Seismic Tomographic Imaging of P Wave Velocity Perturbation Beneath Sumatra, Java, Malacca Strait, Peninsular Malaysia and Singapore

J. Earth Syst. Sci. (2021) 130:23 Ó Indian Academy of Sciences

https://doi.org/10.1007/s12040-020-01530-w (0123456789().,-volV)( 0123456789().,-vol V)

Seismic tomographic imaging of P wave velocity perturbation beneath Sumatra, Java, Malacca Strait, Peninsular Malaysia and Singapore

1, 2 ABEL UYIMWEN OSAGIE * and ISMAIL AHMAD ABIR 1Department of Physics, University of Abuja, P.M.B 117, Abuja, Nigeria. 2School of Physics, Universiti Sains Malaysia, 11800, Pulau Penang, Malaysia. *Corresponding author. e-mail: [email protected] MS received 6 March 2020; revised 20 July 2020; accepted 24 October 2020

P wave tomographic imaging of the crust down to a depth of 90 km is performed beneath the region encompassing Sumatra, Java, Malacca Strait, peninsular Malaysia and Singapore. Inversion is performed with 99,741 Brst-arrival p waves from 16,196 local and regional earthquakes occurred around the Sumatra Subduction Zone (SSZ) between 1964 and 2018. Tomographic results show low-velocity (low-V) anomalies that reCect both accretion and possibly, asthenospheric upwelling associated with subduction of the Australian Plate beneath Eurasia around the SSZ. The prominent low-V anomaly is thickest around the Conrad, extending beneath Straits of Malacca and parts of peninsular Malaysia, but disap- pears around the Moho in the region. Below the Moho, the subducting slab, represented by a high-velocity (high-V) anomaly, trends in the orientation of Sumatra. At these depths, the eastern shorelines of Sumatra, most parts of Malacca Strait and the west coast of peninsular Malaysia show varying degrees of positive velocity anomalies. We consider that asthenospheric upwelling around the SSZ may provide heat source for the 40 or more hot springs distributed north–south in peninsular Malaysia. Different east–west and north–south cross-sections reveal the subsurface anomalies at various parts of the region. The predominant low-V anomaly is less than 35 km in depth, but other low-V anomalies are deeper. Keywords. Seismic tomography; regional; p waves; Brst-arrivals; Sumatra Subduction Zone.

1. Introduction apparent around Sumatra, but seldom destructive around countries like Malaysia and Singapore The Sumatra Subduction Zone (SSZ) is located which are located on the stable Indochina–Sunda- along where the Indo-Australia Plate subducts land. As a result, there is little appetite for deep beneath the Eurasian Plate (Bgure 2, green and structural investigations. Although located on the brown lines). The Brst record of an earthquake in stable Sunda continental shelf, seismic activities the archive of the bulletin of International Seis- around the SSZ have historically aAected these mological Center (ISC) dates back to 1907. The other countries. absence of seismic stations at that time is evident Since the 1970s, many large earthquakes have in the scanty dataset prior to the mid-1970s. occurred around the SSZ. Some earthquakes, e.g., However, increased global seismic networks in the the 1976 (mb 6.4), 2004 (Mw 9.15), 2005 (Mw 8.6) 1970s and 1980s provided more datasets for the and 2007 (Mw 8.4) aAected areas far beyond their region. The devastating eAects of earthquakes is epicenters. For example, the 26th December 2004 23 Page 2 of 13 J. Earth Syst. Sci. (2021) 130:23

