Search

Peak ground acceleration (PGA) can be used to estimate the seismic intensity. However, using P-wave features to estimate PGA is a challenging task. One of the reasons for that is that a seismic wave commonly undergoes modification due to various site effects, consequently leading to uncertainty in the predicted PGA. In order to accommodate site effects using site parameters together with P-wave parameters, this paper takes advantage of machine learning to consider multiple parameters simultaneously. Several artificial neural network (ANN) models considering different site effect parameters are constructed. The performances of these ANN models were investigated and compared. In total, 53531 ground motion data obtained from the Taiwan Strong Motion Instrumentation Program were utilized to develop the proposed approach. It was found that the proposed ANN model with horizontal-to-vertical spectral ratio parameters effectively reduces the error of the estimated PGA when compared with either the ANN model without site parameters or the ANN model with other site parameters.

The strong ground motion of 6 February 2018 M 6.2 Hualien Earthquake triggered a series of co-seismic geomagnetic fluctuations and seismo-traveling atmospheric disturbances (STADs) signatures in infrasonic waves and micro-pressures upon the seismic wave arrival. Networks of 9 QuakeFinder systems, 3 infrasound systems, 2 tiltmeters, 2 micro-barometers, and 11 co-located seismometers are used in this study. Each QuakeFinder system consists of a 3-axes induction magnetometer, an air conductivity sensor, a geophone, and temperature/relative humidity sensors. Co-seismic signatures clearly appear in the induction magnetometers, infrasound systems, and micro-barometers data. The magnetometers register both high- and lowfrequency pulsations. Geomagnetic fluctuations occur upon the seismic wave arrival but last a longer duration, while the STADs lag their co-located seismic waves by about 15 - 45 s. The long-lasting fluctuations recorded by both induction and fluxgate magnetometers suggest that the ground/underground water motion play an important role, which is further conformed by low-frequency fluctuations in the tiltmeter data. In general, the amplitude of geomagnetic fluctuations decays as away from the epicenter. However, unusual large co-seismic geomagnetic fluctuations are detected over areas of the abnormal seismic intensity level and/or the magnetic underground structure with anomalously high susceptibilities.

This study used the results from 45 microtremor array measurements to construct a shallow shear wave velocity structure in the western plain of Taiwan. We constructed a complete 3D velocity model based on shallow and tomography models for our numerical simulation. There are three major subsurfaces, engineering bedrock (V_S = 600 m s^(-1)), Pliocene formation and Miocene formation, constituted in the shallow model. The constant velocity is given in each subsurface. We employed a 3D-FD (finite-differences) method to simulate seismic wave propagation in the western plain. The aim of this study was to perform a quantitative comparison of site amplifications and durations obtained from empirical data and numerical modelling in order to obtain the shallow substructure soil response. Modelling clearly revealed that the shallow substructure plays an important role in strong ground motion prediction using 3D simulation. The results show significant improvements in effective shaking duration and the peak ground velocity (PGV) distribution in terms of the accuracy achieved by our developed model. We recommend a high-resolution shallow substructure as an essential component in future seismic hazard analyses.

It is seen that strong motion generation area plays an important role in the shaping of strong motion records at the observation point. Strong motion generation areas identified within the rupture plane of the 27 March 2013 Nantou, Taiwan, earthquake (Mw = 5.9) have been modelled in this work. It is seen that all available records are at the surface which include site amplification terms. The modified semi-empirical technique effectively simulates records at rock site. The site amplification terms in all records have been removed using SHAKE 91 program and velocity input at each site. The observed records corrected for site amplification terms are further used for comparison with simulated record at the bedrock from several models. Once the observed records at soil sites are transferred at the bedrock, the next task is selection of final model that gives best fit records. Peak ground acceleration from simulated record at four sites is compared with that from corrected observed data. Since the semi-empirical technique of simulation is strongly dependent on various modelling parameters such as dip, strike, rake, rupture velocity and starting points of rupture, these parameters change iteratively in a specified range in a heuristic way to obtain best modelling parameters. The model giving minimum root mean square error (RMSE) is retained at final model. It is seen that minimum root mean square error of the wave form comparison has been obtained at four stations for the source model having single strong motion generation area. Strong motion records have been simulated at four different recording stations. Comparison of observed and simulated records has been made in terms of RMSE between simulated and observed acceleration records, velocity records and the response spectra at each of four stations. Comparison of waveforms and parameters extracted from observed and simulated records confirms the efficacy of the modified technique to model earthquake characterized by SMGAs.

