Seismic velocity structure of the Jakarta Basin, Indonesia ...

Geophys. J. Int. (2018) 215, 431–449 doi: 10.1093/gji/ggy289Advance Access publication 2018 July 17GJI Seismology

Seismic velocity structure of the Jakarta Basin, Indonesia, usingtrans-dimensional Bayesian inversion of horizontal-to-verticalspectral ratios

A. Cipta,1,2 P. Cummins,1,3 J. Dettmer,4 E. Saygin,5 M. Irsyam,6 A. Rudyanto7 andJ. Murjaya7

1Research School of Earth Sciences, Australian National University, Canberra, ACT 0200, Australia. E-mail: [email protected] Agency of Indonesia, Bandung, Jawa Barat 40122, Indonesia3Geoscience Australia, Canberra, ACT 2601, Australia4Department of Geoscience, University of Calgary, Calgary, AB T2N 1N4, Canada5The Commonwealth Scientific and Industrial Research Organisation (CSIRO), University of Western Australia, Canberra, ACT 2601, Australia6Faculty of Civil and Environment Engineering, Bandung Institute of Technology, Bandung 40132, Indonesia7Indonesian Agency for Meteorology Geophysics and Climatology, DKI Jakarta 10720, Indonesia

Accepted 2018 July 15. Received 2018 May 9; in original form 2017 November 29

S U M M A R YCharacterizing the interior structure of the Jakarta Basin, Indonesia, is important for the im-provement of seismic hazard assessment there. A dense-portable seismic broad-band network,comprising 96 stations, has been operated between October 2013 and February 2014 coveringthe city of Jakarta. The seismic network sampled broad-band seismic noise mostly originatingfrom ocean waves and anthropogenic activity. We used horizontal-to-vertical spectral ratio(HVSR) measurements of the ambient seismic noise to estimate fundamental-mode Rayleighwave ellipticity curves, which were used to infer the seismic velocity structure of the JakartaBasin. By mapping and modelling the spatial variation of low-frequency (0.124–0.249 Hz)HVSR peaks, this study reveals variations in the depth to the Miocene basement. These vari-ations include a sudden change of basement depth from 500 to 1000 m along N–S profilethrough the centre of the city, with an otherwise gentle increase in basin depth from southto north. Higher frequency (2–4 Hz) HVSR peaks appear to reflect complicated structure inthe top 100 m of the soil profile, possibly related to the sediment compaction and transitionsamong different sedimentary sequences. In order to map these velocity profiles of unknowncomplexity, we employ a trans-dimensional Bayesian framework for the inversion of HVSRcurves for 1-D profiles of velocity and density beneath each station. Results show that verylow-velocity sediments (<240 m s−1) up to 100 m in depth cover the city in the northern tocentral part, where alluvial fan material is deposited. These low seismic velocities and thevery thick sediments in the Jakarta Basin will potentially contribute to seismic amplificationand basin resonance, especially during giant megathrust earthquakes or large earthquakes withepicentres close to Jakarta. Results have shown good correlation with previous ambient seismicnoise tomography and microtremor studies. We use the 1-D profiles to create a pseudo-3-Dmodel of the basin structure which can be used for earthquake hazard analyses of Jakarta, amegacity in which highly variable construction practices may give rise to high vulnerability.The methodology discussed can be applied to any other populated city situated in a thicksedimentary basin.

Key words: Seismic noise; surface waves; site effects.

1 I N T RO D U C T I O N

This study aims to develop a model for the basin structure of Jakarta,the capital city of Indonesia. Jakarta is one of the world’s megacities,

with at least 10 million inhabitants in Jakarta itself and over 30million people living in the greater Jakarta area. Jakarta’s averagepopulation density of about 14 000 km−2 indicates it is an areaof high potential risk, and therefore it is important to carefully

C© The Author(s) 2018. Published by Oxford University Press on behalf of The Royal Astronomical Society. 431

Dow

nloaded from https://academ

ic.oup.com/gji/article/215/1/431/5055145 by guest on 31 M

ay 2022

mailto:[email protected]

432 A. Cipta et al.

assess its seismic hazard. In terms of geotectonic and demographicconditions, Jakarta is similar to Mexico City, with both cities sittingon deep sedimentary basins and 250–350 km distant from the Javaand Mexican subduction zones, respectively. Therefore, it seemsreasonable to expect that to first order the cities might experiencesimilar impacts from a major subduction zone earthquake.

The 1985 Mw 8.1 Michoacan earthquake occurred at the Mexicansubduction zone more than 300 km away from Mexico City. Despitethe relatively large distance, this event resulted in about 10 000deaths and extensive damage of more than 2000 structures (Esteva1988; Padilla y Sanchez 1989). Poorly consolidated lake sedimentsand the geometry of the Mexico City Valley acted to amplify andincrease the duration of shaking, the combined effect of which hada critical influence on the extent of damage (Kawase & Aki 1989;Padilla y Sanchez 1989; Furumura & Kennett 1998).

Previous studies in the basins of southern California (Magistraleet al. 2000) and the Lower Rhine Embayment (Ewald et al. 2006)showed the importance of 3D basin geometry in modeling the char-acteristic of seismic waves propagation in basins. Data from 3Dsimulations of wave propagation showed the importance of wavefront focusing due to low velocity zones or edge-diffracted waves(Ewald et al 2006), basin resonance (Castellaro et al. 2014) andlocal-scale multi-scattering and prolonged ground motion (Olsen etal. 2006; Furumura & Hayakawa 2007; Denolle et al. 2014; Cruz-Atienza et al. 2006; Viens et al. 2016). On the other hand, Roten etal. (2014) found that non-cohessive shallow low-velocity sedimentsof Los Angeles Basin may reduce seismic shaking due to earthquakerupturing the San Andreas Fault. Accordingly, a rigorous quantifi-cation of seismic properties and geometry of the Jakarta Basin is aprerequisite to generating a robust seismic hazard map for the city.

At any given site, the dynamic soil properties and basin geometryare the principal components that account for site amplification andprolonged duration of ground motion. Understanding the seismicshear velocity profile is important for reliable forecasts of groundresponse for earthquake scenarios and seismic hazard evaluation.In Europe, America and Japan, models for sedimentary basins areavailable such as the Mygdonian Basin in Greece (Manakou et al.2010), the Lower Rhine Embayment in Germany (Ewald et al.2006), the Santiago de Chile Basin in Chile (Pilz et al. 2010), theOsaka Basin in Japan (Kagawa et al. 2004; Iwaki & Iwata 2011),the Tagus Basin in Portugal (Borges et al. 2016), the Mexico ValleyBasin in Central Mexico (Cruz-Atienza et al. 2016) and the Po Plainin Northern Italy (Berbellini et al. 2017). For the case of the JakartaBasin, however, very few geophysical studies are available (Ridwan2016; Ridwan et al. 2016; Saygin et al. 2016, 2017). Furthermore,only a few measured boreholes have reached the geological bedrockdeemed to be the Pliocene–Pleistocene boundary (Delinom et al.2009).

Jakarta’s sedimentary basin is undergoing rapid subsidence, witha maximum rate of about 26 cm yr−1, mainly due to groundwaterextraction (Abidin et al. 2011; Ng et al. 2012). From this fact aloneit can be deduced that Jakarta Basin is filled with highly saturatedpoorly consolidated sediments which can accommodate a huge vol-ume of groundwater. This suggests that, in addition to increasing theintensity and duration of seismic wave motion, Jakarta Basin mayalso be prone to liquefaction if it were to experience earthquake-generated strong ground motion.

Hence, in order to ascertain the extent to which future earth-quakes may affect the megacity of Jakarta, it is vital to develop arobust model for the geometry and physical properties of the JakartaBasin. Since the basin is a densely populated urban area, passiveseismic methods are the most viable approach to determining basin

structure. Among passive seismic methods, spatial autocorrelation,known as SPAC (Aki 1957; Okada 2003; Asten 2006), ambient seis-mic noise tomography (ANT; Shapiro & Campillo 2004; Shapiroet al. 2005) and horizontal-to-vertical spectral ratio (HVSR, see be-low) are the most common techniques applied in urban areas. WhileSPAC and ANT utilize interstation cross-correlation of ambientseismic noise as estimates of surface wave dispersion curves, HVSRmakes use of single-station HVSR as estimates of the Rayleigh waveellipticity. In this study, a trans-dimensional (trans-D) Bayesian ap-proach is applied for the ellipticity curve inversion to obtain 1-Dvelocity profiles at each station. Subsequently, a 3-D basin model isconstructed by interpolating these 1-D profiles.

2 G E O T E C T O N I C S E T T I N G A N DH I S T O R I C A L E A RT H Q UA K E S

Two major tectonic components comprise the Indonesianarchipelago: the Sundaland or Sunda block in the west and mi-croblocks in the east. The former is composed of Java, Sumatra,Kalimantan, Sulawesi and Nusa Tenggara Islands. Jakarta is lo-cated on northwest Java as highlighted in Fig. 1. Estimates of thenorthward motion of Australia with respect to Java range from67 mm yr−1 (Simons et al. 1997) to 70 mm yr−1 (Hall 2009), withdirection almost normal to the Java Trench. This megathrust plateboundary located approximately 250 km south of Jakarta and thesubducting slab beneath Java pose a high seismic hazard that mayseriously affect the city when large earthquakes occur. In addition,crustal faults including the Cimandiri, Lembang and Baribis faults,located within the vicinity of Jakarta, also contribute to the high-hazard level in the city.

Using borehole and geohydrological data, Lubis et al. (2008)described Jakarta Basin as separated from Depok in the south bynormal faults and underlain by geological bedrock composed ofTertiary rock. The depth of the bedrock near Depok was estimated at50 m, undergoing an abrupt change 3 km from the basin’s southernrim, up to a depth of 250 m. The basin depth gradually deepens fromthat point to central Jakarta to a depth of at least 350 m, and the basinfloor was thought to flatten from central Jakarta to Jakarta Bay in thenorth. However, a recent ANT study suggested a basin depth of morethan 1000 m (Saygin et al. 2016) while a microtremor array studyestimated a maximum depth to engineering bedrock (where shearwave velocity VS ≥ 900 m s−1) of 725 m (Ridwan 2016; Ridwanet al. 2016).

The surface geology of Jakarta has Quaternary sedimentary unitslike alluvium, alluvial fan and beach ridge deposits (Turkandi et al.1992). Alluvium is exposed from the coastline to 6 km southwardsin the city centre, which covers 35 per cent of Jakarta. Further south-ward, the surface geology is dominated by alluvial fan deposits witha patch of Tertiary volcanic material deposited in the west. The al-luvial fan sediments are deposited from elevated areas in the southof Jakarta, piling up at the centre with up to 300 m thickness. ThesePlio–Pleistocene sediments, mostly composed of clayey-sand, un-conformably overlie the Miocene formations which are croppingout and encircling the basin in the west, south and east. These up-lifted Tertiary formations are known as the Tangerang High in thewest, Depok High in the south and Rengasdengklok High in the east(Delinom 2008).

