The Origin and Mantle Dynamics of Quaternary Intraplate Volcanism in Northeast China From Joint Inversion of Surface Wave and Body Wave

Abstract

We present a 3-D model of NE China by joint inversion of body and surface waves. The joint inversion significantly improves the resolution at shallow depths compared with body wave tomography alone and provides seismic evidence for the origin of Quaternary volcanism in NE China. Our model reveals that the mantle upwelling beneath the Changbaishan volcano originates from the transition zone and extends up to ~60 km, and spreads at the base of the lithosphere with the upwelling head ~5 times wider than the raising tail in the lower upper mantle. However, low velocities beneath the Halaha and Abaga volcanoes in the Xingmeng belt are confined to depths shallower than 150 km, suggesting that magmatism in the Xingmeng belt is more likely caused by localized asthenospheric upwelling at shallow depths rather than from the common deep source. A small-scale sublithospheric mantle convection may control the spatial and temporal distribution of Quaternary magmatism in NE China; that is, the upwelling beneath the Changbaishan volcano triggers the downwelling beneath the southern Songliao basin, where the high velocity imaged extends to ~300 km. The downwelling may further induce localized upwelling in the surrounding areas, such as the Halaha and Abaga volcanoes. Thanks to the joint constraints from both surface and body waves, we can estimate the dimension of the convection cell. The convection cell is located between 42°N and 45°N, spreads around ~500 km in the W-E direction measured from the distance between centers of downwelling and upwelling, and extends to ~300 km vertically.

Key Points

• A 3-D velocity model of NE China is obtained by joint inversion of surface and body waves
• The small-scale sublithospheric convection is responsible for the intraplate volcanism in NE China
• The size of the upper mantle convection cell is estimated more precisely from the new joint inversion model

1 Introduction

Plate tectonics has successfully explained the origins of most volcanoes occurring at plate boundaries, such as those along the mid-ocean ridges and subduction zones (e.g., Morgan, 1968). However, there are still a lot of volcanoes erupting remotely from the plate boundaries, so-called intraplate volcanoes, which cannot be explained by the Plate Tectonics theory (King & Ritsema, 2000; Morgan, 1971). Some of these intraplate volcanoes sit on top of deep-rooted mantle plume, such as those along the Hawaii volcanic chain. These groups of volcanoes are hot spot volcanoes, and their origin can be explained by the mantle plume theory (French & Romanowicz, 2015; Morgan, 1971). However, most of intraplate volcanoes are not associated with any deep-rooted mantle plume, and their origins are still not fully understood.

Northeast China (hereafter abbreviated as NE China), located between the North China Craton (NCC) and the Siberia Craton (Figure 1), is characterized by widespread intraplate volcanism spanning from the late Cretaceous to the present and occurring at areas more than 1,000 km away from the Japanese Trench where the Pacific Plate is subducting into the mantle (Liu et al., 2001). The Cenozoic intraplate volcanism in NE China is part of the volcanic belt in the western Pacific rim (Liu et al., 2001). Intraplate volcanoes are widely distributed surrounding the Songliao basin, mainly characterized by alkali basaltic eruptions (Chen et al., 2007). The Changbaishan volcano, located at the border between China and North Korea, is the largest and most well-known volcanic center in NE China, which started to activate since 28 Ma. The so-called “Millennium eruption” that took place at the Changbaishan volcano during CE 946 is one of the largest ever recorded volcanic events in human history (Xu et al., 2012). The lava shield from that eruption covers an area of 2 × 10⁴ km² with the maximum elevation of nearly 2,800 m (Wei et al., 2003). The volcanic unrest of the Changbaishan in recent years arouses concerns of the eruptions in the near future (Xu et al., 2012). Other than the Changbaishan volcano, the Jingpohu and Longgang volcanoes (LGV) along the Tanlu fault zone in the Changbaishan mountain region (CBM) were also active during the Holocene (Figure 1) (Liu et al., 2001).

Figure 1

Distribution of seismic stations and geological settings. Blue triangles are 127 NECESSArray stations. Red points = Cenozoic volcanic groups in NE China. Volcanic symbols = Quaternary volcanoes in NE China. XMOB = Xing'an-Mongolia orogenic belt; SLB = Songliao basin; CMB = the Changbai mountain region. White lines outline major Mesozoic basins in NE China. Yellow lines outline tectonic blocks in eastern China. CBV = Changbaishan volcano; JPHV = Jingpohu volcano; LGV = Longgang volcano; WDLCV = Wudalianchi volcano; HLHV = Halaha volcanoes; ABGV = Abaga volcanoes. NCC = North China Craton China; TLFZ = Tanlu fault zone, respectively. The right bottom inset shows the tectonic settings of NE China in a larger scale with the westward subduction of the western Pacific Plate indicated by the black arrow.