(Mw 9.15) earthquake triggered a tsunami that tomographic inversion requires four major aAected not only Indonesia, but also Sri Lanka, computational stages – model parameterization, Thailand and Malaysia which killed thousands of forward calculation, inversion and analysis of people in its wake (Komoo and Othman 2005). the model resolution (e.g., Rawlinson and The event set the scene for a large-scale eAort by Sambridge 2003;Zhao2015). In other words, a the Malaysian Government to set up the Malay- tomographic inversion algorithm is a suite of sian National Tsunami Early Warning System different routines that usually contains the (MMD 2012). By causing deformation of the above components or more. For example, Sundaland core as a result of intraplate stress tomog3D (Zhao 2004) uses pseudo-bending ray build-up after the event, previously inactive faults tracing technique for the forward calculation have been reactivated within peninsular Malaysia stage and the LSQR algorithm (Paige and (Shuib 2009). Saunders 1982) in the inversion stage. The Global one-dimensional (1D) reference velocity painstaking task of understanding the data models are good approximations for the purpose of preparation stage for use in tomog3D is com- locating earthquakes, but insufBcient to satisfy pounded by the need to develop a technique to velocity variations at local and regional distances. visualize the Bnal output. This challenge Three-dimensional (3D) seismic investigation of informed the decision to develop an inversion the crust and upper mantle will not only provide scheme. Afterall, every inversion is validated by better understanding about the regional p wave a test of model resolution. velocity distribution within the SSZ, but also for In this work, a ray tracing scheme in 3D is the surrounding regions. Seismic tomography pro- developed using the idea behind an earlier 1D vides the ability to obtain cross-sectional images of scheme (Kim and Baag 2002) which has been used the subsurface due to variations in rock properties. by some authors (e.g., Kang et al. 2013; Osagie At crustal dimensions, variations in p wave veloc- and Woohan 2013; Hong et al. 2017; Osagie et al. ities can be mapped to variations in rock properties 2017). Accuracy and speed (fast convergence rate) from which subsurface stratigraphy or anomalies in the 3D ray tracing scheme is germane to suc- can be deduced. At shallow depths, information cessful inversion. The ray tracing scheme con- about the subsurface structure and strength char- verges rapidly, saving computational time and can acteristics (e.g., bulk modulus, shear modulus and accommodate a model with as many layers as density) can be useful for seismic hazard assess- required for traveltimes computation at regional, ments. This can provide useful information in the local or Beld work subsurface investigations. The construction of infrastructural projects (e.g., tomographic inversion scheme incorporates the bridges, tunnels, dams and sky-scrapers). The LSQR algorithm with enhancement by damping ultimate goal of seismic tomography is to show 3D and smoothing (Zhao 2004). The scheme can read variation in seismic velocity within a model. direct arrival time downloads from the bulletin of Whether the variations observed is due to International Seismological Center (BISC), sum- (a) temperature, (b) composition or (c) physical mary Ble from the Incorporated Research Insti- properties is an open debate (Rawlinson et al. tutions for Seismology (IRIS) and s-Ble from 2014). SEISAN software (Havskov and Ottemoller 1999). The aim of this study is to image the sub- Apart from information about the regional p wave surface of the region using Brst-arrival p wave velocity distribution, the result of this study will data from local and regional earthquakes that provide a platform for near-surface tomographic have occurred within and around the SSZ. studies in the surrounding regions. However, this Compared to later phases, Brst-arrival p waves study is limited in scope to local and regional are easier to identify and are often the best seismic tomography. Traveltime accuracy of the dataset available for local and regional tomo- ray tracing algorithm is limited to epicentral dis- graphic studies. More so, the use of Brst-arrival tances of about 1200±200 km. This is due to the p waves diminishes concerns about signal-to- shooting method of ray tracing used for the for- noise ratio associated with the use of later ward calculation. At epicentral distances beyond arrivals with amplitudes comparable to the 1400 km, the estimated takeoA angle for crustal background noise. To achieve the aim of this events are very close to 90°, at which takeoA study, a seismic tomographic inversion angle, rays cannot computationally get to scheme is developed. In general, seismic stations. J. Earth Syst. Sci. (2021) 130:23 Page 3 of 13 23