Throughout Jakarta’s history, at least three large earthquakes havedevastated the city. The 1699 earthquake caused 28 casualties (Reid2012) and triggered a number of landslides within the vicinity ofBogor (Nata & Witsen 1700), a city located about 50 km south of

Dow



ay 2022

Seismic velocity structure of the Jakarta Basin 433

Figure 1. (a) Simplified tectonic setting of the Indonesian region and (b) western Java, detail of inset area indicated in (a). The subduction boundary is shownas toothed curve while the study area is the orange shaded area in (b). Motion of the Australian and Eurasian plates are indicated by back arrows, with relativemovement at the Java Trench resolving to nearly normal convergence at a rate of 7 cm yr−1. Major faults are indicated by blue lines, while black toothed,red and green dashed lines denoted subduction, microcontinent boundaries and Benioff counters, respectively. The blue dashed line is the continuation of theBaribis fault to the west and east as described in Simandjuntak & Barber (1996).

Jakarta. The largest earthquake to impact the city had a magnitudeof Mw 8.5 and occurred on 1780 January 22 (Albini et al. 2013).Ground shaking caused 27 sheds and houses to collapse in Zandseeand Moorish gracht (canal), located in present-day Central Jakartawhere Jakarta Cultural Centre is now standing (Wichmann 1918translated by Haris & Major 2016). Half a century later, on 1834October 10, an Mw 7.0 earthquake was associated with seismic in-tensity considered to be the highest to strike the region (Musson2012). With 30 million people inhabiting Greater Jakarta, the fatal-ity count could be very high should any of these historical eventsre-occur today (Nguyen et al. 2015).

3 H V S R M E A S U R E M E N T S

Microtremor survey methods to evaluate the shear wave velocityprofile are gaining in popularity because of their applicability inurban areas. These methods are non-invasive, utilizing continuousenergy produced by both natural phenomena and human activity.Kanai (1957) observed that the amplitude–frequency relation ofearthquake motion exhibited peaks whose period depended on ‘eachkind of ground’—short period (0.1–0.4 s) on hard ground and longerperiod (0.4–0.8 s) on softer ground—and noted that the same peakperiods were observed in ambient seismic noise (microtremors). Aki(1957), using microtremors recorded at night (6–10 p.m.), also ob-served that dispersion was site-dependent, and appeared to be char-acteristic of the underlying medium. Although the HVSR methodwas first applied for seismic microzonation by Nogoshi & Igarashi(1971), Nakamura (1989) is more widely cited for his promotionof the method and for proposing that the fundamental resonancefrequency be defined as the peak of the HVSR curve, defined as

HVSR(ω) =√

HEW(ω) × HNS(ω)

V (ω), (1)

where HEW(ω) and HNS(ω) denote Fourier amplitude spectra ineast–west and north–south directions, respectively, while V(ω) isthe Fourier amplitude spectrum for the vertical component. The

two horizontal components are combined using the geometric av-erage, then divided by V(ω) to obtain the measured HVSR curve(Harutoonian et al. 2013).

Since this early work, the HVSR technique has been widelyused to characterize site response in seismic microzonation studies.The method as proposed by Nakamura (1989) equated both theHVSR peak amplitude and period with those of the S-wave transferfunction. Subsequent studies found that, while the period of theHVSR peak coincides with the resonant period of S waves in thesediment column, the amplitude of the peak often does not matchthat of the S-wave transfer function (Lermo & Chavez-Garcia 1993;Lachet & Bard 1994; Bonilla et al. 1997; Lunedei & Albarello2010). Several studies have used HVSR to measure this resonantperiod in the range 0.1–2.0 s to infer site class (Zhao et al. 2006;Ghofrani & Atkinson 2014) and to explain the spatial patterns ofearthquake damage (Gosar 2010). HVSR measurements of resonantperiod have also been used to infer depth of sediments when velocityis known from, for example, surface wave dispersion (Ibst-von Seht& Wohlenberg 1999; D’Amico et al. 2008) or average velocity whendepth is known from geophysical surveys (Bodin et al. 2001).

Other studies have confirmed the original result of Nogoshi &Igarashi (1971) that, although the peak in the HVSR curve coincideswith the S-wave resonant period, the HVSR curve itself is closelyrelated to the ellipticity of Rayleigh waves (Konno & Ohmachi 1998;Arai & Tokimatsu 2000). This is true not only when the ambientseismic noise was confirmed to be dominated by Rayleigh waves(Scherbaum et al. 2003) but also in numerical studies that used fullwavefield modelling of the ambient seismic noise (Field & Jacob1993; Lachet & Bard 1994; Lunedei & Albarello 2010). This hasled to a number of studies that used the relationship between HVSRand Rayleigh wave ellipticity to invert for S-wave velocity profiles(Fah et al. 2003; Scherbaum et al. 2003; Arai & Tokimatsu 2004;Parolai et al. 2005), and this is the approach we adopt in this study.

The Jakarta Basin is thought to have a relatively long resonantperiod, with the results of (Saygin et al. 2016) indicating an av-erage S-wave velocity of 500 m s−1 extending to an average depthof 500 m. This suggests an S-wave resonant period of about 4 ×

Dow



ay 2022

434 A. Cipta et al.

h/VS = 4 s, with even longer periods possible in the deepest partof the basin beneath northern Jakarta. Although most of the afore-mentioned studies have involved S-wave resonant peak periods inthe range of 0.1–1.0 s, typically associated with S-wave velocitystructure at 100 m depth or less, a few involved situations similarto the Jakarta Basin. Yamanaka et al. (1994) used HVSR measure-ments with peak periods of 2–8 s to infer basin depths in the KantoPlain of 1550 m. Bodin et al. (2001) used HVSR measurementswith peaks in the range 2–5 s to infer average basin velocities of600–1000 m s−1 in the Mississippi Embayment, for which previousstudies had established a maximum depth of about 1000 m.

In order to make HVSR measurements covering long resonantperiods, passive seismic measurements have been carried out us-ing three-component Trillium Compact sensors, with sensitivity tovelocity flat in the frequency range 0.05–100 Hz. These sensorsrecorded background noise at 96 sites distributed over the JakartaBasin (Fig. 2a). Ambient seismic noise was recorded for at least1 month at each site and processed according to the details givenbelow.

3.1 Assumptions

Foremost among our assumptions is that the HVSR curves we es-timate are determined by the ellipticity of the fundamental modeRayleigh wave. It is important to note that in general the ambi-ent seismic noise wavefield consists of body waves (P and S) andsurface waves (Rayleigh and Love; Bonnefoy-Claudet et al. 2006).Indeed, the original explanation of the HVSR method as proposed byNakamura (1989) was that it is determined solely by the SH-waveresonance. However, subsequent studies showed that the HVSRcurve is closely linked with Rayleigh wave ellipticity (Lachet &Bard 1994; Lermo & Chavez-Garcia 1994; Fah et al. 2001, 2003).Failing to account for other, non-Rayleigh wave components of theambient seismic noise can lead to an overestimation of the Rayleighwave ellipticity (Poggi et al. 2012; Hobiger et al. 2013). Followingthe conclusions of the comprehensive review of the nature of theseismic ambient seismic noise wavefield by Bonnefoy-Claudet et al.(2006), our assumption that the ambient seismic noise wavefield isdominated by the fundamental mode Rayleigh wave is supported bythe proximity of the ocean, both the Java Sea to the north and theIndian Ocean to the south. This suggests that the dominant sourcesof seismic noise will be associated with Rayleigh waves excitedby ocean swell (Longuet-Higgins 1950; Yamanaka et al. 1993). Thelong period, 5–7 s, of the main peak in our HSVR curves (see below),and even the higher frequency peaks at 2–4 Hz are likely associatedwith this natural source of Rayleigh wave energy, especially sincewe used measurements made at night and verified their stability.

Our assumption that the fundamental mode is dominant is sup-ported by the ANT study of Saygin et al. (2016), who used this samedata set and noted that possible higher order Rayleigh wave modesarriving before the dominant fundamental mode had a much lowersignal-to-noise ratio. While several studies that use HVSR to invertfor VS profiles have obtained satisfactory results using only the fun-damental mode Rayleigh wave (Yamanaka et al. 1994; Konno &Ohmachi 1998; Scherbaum et al. 2003), other studies have foundthat higher modes can make an important contribution to the HVSRcurve, particularly when low-velocity zones are present (Arai &Tokimatsu 2004; Parolai et al. 2005; Savage et al. 2013; Rivet et al.2015). The emergent higher mode Rayleigh waves are usually re-lated to peaks above the S-wave resonance frequency, hence the

secondary and later peaks should be treated carefully because theymay be contaminated with higher mode energy (Asten et al. 2004)

Finally, we note that we assume the velocity structure beneaththe station can be approximated by a set of horizontal layers, eachuniform in seismic velocities and density. Previous studies (Rid-wan et al. 2016; Saygin et al. 2016, 2017) assumed a locally 1-Dapproximation and suggested that the seismic velocity structure inthe Jakarta Basin is slowly varying, except for a possible sharpincrease in depth from about 300–500 m in central Jakarta alongthe south–north profile (see fig. 13.1 of Saygin et al. 2016). Theinfluence of 3-D structure on HVSR curves has been studied byUebayashi et al. (2012a,b) for Osaka Basin and Gueguen (2007)for the Grenoble Basin, who have noted that the 3-D structure canlead to peaks in HVSR curves that are broader than those predictedfrom the 1-D velocity profile beneath the station, and can also shiftthe peak HVSR frequencies by 10–20 per cent or more. Thus, careshould be taken in interpreting our HVSR curves both at the edgesof the Jakarta Basin and at the putative sharp increase in depth nearcentral Jakarta.

3.2 Data

Prior to the seismometer deployment whose data are used here, theonly previous seismographic study covering the Jakarta Basin wasthat of Ridwan (2016) and Ridwan et al. (2016), who used shorterperiod sensors to estimate surface wave dispersion curves in thefrequency range 0.2–3.0 Hz using the SPAC method (Aki 1957).The considerable depth to engineering basement (>700 m) and thelow velocities of the sediments comprising the basin (≈500 m s−1)found by Ridwan (2016) and Ridwan et al. (2016) suggested thatthe seismic records from broad-band seismometers may be moreuseful in modelling its interior geometry. To this end, 52 three-component, broad-band sensors were deployed at various sitesthroughout Jakarta from October 2013 to January 2014 (Sayginet al. 2016). In order to cover the basin at an average interstationspacing of about 2 km, 26 seismometers were re-deployed succes-sively at 44 locations every 3 months. In total, 96 seismic stationswere installed, with ambient seismic noise recorded over intervalsranging from 1 to 3 months.

The data from this broad-band experiment were used in an ANTstudy to image the shear wave velocity structure of the Jakarta Basin(Saygin et al. 2016), as well as in a study of P-wave reflectivity fromautocorrelation of seismic noise (Saygin et al. 2017). However, bothof these studies have limiting resolving power for the basement: thelong wavelengths of the ANT study precluded imaging of sharp dis-continuities and the P-wave reflectivity method lacked an advancedmodelling framework. In this study, these same ambient seismicnoise recordings are used to compute HVSR curves, and an inver-sion is applied to these curves to obtain a velocity profile at eachstation.