The magmatism of Changbaishan volcano is generally believed to be associated with the deep subduction of the Pacific plate, rather than being associated with deep mantle plumes as no low seismic velocities are revealed in the lower mantle (French & Romanowicz, 2015). However, the exact origin and mantle dynamic that is responsible for the intraplate volcanism in Changbaishan is still controversial (Chen & Pei, 2010; Guo, Chen, et al., 2016; Lei & Zhao, 2005; Li et al., 2012; Tang et al., 2014; Tian et al., 2016). For example, Lei and Zhao (2005) proposed that the intraplate volcanism in Changbaishan is caused by the upwelling of hot and wet asthenospheric mantle due to the deep dehydration of the stagnation of Pacific slab in the mantle transition zone. However, Tang et al. (2014) recently suggested that the Changbaishan volcano is fed by the upwelling of hot and buoyant sublithospheric mantle, which has been carried down to the transition zone by the subducting Pacific Plate and escaped through a slab “gap” in the transition zone beneath the area of the Changbaishan volcano.

Furthermore, the Halaha and Abaga volcanoes in the Xingmeng belt are less-documented Cenozoic intraplate volcanoes compared with the Changbaishan volcano. The Abaga volcanoes that erupted between the Miocene and Quaternary are composed of more than 300 volcanic cones and cover a ~10,000 km² lava plateau (Chen et al., 2015; Ho et al., 2008). The onset of the volcanic activities of the Halaha volcanoes occurred at ~2.0 Ma and lasted until the Holocene (Liu et al., 2001). The lava flows spread along the Halaha Rivers and produced a low lava platform (Ho et al., 2013). Several dynamic models have been proposed to explain the origin of Xingmeng magmatism group, including the mantle upwelling generated by piling up and thickening of stagnant Pacific slab in the mantle transition zone (Zou et al., 2008), the influx of fluid released from the stagnant slab in the mantle transition zone (Huang & Zhao, 2006), and the mantle convection at the craton edge (Niu, 2005).

High-resolution image of crust and upper mantle structures is the key to discriminating among the competing models for the origin of magmatism in NE China because low seismic velocities are usually expected in the upper mantle beneath volcanoes, and constraints on the depths and scales of low-velocity anomalies would contribute to the understanding of the origins of these volcanoes (Guo et al., 2015; Guo, Chen, et al., 2016; Tang et al., 2014; Zheng et al., 2011). Most previous seismic tomography in NE China used teleseismic body wave tomography to image the upper mantle structures (Duan et al., 2009; Lei & Zhao, 2005; Tang et al., 2014). However, ray paths of teleseismic body waves are nearly vertical at shallow depths, leading to the fact that crustal and uppermost mantle velocity structures cannot be well constrained in teleseismic body wave tomography; while fundamental surface waves are most sensitive to shallow structures (<300 km) but become almost insensitive to deep structures (>300 km). Thus, joint inversion of body wave and surface wave can provide complementary constraints on structures at different depths, reducing the nonuniqueness inherent in individual seismic tomography and achieving a more realistic model compared with body wave or surface wave tomography alone (Fang et al., 2016; Nunn et al., 2014; Obrebski et al., 2011; Ritsema et al., 2011; Syracuse et al., 2015; West et al., 2004; Zhang, Maceira, et al., 2014).

In this study, we construct a 3-D crustal and upper mantle model from the surface down to 800 km beneath NE China by joint inversion of surface wave phase velocities and body wave traveltimes. Our joint inversion model with unprecedentedly high resolution reveals detailed low-velocity anomalies with distinct magnitudes and depth ranges beneath the Halaha, Abaga, and Changbaishan volcanoes and a prominent high-velocity anomaly beneath the southern Songliao basin, which sheds new light on understanding the geodynamic origin of the intraplate volcanism in NE China.

2 Geological Background

NE China is the eastern part of the Central Asian Orogenic Belt, one of the largest Phanerozoic orogenic belts on the Earth (Zhou & Wilde, 2013). NE China is bounded by the Mongol-Okhotsk suture to the north and the Solonker suture to the south (Figure 1). To the east of NE China is the subducting Pacific plate (Figure 1). NE China can be tectonically divided into the Changbaishan mountain belt (CMB) in the east, the Songliao basin in the middle, and the Xingmeng belt in the west (Figure 1).