2. Earthquake data about 7 cm. The SFS is among the world’s major strike-slip faults with different convergence rate for The data used in this work is obtained from the different segments of the fault (Molnar and Dayem IRIS and the BISC with the bulk of the data being 2010). The Sumatra shear (dextral strike slip) fault arrival time picks from the BISC. The earthquakes is approximately 1500 km, cutting through the (magnitude C 3.0 mb) are conBned to the top 100 entire Sumatra Island. The mechanism for the km in focal depths, latitude ranges from 10°Nto Sumatra shear zone is an open debate. Shuib 10°S and longitude spans 93°–115°E. In 2018, this (2009) proposed extrusion resulting from collision study obtained all available arrival time data from between India and Eurasia, while Hutchison (2010) the BISC between 1964 and 2016. The BISC data believes Oroclines mechanism is responsible. The ended in 2016 as at the time of data collection for fundamental difference in their proposed mecha- this study. Review of the BISC data is usually nism is in relative plate motion and velocity of about 24 months behind time because arrival times subduction. The subduction velocity is less than are manually checked by analysts at the BISC. the plate velocity, resulting in accretion and Later in 2019, the revised BISC dataset until mid- shearing. 2017 was added to the dataset. Waveform data According to the report of the Malaysian from the IRIS free online data archive between Meteorological Department (MMD and ASM 2006 and 2018 are retrieved with a software 2009), the tectonic framework for the whole of (JWEED) developed by seismology department at Malaysia covers between longitudes 90°–140°E the University of South Carolina (USC 2012). This and latitudes of 20°–12°S. The study area is was the period of good waveform data during the within this framework. The nearest segment of the time of data collection. Time window of the same SFS is about 300 km away from Singapore size is taken before and after the Brst-arrival onset (Macpherson et al. 2013) which is located at the of p wave. The mean amplitude for both is calcu- southern end of peninsular Malaysia. The study lated and a signal-to-noise ratio (above 3.0) is area also falls within the Malay Basin, which selected for each waveform. Hand-picked phases according to Tjia (2010) originated in the Late from the IRIS waveforms between 2017 and 2018 Cretaceous as a major aulacogen on the Malay are added to the dataset from the BISC to obtain dome and developed structurally through modiB- a total of 16,196 events with 109,260 Brst-arrival cations by differential extrusion of Indosinian p waves. crustal slabs. Tjia (2010) also provided evidence that suggests occasional crustal movements in parts of Sundaland. The 2004 (Mw 9.15) earth- 3. Tectonic setting of the study region quake disturbed the surrounding plate and deformed the core of Sundaland causing both co- The region under investigation covers peninsular seismic and post-seismic deformations for the Malaysia, Singapore, Straits of Malacca, Sumatra whole of southeast Asia (Marto et al. 2013). By and Java in Indonesia (Bgure 2). The predominant causing deformation of the Sundaland core as a tectonic structures in the area are the SSZ which is result of intraplate stress build up after the event, part of a longer Sunda Trench and the Sumatran it is reported to have reactivated previously Fault System (SFS). The SFS runs parallel to the inactive faults within peninsular Malaysia (Shuib SSZ (Bgure 2) and consists of 19 segments (Huchon 2009). These are a recipe for more tectonic events and Le Pichon 1984; Sieh and Natawidjaja 2000; around the region. Irsyam et al. 2008; Lin et al. 2009). The segments More than 40 hot springs with NNW–SSE are separated by dilatational and contractional tectonic trend identiBed in peninsular Malaysia are step-overs and abrupt changes in trend (Sieh and associated with tectonic activities (Samsudin et al. Natawidjaja 2000). The SSZ is an example of a 1997). Baioumy et al. (2015) identiBed 20 addi- partitioned subduction zone, where much of the tional hot springs with similar trend. The origins of trench-normal plate motion is accommodated by heat sources of the springs are unclear. Both the megathrust plate boundary, but much of the Samsudin et al. (1997) and Baioumy et al. (2015) trench-parallel motion is accommodated by the have proposed different models to explain the heat SFS (AmalB et al. 2016). According to Huchon and distribution mechanism for the hot springs. These Le Pichon (1984), the SSZ has a convergence models may be difBcult to validate without the direction of N10°E with a yearly convergence of means to image the subsurface. 23 Page 4 of 13 J. Earth Syst. Sci. (2021) 130:23

4. Methodology parameterization in this study is driven by conve- nience and computational capabilities. The research follows the standard tomographic processes of model parameterization, forward cal- 4.2 Forward traveltime calculation in 3D culation, inversion and analysis of the solution (e.g., Rawlinson and Sambridge 2003; Zhao 2015). The station distribution spans latitudes 10°N–10°S and longitudes 93°–115°E. A total of 132 seismic 4.1 Model parameterization stations are used for this study. Earthquake epicenters are distributed over an area approximately The iasp91 (Kennett and Engdahl 1991) 1D model 1,970,000 km2 which is roughly the size of Mexico (Bgure 1) with a Moho depth of 35 km is used to (Bgure 2). The hypocentral parameters reported by construct a 3D grid by adapting the routine of both the BISC and IRIS are used to compute Zhao (2001). Latitude range from 10°N–10°S, while traveltimes and raypaths. The iasp91 model is used longitude is from 93°–115°E. The grid spacing is set as the initial velocity model. The 3D ray tracing at 1° in both latitude and longitude directions. In scheme which is based on two-point ray shooting depth, 13 grid mesh layers (1, 2, 5, 10, 15, 20, 25, method is used for the forward calculation. Travel- 30, 35, 40, 50, 70 and 90 km) are considered. The times and raypaths for direct and refracted rays are model dimensions are 20923913 nodes (yellow computed separately within the scheme. Rays spheres in Bgure 3) along the latitude, longitude refracted from existing boundaries are constrained and depth directions, respectively. The spacing is to bottom in each layer between interfaces. The for computational considerations and also intended scheme selects the raypath with the shortest travel- to make each grid cell large enough to record the time between direct and refracted rays. A vari- minimum number of ray segments (hit-count) able is deBned to provide information (latitude, required for inclusion in the inversion process. The longitude, depth, velocity, time from source, ray- velocity perturbation at any point within a grid cell path length from source and the angle of ray travel) is obtained from trilinear interpolation of pertur- at various points along the raypath. The various bations at the eight grid nodes surrounding that points along raypaths are determined by a prede- point (Zhao 2001). The method of parameteriza- Bned value of step-length from the earthquake tion can impose limitations on the kinds of sub- source to stations. Figure 3 shows 4,693 grid nodes surface structure. Rawlinson et al. (2014) argues (yellow spheres), earthquake hypocenters (red that the choice of parameterization can sometimes spheres), seismic stations (blue triangles), raypaths lead to the appearance of artefacts instead of for Brst-arrival direct p waves (black spots) and the target subsurface structures. The choice of refracted p waves (green spots) within the model volume (please note that the depth axis in Bgure 3 is about 55 times exaggerated compared with the distance dimension of latitude and longitude values).