Since both the ANT study of Saygin et al. (2016) and the SPACstudy of Ridwan et al. (2016) suggest that low shear wave velocitiesextend to a basin depth of 500–700 m or more, it was important forthe data processing to resolve the HVSR at periods of 5 s or longer.While HVSR studies typically use a time window of around 100 sduration (SESAME 2004), we tried to ensure stability of our long-period HVSR curves by using time windows of 1000 s duration(see Figs 2a–f). We note that Langston & Horton (2004) suggesteda 600 s window to retrieve fundamental frequencies near 0.1 Hz inthe Mississippi Embayment. We used the Geopsy software (GeopsyTeam 2004) to compute the HVSR curves in the frequency range

Dow



ay 2022


Figure 2. (a) Distribution of stations indicated as black dots while blue dots indicate stations mentioned in the text and red diamonds are stations with profilesare shown in Supporting Information Fig. S4. The horizontal (WE) and vertical (SN) lines indicate the transect lines used in Fig.5. (b) The variation in peakfrequency over 24 hr, the longer the bar, the higher the peak frequency (in range 0.0–0.145 Hz), and numbers 0–23 indicate hours. (c and d) Comparison betweenelipticity curves recorded at quiet (10 p.m.–4 a.m.) and busy (4 a.m.–10 p.m.) times, respectively, at JKA07. Dashed line indicates average peak frequencycomputed from quiet times in a week. Each colour represents the time of measurement. (e and f) For most cases, in order to measure reliable HVSR curves atabout 0.1 Hz, a window length of 100 s as suggested by SESAME is enough. However, for a few cases, window lengths of 750 s or longer are needed. Colourrepresents window length used in the analyses.

from 0.1 to 10 Hz as shown in Fig. 2. In general, the observed peakfrequencies and amplitudes at each station changed slightly withtime, but using measurements at quiet times (i.e. from 10 p.m. to 4a.m., see Fig. 2) resulted in very stable curves, for peak frequencyas well as amplitude. For most stations, a 100 s time-series windowresulted in reliable HVSR curves. However, for stations JKA54 andJKA42, a window length of 750 s or higher was needed. Therefore,for consistency, a 1000 s window length was applied for all stations.

Raw spectra often show narrow modulations and spikes that leadto extreme values of HVSR. To mitigate this effect, the smoothingoperator proposed by Konno & Ohmachi (1998) was introduced.Also, the ratio between the average level of signal amplitude over ashort period of time (STA) and long term average (LTA) is used as

a parameter to exclude transient signals. The STA and LTA are setto 1 and 30 s, respectively, and only windows with STA/LTA ratiobetween 0.2 and 2.5 were used for the HVSR computation.

4 H V S R C U RV E S

HVSR curves show peaks for certain dominant periods that arediagnostic of the impedance contrasts in the underlying velocitystructure. Peaks in the HVSR curve are typically associated withzeros in the vertical component Rayleigh wave, at frequencies wherethe sense of motion switches from prograde to retrograde or viceversa, and the Rayleigh wave is horizontally polarized (Konno &

Dow



ay 2022

436 A. Cipta et al.

Ohmachi 1998; Berbellini et al. 2016). Such peaks typically coin-cide with the fundamental resonant frequency f0 of S waves in thesedimentary column, given by period f0 = 1/4 VS/h, where h is thedepth to a strong impedance contrast and VS is the shear velocity ofbasin sediments.

In this study, we found that the HVSR curves were typicallycharacterized by two peaks, one a low-frequency peak in the range0.1–0.3 Hz which we denote f l

0 and a high-frequency peak above1 Hz, which we denote f h

0 . We found that our observations couldbe broadly classified into four groups: (1) the absence of any peak(i.e. a flat HVSR curve or no peak with amplitude >2) which mayreflect the absence of a strong impedance contrast due to gradualcompaction of the rock during self-loading, as observed at siteJKC13 (Fig. 3a); (2) a sharp and high peak at an f l

0 , followed byanother sharp and quite high peak at f h

0 between 1 and 2 Hz, asobserved at site JKA49 (Fig. 3b) recorded in the northern basinwhere the Holocene alluvium overlies older alluvial sand; (3) asharp and high peak at an f l

0 , followed by a second peak at f h0

higher than 2 Hz (Fig. 3c), observed in stations near alluvial fandeposits or along an alluvium boundary where river channel depositsslice through alluvial fan or where beach ridges are deposited overalluvium; and (4) a single peak at f l

0 , as observed at alluvial fansites such as JKC01 as shown in Fig. 3(d).

Types 2–3 are all characterized by sharp f l0 peaks, while Type

4 includes both sharp and broad f l0 peaks. We distinguish between

these sharp and broad f l0 peaks in the following section. Overall,

most of the stations having lowest f l0 (0.12–0.13 Hz) are located

in areas predominantly composed by recent alluvium, however, aless clear correlation between f l

0 and surface geology is shown inFig. 3(e).

4.1 Sharp f l0 peaks

Almost all the stations recorded sharp HVSR peaks in the frequencyrange 0.1–0.2 Hz, except for the southernmost stations. These sharpf l0 peaks suggest a strong and deep impedance contrast. A high and

narrow peak indicates a strong and sharp impedance contrast (Gosar2010). The greater the magnitude and sharpness of the contrast inlithology, the higher and sharper the peak in the HVSR curve isexpected to be (Tarabusi & Caputo 2016).

While observed f l0 range from 0.124 to 0.249 Hz, the frequencies

of the f h0 peaks range between 1 and 6 Hz or are absent. Generally,

the stations in the south have f l0 peaks that are lower and broader,

and centred at higher frequencies than the f l0 peaks for stations

in the north. The broad peak amplitude may reflect the irregularstructure of the basin floor, the stratigraphic complexity at depthor the unconformity between younger volcanic and older marineformations. At JKC18 where the Plio–Pleistocene volcanic rocksand alluvial fan are underlain by Tertiary marine formations, a broadHVSR curve was recorded.

In the north, the absence of Pleistocene volcanic rocks shouldresult in a significant impedance contrast between the Tertiarybasement and overlying soft sediment. Near the surface, the broadvariability of f h

0 indicates a corresponding variability of surficialsediment thickness. This also suggests variability in short-periodamplification level. On the other hand, the absence of an HVSRpeak as shown in Fig. 3(a) may be related to a shallow stratigraphichorizon or compaction that is gradual with depth and results inweak basement impedance contrast (it might also be due to bedrockoutcrop, but little or no such outcrop exists in our study area).

This implies that little or no amplification will occur (Castellaro &Mulargia 2009a,b).

Cipta et al. (2018) have used 2-D waveform modelling along anNS cross-section of the Jakarta Basin model developed here to showthat very large amplification of ground motion (up to 1500 per centPGV) can occur for a large scenario earthquake on the Java Trench,with the thicker sediments in the north experiencing the largestamplification. However, the wedged shape of the basin edge as itshallows to the south also plays an important role in amplifying seis-mic waves (up to 1200 per cent PGV), particularly for the intraslabearthquake scenario considered by Cipta et al. (2018). For both themegathrust and intralsab scenarios, the spectral amplifications weremost prominent in the period range 5–7 s, similar to that of theobserved f l

0 HVSR peaks.

4.2 Broad f l0 peaks

Lower amplitude and broader peak could reflect a weaker contrastor a more gradual discontinuity. A high variability of surface lithol-ogy may also cause a lower peak amplitude. A broad peak is alsoindicative of irregularity of subsurface structure (Uebayashi 2003)and a broad asymmetric peak indicates variability in rock formation(Gosar 2010).

Broad HVSR peaks at f l0 as observed at stations JKA55 and

JKC02 (Fig. 4) may be related to steeply dipping bedrock underlyingshallow sediments (Ozalaybey et al. 2011) or variability of bedrockformation (Uebayashi 2003; Bonnefoy-Claudet et al. 2009; Gosar2010). Broad HVSR peaks may also correspond to meaningful 2-Dor 3-D variation in the bedrock–sediment interface. In such a situa-tion, diffracted waves including body and surface waves, includingLove and higher order Rayleigh waves may be present, so that ourinterpretation of the HVSR curves in terms of fundamental modeRayleigh waves in a 1-D structure may not be reliable (Bonnefoy-Claudet et al. 2009). In any case, the study of such small-scale 3-Dstructure is beyond our scope.

Broad HVSR peaks at f l0 are recorded at the southernmost sta-

tions where the bedrock depth is expected to be shallow. The slightlybumpy topography decorated by low hills reflects the variation ofunderlying bedrock structure. In accordance to its proximity to thestream heads, this area is relatively closer to the basin rim, and as aconsequence the bedrock depth is shallow. This geographic condi-tion may lead to the emergence of broad f l

0 peaks at stations JKA55and JKC02, as shown in Fig. 4. Similarly, broad HVSR curves areobserved at stations in elevated areas in southwest Izmit Bay basinwhere the basin depth is about 200 m based on a gravimetric study(Ozalaybey et al. 2011).

5 H V S R P E A K F R E Q U E N C I E S A N DA M P L I T U D E S

The HVSR technique is able to furnish estimates of the frequenciesf l0 and the corresponding HVSR peak amplitudes at each site. Both

of these can potentially be used as proxies to map basement depth,and in Figs 5(a) and (b) we have mapped the interpolated values off l0 and peak amplitudes, respectively. These both suggest a pattern

of deepening basement from south to north that is very similar to theresults of Saygin et al. (2016). It is important to note that the simplekriging method is used to interpolate data sets and produce thosemaps. The kriging method considers the distance and the degreeof variation between known data points when estimating values

Dow



ay 2022


Figure 3. Ellipticity curves measured in Jakarta: (a) flat, (b) sharp peak at low frequency (<0.25 Hz) and another at a frequency between 1 and 2 Hz, (c) peakat low frequency (<0.25 Hz) and at frequency above 2 Hz and (d) peak only at low frequency (<0.25 Hz). Variability of the second peak (or absence of thishigh-frequency HVSR peak) roughly reflects surface geology. Panels (e) and (f) show the distribution of low- and high-frequency HVSR peaks, respectively,plotted on a geological map. Thin blue lines indicate Cisadane, Ciliwung and Kali Bekasi faults as mentioned by Moechtar (2003) and Putra et al. (2016).Location of stations is shown in Fig. 2(a).

in unknown areas. However, at the data points themselves it mayproduce values that are slightly different from the observed data.

Maps of f l0 and corresponding peak amplitudes (Figs 5a and b,

respectively) show two areas that exhibit both pronounced lows inpeak frequency f l

0 and high-peak amplitude: (1) the deepest part ofthe basin located in the northeast; and (2) the high-peak amplitudepassage crossing the city from the northwest to the southeast. Theformer coincides with high rates of subsidence (Abidin et al. 2011)while the latter may reflect a fault or anticline axis.

Supplementary to f l0 , higher frequency peaks at f h

0 emerge at33 stations. Stations that exhibit f h

0 peaks are distributed mostly in

northern Jakarta where thin marine and terrestrial sediments overliethe stiffer and older alluvium. Recently deposited and intercalatedmarine and terrestrial sediments may form a relatively thin sedimentlayer which may be responsible for the f h

0 in the frequency ranges1–2 Hz (Fig. 2f). We surmise that the high amplitude of these f h

0

peaks indicates a high impedance contrast between relatively youngsediments and underlying stiffer alluvium in Figs 3(b)–(f) and 6(a)–(d).

Higher frequency ( f h0 ) peaks also emerge at stations in the south–

north (SN) cross-section indicated in Fig. 2(a) where alluvial fandeposits dominate the surface geology. In these stations, f h

0 appears

Dow



ay 2022

438 A. Cipta et al.

Figure 4. Curves showing broad peaks at f l0 at stations JKA55 and JKC02, both located in the southernmost of the city where basin depth is shallow. Location

of stations is shown in Fig. 2(a).

