The horizontal elastic response spectrum of earthquake ground shaking recorded at TMDA recorded from all six considered events in Table 2 from 2007 to 2016 are displayed in Fig. 8. The thick black lines show the pseudo-spectral acceleration of TMDA (surface) while the thick black dash lines show horizontal elastic response spectra of TMDB (borehole) for the RotD50 component of horizontal SA directions. In addition, the grey lines show spectrum at SRDT and PRAC stations. It could be clearly observed that the spectrum in Bangkok is much larger than those from outside basin stations located at a similar distance for most structural periods. Furthermore, the horizontal spectra of TMDA show high energy at long periods observed between 0.5 and 2 s, similar to those reported using microtremor observations by Bidhya et al. (2021). It is interesting to note also that the amplification at spectral ordinate between 4 and 6 s could be observed from TMDB records from both 2008 and 2009 events. However, the spectral amplification at other periods, between 0.5 and 0.7 s, observed from all considered events in TMDA (surface) are missing from TMDB (borehole) spectrums. These results indicate that the top 47-m surface layer plays a key role in determining the pattern of the observed spectrum at short structural periods.

In order to investigate the long period energy content within the basin, the calculated horizontal spectrum of the TMDA station using different parts of accelerogram are also considered. The thick dotted lines showing horizontal spectrums for selected window contains only the body waves (without surface wave) which do not show any peak for spectral ordinates greater than 2 s. This evidence shows that the long period energy content is related to the arrival of surface waves observed within the basin. Figure 8 also compares the estimated median horizontal spectrum computed from Boore et al. (2014) and Zhao et al. (2006) equations for shallow and intra slab earthquakes. The effect of basin amplification terms in BSSA14 tends to provide greater median values for moderate to long spectral ordinates (0.5 to 2 s) than the observed horizontal spectrums at TMDA station. However, the Zhao et al. (2006) equation with only soil amplification through Vs₃₀ term seem to underestimate most of long structural periods (T > 1 s). It is interesting to mention also that the peak spectral ordinates related to surface waves seem to increase with increasing earthquake magnitude. The highest SA has been observed at a period around 5 s from the 2009 event with observed spectral ordinate around 6 × 10⁻³ g. In addition, the characteristic of the vertical spectrum (black line) is limited since there is low spectrum ordinates greater than 1 s.

Based on previous analysis, it is become clear that long period ground motion (T > 1 s) is due to the basin effect rather than source effect (Tuladhar et al., 2004; Poovarodom and Plalinyot 2013; Jirasakjamroonsri et al. 2018; and Subedi et al., 2021). To assess the time dependence and its effect on the spectrum amplitude for Bangkok basin stations, spectrograms are determined with a moving window of 10 s and 5-s overlapping time window. Additional file 1: Fig. S2 shows the spectral analysis for the 2019 Mw 6.2 event recorded at BKSI station. However, a similar pattern could also be observed from other Bangkok basin stations. After the arrival of the surface wave, the spectrogram shows that the low frequency content (from 1 to 0.3 Hz) dominates the spectral content for 3 and 4 min. These phases could be seen for all three directions in ground motion records. In addition, lower frequency phases (between 0.1 and 0.2 Hz) continue to dominate for the rest of the duration, especially between 3 and 6 min.

To gain better understanding regarding the characteristics of long period ground motion recorded inside the Bangkok basin stations (i.e. TMDA, SIRA, PWSA, PWNA, KMUT, and PTNA), in the current study, we analyst ground motion records from these 8 events and compute horizontal/vertical spectral ratios (HVSR) for seismic stations located inside and outside the Bangkok basin. Following original work by Nakamura (1989) and Mase and Sugianto (2021) (using microtremor data) and Lermo and Chavez-Garcia (1993) (using recorded ground motion), the HVSR analysis has been implemented extensively to determine basin-induced amplification, using both microtremor data and ground motion recordings. Following the SESAME guidelines (Bard 2005), only mean ± one standard deviation HVSR peaks and the standard deviation of the frequency of the HVSR peak, is derived for TMDA station. The HVSRs of Fourier amplitude spectra are developed for the current work.

