melgar for assistance in the use of MudPY and the GeoClaw tsunami propagation software. We thank Junle Jiang for his helpful discussions and assistance regarding the tsunami modeling software. We thank Editor Gavin Hayes, Tim Melbourne, and an anonymous reviewer for their constructive feedback and reviews. We appreciate the technical support from the IGPP Help Desk staff. We used the following software: fk code for computing the Green's functions (www.eas.slu.edu/People/LZhu/downloads/fk3.2.tar), MudPy slip inversion code (github/dmelgarm/MudPy), FakeQuakes kinematic forward modeling code (github/dmelgarm/FakeQuakes), and GeoClaw tsunami modeling software (geoclaw.org). The figures for this paper were made using Geodetic Mapping Tools and Python. Synthetic data used in this work can be found at http://agsweb.ucsd.edu/seismogps/cascadia/.