Figure 5. (a) Map of interpolated f l0 frequencies, showing two prominent low-frequency areas (coloured red) in the northeast corner and along a northwest–

southeast passage. Coloured circles indicate stations at which the corresponding peak frequencies were observed. (b) Map of interpolated f l0 amplitudes, with

coloured circles indicting the stations where the corresponding amplitudes were observed.

at higher frequency (4–6 Hz) than those that emerge at northernstations. We speculate that self-compaction and near-surface weath-ering are the cause of the emergence of higher frequency f h

0 peaks.Low amplitude (≤3) at f h

0 (4–6 Hz) implies this geological phe-nomenon, where thin degraded alluvial fan is overlying the firmersoil (Figs 4b, 3f, 6a–d).

In Figs 6(a) and (b), 2-D transects have been constructed whichshow the correlation of f l

0 and f h0 HVSR peaks between adjacent

stations, and illustrate how they vary in SN and west–east (WE)directions. A sudden decrease of f l

0 is clearly visible in the SNtransect. In particular, f l

0 decreases from 0.223 Hz (JKC02) to0.146 Hz (JKA12), indicating an abrupt change in basin depth overa distance of only 2.6 km. At the northernmost stations (e.g. JKA24),f l0 has increased slightly to 0.154 Hz, over a distance from JKA12

of around 14.6 km. This suggests an abrupt increase in basementdepth northward between JKC02 and JKA12, the basement thenflattening up to the central part of Jakarta and becoming slightlyshallower to the north.

The WE transect shows a complexity of near-surface layering,as reflected by the f h

0 peaks. Sites JKA44 and JKB14, located nearand on Plio–Pleistocene tuff deposits, respectively, do not show clearf h0 (Figs 6c and d). The absence of f h

0 peaks suggests there are nostrong impedance contrasts in the near-surface layering. Further to

the east, as the surface geology changes from tuff to alluvial fan andalluvium with locally intercalated beach deposits, the f h

0 peaks aremore pronounced. These f h

0 peaks appear at a frequency of 4 Hzexcept for stations JKA32 and JKA07 where they appear near 2 Hz.In addition, a smooth decrease of f l

0 is shown in the WE transect.The f l

0 at the westernmost station decreases towards the centre,fluctuates and then decreases eastward. This pattern indicates ashallower basement in the west, undulating in the centre and deeperin the east (Figs 6c and d).

The peak frequency data from 93 stations covering Jakarta haveindicated significantly low f l

0 in the northeast corner of the city,high f l

0 in the west and south and fluctuating values of f l0 at the

centre. A low f l0 patch along the NW–SE portion of central Jakarta

is sandwiched within the three narrow patches of high f l0 in the

west, south and east. This gully-like feature is situated at high-peak HVSR, with surrounding low-peak HVSR within the vicinity.In general, the low f l

0 values in Jakarta corresponds to high-peakHVSR and vice versa as depicted in Fig. 5.

6 I N V E R S I O N O F H V S R C U RV E S

Quantitative estimates of depth profiles of elastic properties suchas shear wave velocity VS can be obtained through HSVR curve

Dow



ay 2022


Figure 6. Comparison of HVSR spectra along the SN (a and b) and WE (c and d) transects indicated in Fig. 2(a). The SN transect (a and b) shows that thefrequency f l

0 of the low-frequency HVSR peak abruptly decreases at JKA12, then gradually increases at JKA47. This graph also illustrates the effect of theimpedance contrast at shallower depth on the high-frequency peak amplitude in the ellipticity curve. The WE transect (c and d) shows highest f l

0 at the westernend, lowest f l

0 at the eastern end and variable f l0 in the middle. Blue and red dashed lines are the highest and lowest peak frequencies (peak f l

0 ), respectively, atcorresponding cross-sections. The green solid lines show lineation of peak frequency and blue dots are stations indicated in Fig. 2(a). The x-axis is the relativelocation of stations in the SN (b) and WE (d) directions, as indicated by blue dots in (a) and (c).

Dow



ay 2022

440 A. Cipta et al.

inversion, wherein it is typically assumed that the shape is de-termined by the Rayleigh wave ellipticity. However, the nonlineardependence of the HVSR curve on the depth profile of elastic pa-rameters makes the inversion challenging. Scherbaum et al. (2003)used a simple model in which VS having a power-law dependenceon depth overlies a half-space to invert HVSR curves for model pa-rameters using a grid search. Arai & Tokimatsu (2004) developeda nonlinear least-squares approach in inverting HVSR curves usingmultimodel Rayleigh wave ellipticity to resolve VS profiles includ-ing low-velocity zones. More recently sampling methods, with afixed number of layers, are becoming widely used (Fah et al. 2003;Wathelet et al. 2004; Parolai et al. 2005; Hobiger et al. 2013). Allof these methods use simplifying assumptions on the model, forexample, power-law depth dependence (Scherbaum et al. 2003),fixed number of layers (Fah et al. 2003; Parolai et al. 2005; Ho-biger et al. 2013), fixed VP and/or density (Scherbaum et al. 2003),restricted Poisson’s ratio (Hobiger et al. 2013) or fixed bedrockvelocity (Parolai et al. 2005).

A priori constraints on the complexity of velocity profiles inthe Jakarta Basin present a problem because we do not know howcomplicated they might be. The surface geology is dominated bypoorly consolidated, water-saturated sediments with a basement athundreds of metres depth. Thus, we expect the velocity profile tobe complicated in the topmost 100 m of the profile, where watersaturation will lead to a high Poisson’s ratio that will decreaserapidly with depth along with an increase in seismic velocity due tocompaction. Deeper than 100 m, there may be intervening layers ofmarine Pliocene and Quaternary sand and deltaic sediments untilthe basement is reached at several hundred metres depth or more.We would like to find the simplest model that can fit the data, butwe neither know what minimum complexity is required nor whatthe uncertainty is in our HVSR curves.

Bayesian inversion provides a framework to combine informationon model parameters we have before making an observation with aprobabilistic expression of data information to obtain an a posterioriprobability density function (PDF) for the model parameters. Todefine the posterior distribution, Bayes’ theorem (Bayes 1763; Sivia& Skilling 2006) is used, which is expressed as

posterior = likelihood × prior

evidence.

More specifically, if we consider a model space consisting of a fixednumber k of horizontal layers, each having a uniform distributionof S-wave velocity VS

k , ratio of P- to S-wave velocity (VP/VS)k, anddensity ρk, Bayes’ theorem can be expressed for a data vector d andmodel vector mk as:

P(mk |d) = P(d|mk)P(mk)∫M P(d|m′

k)dm′k

, (2)

where P(mk |d) is the conditional probability of the model giventhe data (the ‘posterior’), P(d|mk) is the conditional probability ofthe data given the model (the ‘likelihood’), P(mk) is the a prioriprobability of the model (the ‘prior’) and

∫M P(d|m′

k)dm′k is the

probability of the data (the ‘evidence’). We note that in Bayesian in-version the evidence integral is typically regarded as a normalizationconstant and ignored, and the numerator of the posterior PDF canbe explored using a Markov chain (Mosegaard & Tarantola 1995),which would sample over the space of models with k uniform layers.While it might be possible to perform multiple inversions for dif-ferent values of k, and test the hypothesis that a model with k′ layersis more plausible than one with k layers using the ratio of evidence

values (Bayes’ Factor), in practice computation of the evidence in-tegral

∫M P(d|m′

k)dm′k is very time consuming, and this would not

address the problem of assessing how our lack of knowledge of anappropriate value for k contributes to model uncertainty.

Green (1998) showed that Bayes’ rule can be written for aBayesian hierarchical model to include a hyperparameter k:

P(k, mk |d) = P(k)P(d|k, mk)P(mk |k)∑k′∈K′

∫M P(k ′)P(d|k ′, m′

k′ )dm′k′

, (3)

where k can be interpreted as indexing possible choices of models,in our case models with different numbers of layers. As with eq. (2),the denominator in eq. (3) can be regarded as a normalizing constantand ignored, and the posterior P(k, mk |d) can be explored by usinga Markov chain Monte Carlo approach to sample the numerator, asdescribed in Malinverno (2002), Sambridge et al. (2006), Dettmeret al. (2010) and elsewhere. This ‘trans-D’ approach allows a groupof model parametrizations, in our case models with varying numberof layers, to be considered simultaneously for analysis. Becauseall models are considered in the analysis, the trans-D approachallows us to account for how limited knowledge about the numberof layers affects the parameter and uncertainty estimates based onthe posterior. Inferences obtained in this manner avoid the inherentbiases involved in selecting a single fixed number of layers and arehence more realistic and reflective of the state of knowledge aboutmodel parameters.

We invert our HVSR curves for depth profiles of elastic parame-ters using trans-D Bayesian inversion (Malinverno 2002; Sambridgeet al. 2006; Dettmer et al. 2010). Since this statistical samplingmethod requires very rapid forward computations, we make useof the highly optimized forward computations of Wathelet et al.(2004), which parametrize the velocity structure as a stack of ho-mogeneous layers overlying a half space. We invert for the numberof layers k as well as for VS, VP/VS, ρ and thickness h in each layer.In addition, we use a hierarchical approach (e.g. Bodin et al. 2012;Dettmer et al. 2012) to estimate the noise (i.e. the value of σ ) aspart of the inversion.

Following Dettmer et al. (2012) eq. (6), we use the likelihoodfunction:

P(d|k, mk) = 1

(2π )N/2|Cd |1/2

× exp

(−1

2(d − d(k, mk))T C−1

d (d − d(k, mk))

),

(4)

where Cd = σ 2 ∗ I, with I the identity matrix and σ the noise stan-dard deviation (the magnitude of the noise) and the data vectord = [HVSR(ω0), HVSR(ω1), ..., HVSR(ωn)]. The modelled datavector d(k, mk) is the value of HVSR calculated at frequenciesω0, ω1, ..., ωn for the model having number of layers k and modelvector mk consisting of the layer parameters hi, V i

S , (VP/VS)i and ρ i

, where i = 1, ..., k. These layer parameters, as well as the number oflayers k and the standard deviation σ , are unknowns in the inversion.

As described in more detail by Dettmer et al. (2012), this trans-D approach to model selection lets the data infer its own noisemagnitude and model complexity. Although it involves no regu-larization such as model smoothing, Malinverno (2002) has shownthat Bayesian model selection tends to ‘parsimony’ in the sense thatmodels with more complexity than is required to fit the data tend tobe assigned low postiori probability.

Dow



ay 2022


6.1 Synthetic tests

The importance of model selection using the trans-D method foruncertainty quantification is illustrated with a simulation examplein Fig. 7 and Supporting Information Figs S1–S3. Data were sim-ulated by calculating an HVSR curve for a 10-layer model whichis representative of the expected complexity in the Jakarta Basin.Gaussian-distributed random noise with σ = 0.07, in logarithmicscale, was added to the HVSR curve, which appears close to thenoise level in the observed data.

In the true model, velocity rapidly increases with depth in the top100 m due to compaction, after which there is little variation until abasement is reached at 900 m depth (the white curve in Fig. 7). Thisstructure results in two peaks in the fundamental-mode Rayleighwave ellipticity spectrum at 3.0 and 0.16 Hz, corresponding to thevelocity increase in the top 100 m and the 900 m basin depth,respectively. Three inversions were carried out: (1) Fig. 7(a) (seealso Supporting Information Fig. S1) shows the results for trans-Dinversion, where the number of layers in the model is treated asunknown, with a prior set as uniform between 3 and 20 layers; (2)Fig. 7(b) (see also Supporting Information Fig. S2) shows resultsfor an inversion were the number of layers was fixed at three, whichwould generally be regarded as a reasonable choice, but representsan underparametrized inversion; (3) Fig. 7(c) (see also SupportingInformation Fig. S3) shows results for an inversion where the num-ber of layers was fixed at 20, which represents an over-parametrizedinversion and is expected to result in spurious, unconstrained struc-ture.

