G24A-01 INVITED
Towards Real-time Recognition of Near-Field Tsunamigenic Earthquakes
Whether a tsunami has been generated from a large earthquake immediately offshore cannot be determined within a crucial time window from traditional seismic or tide gauge data alone. Geodetic data show that coseismic motions of the Earth's surface, even hundreds of kilometres from the earthquake, can be used to determine the nature of the rupture and whether the earthquake is likely tsunamigenic. High-rate data (1 Hz or greater) from continuous GPS stations located along the coastal margin can provide on-line streamed data that can be analyzed in real time to provide relative positions to an accuracy of 1-2 cm horizontally and 3-5 cm vertically with a latency of a few seconds. Regional ground displacements at the time of a major offshore earthquake could therefore point to the certainty of a tsunami within several minutes. Likewise, a coastal network of strong- motion instruments transmitting data in real-time can provide an immediate estimate of rupture propagation and rupture length and thereby possesses the same rapid tsunami discrimination potential. The Geological Survey of Canada, in cooperation with the Canadian Hydrographic Service has set up a small network of geodetic quality GPS installations with continuous on-line IP communications. This facilitates real-time positioning along the coast of the Canadian segment of the Cascadia subduction zone. The aim is to evaluate the realizability and effectiveness of automatically determining major vertical and horizontal motion at coastal versus inland GPS stations that would unambiguously and rapidly indicate tsunami generation. In addition, a prototype network of smart strong-motion seismographs is being deployed in the coastal region. Determining the length and character earthquake of rupture in the immediate offshore region using on-scale seismic data should also provide real-time tsunami discrimination. It is hoped that these relatively low cost techniques can become mainstream tools of tsunami warning systems worldwide.
G24A-02 INVITED
The 28th September 2004 Parkfield earthquake revisited through high-rate GPS data inversion.
Increasingly, Global Positioning System (GPS) data can also be used in real time to complement seismic data in providing robust real-time continuous earthquake information, and potentially, early warning. The occurrence of the Parkfield earthquake on the 28th of September 2004 provides an opportunity to test the sensitivity and reactivity of network, data processing algorithms, and the implementation of GPS static and temporal solutions in finite-source inversions. Incorporation of GPS data in realtime processing algorithms is important for several reasons. First, static deformation with adequate station coverage can be used to independently determine the orientation and dimension of fault rupture, as well as the scalar seismic moment. This processing complements routine moment tensor (MT) processing, providing needed redundancy, but goes beyond the MT with the potential for causative fault plane identification and determination of fault rupture dimensions. The dimensions of the rupture plane derived from GPS data can then be used to improve ShakeMap by accounting for rupture finiteness. Second, the rapidly determined deformation may also be integrated into joint inversions with seismic waveform data for kinematic rupture models. This can be accomplished using static deformation estimates, as well as displacement time series derived from high-rate GPS data. Thirdly, GPS provides a strong motion displacement meter capability for the largest earthquakes. Double integration of acceleration to displacement to recover the broadband time series with static offset can be problematic, whereas GPS potentially can measure it directly. We first present the calibration of a GPS time series by comparing it with records from seismic sensors for the Parkfield event, validating the use of the GPS in the near-field during a large event. Finite-source inversion results based on static GPS, GPS time series, as well as combinations of these data with seismic records will be compared. The sensitivity of the inversion results to the distribution of the near-field GPS sites will be presented. Realtime implementation strategies at the Berkeley Seismological Laboratory will be discussed.
G24A-03 INVITED
Accuracy assessment of high-rate GPS measurements for seismology
Analysis of GPS measurements with a controlled laboratory system, built to simulate the ground motions caused by tectonic earthquakes and other transient geophysical signals such as glacial earthquakes, enables us to assess the technique of high-rate GPS. The root-mean-square (rms) position error of this system when undergoing realistic simulated seismic motions is 0.05~mm, with maximum position errors of 0.1~mm, thus providing "ground truth" GPS displacements. We have acquired an extensive set of high-rate GPS measurements while inducing seismic motions on a GPS antenna mounted on this system with a temporal spectrum similar to real seismic events. We found that, for a particular 15-min-long test event, the rms error of the 1-Hz GPS position estimates was 2.5~mm, with maximum position errors of 10~mm, and the error spectrum of the GPS estimates was approximately flicker noise. These results may however represent a best-case scenario since they were obtained over a short (~10~m) baseline, thereby greatly mitigating baseline-dependent errors, and when the number and distribution of satellites on the sky was good. For example, we have determined that the rms error can increase by a factor of 2--3 as the GPS constellation changes throughout the day, with an average value of 3.5~mm for eight identical, hourly-spaced, consecutive test events. The rms error also increases with increasing baseline, as one would expect, with an average rms error for a ~1400~km baseline of 9~mm. We will present an assessment of the accuracy of high-rate GPS based on these measurements, discuss the implications of this study for seismology, and describe new applications in glaciology.
G24A-04
The Importance of Data Equalization for Reducing 'Repeatable' Errors in High-Rate GPS Positions
When GPS position solutions are computed stochastically, positioning error that would normally be distributed to the solution residuals instead appears in the position estimates. These errors are often several centimeters in magnitude and can easily disguise any geophysical signal, consequently limiting the applicability of raw high-rate GPS positions to geophysical monitoring. Therefore, these position errors must be mitigated for proper integration of high-rate GPS and real-time seismic data. A good deal of recent high-rate GPS research has centered on mitigating high-rate positioning error at the post-processing analysis stage, i.e. removing errors from position timeseries. Many of these techniques (sidereal filtering, modified sidereal filtering, aspect repeat time adjustment) compute positions on one or more days to develop an error profile, align this error profile with a day of interest, and subtract the error profile. The development of these high-rate error reduction techniques has concentrated on determining the perfect time shift to align the error profile with the day of interest. However, proper determination of repeat period is negated if identical data sources are not used in the least squares adjustment - for the satellite-antenna geometry to repeat between two different solutions, the same satellites must be used to compute those solutions and the same number of biases must be resolved. Therefore, for position errors to be truly repeatable, the same satellites and biases must be used in comparable solutions. This paper addresses data equalization, the process by which GPS data on two separate days are brought into spatial alignment and forced to be truly repeatable. We discuss methods for equalizing data before solving for position, and provide examples of data equalization and its effects on error reduction in high-rate GPS positions. Additional data equalization complications are discussed, including equalizing by data quality and accounting for automated data editing algorithms.