One decade after the 1999 Chi-Chi earthquake, central Taiwan experienced more strong ground shaking [Central Weather Bureau (CWB), intensity VII] induced by a ML 6.2 earthquake on 5th November 2009. This earthquake occurred in the Mingjian Township of Nantou County, only 12 km southwest of the Chi-Chi earthquake epicenter. The broadband microearthquake monitoring network operated by the National Center for Research on Earthquake Engineering (NCREE) observed numerous aftershocks in the five days following the mainshock. The relocated aftershocks and the mainshock focal mechanism indicated a NE-SW striking fault dipping 60¢X toward the northwest. This fault plane is inside the pre-Miocene basement and the rupture extends from the lower crust to 10 km depth just beneath the basal décollement of the thin-skinned model that is generally used to explain the regional tectonics in Taiwan. The fault plane is vertically symmetrical with the Chelungpu fault by the basal décollement. The NW-SE compressive stress of plate collision in Taiwan, as well as the deep tectonic background, resulted in the seismogenic structure of the Mingjian earthquake at this location.

The western coastal plain of Taiwan borders the Western Foot hills, which is a fold-and-thrust zone and one of Taiwan's major seismic zones. Earth quakes from the Western Foot hills are a frequent cause of disaster in the western coastal plain. In this study, array measurements of microtremors were conducted at seven sites through out the western coastal plain. At each site, four arrays (S, M, L, and XL arrays) of different sizes were considered. The largest radius of each array ranged from 32 m to 1 km. The Maximum Likelihood Method of Frequency-Wavenumber (F-K) analyses was used to obtain phase velocity dispersion curves. After the inversion of the dispersion curves with the initial models decided by Genetic Algorithm (GA) searching, shallow S-wave velocity structures for the western coastal plain were estimated.

In the Tatun mountain area of northern Taiwan are two strong motion stations approximately 2.5 km apart, TAP056 and TAP066 of the TSMIP network. The accelerometer at station TAP056 is often triggered by earthquakes, but that at TAP066 station is not. Comparisons of vertical and horizontal peak ground accelerations reveal PGA in the vertical, east-west, and north-south components at TAP056 station to be 3.89, 7.57, and 5.45 times those at station TAP066, respectively. The PGA ratio does not seem to be related to earthquake source or path. Fourier spectra of earthquake records at station TAP056 always have approximately the same dominant frequency; however, those at station TAP066 are different due to different sources and paths of different events. This shows that spectra at TAP056 station are mainly controlled by local site effects. The spectral ratios of TAP056/TAP066 show the S-wave is amplified at around 8 ~ 10 Hz. The horizontal/vertical spectral ratios of station TAP056 also show a dominant frequency at about 6 and 8 ~ 10 Hz. After dense microtremor surveying and the addition of one accelerometer just 20 meters away from the original observation station, we can confirm that the top soft soil layer upon which the observation station is constructed generates the local site response at station TAP056.

The horizontal-to-vertical spectral ratio has become increasingly popular in studies of site effect and determination of the predominant period of a site. In this study, this method is extended to identify nonlinear soil responses. To establish this phenomenon, borehole array records that already showed nonlinear site responses using spectral ratios between surface and borehole station pairs were analyzed. Moreover, in this study, the horizontal-to-vertical spectral ratio method was applied to weak and strong motion records from the same dataset. The results demonstrate that nonlinear site responses can be evaluated using horizontal-to-vertical spectral ratios of surface recordings at a single station.