In all three cases eight Markov chains were run in parallel andmore than 100 000 posterior samples were recorded. The total num-ber of steps was much larger, since only every 100th step wasrecorded and half the initial steps were discarded as ‘burn-in’. Theprior was uniform and is given by the plot boundaries in SupportingInformation Figs S1–S3. The bounds are wide so that data informa-tion predominantly constrains the solution. Convergence for theseinversions was judged based on examining the chain history forall eight chains. Since the chains are independent but sample theparameters in highly similar manner, convergence is likely.

The trans-D results in Fig. 7 show excellent agreement with thetrue model. Importantly, uncertainty estimates appear reasonableand increase substantially with depth, which is commonly observedfor diffusive wave fields, such as Rayleigh waves. The data also ap-pear to provide only limited information about selecting the numberof layers for this site which results in substantial uncertainty in kand models with 7 to 15 layers all fit the data acceptably well. How-ever, the posterior also shows clearly that the VS marginal profile isparsimonious in that it captures an appropriate level of complexitywithout introducing spurious layers. The trans-D result also providesan excellent data fit and the noise standard deviation estimated bythe inversion is close to the true value with some uncertainty from0.06 to 0.10. Importantly, the ability of HVSR curve inversion toresolve multiple VS layers is encouraging for our application. Whilethe HVSR curve is mostly sensitive to VS, its sensitivity to VP/VS

and ρ is in general significant (Berbellini et al. 2016). For this rea-son we included VP/VS and ρ as well as VS in the inversion, butsince the resolution of VP/VS and ρ was poor we will not interpretthem further.

In the three-layer case (Fig. 7b and Supporting Information Fig.S2), an apparently reasonable and well-constrained model is ob-tained. However, the estimated posterior bears little resemblanceto the true model and uncertainties are estimated to be low andincrease only slightly with depth. In addition, the estimated noise

standard deviation is four times larger than the true value, a knownissue for inversions with underparametrized models (Dettmer et al.2009). The large noise standard deviation is also reflected in thepoor data fit (Supporting Information Fig. S2).

In the 20-layer case (Fig. 7c and Supporting Information Fig. S3),the data fit is excellent and the noise standard deviation is estimatedclose to the true value. Such good data fits commonly boost confi-dence in the inversion results. However, the posterior estimate doesnot represent the true model well. Rather, the results are plagued byvery large uncertainties and erroneous, unconstrained structure.

In conclusion, the trans-D results are far superior and thisparametrization is appropriate for HVSR data. Importantly, the abil-ity of HVSR curve inversion to simultaneously resolve both the finescale VS layering in the upper 100 m and the coarser scale layeringassociated with a basement depth of several 100 m is encouragingfor our application.

6.2 Inversion of Jakarta’s HVSR curves

Out of the 96 stations with data, 93 stations were deemed useful inmapping the geometry of the Jakarta Basin. The trans-D inversionis applied to HVSR curves from these stations to obtain estimatesof VS profiles in this deep basin with an irregular bedrock-sedimentinterface. The HVSR curves are assumed to be a representationof fundamental-mode Rayleigh wave ellipticity over the frequencyband of 0.1–10 Hz.

To assign the prior, we choose upper and lower limits for pa-rameter values (k and VS), based on existing geological knowledge(thickness and age of rock formations), boreholes and hydrogeo-logical investigations (depth and thickness of aquifers). The hydro-geological study of Delinom et al. (2009) found that underneaththe Jakarta Basin, there are three groundwater aquifers at depths of0–40, 40–140 and 140–250 m, respectively. This may indicate thatthere are at least three distinguishable layers in the top 250 m ofsoil. The series of volcanic products and by-products, such as al-luvial fans, that were deposited after the Plio–Pleistocene thrusting(Kloosterman 1989), marked the early Quaternary sedimentationand likely have different density from underlying Late Miocene for-mations. In addition, an interface due to the Oligocene–Miocenetectonic period (Turkandi et al. 1992) is expected. Therefore, we seta uniform prior on k between 3 and 20 layers, ensuring that the ex-pected structure fits well within this prior specification. The prior forinterfaces to occur is chosen to be uniform from 0 to 3000 m depth.The prior on VS is chosen uniform between 100 and 4000 m s−1 toinclude a wide variety of sediment and rock types. Both density andVP/VS are treated as unknown nuisance parameters in the inversion.For density, the prior is set uniform between 1.5 and 4.0. Finally,the VP/VS prior is set uniform from

√2 to 8, to include the potential

of poorly consolidated, highly water-saturated sediments.Fig. 8 shows an example of inverted VS profiles and observed

and computed HVSR curves for station JKB18. Log likelihood,optimum number of layers and standard deviation are also presented.Fig. 9 shows VS results and HVSR curves for stations JKB20 andJKA12. The slim coloured bands indicate the posterior probability(normalized at each depth level) and show that the prior informationwas substantially updated with data information which resulted in anarrow range of acceptable velocity-depth profiles. The data fits forthese stations are also shown and show good fits to both the low-and high-frequency HVSR peaks.

The inversion procedure was repeated for the remaining 93 sta-tions in identical manner. The maximum a posteriori VS profiles

Dow



ay 2022

442 A. Cipta et al.

Figure 7. VS results of Bayesian inversion of synthetic data for which random noise has been added to forward calculations of an HVSR curve calculatedusing the 10-layer model indicated by the white solid curve. Three inversions have been carried out: (a) a trans-dimensional inversion that accounts for modelswith variable numbers of layers and inversions in which the number of layers is fixed at 3 and 20 (b and c, respectively). The insets at the top right of each panelexpand the details of the velocity profile in the top 200 m.

from each station were then gridded with linear interpolation in 3-Dspace. Since all models featured an abrupt change in VS from a valueof 700–900 to 1500–2100 m s−1 at some depth that we interpret asbasement, an iso-surface at VS 1500 m s−1 was used to construct abasin depth model. On average, the depth of this iso-surface rangesfrom 700 to 900 m, although some western and eastern parts of theJakarta Basin are significantly deeper than 1250 m, while shallowerbasement depths of ∼300 m or less are estimated in the south. Ingeneral, the southern part of the basin is shallower than the northernpart.

An SN transect of basin structure constructed by interpolatinginversion results along the line indicated in Fig. 2(a) shows the vari-ation in basement geometry (Fig. 11a). A zone of very low velocity(300 m s−1 or less) dominates the top ∼100 m, while the shearwave velocity above the basement is less than 900 m s−1, abruptlyincreasing to 1500–2100 m s−1 in the basement. Taking this as thebasement–sediment boundary, in the southern half of the city it issignificantly shallower than the northern, with an abrupt increase indepth at around 6.24◦S as depicted in Fig. 11(a). These velocity pro-files are in good agreement with the distribution of low-frequencyHVSR peaks. The stations in the south, including JKC18, JKC14and JKC02 recorded higher f l

0 (0.191–0.224 Hz), while the centralstations, such as JKA12, JKC16 and JKC01, recorded significantlylower f l

0 (0.138–0.146 Hz), the northern stations, including JKA38,JKB16, JKA47, JKB20 and JKA24, recorded higher f l

0 (0.154 Hz)compared to the stations at the centre, as shown in the same figure.

A shallow basin depth in the south has also been inferred from thesurface geological data. At the head of the alluvial fan, the depositsare thinner with coarser grain-sized components. Further from thehead, along the gentle slope, thicker and finer grain-sized sedimentswere deposited. This may explain the basement depth reaching amaximum depth in central Jakarta. The accumulation of youngermarine sediments overlies the alluvial fan on the northern part of thecity, which together with the alluvial fan and Pliocene–Pleistocene

tuff fill the basin up to a thickness of 1300 m. The tectonic highknown as Thousand Islands High in the north of the bay of Jakartalimits the sedimentation rates along the coastal plain. Consequently,sediment deposits are gradually thinning northwards. Prograde de-positions, both from the south and north, caused thick sedimentationin the centre while thinning southwards and northwards as indicatedin the basin depth map in Fig. 10.

7 D I S C U S S I O N

Our study has shown that the Jakarta Basin, a Miocene basin filledwith Pliocene–Holocene sediments, has thickness ranging from 300m up to 1350 m. It is located approximately 250 km from the Javasubduction zone. The similarity in geology and tectonics suggeststhat Jakarta may experience significant building damage analogousto Mexico City if a high-magnitude subduction zone or intraslabearthquake occurs. Further complicating the situation for Jakartaare 69 tall buildings (40+ storeys) that are currently in use, rankingJakarta as 14th among the cities with the greatest number of build-ings of at least 150 m height (CTBUH 2017; Emporis 2017). Suchhigh-rise buildings are prone to high shaking at 3 s or longer periodsas was observed in Mexico City during the Michoacan Earthquake(Padilla y Sanchez 1989; Flores-Estrella et al. 2007).

Previous studies in the basins of southern California (Magistraleet al. 2000) and the Lower Rhine Embayment (Ewald et al. 2006)showed the importance of 3-D basin geometry in modelling the char-acteristic basin propagation of seismic waves. Data from the 3-Dsimulations of wave propagation are available including wave frontfocusing due to low-velocity zones or edge-diffracted waves (Ewaldet al. 2006), basin resonance (Castellaro et al. 2014) and local-scalemultiscattering and prolonged ground motion (Olsen et al. 2006;Furumura & Hayakawa 2007; Denolle et al. 2014a; Cruz-Atienzaet al. 2016; Viens et al. 2016). Interesting research of Roten et al.(2014) found that non-cohesive shallow low-velocity sediments of

Dow



ay 2022


Figure 8. Trans-dimensional inversion that allows the data to determine the number of layers. The 11-layer model can recover the ellipticity curve well andshows a strong impedance contrast at a depth of about 950 m that may indicate basement depth. The location of JKB 18 is shown in Fig. 2(a).

Los Angeles Basin may reduce seismic shaking generated by SanAndreas Fault. Accordingly, a rigorous quantification of seismicproperties and geometry of the Jakarta Basin is a prerequisite togenerating a robust seismic hazard map for the city.

Given the basin model presented in Fig. 10, it is clear that there aresome regions flanked by deeper depressions. A ridge-like structure(light green: shallower depth), extending from the central–northcoast to the southwest, is sandwiched by two deeper bedrock regions(yellow–red) on its northwest and southeast sides (dashed lines). Thesouthern basin is shallower but also exhibits significant topography,including a shallow ‘bump’ in the basement that is surroundedby deeper sediment. In the case of the Lower Rhine Embayment,prolonged shaking is recorded in an area situated between two deepdepressions. The dependence of seismic wave amplitude on basindepth was also observed in the Lower Rhine Embayment (Ewaldet al. 2006). In Jakarta, this phenomenon may occur in the northwestand northeast where the basin is very deep.