4.3 Synthetic test in 3D

The Bnal model will be subjected to scrutiny without analyzing the robustness of the resolution. Synthetic test reconstructs a test-model using the same source–receiver conBguration as the model to be determined. The so-called checkerboard resolution test (CRT) is a commonly used method for synthetic tests. It consists of an alternating pattern of positive and negative anomalies relative to the background model. The resolution limits of the dataset are inferred from differences between the reconstruction and the known structure. In this Figure 1. iasp91 one-dimensional velocity model. study, the CRT (Zhao et al. 1992;Leveque et al. J. Earth Syst. Sci. (2021) 130:23 Page 5 of 13 23

Figure 2. Bathymetry/topography map of the study area showing the distribution of 16,196 earthquake epicenters (circles increasing with magnitude between Mw 3.0 and 7.8) and 132 seismic stations (blue triangles).

1993) is used to create alternating positive and partly to control the colour spectrum of velocity negative velocity perturbations of 4% at selected variation within the model space. depths in both horizontal and vertical directions (Bgure 4). Random error with a standard deviation of 0.1 sec is added to the synthetic data. The result 4.4 3D inversion is expected to provide information about spatial resolution of the study area. According to Zhao Inversion is carried out with only Brst-arrival et al. (1992), the lengths of the segments in the p waves. The Brst-arrival s waves are insufBcient upper and the lower crust are only about several and less reliable. A total of 99,741 Brst-arrival tens of kilometers and the velocity perturbation is p waves met the traveltime residual threshold of about 6–7% for p wave and about 10% for s wave. ±2.5 sec. The LSQR algorithm (Paige and Saun- Different perturbation values are used depending ders 1982) is used to solve the large and sparse on the author’s preference. For example, Van der system of equations associated with traveltimes Hilst et al. (1997) used 0.5%, Huang and Zhao and the unknown velocity parameters. Damping (2006) used 2%, Tian et al. (2009) used 3% and and smoothing (Zhao 2004) is also applied. Differ- Rawlinson et al. (2010) used 6%. This study uses ent damping factors is tested and a damping factor 4% perturbation to p wave velocity. The decision is of 30.0 was best suited for the tradeoA between due partly to computational considerations and data variance reduction and smoothness of the 23 Page 6 of 13 J. Earth Syst. Sci. (2021) 130:23 obtained 3D velocity model. A minimum of 10 5. Results and discussion observations (hit-count) within any grid cell is used in the inversion. Stations in peninsular Malaysia 5.1 Result of checkerboard test and Java have less azimuthal coverage compared with the stations in Sumatra, but there is com- Figure 5 shows the results of CRT at 12 represen- pensation by the relatively large number of arrival tative depths when a 1° 9 1° grid is adopted at times. ±4% perturbation. Black and white squares denote slow and fast velocities respectively with respect to the absolute velocity values. The inverse problem is solved for velocity perturbations at grids cell with 10 or more hit-counts. Given the azimuthal coverage, the resolution appears good at most depths beneath Sumatra, West Java, Malacca Strait, peninsular Malaysia and Singapore. How- ever, at 1 km, there is no resolution beneath peninsular Malaysia. Also, there is no resolution above latitude 6°N around the border between Thailand and Malaysia, and most parts of East Malaysia. The situation is the same between longitudes 109°–111°E including central and East Java, although resolution appears good in Malang and Surabaya. These are due either to lack of stations or insufBcient raypath information to perform inversion. Therefore, the actual inversion results may show similar limitations.

Figure 3. Grid nodes (4,693 yellow spheres) within the model 5.2 Result of inversion volume, earthquake hypocenters (red spheres), seismic stations (blue triangles), raypath for Brst-arrival direct p waves Tomographic maps beneath the study region at (black spots) and refracted p waves (green spots). Raypaths different depths (Bgure 6) show both low velocity are calculated using the iasp91 model (depth axis is 55 times (low-V) and high velocity (high-V) regions. At exaggerated). 1 km depth slice (Bgure 6a), a low-V anomaly is

Figure 4. Initial checkerboard model with a 1° 9 1° grid space at ± 4% perturbation. (a) For depths at 1, 5, 15, 25, 35, 50 and 90 km and (b) for depths at 2, 10, 20, 30, 40 and 70 km. J. Earth Syst. Sci. (2021) 130:23 Page 7 of 13 23

Figure 5. Results of checkerboard resolution tests at 12 representative depths (depth values above each Bgure) with a 1° 9 1° grid at ± 4% perturbation. Black and white squares denote low and high velocities relative to the absolute velocity values. observed around the forearc region, west of the thinner at 5 and 30 km when compared with 20 km northernmost part of Sumatra. This may be due to may be due to asthenospheric upwelling associated the formation of accretionary prism as a result of with subduction of the oceanic Australia plate the oceanic Australia subducting plate beneath beneath the Eurasian plate around the SSZ. The that region of SSZ. Depth slices down to 30 km low-V appears to extend as far as peninsular (Bgure 6b–g) show NW–SE trending low-V Malaysia and Singapore (Bgure 6e). We consider anomalies beneath most parts of Sumatra Island the possibility of diapirism (the rise of less dense and oA the western shores. The anomaly which is rocks through buoyant forces) at the upper crust. 23 Page 8 of 13 J. Earth Syst. Sci. (2021) 130:23

Figure 6. P wave velocity images at 12 representative depths (depth values above each Bgure) with 1° 9 18 grid at ±4% perturbation. Red and blue colours denote low and high velocities relative to the absolute velocity values.