From the Paleozoic to late Mesozoic, NE China is formed by the accretion of a series of island arc and microcontinents, including the Erguna Block, Xing'an Block, Songliao Block, and Khanka-Jiamusi-Bureya Massif (Zhou & Wilde, 2013). However, there are still many contentious interpretations on when and where the accretion occurred (Zhou & Wilde, 2013). It is generally believed that the collision of NE China and NCC along the Solonker suture during the early Mesozoic (~230 Ma) and the closure of the Mongal-Okhotsk Ocean along the Mongol-Okhotsk suture during the middle Jurassic (~160 Ma) concluded the final docking of NE China between the Siberia Craton and NCC (Sengor et al., 1993).

Since the late Mesozoic, the tectonic evolution of NE China is largely dominated by the subduction of the western Pacific plate (Wu et al., 2005). The Songliao basin, which is one of the largest hydrocarbon provinces in China, has experienced significant tectonic extension during the late Jurassic and accumulated very thick sedimentary deposits during the postrift stage in the early Cretaceous (Wei et al., 2010). The Mesozoic crustal extension, which is manifested by widespread granitic rocks, metamorphic core complexes, and the development of rifting basins, is most likely due to the rollback of the Paleo-Pacific plate from the late Jurassic to the early Cretaceous (Wang et al., 2006). Geophysical studies show that the current lithosphere is thin (less than 100 km) beneath the Songliao basin (Guo et al., 2014; Guo, Chen, et al., 2016; Zhang, Wu, et al., 2014).

3 Data and Method

Seismic data used in this study are recorded by the NECESSArray, an international collaborative temporal seismic experiment operated in NE China between September 2009 and August 2011 (Guo et al., 2015; Guo, Chen, et al., 2016; Guo, Yang, et al., 2016; Tang et al., 2014; Tao et al., 2014). All the seismic stations were deployed in the field for at least 1.5 years to make sure that they can record sufficient teleseismic events. A total of 127 broadband seismic stations as shown in Figure 1 are used to obtain surface wave phase velocity and body wave traveltime data.

Surface Wave Dispersion Data

Local Rayleigh wave dispersion curves at 6–140 s period are extracted from Rayleigh wave phase velocity maps constructed by ambient noise tomography (ANT) at 6–40 s period and two-plane-wave tomography (TPWT) at 25–140 s period from Guo, Chen, et al. (2016). The details about the data processing and tomography methods in constructing these phase velocity maps from ANT and TPWT are given in our previous surface wave tomographic studies (Guo et al., 2015; Guo, Chen, et al., 2016). So, here, we only briefly describe the data processing and tomography procedures.

We process 2 years of ambient noise data recorded by NECESSArray following the procedures of Bensen et al. (2007) to obtain cross correlations between all station pairs of NECESSArray. Then, we measure Rayleigh wave phase velocity dispersion curves at 8–40 s periods from these cross correlations by adopting the frequency-time analysis method (Levshin & Ritzwoller, 2001; Lin et al., 2007). Finally, using the measured phase velocities between a total of 4,218 interstation ray paths, we generate phase velocity maps at 8–40 s by utilizing a ray theory-based tomography method (Barmin et al., 2001). To generate phase velocities at longer periods, we collect a total of 125 teleseismic earthquakes with the magnitudes larger than 5.5 Mb (Figure 2a). We isolate the teleseismic Rayleigh waves from this event and then measure the amplitudes and phases of Rayleigh waves using frequency-time analysis method. Then, we adopt TPWT method (Yang & Forsyth, 2006) to invert the measured phase and amplitude data to generate phase velocity maps at 20–140 s. In the tomography, 2-D finite frequency sensitivity kernels (Zhou et al., 2004) are employed to represent the sensitivities of surface waves to structural heterogeneities (Li, 2011; Yang & Forsyth, 2006).

Figure 2

(a) Distribution of earthquakes used in the two-plane-wave tomography (epicentral distances between 30° and 120°); and (b) teleseismic earthquakes used to obtain body wave traveltime (epicentral distances between 30° and 90°). Red star denotes the center of our study area. Blue dots represent teleseismic earthquakes used to obtain surface wave or body wave. Red circles mark the epicentral distance of 30°and 90°.

In Figure 3, we show some examples of phase velocity maps at several periods. The uncertainties of phase velocities in most of our study region are between 10 and 55 m/s, with the specific values dependent on period. The lateral variations of phase velocity maps are described in detail by Guo, Chen, et al. (2016).