Profiling of velocity structure, sediment thickness and basin to-pography can also be accomplished via the joint inversion of HVSR

and the surface wave dispersion curves, where the latter can be gen-erated using autocorrelation (Claprood et al. 2012) or multichannelanalysis (Gorstein & Ezersky 2015; Pandey et al. 2016). As shownin Fig. 11(a), our direct inversion of the ellipticity curve resulted ina velocity profile that is comparable with the results obtained usingANT conducted by Saygin et al. (2016). The advantage of ANT isits treatment of lateral variability, while in our HVSR analysis wehave assumed locally 1-D structure. However, the vertical resolutionof our results is higher than the results presented by Saygin et al.(2016).

8 C O N C LU S I O N

We have shown that the HVSR of seismic ambient seismic noisein the Jakarta Basin can be used to make robust estimates of theHVSR peak periods, which has been shown to coincide with theS-wave resonant period in many studies (Lermo & Chavez-Garcia1993; Lachet & Bard 1994; Bonilla et al. 1997; Lunedei & Al-barello 2010). In Jakarta, these peak periods are in the range 4–8 s,

Dow



ay 2022

444 A. Cipta et al.

Figure 9. Inverted velocity profile at stations JKB20 (a) and JKA12 (b), the relatively thin coloured bands near surface and broader in the deeper part indicatestandard deviation broader as depth increasing, as expected. The very thin bar at depth more than 2300 m indicates that inversion cannot resolve the data well(b). Computed ellipticity curves (green) at stations JKB20 (c) and JKA12 (d) fit observed ellipticity curves (blue) well.

with generally shorter periods in the south and longer periods inthe north, in agreement with previous studies suggesting very lowvelocities extending to a basement at several hundred metres depththat deepens northwards (Ridwan 2016; Ridwan et al. 2016; Sayginet al. 2016). S-wave resonant periods in this range are potentially aconcern for very tall (40–80 storey) buildings, which are prevalentin Jakarta currently and whose number is expected to increase inthe next decade.

Assuming the HVSR curve measured from ambient seismic noisein the Jakarta Basin is dominated by the fundamental mode Rayleighwave, and is therefore an expression of its ellipticity (Lachet & Bard1994; Scherbaum et al. 2003; Arai & Tokimatsu 2004) we have alsoinverted this curve to estimate the S-wave velocity structure of thebasin. In order to resolve both shallow velocity structures due tocompaction of unconsolidated sediments as well as the deep basinarchitecture, we used trans-D Bayesian inference, which not onlyallows the complexity of the model (in our case the number oflayers) to adapt to fit the data but also accounts for uncertaintyin the model due to the unknown number of layers. At the sametime the ‘Bayesian parsimony’ feature of Bayesian model selection(Malinverno 2002) assigns low a posteriori probabilities to modelswhose complexity is not required by the data. The dense seismicnetwork we deployed in the Jakarta Basin enabled the mapping ofbasement topography, whose depth varies within the range 300–1400 m in a pattern similar to that obtained in previous studies

using ANT (Saygin et al. 2016) and SPAC (Ridwan et al. 2016),respectively.

Previous studies using HVSR to infer basin structure have notedlimitations of the type of HVSR analysis presented here. Severalhave noted the importance of using higher order Rayleigh modesin fitting observed HVSR curves, particularly when low-velocityzones may present (Arai & Tokimatsu 2004; Asten et al. 2004;Parolai et al. 2005; Poggi et al. 2012; Rivet et al. 2015). Othershave discussed the importance of also taking into account Loveand body wave energy when interpreting the HVSR curve (Arai &Tokimatsu 2004; Sanchez-Sesma et al. 2011; Salinas et al. 2014)as well as the importance of jointly inverting HVSR data withother data sets such as surface wave dispersion curves to avoidsome of the strong trade-offs in thickness versus velocity modellayers (Scherbaum et al. 2003; Parolai et al. 2005; Hobiger et al.2013). Finally Gueguen et al. (2007) have discussed the difficultyin applying HVSR to basins with small aspect ratio. We agreethat all of these are potential shortcomings of the approach wehave taken in this study, and for this reason regard our model ofthe Jakarta Basin as preliminary pending further investigation ofthese limitations. However, we believe that we have shown thatthe trans-D Bayesian approach to inversion of HVSR data is aneffective approach to this highly nonlinear inverse problem that isworthy of further study, especially in deep sedimentary basins likeJakarta.

Dow



ay 2022


Figure 10. Considering velocity = 1300m s−1 as the basement, we can map the geometry of the basin. The basin depth ranges from 300 m in the southeast tomore than 1300 m in the west and east. Coloured circle is actual data point resulted from inversion of HVSR, while areas surrounded by deeper bedrock-depthis indicated by black dashed lines.

Figure 11. A drastic velocity change is apparent as velocity reaches about 1300m s−1. We consider this boundary to be the basement (white stripe).Considering this white stripe as basement, it can be concluded that in southern Jakarta the sediment thickness reaches 700 m and is the thickest sediment(1300 m) accumulated in the centre of Jakarta. A normal fault-like structure delimits the thinner sediment in the south from the thicker sediment in the north.Comparing between (a) HVSR and (b) ANT, in general, both methods resulted similar patterns, with the southern half of the city having significantly shallowerbasement than the northern half. An abrupt topographic change in basement depth appears in the centre of the cross-section. ANT and HVSR results weregridded using the simple kriging interpolation method. The location of the cross-section line (and its S–N direction) and measurement points are presented inFig. 2(a).

Dow



ay 2022

446 A. Cipta et al.

A C K N OW L E D G E M E N T S

Computations were performed on the Terrawulf cluster, a com-putational facility supported through AuScope and the AustralianGeophysical Observing System (AGOS). We used geopsy softwareto estimate the HVSR curve, the source code can be downloadedfrom http://www.geopsy.org/. Most of the figures are drawn usingGeneric Mapping Tools (Wessel & Smith 1998), python and QGIS(QGIS Development Team). We also thank Laurence Davies forcrucial advice on the use of the Wathelet[2005] code for HVSRcomputation. We wish to express our gratitude to Muriel Naguitfor comments that helped improve the manuscript. We would like tothank Andrea Berbellini and anonymous reviewer for their insightfulcomments and suggestions on the manuscript that led us to improveour work. Our revisions reflect all reviewers’ suggestions and com-ments. This work was partially supported by Australian Departmentof Foreign Affairs and Trades Grant 91982 and Australian ResearchCouncil (ARC) Linkage Grant LP110100525. AC was supported bya scholarship from the Indonesian Ministry of Energy and MineralResources (MAK 020.01.01.1881.002.001.012 A.521219).

R E F E R E N C E SAbidin, H.Z., Andreas, H., Gumilar, I., Fukuda, Y., Pohan, Y.E. & Deguchi,

T., 2011. Land subsidence of Jakarta (Indonesia) and its relation withurban development, Nat. Hazards, 18, 232–224.

Aki, K., 1957. Space and time spectra of stationary stochastic waves, withspecial reference to microtremors, Bull. Earthq. Res. Inst., 35, 415–456.

Albini, P., Musson, R.M.W., Gomez Capera, A.A., Locati, M., Rovida, A.,Stucchi, M. & Vigano, D., 2013. Global Historical Earthquake Archiveand Catalogue (1000-1903), GEM Technical Rep. 2013-01 V1.0.0, GEMFoundation, 202 p.

Arai, H. & Tokimatsu, K., 2000. Effects of Rayleigh and Love Waves onmicrotremor HVSR s, in The 12th WCEE, New Zealand.

Arai, H. & Tokimatsu, K., 2004. S-Wave Velocity Profiling by Inversion ofMicrotremor HVSR Spectrum, Bull. seism. Soc. Am., 94(1), 53–63.

Asten, M.W., 2006. On bias and noise in passive seismic data from finitecircular array data processed using SPAC methods, Geophysics, 7(6),V153–V162.

Asten, M.W., Dhu, T. & Lam, N., 2004. Optimised array design for mi-crotremor array studies applied to site classification–observations, resultsand future use, in The 13th WCEE, Paper No. 2903, Vancouver, B.C.,Canada.

Bayes, T., 1763. An Essay Towards Solving a Problem in the Doctrine ofChances, ed. Davis, C., Royal Society of London.

Berbellini, A., Morelli, A. & Ferreira, A.M.G., 2016. Ellipticity of Rayleighwaves in basin and hard-rock sites in Northern Italy, Geophys. J. Int., 206,305–407.

Berbellini, A., Morelli, A. & Ferreira, A.M.G., 2017. Crustal structure ofnorthern Italy from the ellipticity of Rayleigh waves, Phys. Earth planet.Inter., 265, 1–14.

Bodin, P., Smith, K., Horton, S. & Hwang, H., 2001. Microtremor obser-vation of deep sediment resonance in metropolitan Mamphis, Tennessee,Eng. Geol., 117(B2), 2156–2202

Bodin, T., Sambridge, M., Tkalcic, H., Arroucau, P., Gallagher, K. & Rawl-inson, N., 2012. Transdimensional inversion of receiver functions andsurface wave dispersion, J. geophys. Res., 62, 159–168.

Bonilla, L.F., Steidl, J.H., Lindley, G.T., Tumarkin, A.G. & Archuleta, R.J.,1997. Site amplification in the San Fernando Valley, California: variabilityof site-effect estimation using the S-wave, coda, and HVSR methods, Bull.seism. Soc. Am., 87(3), 710–730.

Bonnefoy-Claudet, S., Baize, S., Bonilla, L.F., Berge-Thierry, C., Pasten, C.,Campos, J., Volant, P. & Verdugo, R., 2009. Site effect evaluation in thebasin of Santiago de Chile using ambient seismic noise measurements,Geophys. J. Int., 176(3), 925–937.

Bonnefoy-Claudet, S., Cotton, F. & Bard, P.-Y., 2006. The nature of noisewavefield and its applications for site effects studies. A literature review,Earth-Sci. Rev., 79, 205–227.

Borges, J.F., Silva, H.G., Torres, R.J.G., Caldeira, B., Bezzeghoud, M., Fur-tado, J.A. & Carvalho, J., 2016. Inversion of ambient seismic noise HVSRto evaluate velocity and structural models of the Lower Tagus Basin, Por-tugal, J. Seismol., 20, 875–887.

Castellaro, S. & Mulargia, F., 2009a. Estimates of Vs30 based on constrainedH/V ratio measurements alone, in Increasing Seismic Safety by Combin-ing Engineering Technologies and Seismological Data, pp. 85–97, edsMucciarelli, M., Herak, M. & Cassidy, J., Springer.

Castellaro, S. & Mulargia, F., 2009b. VS30 estimates using con-strained HVSR measurements, Bull. seism. Soc. Am., 99(2A),761–773.

Castellaro, S., Padron, L.A. & Mulargia, F., 2014. The different responseof apparently identical structures: a far-field lesson from the Mirandola,Bull. Earthq. Eng., 12(5), 2481–2493.

Cipta, A., Cummins, P., Irsyam, M. & Hidayati, S., 2018. Basinresonance and seismic hazard in Jakarta, Indonesia, Geosciences,8(4).

Claprood, M., Asten, M.W. & Kristek, J., 2012. Combining HVSR mi-crotremor observations with the SPAC method for site resonance study ofthe Tamar Valley in Launceston (Tasmania, Australia), Geophys. J. Int.,191(2), 765–780.

Council on Tall Buildings and Urban Habitat (CTBUH), 2017. ‘Theskyscrapper center: the global tall building database of CTBUH’. Available at: http://www.skyscrapercenter.com/index.php/city/jakarta,accessed on 2017 August .