Marschall and Schumacher (2012) have proposed trend observed at shallower slices (Bgure 6a–g) migration of diapirs from the surface of a sub- appear to truncate before the 35 km depth slice ducting slab which can be transported upwards. (Bgure 6h). This gives an indication that the Moho The diapiric upwelling model of subduction inves- is somewhere between 30 and 35 km. From 35 km tigated by Hall and Kincaid (2001) suggests the downwards (Bgure 6h–l), the subducting Aus- creation of low-density conduits that allow buoy- tralian plate is represented by a high velocity ant Cow. The anomalous velocity distribution (high-V) anomaly trailing the shorelines of J. Earth Syst. Sci. (2021) 130:23 Page 9 of 13 23

Sumatra and Java and between the SSZ and SFS. low-V anomaly and the SSZ from 6°Nto8°S(Bgure 7). However, a low-V anomaly is observed around the The EWVC at 6°N, beneath the Island of east coast of Sumatra at 35, 40 and 50 km depths. We (Bgure 7a), the oceanic slab is imaged by the Fifteen east-west vertical cross-sections (EWVC) high-V at longitudes\96°E and a possible astheno- show the movement from left to the right of the spheric upwelling is represented by the low-V anomaly.

Figure 7. P wave velocity images at 15 east–west vertical cross-sections (latitude values below each Bgure). Red and blue colours denote low and high velocity anomalies relative to the absolute velocity values. Ac (Aceh city), Ba (Bawa Island), BBI (Bangka Belitung Islands), CJ (Central Java), CS (Central Sumatra), Is (Some Islands), K (Kluang), L (Lingga Island), M (Masa Island), Ni (Nias Island), PM (Peninsular Malaysia), SbI (Siberut Island), SM (Straits of Malacca), SpI (Sipura Island), SS (South Sumatra), ST (Sunda trench), SZ (seismogenic zone), TC (Tanjung Balai city), We (We Island), WJ (West Java), WS (West Sumatra). 23 Page 10 of 13 J. Earth Syst. Sci. (2021) 130:23

Figure 8. P wave velocity images at 15 north–south vertical cross-sections (longitude values below each Bgure). Red and blue colours denote low and high velocities relative to the absolute velocity values. Ac (Aceh city), AS (Andaman Sea), CB (Campell Bay), Islands (e.g., Lekon, Simeulue Tjut and Tapan), PM (Peninsular Malaysia), SM (Straits of Malacca), SS (South Sumatra), ST (Sunda trench), SZ (seismogenic zone), WJ (West Java), WS (West Sumatra).

Moving southwards, the EWVC at 5° and 4°N Aceh (Ac, Bgure 7c and d) and the northern part of (Bgure 7b and c), the low-V anomaly is probably due peninsular Malaysia (PM, Bgure 7d). About 100 km to accretionary prism or/and upwelling beneath due south the EWVC at 3°N shows high-V structure J. Earth Syst. Sci. (2021) 130:23 Page 11 of 13 23