Figure 3

Phase velocity maps plotted as perturbations at periods between 6 s and 140 s from (a, b) ambient noise tomography (ANT) and (c–f) two-plane-wave tomography (TPWT).

Body Wave Traveltimes

S wave differential traveltimes are obtained from 99 teleseismic events (Figure 2b) with their magnitudes larger than 5.5 Mb and epicentral distances between 30° and 90°. Numerous studies have demonstrated that body wave traveltimes are frequency dependent as a result of diffractive wavefront healing (Hung et al., 2004; Hung et al., 2011). To account for the dispersion of body wave delays times, we measure traveltimes at several different frequency bands. Following Tang et al. (2014), we measure differential traveltimes of S waves at station pairs whose distances are less than 200 km at the low-, intermediate-, and high-frequency bands of 0.02–0.05 Hz, 0.05–0.1 Hz and 0.1–0.5 Hz, respectively. We obtain differential traveltimes based on a multichannel cross-correlation method (VanDecar & Crosson, 1990). Finally, we use a total of 4,281 S wave ray paths at the low-frequency band, 6,174 at the intermediate one, and 3,680 at the high one for body wave tomography.

Joint Inversion

We first obtain a reference 1-D S wave velocity model by inverting the regional average 1-D phase velocity dispersion curve as shown in Figure 4. The inversion takes the preliminary reference Earth model (Dziewonski & Anderson, 1981) as the starting model. A linearized, iteratively updated inversion method (Guo et al., 2015) is employed to invert for the reference model using LSQR method (Paige & Saunders, 1982) through 20 iterations.

Figure 4

(a) Black line = global preliminary reference Earth model (PREM) (Dziewonski & Anderson, 1981); red line = 1-D regional reference model from the inversion of regional-averaged phase velocity. (b) Black line = phase velocity from PREM; blue points = regional-averaged phase velocity; red line = synthetic phase velocity from the 1-D regional reference model (red line in Figure 4a).

We parameterize the crust and mantle of NE China into a 3-D spherical cap with the center located at 124.5°E and 39.5°N. The model space extends from 113°E, 37°N to 136°E, 51°N horizontally, and from the surface down to 1,400 km depth vertically, yielding a grid spacing of ∼40 km (47 × 47 nodes) in the horizontal direction and 10 km in the vertical direction (141 nodes), respectively. We use the following weighting scheme to integrate the body wave traveltimes and surface wave dispersion curves in the joint inversion:

Here δv indicates S wave velocity perturbation; K_t represents the 3-D sensitivity kernel of S wave differential traveltime; K_c is the partial derivative kernel of local dispersion curve at a geological node; d_t and d_c are body wave measurements and phase velocity dispersion curves, respectively. Here the body wave and surface wave data are weighted by the data uncertainties of σ_t and σ_c weighting coefficients of α and β, respectively. We use 3-D finite-frequency-sensitive kernels to relate S wave traveltimes to S wave velocity model (Hung et al., 2004). The finite-frequency kernel is computed by a paraxial approximation theory following Dahlen et al. (2000) using regional reference 1-D model (the red line in Figure 4a). First, a two-point ray tracing by a Bisection shooting method is launched for each event-station pair. Then, the location of each ray path inside the tomographic area is used to construct the 3-D sensitivity kernel. The regional reference model (the red line in Figure 4a) is used to calculate the 1-D sensitivity kernel of surface wave at each geological grid. An appropriate weighting scheme (Guo et al., 2015; Julià et al., 2000; Obrebski et al., 2011) is important to balance the contribution of each data set in the joint inversion. The conventional trade-off curve between surface wave and body wave residuals is used to select optimal weighting coefficients (Figure 5) after a series of inversions. Finally, we use the weighting parameters α = 1 and β = 5 in our joint inversion. Here the standard damped LSQR method (Paige & Saunders, 1982) is used to solve this noniterative inversion problem, since the inversion starts from a good regional reference model, which is optimized to fit both data sets. Furthermore, the data variance reduction of both data sets is significant after one iteration with the differences between the synthetic and observed data within data uncertainties (see section 4.1).

Figure 5

Root-mean-square (RMS) misfit of surface wave phase velocity versus RMS misfit of body wave traveltime as a function of weight parameter. The number along the curve is the ratio of β/α. We select β/α = 5 as the weight parameter for the joint inversion.