Cruz-Atienza, V.M., Tago, J., Sanabria-Gomez, J.D., Chaljub, E., Etienne,V., Virieux, J. & Quintanar, L., 2016. Long duration of ground mo-tion in the paradigmatic Valley of Mexico, Nat. Sci. Rep., 6(38807),doi:10.1038/srep38807.

Delinom, M.R., 2008. Groundwater management issues in the Greater Area,Indonesia. in Procedings: International Workshop on Integrated Water-shed Management for Sustainable Water Use in a HumidTropical Region,JSPS-DGHE Joint Research Project, Bull. TERC Univ. Tsukuba, 8(2),40–54.

Delinom, M.R., Abidin, H.Z., Taniguchi, M., Suherman, D., Lubis, R.F. &Yulianto, E., 2009. The contribution of human activities to subsurfaceenvironment degradation in Greater Jakarta Area, Indonesia, Sci. TotalEnviron., 407, 3129–3141.

Denolle, M.A., Miyake, H., Nakagawa, S., Hirata, N. & Beroza,G.C., 2014a. Long-period seismic amplification in the KantoBasin from the ambient seismic field, Geophys. Res. Lett., 41(7),2319–2325.

Dettmer, J., Dosso, S.E. & Holland, C.W., 2009. Model selection andBayesian inference for high-resolution seabed reflection inversion, J.acoust. Soc. Am., 125(2), 706–716.

Dettmer, J., Dosso, S.E. & Holland, C.W., 2010. Trans-dimensional geoacoustic inversion, J. acoust. Soc. Am., 128(6),3393–3405.

Dettmer, J., Molnar, S., Steineger, G., Dosso, S.E. & Cassidy, J.F., 2012.Trans-dimensional inversion of microtremor array dispersion data withhierarchical autoregressive error models, Geophys. J. Int., 188(2), 719–734.

D’Amico, V., Picozzi, M., Baliva, F. & Albarello, D., 2008. Ambi-ent noise measurements for preliminary site-effects characterizationin the urban area of Florence, Italy, Bull. seism. Soc. Am., 98(3),1373–1388.

Emporis, GmbH, 2017. Building Directory: Jakarta. Available at: https://www.emporis.com/city/100259/jakarta-indonesia, accessed on 2017 August2.

Esteva, L., 1988. Mexico earthquake of September 19, 1985–consequences,lessons, and impact on research and practice, Earthq. Spectra, 4(3), 413–426.

Ewald, M., Igel, H., Hinzen, K.-G. & Scherbaum, F., 2006. Basin-relatedeffects on ground motion for earthquake scenarios in the Lower RhineEmbayment, Geophys. J. Int., 166(1), 197–212.

Dow



ay 2022

http://www.geopsy.org/

http://www.skyscrapercenter.com/index.php/city/jakarta

https://www.emporis.com/city/100259/jakarta-indonesia


Field, E.H. & Jacob, K.H., 1993. A comparison and test of various site-response estimation techniques, including three that are not reference-sitedependent, Bull. seism. Soc. Am., 85(4), 1127–1145.

Flores-Estrella, H., Yussim, S. & Lomnitz, C., 2007. Seismic response of theMexico City Basin: a review of twenty years of research, Nat. Hazards,40, 357–372.

Furumura, T. & Hayakawa, T., 2007. Anomalous propagation of long-periodground motions recorded in Tokyo during the 23 October 2004 Mw 6.6Niigata-ken Chuetsu, Japan, Bull. seism. Soc. Am., 97(3), 863–880.

Furumura, T. & Kennett, B.L.N., 1998. On the nature of regional seismicphases III. The influence of crustal heterogeneity on the wave-field forsubduction earthquakes: the 1985 Michoacan and 1995 Copala, Guerrero,Mexico earthquakes, Geophys. J. Int., 135(3), 1060–1084.

Fah, D., Kind, F. & Giardini, D., 2001. A theoretical investigation of averageH/V ratios, Geophys. J. Int., 145(2), 355–549.

Fah, D., Kind, F. & Giardini, D., 2003. Inversion of local S-wave velocitystructures from average H/V ratios, and their use for the estimation ofsite-effects, J. Seismol., 7(4), 449–467.

Geopsy Team, 2004. ‘Geopsy source code ’. Available at: http://www.geopsy.org/, accessed on 2015 December .

Ghofrani, H. & Atkinson, G.M., 2014. Site condition evaluation usinghorizontal-to-vertical response spectral ratios of earthquakes in the NGA-West 2 and Japanese databases, Soil Dyn. Earthq. Eng., 67, 30–43.

Gorstein, M. & Ezersky, M., 2015. Combination of HVSR and MASWmethods to obtain shear wave velocity model of subsurface in Israel, Int.J. Geohazards Environ., 1(1), 20–41.

Gosar, A., 2010. Site effects and soil-structure resonance study in the Ko-barid basin (NW Slovenia) using microtremors, Nat. Hazards Earth Syst.Sci., 35, 415–456.

Green, P.J., 1995. Reversible jump Markov chain Monte Carlo computationand Bayesian model determination, Biometrika, 82, 711–732.

Gueguen, P., Cornou, C., Garambois, S. & Banton, J., 2007. On the limitationof the H/V Spectral Ratio using seismic noise as an exploration tool:application to the Grenoble Valley (France), a small apex ratio basin,Pure appl. Geophys., 164(1), 115–134.

Hall, R., 2009. Hydrocarbon basins in SE Asia: understanding why they arethere, Pet. Geosci., 15, 131–146.

Haris, R. & Major, J., 2016. Waves of destruction in the East Indies: theWichmann catalogue of earthquakes and tsunami in the Indonesian re-gion from 1538 to 1877, in Geohazards in Indonesia: Earth Science forDisaster Risk Reduction. pp. 9–46, eds Cummins, P.R. & Meilano, I.,Geological Society London, Special Publications.

Harutoonian, P., Leo, C.J., Tokeshi, K., Doanh, T., Castellaro, S., Zou, J.J.,Liyanapathirana, D.S. & Wong, H., 2013. Investigation of dynamicallycompacted ground by HVSR-based approach, Soil Dyn. Earthq. Eng., 46,20–29.

Hobiger, M. et al., 2013. Ground structure imaging by inversions of Rayleighwave ellipticity: sensitivity analysis and application to European strong-motion sites, Geophys. J. Int., 192(1), 207–229.

Ibs-von Seht, M. & Wohlenberg, J., 1999. Microtremors measurements usedto map thickness of soft soil sediments, Bull. seism. Soc. Am., 89, 250–259.

Iwaki, A. & Iwata, T., 2011. Estimation of three-dimensional boundaryshape of the Osaka sedimentary basin by waveform inversion, Geophys.J. Int., 186(3), 1255–1278.

Kagawa, T., Zhao, B., Miyakoshi, K. & Irikura, K., 2004. Modeling of 3Dbasin structures for seismic wave simulations based on available informa-tion on the target area: case study of the Osaka Basin, Japan, Bull. seism.Soc. Am., 94(4), 1353–1368.

Kanai, K., 1957. Semi-empirical formula for the seismic characteristics ofthe ground, Bull. Eearthq. Res. Inst., 35, 309–325.

Kawase, H. & Aki, K., 1989. A study on the response of a soft basin forincident S, P, and Rayleigh waves with special reference to the longduration observed in Mexico City, Bull. Seism. Soc. Am., 79(5), 1361–1382.

Kloosterman, F.H., 1989. Groundwater flow systems in the Northern CoastalLowlands of West- and Central Java, Indonesia, Dissertation at VrijeUniversity, Amsterdam, the Netherland.

Konno, K. & Ohmachi, T., 1998. Ground-motion characteristics estimatedfrom spectral ratio between horizontal and vertical components of mi-crotremor, Bull. seism. Soc. Am., 88(1), 228–241.

Lachet, C. & Bard, P.Y., 1994. Numerical and theoretical investigations onthe possibilities and limitations of Nakamura’s technique, J. Phys. Earth.,42(5), 377–397.

Langston, C.A. & Horton, S.P., 2004. Three-Dimensional seismic-velocitymodel for the unconsolidated Mississippi embayment sediments fromHVSR ambient noise measurements, Bull. seism. Soc. Am., 104(5), 2349–2358.

Lermo, J. & Chavez-Garcia, J., 1993. Are microtremors useful in the siteresponse evaluation? Bull. seism. Soc. Am., 84(5), 1350–1364.

Lermo, J. & Chavez-Garcia, J., 1994. Site effect evaluation usingspectral ratios with only one station, Bull. seism. Soc. Am., 83(5),1574–1594.

Longuet-Higgins, M.S., 1950. A theory of the origin of microseisms, Phil.Trans. R. Soc. Lond. A., 243(857), 1–35.

Lubis, R.F., Sakura, Y. & Delinom, R., 2008. Groundwater recharge and dis-charge processes in the Jakarta groundwater basin, Indonesia, Hydrogeol.J., 16, 927–938.

Lunedei, E. & Albarello, D., 2010. Theoretical HVSR curves from fullwavefield modelling of ambient vibrations in a weakly dissipative layeredEarth, Geophys. J. Int., 18(2), 1093–1108.

Magistrale, M., Day, S., Clayton, R.W. & Graves, R., 2000. The SCECSouthern California reference three-dimensional seismic velocity modelversion 2, Bull. seism. Soc. Am., 90(6B), S65–S76.

Malinverno, A., 2002. Parsimonious Bayesian Markov chain Monte Carloinversion in a nonlinear geophysical problem, Geophys. J. Int., 151(3),675–688.

Manakou, M.V., Raptakis, D.G., Chavez-Garcıa, F.J., Apostolidis, P.I. &Pitilakis, K.D., 2010. 3D soil structure of the Mygdonian basin for siteresponse analysis, Soil Dyn. Earthq. Eng., 30(11), 1198–1211.

Moechtar, H., 2003. Siklus Sekuens Stratigrafi Cekungan Kuarter Jakarta(Sequence Cycle of Quaternary Basins of Jakarta). Puslitbang Geologidan IAGI Pusat, Bandung, Indonesia.

Mosegaard, K. & Tarantola, A., 1995. Monte Carlo sampling of solutions toinverse problems, J. geophys. Res., 100(B7), 12 431–12 447.

Musson, R.M.W., 2012. A provisional catalogue of historical earthquakes inIndonesia, Open File Rep, OR/12/073, British Geological Survey.

Nakamura, Y., 1989. A method for dynamic characteristics estimation ofsubsurface using microtremor on the ground surface, Q. Rep. RTRI, 30(1),25–33.

Nata, T.P. & Witsen, B., 1700. A relation of the bad condition of the moun-tains about the Tungarouse and Batavian Rivers, having their source fromThence, occasioned by the earthquake between the 4th and 5th of Jan-uary, 1699. Drawn up from the account given by the Tommagon PorboNata, (who hath been there) and sent to the Burgermaster Witsen, whocommunicated It to the R. Society, of which he is a member, Phil. Trans.,22(260-276), 595–598.

Ng, A.H.-M., Ge, L., Li, X., Abidin, H.Z., Andreas, H. & Zhang, K., 2012.Mapping land subsidence in Jakarta, Indonesia using persistent scattererinterferometry (PSI) technique with ALOS PALSAR, Int. J. Appl. EarthObs. Geoinfo., 18, 232–242.

Nguyen, N., Griffin, J., Cipta, A. & Cummins, P., 2015. Indonesia’s historicalearthquakes: modelled examples for improving the national hazard map,Geoscience Australia, Record No. 2015/23, 79 p.