(Bgure 7d) beneath Straits of Malacca (SM) and sediment accretion possibly containing remnant close to the eastern shores of Tanjung Balai city fragments of oceanic crust (Kopp et al. 2002). (TC). The EWVC at 2°N(Bgure 7e) shows upwel- Around here, the Australian plate subducts at a ling from a depth of about 40 km which appears to rate of 67 mm/yr in a direction N11°E(Kopp reach the surface. This may be responsible for et al. 2002). distribution of the small islands (Is) like Lasia, Babi, Fifteen north–south vertical cross-sections Bankaru, Tuangku and Udjung Batu around that (NSVC) show a similar trend in the low-V distri- location. Another upwelling between longitudes bution (Bgure 8). The NSVC at longitude 94°E 100°–104°E(Bgure 7e), cutting across North (Bgure 8a) shows a high-V between 40 and 60 km Sumatra (NS), Straits of Malacca (SM) and Kluang in depth which may be responsible for the upwel- (K) in the southern part of peninsular Malaysia ling represented by the low-V beneath Campell appears to start at depths greater than 90 km, but Bay (CB) oA the northern shores of North Suma- stops below the Moho. Starting from about 96°E, the tra. About 100 km due east (Bgure 8b) towards EWVC at 1°N(Bgure 7f) shows a low-V suspected to Aceh (Ac), the high-V anomaly appears to descend be due to upwelling reaching the surface. This may deeper. The high-V northwards from about lati- be responsible for the Islands of Nias (Ni) and Bawa tude 5°N is part of the oceanic Coor beneath (Ba) among other Islands around that coordinate. Andaman Sea (Bgure 8b, AS). Some islands like The cross-section (Bgure 7f) represents some struc- the Lekon, Simeulue Tjut and Tapan (together tural heterogeneity which is associated with the illustrated as Islands in Bgure 8c) may be the locked portion of the subduction interface where consequence of the upwelling represented by the the 2002 (Mw 7.2), 2008 (Mw 7.3) and 2010 (Mw 7.2, low-V anomaly in the NSVC at longitude 96°E. Mw 7.8) ruptures have occurred (Philibosian et al. The NSVC at 96°–98°E(Bgure 8c–e) from latitudes 2014). 5°N to the equator show a similar pattern and The EWVC at the equator (Bgure 7g) shows the evidence of upwelling that manifests in the outcrop low-V anomaly stretching between longitudes of many islands. The NSVCs (Bgure 8f and g) also 97°–105°E cutting through west Sumatra (WS) show a cause for the upwelling observed in that which may be responsible for islands like Masa location. (M) and Lingga (L), respectively, west and east of From the NSVC longitude 101°E(Bgure 8h) Sumatra. The pattern is almost similar for the beneath peninsular Malaysia (PM) (latitudes EWVC at 1°–5°S(Bgure 7h–l), central to south 6°–3°N), a subsurface low-V anomaly is imaged at Sumatra (CS, SS), approximately 500 km due the crust with the deepest point at about 35 km from south, leaving in its wake, islands like Siberut (SbI, the surface. Asthenospheric upwelling may be the Bgure 7h), Sipura (SpI, Bgure 7i), Siruamata and source of heat to 40 or more hot springs with a Pagai-selatan (Bgure 7h) to the west of Sumatra. north–south distribution from the Kedah in the Islands to the east of Sumatra which may be the north to Johor in the southern part of the peninsular result of upwelling include Bangka Belitung Islands (Samsudin et al. 1997; Baioumy et al. 2015). The (BBI, Bgure 7l) among several others. With an low-V observed beneath peninsular Malaysia may average depth of about 80 km, the EWVC at be due to lower densities as a result of heat. How- 6°S(Bgure 7m) shows one of the deepest roots ever, there appears to be discontinuity (Bgure 8h) ([80 km) of the low-V anomaly at the northern tip beneath Straits of Malacca (SM) after which a dee- of west Java (WJ). The EWVC at 7°S(Bgure 7n) per and wider low-V distribution which may be cuts across west and central Java (WJ, CJ). responsible partly for the formation of western Between longitudes 102° and 104°E(Bgure 7n), is Sumatra (WS) is observed. The NSVC at longitude part of Sunda trench (ST) (Molnar and Dayem 102°E(Bgure 8i) shows no evidence of the low-V 2010), where the upwelling is obvious. Between anomaly beneath peninsular Malaysia. In light of longitudes 106° and 109°E(Bgure 7n) is west Java this observation, this study suggests a possible end (WJ) and central Java (CJ) beneath which the to the horizontal spread of the upwelling around the deepest part of the upwelling may be more than middle part of peninsular Malaysia. However, low-V 50 km below the surface. The EWVC at 8°S is apparent beneath southern Sumatra (SS) at this shows a high-V anomaly between 106° and 109°E cross-section. Again, the southward movement of which is below the southern shoreline of west the low-V anomaly (Bgure 8j–o) as the slice progress Java (Bgure 7o), representing the