The HVSR of the recorded ground motion uses the geometric-mean of the horizontal components of the north–south and east–west components and the selected 10 min signal long with signal to noise ratio greater than 5 by correcting any obvious noise from the recorded ground motion. The H/V spectrum results are then given smoothing again with the Konno-Ohmachi algorithm with a smoothing constant of 20, and a window sample of 40 percents. A different smoothing constant of 40 was also investigated, but we did not see a large deviation of HVSR results. The HVSR of these recorded motions in Table 2 are shown in Fig. 9. Though the peak amplitudes of these HVSR curves at TMDA station vary from 7 to 11, the mean HVSR and its standard deviations (± s) seems to have been typically high within the period between 5.1 and 5.5 s (corresponding to 0.19–0.18 Hz, respectively) with smaller peaks between 0.5 and 2 s (corresponding to 2 and 0.5 Hz, respectively). A double peak HVSR spectrum exhibiting two different peaks would normally indicate that there are two high impedance contrasts below the station at two different levels: one for a dense layer and another for a narrow layer. Similarly, double-peak mean HVSR curves could also be observed for other Bangkok basin stations. However, due to the small number of records, the data is still insufficient to draw a reliable dominant site period for each station. Nevertheless, we noticed that for stations located within the central part of Bangkok (i.e. SIRA, PWSA, and PWNA), which have the deepest basement depth, about 600–850 m, the pre-dominant periods vary between 6.5 and 7.5 s. Further studies should investigate the soil amplification effect of these long-period predominant peaks with reliable low-frequency seismometers through large-array observations. The comparison between peak periods of the HVSR curves and those recorded elastic spectra are consistent. In contrast, at PRAC and SRDT stations, HVSR curves are close to 1, indicating very low amplification, Additional file 1: Fig. S3. This conforms to the previous soil profile information of these stations situated at the stiff soil or soft rock sites.

Based on previous analysis, it was become clear that long period ground motion between 0.1 and 0.3 Hz is due to the basin effect rather than source effect. However, it is still not clear what the cause of this mechanism is. In order to analyze the long period characteristics, velocity ground motion at TMDB station was band-pass filtered in the range of 0.1–0.3 Hz using a second order zero-phase Butterworth filter for the Mw 7.9 12 May 2008 event, Fig. 10. The time sequence of the particle motion (Hodogram plot) is illustrated in the lower half of the figure, in which the x- and y-axes of the graph show a series of plan views (EW and NS, respectively). The top trace of the velocity seismograms, east–west (EW) or almost transverse-component, indicates clearly dispersed wave train around the travel time 530–570 s, where the amplitude is the largest in the transverse direction. The particle motion of the dispersed waves indicates that these waves are almost transversely polarized in the horizontal plane around the above-mentioned travel time and hence these prominent phases are most probably Love waves. However, detailed inspection suggests a small deviation from purely polarized motion in the SH-direction. This deviation from the linear polarization is more remarkable after the travel time of around 570 s; the Rayleigh-type ground motion and other phases such as scattered waves may be coming after this time. It is worth mentioning also that a similar pattern of particle motion could be observed for recorded ground motion both at the surface and borehole (Figs. 5 and 10a) indicating long period energy developed through the deep soil profile.

As seen in Fig. 10, it is obvious that the dominated low-frequency (between 0.1 and 0.3 Hz) ground motion in the Bangkok basin is affected by the locally generated surface waves. These observations, in conjunction with observations reported in other sedimentary basins (Pacor et al. 2007; Yoshimoto and Takemura 2014; Tsai et al. 2017) show that the behavior of low-frequency ground motions, amplified by the basin-induced fundamental Love waves, is governed largely by the deep alluvial deposits. Predominant frequencies of low-frequency ground motion in the Bangkok basin show the tendency to decrease with the depth of bedrock, whereas it is nearly similar (approximately between 0.3 and 0.18 Hz) in the shallower part of the basin. With the possibility of great earthquakes at a much closer distance from the Three Pagodas Fault (M > 6.5 at 100 km), Sagiang Fault (M > 7.9 at 500 km), the significant and as yet still unquantified long period ground motion in the Bangkok Basin requires further investigation through maintain broadband and strong motion networks for better quantitative understanding how much long period ground motions could able to amplifiedyIn addition, a large aperture array with reliable low frequency (less than 0.3 Hz) seismometers should be deployed with longer collecting time periods than used in the current studies in order to clarify the long period behavior of the deep alluvial basin. In addition, further ground motion modelling should also take into account the effect of surface waves in Bangkok basin. Since there is a clear presence of surface waves both in the ground surface and borehole and the complicated S-wave amplification, a simple 1D ground response analysis might not be enough to model both observed Love and Rayleigh waves. Further studies considering the 2D/3D basin structure are necessary but are beyond the scope of this preliminary work.