Nogoshi, M. & Igarashi, T., 1971. On the amplitude characteristics of am-bient seismic noise (Part 2), J. Seismol. Soc. Japan, 24, 26–40.

Okada, H., 2003. The Microtremors Survey Method, AGU, GeophysicalMonograph.

Olsen, K.B. et al., 2006. Strong shaking in Los Angeles expectedfrom southern San Andreas earthquake, Geophys Res. Lett., 33(7),doi:10.1029/2005GL025472.

Padilla y Sanchez, R.J., 1989. Geology and tectonics of the basin of Mexicoand their relationship with the damage caused by the earthquakes ofSeptember 1985, Intl. J. Min. Geol. Eng., 7(1), 17–28.

Pandey, B., Jakka, R.S., Kumar, A. & Mittal, H., 2016. Site characterizationof strong motion recording stations of Delhi using joint inversion of phase

Dow



ay 2022

http://www.geopsy.org/

448 A. Cipta et al.

velocity dispersion and HVSR curve, Bull. seism. Soc. Am., 106(3), 1254–1266.

Parolai, S., Picozzi, M., Richwalski, S.M. & Milkereit, C., 2005. Joint in-version of phase velocity dispersion and H/V ratio curves from seismicnoise recordings using a genetic algorithm, considering higher modes,Geophys. Res. Lett., 32(1), doi:10.1029/2004GL021115.

Pilz, M., Parolai, S., Picozzi, M., Wang, R., Leyton, F., Campos, J. & Zschau,J., 2010. Shear wave velocity model of the Santiago de Chile basin derivedfrom ambient seismic noise measurements: a comparison of proxies forseismic site conditions and amplification, Geophys. J. Int. 182(1), 355–367.

Poggi, V., Fah, D., Burjanek, J. & Giardini, D., 2012. The use of Rayleigh-wave ellipticity for site-specific hazard assessment and microzonation:application to the city of Lucerne, Switzerland, Geophys. J. Int., 188(3),1154–1172.

Putra, S.D.H. & Suryantini Srigutomo, W., 2016. Thermal modeling andheat flow density interpretation of the onshore Northwest Java Basin,Indonesia, Geotherm. Energy, 4(12), doi:10.1186/s40517-016-0052-x.

Reid, A., 2012. Historical evidence for past tsunamis in the Java subductionzone, Working Paper No. 178, Asia Research Institute.

Ridwan, M., 2016. Development of an engineering bedrock map beneathJakarta based on microtremor array measurements. Study of ground sub-surface in Jakarta by using microtremor array method: identification ofengineering bedrock depth and site class, Dissertation at Institut TeknologiBandung, Bandung, Indonesia.

Ridwan, M., Widiyantoro, S., Irsyam, M. & Afnimar Yamanaka, H., 2016.Development of an engineering bedrock map beneath Jakarta based onmicrotremor array measurements, Geol. Soc., London, Special Publica-tions, 7(1), 44.

Rivet, D., Campillo, M., Sanchez-Sesma, F., Shapiro, N. & Singh, S.K.,2015. Identification of surface wave higher modes using a methodologybased on seismic noise and coda waves, Geophys. J. Int., 203(2), 856–868.

Roten, D., Olsen, K.B., Day, S.M., Cui, Y. & Fah, D., 2014. Expected seismicshaking in Los Angeles reduced by San Andreas fault zone plasticity,Geophys. Res. Lett., 8, 2769–2777.

Salinas, V. et al., 2014. Using diffuse field theory to interpret the H/VSpectral Ratio from earthquake records in Cibeles seismic station, MexicoCity, Bull. seism. Soc. Am., 104(2), 995–1001.

Sambridge, M., Gallagher, K., Jackson, A. & Rickwood, P., 2006. Trans-dimensional inverse problems, model comparison and the evidence, Geo-phys. J. Int., 167(2), 528–542.

Savage, M.K., Lin, F.-C. & Townend, J., 2013. Ambient noise cross-correlation observations of fundamental and higher-mode Rayleigh wavepropagation governed by basement resonance, Geophys. Res. Lett.,40(14), 3556–3561.

Saygin, E., Cummins, P.R. & Lumley, D., 2017. Retrieval of the P wavereflectivity response from autocorrelation of seismic noise: Jakarta Basin,Indonesia, Geophys. Res. Lett., 44(2), 792–799.

Saygin, E. et al., 2016. Imaging architecture of the Jakarta Basin, Indonesiawith transdimesnional inversion of seismic noise, Geophys. J. Int., 204(2),918–931.

Scherbaum, F., Hinzen, K.-G. & Ohrnberger, M., 2003. Determination ofshallow shear-wave velocity profiles in the Cologne, Germany area usingambient vibrations, Geophys. J. Int., 152(3), 597–612.

SESAME Team, 2004. Guidelines for the implementation of the HV Spec-tral Ratio technique on ambient vibrations; measurements, processing,and interpretation, WP12 European Commission - Research General Di-rectorate Project No. EVG1-CT-2000-00026 SESAME, Report NumberD23.12.

Shapiro, N.M. & Campillo, M., 2004. Emergence of broadband Rayleighwaves from correlations of the ambient seismic noise, Geophys. Res. Lett.,31(7), doi:10.1029/2004GL019491.

Shapiro, N.M., Campillo, M., Stehly, L. & Ritzwoller, M.H., 2005. High-resolution surface-wave tomography from ambient seismic noise, Science,307(5715), 1615–1618.

Simandjuntak, T.O. & Barber, A.J., 1996. Contrasting tectonic styles inthe Neogene orogenic belts of Indonesia, Geol. Soc. Lond. Spec. Publ.,106(1), 185–201.

Simons, W.J.F. et al., 1997. A decade of GPS in Southeast Asia: re-solving Sundaland motion and boundaries, J. geophys. Res., 112(B6),doi:10.1029/2005JB003868.

Sivia, D. & Skilling, J., 2006. Data Analysis: A Bayesian Tutorial, OxfordUniversity Press.

Sanchez-Sesma, F.J. et al., 2011. A theory for microtremor H/V spec-tral ratio: application for a layered mediu, Geophys. J. Int., 186(1),221–225.

Tarabusi, G. & Caputo, R., 2016. The use of HVSR measurements forinvestigating buried tectonic structures: the Mirandola anticline, NorthernItaly, as a case study, Int. J. Earth Sci., 106(1), 341–353.

Turkandi, T., Agustiyanto, D.A. & Purbo-Hadiwidjojo, M.M., 1992. Geo-logic map of Jakarta and Kepulauan Seribu Quadrangles, Java, GRDC,Bandung, Indonesia.

Uebayashi, H., 2003. Extrapolation of irregular subsurface structures usingthe Horizontal-to-Vertical Spectral Ratio of long-period microtremors,Bull. seism. Soc. Am., 93(2), 570–582.

Uebayashi, H., Kawabe, H. & Kamae, K., 2012a. Temporal variation andreproduction of microseism HVSR spectral features in the Osaka Sedi-mentary Basin with an irregular-shaped interface, in Proceedings at the15th WCEE, Lisbon, Portugal.

Uebayashi, H., Kawabe, H. & Kamae, K., 2012b. Reproduction of micro-seism HVSR spectral features using a three-dimensional complex topo-graphical model of the sediment-bedrock interface in the Osaka sedimen-tary basin, Geophys. J. Int., 189(2), 1060–1074.

Viens, L., Koketsu, K., Miyake, H., Sakai, S. & Nakagawa, S., 2016. Basin-scale Green’s functions from the ambient seismic field recorded by MeSO-net stations, J. geophys. Res., 121(4), 2507–2520.

Wathelet, M., Jongmans, D. & Ohrenberger, M., 2004. Surface-wave in-version using a direct search algorithm and its application to ambientvibration measurements, Near Surf. Geophys., 2, 211–221.

Wichmann, A., 1918. Die Erdbeben des Indischen Archipels bis zum Jahre1857, J. Muller, Amsterdam.

Yamanaka, H., Dravinski, M. & Kagami, H., 1993. Continuous measure-ments of microtremors on sediments and basement in Los Angeles, Cali-fornia, Bull. seism. Soc. Am., 83(5), 1595–1609.

Yamanaka, H., Takemura, M., Ishida, H. & Niwa, M., 1994. Characteristicsof long-period microtremors and their applicability in exploration of deepsedimentary layers, Bull. seism. Soc. Am., 84(6), 1831–1841.

Zhao, J.X. et al., 2006. An empirical site-classification method for strong-motion stations in Japan using H/V Response Spectral Ratio, Bull. seism.Soc. Am., 96(3), 914–925.

Ozalaybey, S., Zor, E., Ergintav, S. & Tepirdamaz, M.C., 2011. Geology andtectonics of the basin of Mexico and their relationship with the damagecaused by the earthquakes of September 1985, Int. J. Min. Geol. Eng.,7(1), 17–28.

S U P P O RT I N G I N F O R M AT I O N

Supplementary data are available at GJI online.Figure S1. Results of a synthetic trans-D inversion in which randomnoise with standard deviation of 0.07 has been added to forwardcalculations of an HVSR curve calculated using a 10-layer model(white curve). The top-left panels (a), indicate log likelihood (top),number of layers (middle) and standard deviation (bottom), forindividual steps of the Markov chain used to sample the posteriorPDF. Each of the eight parallel Markov chains is assigned a distinctcolour, so it can be seen that each chain is sampling a consistentdistribution. (only every 100th step in each chain was saved, and halfthe initial steps were discarded as ‘burn-in’, leaving about 7000–10000 steps in each chain which are displayed). In (b), the syntheticdata used as the observed (black curve) HVSR is displayed alongwith the HVSR curve for the model associated with the maximumsampled posterior PDF (magenta curve). The lower panels displayhistograms of the sampled models for VS, VP/VS and density ρ in (c)–(e), respectively. The inset of (c) shows detail for the shallowmost

Dow



ay 2022

https://academic.oup.com/gji/article-lookup/doi/10.1093/gji/ggy289#supplementary-data


200 m of the VS profile. The white curve indicates the VS used toproduce the synthetic HVSR curve.Figure S2. As for Fig. S1, but the number of layers in the inversionis fixed to three. This results in velocity profile far from the initialmodel and the inversion cannot recover the ellipticity curve, hencethe noise level is overestimated (∼0.27 compared with the true valueof 0.07).Figure S3. As for Fig. S1, but in contrast to the three-layer modelof Fig. S2, a 20-layer model recovers the ellipticity curve well.However, the estimated velocity profile is much more complicatedthan the true one. The white curve indicates the velocity model usedto produce the synthetic HVSR curve.

Figure S4. Distribution of stations indicated as black dots whileblue dots indicate stations mentioned in the main text to make SNand WE cross-section plots and red diamonds are stations whichprofiles are shown in Fig. S5.Figure S5. Shear wave velocity profiles to the depth of 40 m ob-tained from HVSR, NSPT and SPAC techniques. Location of sta-tions is presented in Fig. 2Please note: Oxford University Press is not responsible forthe content or functionality of any supporting materials sup-plied by the authors. Any queries (other than missing mate-rial) should be directed to the corresponding author for thepaper.

Dow



ay 2022

Seismic velocity structure of the Jakarta Basin, Indonesia ...

Documents

Transcript of Seismic velocity structure of the Jakarta Basin, Indonesia ...