seismogenic eastwards suggest that upwelling starts at the zone (SZ) (Krabbenhoft€ et al. 2010)with uppermost part of the mantle and continues into 23 Page 12 of 13 J. Earth Syst. Sci. (2021) 130:23 the upper crust in many cases. The seismogenic zone Acknowledgements (SZ) below west Java (WJ) is represented by the blue region (Bgure 8n and o). The authors thank Kim Woohan for the use of his two-point ray tracing codes which was mod- iBed in a subroutine within our algorithm. The 6. Conclusion original codes by Zhao Dapeng was modiBed and used in the model parameterization and travel- Seismic traveltime tomographic inversion of time derivative stages of the inversion process. p wave velocity is obtained beneath the region The authors acknowledge the Bulletin of Inter- covering the Island of Sumatra and Java (In- national Seismological Center (ISC) from where donesia), Malacca Strait, peninsular Malaysia and most part of the arrival time data used in this Singapore. Arrival time record from 132 seismic study is obtained. Also, the data services of stations within the region are obtained from two Incorporated Research Institutions for Seismology data sources: Incorporated Research Institutions (IRIS) were used for access waveforms and for Seismology and the Bulletin of International earthquake catalog which form part of the data Seismological Center. The earthquakes (magni- used in this study. tude C3.0 mb) are conBned to the top 100 km in focal depths, latitude ranges from 10°N–10°S and longitude spans 92°–115°E. A total of 16,196 local Author statement and regional earthquakes are selected. The model grid spacing is set at 1° in both latitude and Osagie Abel: Conceptualization, data curation, longitude directions with 13 depth layers. This methodology, software, visualization, and writing- resulted in a total of 4,693 grid nodes. After reviewing. Ismail Abir: reviewing. selection, 99,741 Brst-arrival p waves met the traveltime residual threshold of ±2.5 sec. The LSQR algorithm with damping and smoothing is References applied to the computed traveltimes for inversion. The results of checkerboard resolution test AmalB O, Phil C, David R and Sri H 2016 Sensitivity analysis provided sufBcient information to support the for probabilistic seismic hazard analysis (PSHA) in the reliability of the inversion. Aceh Fault Segment, Indonesia; Geol. Soc. London, Spec. Tomographic results show prominent low-veloc- Publ. 441 121–131. ity (low-V) anomalies that reCect accretion and Baioumy H, Nawawi M, Wagner K and AriBn M H 2015 possibly asthenospheric upwelling associated with Geochemistry and geothermometry of non-volcanic hot springs in West Malaysia; J. Volcanol. Geotherm. Res. 290 12–22. subduction of the oceanic Australian plate beneath Hall P S and Kincaid C 2001 Diapiric Cow at subduction the Eurasian plate around the Sumatra subduction zones: A recipe for rapid transport; Science 292(5526) zone. The prominent low-V anomaly trends south- 2472–2475. east beneath Sumatra Island and appears to have its Havskov J and Ottemoller L 1999 SEISAN earthquake widest horizontal spread around the Conrad (20 km analysis software; Seismol. Res. Lett. 70(5) 532–534. depth). At this depth, the spread extends beneath Hong T K, Lee J, Kim W, Hahm I K, Woo N C and Park S 2017 The 12 September 2016 ML5. 8 midcrustal earthquake Straits of Malacca and parts of peninsular Malaysia. in the Korean peninsula and its seismic implications; However, the low-V anomaly was not imaged at 35 Geophys. Res. Lett. 44(7) 3131–3138. km, suggesting a source above the Moho. Although Huang J and Zhao D 2006 High-resolution mantle tomography low-V anomalies are observed from east–west and of China and surrounding regions; J. Geophys. Res. north–south vertical cross-sections, these are not 111(B9) B09305, 1–21. Huchon P and Le Pichon X 1984 Sunda Strait and central predominant. This research associates the low Sumatra fault; Geology 12(11) 668–672. velocity distribution observed beneath peninsular Hutchison C S 2010 Oroclines and paleomagnetism in Borneo Malaysia to lower density due to heat which appears and south-east Asia; Tectonophys. 496(1–4) 53–67. to be connected to the subduction activities Irsyam M, Dangkua D T, Hoedajanto D, Hutapea B M, around the Sumatra Subduction Zone. The Kertapati E K, Boen T and Petersen M D 2008 Proposed authors consider asthenospheric upwelling as a seismic hazard maps of Sumatra and Java Islands and microzonation study of Jakarta city, Indonesia; J. Earth possible source of heat to 40 or more hot springs Syst. Sci. 117(2) 865–878. with north–south distribution within peninsular Kang Y A, Kim W and Kang T S 2013 Relocation of Malaysia. earthquakes beneath the East Sea of Korea: Uncertainty of J. Earth Syst. Sci. (2021) 130:23 Page 13 of 13 23

