Methods Of Coseismic Displacement Estimation Using A Single High-rate GNSS Receiver And Combining With Strong-motion Seismometers For Earthquake Early Warning

Earthquake is one of the most common natural disasters on the earth and poses serious risks to human, especially large destructive earthquakes. Currently, no one can accurately predict earthquakes, but earthquake early warning (EEW), which is considered to be a pragmatic and viable way to reduce the damage and casualties during an earthquake, relies on the accurate estimation of displacements and the capability of rapidly detection of the first arrival wave (P-wave) as an earthquake happens. After more than thirty years of exploration and efforts, EEW systems (EEWs) based on the seismic data have been successfully implemented in Japan, Mexico, Taiwan, China, and other nations with varying degrees of sophistication and coverage. Nonetheless such EEWs for large earthquakes still have some deficiencies in the estimation of moment magnitudes (Mw). Earthquake-induced coseismic displacement is the essential information required for the earthquake magnitude. However, due to the rotation, tilt, drift and saturation problem of seismic instruments (strong-motion seismometer and broadband seismometer) and imprecision in the numerical integration process, the integrated displacements are not reliable in real-time. Therefore, the major problem urgent to be solved in EEWs is how to obtain coseismic displacements rapidly and accurately.With the development of the technology, such as high acquisition rate, large data storage capability, etc, high-rate GNSS is proved to be a reliable tool to capture surface displacements including static offsets and dynamic motions in the near field, which does not suffer from the problems as the traditional seismic instrument does. Although high-rate GNSS has advantages in the acquisition of coseismic displacements, there still exist some limitations and challenges when it is applied in real-time earthquake monitoring and early warning. Since GNSS raw measurements are pseudorange, carrier phases and Doppler observations, the real-time high-accuracy data processing approach is needed to obtain displacements and velocities of stations. Other weaknesses of current GNSS measurements are the lower sampling rate and the larger high-frequency noise contribution, and thus the GNSS-derived dynamic motions are not accurate enough to identify P-wave with only millimeter-level amplitude. While seismic sensors are able to sample at very high rates (e.g.200-400 Hz) and perform very well in the high-frequency range as it is much more sensitive to ground motions than GNSS receiver, especially in the vertical direction. The complementary nature of GNSS and seismic sensors for station displacement estimation and P-wave detection are well recognized and the integrated processing of the two dataset is a hot topic in GNSS seismology.This dissertation aims at introducing high-rate GNSS into seismic-based EEWs to solve the magnitude saturation problem and to enhance early warning capability in the face of large earthquakes. Focusing on this core objective, the dissertation studies real-time high-accuracy data processing approach with a single GNSS receiver for coseismic displacement, establishes the combined model applied to a multi-rate Kalman filter of fusing GNSS-derived coseismic displacement with collocated strong-motion data, and realizes the joint application of high-rate GNSS and strong-motion seismometers for EEW. The primary contents and contributions of this dissertation are listed as follows:(1) We systematically investigated two single-receiver approaches to coseismic displacements, the kinematic precise point positioning (PPP) and the precise velocity estimation, including the basic theory, the corresponding model of obtaining coseismic displacements, and the performance in the earthquake events. We did a detailed analysis of the effect of GNSS receiver clock jumps on velocity determination based on carrier phase derived Doppler. The clock jump affected the corresponding relationship between carrier phase measurements and their time tags, which resulted in non-equidistant measurement sampling in time or incorrect time tags. This in turn affected velocity determined by the conventional method which needed equidistant carrier phases to construct the derived Doppler measurements. To overcome the problem of the clock jumps, we devised an improved method of generating Doppler observation taking into account the GNSS receiver clock jumps. Both static and kinematic test results demonstrated that this method was useful to eliminate the impact on velocity determination of GNSS receiver clock jumps.