hypocentral parameters caused by refracted waves; Geosci. Osagie A U and Woohan K 2013 Relocation of earthquakes J. 17(2) 173–182. that occurred beneath ParkBeld Region of California using Kennett B L N and Engdahl E R 1991 Traveltimes for global VELHYPO; IOSR J. Appl. Geol. Geophys. 1(3) earthquake location and phase identiBcation; Geophys. 66–81. J. Int. 105(2) 429–465. Paige C C and Saunders M A 1982 LSQR: An algorithm for Kim W and Baag C E 2002 Rapid and accurate two-point ray sparse linear equations and sparse least squares; ACM tracing based on a quadratic equation of takeoA angle in Trans. Math. Softw. 8(1) 43–71. layered media with constant or linearly varying velocity Philibosian B, Sieh K, Avouac J P, Natawidjaja D H, Chiang functions; Bull. Seismol. Soc. Am. 92(6) 2251–2263. H W, Wu C C, Perfettini H, Shen C, Daryono M R and Komoo I and Othman M 2005 The 26.12.04 Tsunami Disaster Suwargadi B W 2014 Rupture and variable coupling in Malaysia: An Environmental, Socio-economic and Com- behavior of the Mentawai segment of the Sunda megathrust munity Well-being Impact Study; Institut Alam Sekitar dan during the supercycle culmination of 1797 to 1833; J. Pembangunan. Geophys. Res. 119(9) 7258–7287. Kopp H, Klaeschen D, Flueh E R, Bialas J and Reichert C Rawlinson N and Sambridge M 2003 Seismic traveltime 2002 Crustal structure of the Java margin from seismic tomography of the crust and lithosphere; Adv. Geophys. wide-angle and multichannel reCection data; J. Geophys. 46 81–197. Res.: Solid Earth 107(B2) ETG 1-1–ETG 1–24. Rawlinson N, Pozgay S and Fishwick S 2010 Seismic tomo- Krabbenhoft€ A, Weinrebe R W, Kopp H, Flueh E R, Ladage graphy: A window into deep Earth; Phys. Earth Planet. Int. S, Papenberg C, Planert L and Djajadihardja Y 2010 178 101–135. Bathymetry of the Indonesian Sunda margin-relating mor- Rawlinson N, Fichtner A, Sambridge M and Young M K 2014 phological features of the upper plate slopes to the location Seismic tomography and the assessment of uncertainty; and extent of the seismogenic zone; Nat. Hazards Earth Adv. Geophys. 55 1–76. Syst. Sci. 10(9) 1899–1911. Samsudin A R, Hamzah U, Rahman R A, Siwar C, Jani M F M Leveque J J, Rivera L and Wittlinger G 1993 On the use of the and Othman R 1997 Thermal springs of Malaysia and checker-board test to assess the resolution of tomographic their potential development; J. Asian Earth Sci. 15(2–3) inversions; Geophys. J. Int. 115(1) 313–318. 275–284. Lin J Y, Pichon X L, Rangin C, Sibuet J C and Maury T 2009 Shuib M K 2009 The recent Bukit Tinggi earthquakes and Spatial aftershock distribution of the 26 December 2004 their relationship to major geological structures; Bull. Geol. great Sumatra–Andaman earthquake in the northern Soc. Malaysia 55(55) 67–72. Sumatra area; Geochem. Geophys. Geosyst. 10 1–15. Sieh K and Natawidjaja D 2000 Neotectonics of the Sumatran Macpherson K A, Hidayat D, Feng L and Goh S H 2013 fault, Indonesia; J. Geophys. Res. 105(B12) 28,295– Crustal thickness and velocity structure beneath Singa- 28,326. pore’s seismic network; J. Asian Earth Sci. 64 245–255. Tian Y, Zhao D, Sun R and Teng J 2009 Seismic imaging of the Marschall H R and Schumacher J C 2012 Arc magmas sourced crust and upper mantle beneath the North China Craton; from melange diapirs in subduction zones; Nature Geosci. Phys. Earth Planet. Inter. 172(3–4) 169–182. 5(12) 862–867. Tjia H D 2010 Growing evidence of active deformation in the Marto A, Soon T C, Kasim F and Yunus N Z M 2013 Seismic Malay basin region; Bull. Geol. Soc. Malaysia 56 6. impact in Peninsular Malaysia Paper presented at the The USC 2012 JWEED South Carolina Seismology Department at 5th International Geotechnical Symposium, Incheon. the University of South Carolina; http://www.seis.sc.edu/ MMD 2012 The decision and dissemination supports systems software. (DADS) for the Malaysian National Tsunami Early Warn- Van der Hilst R D, Widiyantoro S and Engdahl E R 1997 ing System; Malaysian Meteorological Department (MMD) Evidence for deep mantle circulation from global tomogra- Ministry of Science Technology and Innovation (MOSTI). phy; Nature 386(6625) 578. MMD and ASM 2009 Seismic and tsunami hazards and risks Zhao D, Hasegawa A and Horiuchi S 1992 Tomographic study in Malaysia Final Report; Malaysian Meteorological imaging of P and S wave velocity structure beneath Department (MMD) Ministry of Science Technology and northeastern Japan; J. Geophys. Res. 97(B13) 19,909– Innovation (MOSTI). 19,928. Molnar P and Dayem K E 2010 Major intracontinental strike- Zhao D 2001 New advances of seismic tomography and its slip faults and contrasts in lithospheric strength; Geosphere applications to subduction zones and earthquake fault 6(4) 444–467. zones: A review; Island Arc 10(1) 68–84. Osagie A U, Nawawi M, Khalil A E and Abdullah K 2017 Zhao D 2004 Global tomographic images of mantle plumes and Regional travel-time residual studies and station correction subducting slabs: insight into deep Earth dynamics; Phys. from 1-D velocity models for some stations around Penin- Earth Planet. Inter. 146(1–2) 3–34. sular Malaysia and Singapore; NRIAG J. Astron. Geophys. Zhao D 2015 Methodology of Seismic Tomography; In: 6(1) 19–29. Multiscale Seismic Tomography, Springer, pp. 21–54.

Corresponding editor: ANAND JOSHI