(2) To quickly and accurately capture coseismic displacements with a single GNSS receiver in real time, we proposed a novel approach named the epoch-relative positioning (affiliated with the known coordinates). The proposed approach can overcome the convergence problem of PPP, and also avoid the drift of displacements retrieved from the precise velocity estimation. There are two different strategies for the proposed method. One is the refined variometric (R-variometric) method which calculates the change of position between two adjacent epochs (j, j+1), and then displacements between epoch (j) and epoch (j+n) are obtained by a single integration. The other strategy is the Temporal Point Positioning (TPP) method which directly measures the displacement of the epoch (j+n) with respect to the chosen epoch (j). We discussed the mathematical relationship among the PPP, TPP and R-variometric approaches and verified their equivalence based on two conditions:one is that all the error components in the TPP and R-variometric approaches are carefully considered following the PPP model; the other is that both TPP and R-variometric approaches use accurate known coordinates at the initial epoch (before the earthquake) to eliminate the geometry error. We evaluated the precision of the epoch-relative positioning approach with various error correction schemes and duration time using numerous datasets and demonstrated that few centimeters (horizontal/vertical:3/5 cm) accuracy of coseismic displacements was achievable even for 20 minutes interval after careful corrections of satellite ephemeris, ionospheric delay, tropospheric delay, and geometry errors.(3) We introduced a modified Satellite-specific Epoch-differenced Ionospheric Delay model (MSEID) to compensate for the effect of ionospheric error on single-frequency (SF) receivers, and coseismic displacements of a SF receiver could be accurately calculated using the the epoch-relative positioning in real time. In the MSEID model, the time-differenced ionospheric delays (TDIDs) observed from a regional dual-frequency (DF) GNSS network to a common satellite were fitted to a plane rather than part of a sphere used in the SEID model, and the parameters of this plane were determined by using the coordinates of the stations. Compared with the SEID model, the MSEID model avoided computing the coordinates of IPPs and the coefficients of the model for each satellite at every epoch, making it more efficient for real-time applications. We labored the effectiveness, accuracy and applicability of the proposed approach. There were some conclusions could be drawn from numerous valdation tests:a) the MSEID model was sufficiently precise to allow estimation of TDIDs; b) the distance of the SF station from the surrounding DF stations had a major influence on the accuracy of ionospheric delay compensation using the MSEID method. It was recommended that the average inter-station distance between DF stations was within 80 km when the SF receiver was within the DF receivers-surrounded region, and the distance of SF to DF station was within 50 km when the SF receiver was outside the DF regions; c) the coseismic displacements derived from the proposed method by using SF data were consistent with the results using DF data to 2 cm in horizontal and 4 cm in vertical, which made use of SF receivers feasible to increase the density of the existing high-rate DF GNSS networks which can facilitate timely earthquake early warning and rapid response.(4) We comparatively analyzed the pros and cons of strong-motion seismometers and real-time high-rate GNSS in attaining coseismic displacements. To avoid the defects of using single instrument, we put forward a combined model of fusing GNSS and strong-motion records taking into consideration the corrections of baseline errors, and then applied a multi-rate Kalman filter to analyze the collocated GNSS and strong-motion data. The experimental results showed that a more accurate and reliable broadband displacements could be obtained, which provided the full spectrum of the seismic motion. The broadband displacements had more characteristics of strong-motion results in the early stage of the filter, which could be used to detect the weak movement of stations, and in the late stage of the filer, tightly bound by the GNSS-derived displacements, the broadband displacements did not diverge, and thus accurately recorded the long-period information of seismic waves.(5) We investigated the joint use of the collocated GNSS and strong-motion data for earthquake early warning, and analyzed the methods of the P-wave detection, epiceter determination and magnitude estimation, along with their corresponding effects based on the combined results. The performance was validated based on the data collected during the 2011 Tohoku-Oki earthquake, and here came some conclusion:a) based on the broadband displacements, the arrival time of P-wave could be picked up accurately with a mean value 0.8 s offset by using the STA/LTA and AIC algorithms; b) when P-wave was detected at four near-field GPS/strong-motion seismometer pairs, the epicenter could be determined by using the Geiger method (a least squares method) and the grid search method, and the biases of the estimated epicenter from the reference value determined by USGS were 17 km for the Geiger method and 4 km for the grid search method; c) the moment magnitude was estimated through the empirical scaling relationship between magnitude and the broadband results of either P wave amplitude (Pd) or peak ground displacement (PGD). The first alert information was broadcast at 26.35s after the earthquake onset with the corresponding magnitude Mw 7.4, and the third alert information was set at 102.19 s after the earthquake initiation with the final magnitude Mw 8.9, which was close to the real earthquake magnitude Mw 9.0. These results proved that the combination of GNSS and strong-motion seismometers can be mutually beneficial in both sensitivity and large dynamic displacements, and the broadband displacements were sufficiently accurate to be used for the P-wave detection and the rapid and robust estimation of magnitude without saturation problem, which enhanced early warning capability of the present EEWs in the face of large earthquakes.