It should be noted that, in our joint inversion, Rayleigh wave phase velocity is mainly sensitive to SV wave velocity, and S wave differential times measured at transverse components is sensitive to SH wave velocity. The reason we use Rayleigh wave is that it has deeper depth sensitivities at the same period compared with Love wave and can be measured more reliably than Love wave, since Rayleigh wave measured from vertical components has higher signal-to-noise ratio than Love waves measured from transverse components. Meanwhile, fundamental Love wave can be interfered by high-mode Love wave as shown by Ekström (2011) and Luo et al. (2015), making it hard to separate one mode from another. Furthermore, SH wave measured from transverse components has higher signal-to-noise ratio compared with SV wave measured from vertical/radial components, and therefore, more reliable S wave measurements can be obtained from SH wave than from SV wave. Thus, in our joint inversion, we use Rayleigh wave and SH wave and ignore the effects of radial anisotropy. The effects of radial anisotropy seem not so significant in our study area as the 3-D model (Vsv model) solely constrained from Rayleigh wave data are very similar to the 3-D model (Vsh model) solely constructed from SH waves at the overlapped depth range of 100–300 km for the two separated inversions (see section 4.1).

4 Results

3-D S Wave Velocity Model From Joint Inversion

The resulting 3-D S wave velocity model from the joint inversion of surface wave and body wave is shown in Figure 6 as horizontal slices at different depths. Several vertical cross sections are also presented in Figures 7 and 8. The joint inversion results in a total of ~63% reduction of S wave traveltime variances and a ~88% reduction of surface wave data variances. In comparison, body wave or surface wave inversion alone causes data variance reduction of ~65% or ~93%, respectively.

Figure 6

S wave velocity anomalies at different depths from the joint inversion of surface wave and body wave. The white line outlines the Songliao basin. The upper mantle low velocity beneath the Changbaishan volcano is marked as “LV1”. The low-velocity anomalies in the upper mantle beneath the Halaha volcanoes and Abaga volcanoes are labeled as “LV2” and “LV3”, respectively. HV = the upper mantle high velocity beneath the southern Songliao basin.

Figure 7

Cross sections of joint inversion model along the profile (b) A-A′, (c) B-B′, and (d) C-C′, with their locations shown in Figure 7a.

Figure 8

Same as Figure 7 but for profiles D-D′ and E–E′.

Below, we describe the major velocity features observed in our joint inversion. Since the crustal part of the model is mainly constrained by surface waves data, the crustal features of this study are almost the same as those in Guo, Chen, et al. (2016). Thus, we mainly focus on the description of mantle velocities but briefly describe the major features of crustal structures.

In the shallow crust (Figure 6a), the major features are broad low velocities beneath the Songliao basin, associated with the rather thick sediments (~10 km in the interior), and high velocities in the Xingmeng belt and CBM. The middle-to-lower crust of the Xingmeng belt is characterized by prevalent low velocities, which is in good agreement with previous surface wave studies using NECESSArray and has been interpreted as the consequence of the mafic lower crust delamination during the Late Mesozoic (Guo et al., 2015; Guo, Chen, et al., 2016; Li et al., 2016).

In the uppermost mantle (Figures 6b and 6c), the most prominent feature is the widespread low-velocity anomaly (“LV1”) beneath the CBM (Figures 6-8). The horizontal elongation of the “LV1” closely follows the surficial distribution of the Tanlu fault zone and Cenozoic volcanoes in the CBM. Between 210 and 390 km depth (Figures 6d–6f), the “LV1” becomes weaker and concentrated at the southern CBM, roughly beneath the Changbaishan volcano. In the mantle transition zone, the “LV1” is located ~200 km to the west of the Changbaishan volcano (~126°E, 42°N) with its diameter of ~200–300 km (Figure 7b), which extends to the uppermost part of the lower mantle.

The depth extent of “LV1” is depicted more clearly by the cross sections shown in Figures 7 and 8. The A-A′ and B-B′ profiles (Figures 7b and 7c), which both transect the Changbaishan volcano, display that the “LV1” dips westward with depth and becomes narrower below 300 km depth. Beneath the Changbaishan volcano, the “LV1” continuously extends from the uppermost mantle (~50 km) to depths greater than 660 km. However, to the north of Changbaishan volcano, LV1 can be only traced down to ~200 km (C-C′ and F-F′ profiles). We also image a high-velocity lid with a thickness of 50–70 km atop the “LV1” beneath the Changbaishan volcano (A-A′, B-B′, and F-F′ profiles). Such high-velocity lid is not resolved by previous body wave tomography studies (Lei & Zhao, 2005; Tang et al., 2014).

At the southern Songliao basin, both horizontal slices and vertical profiles reveal a pronounced, W-E striking high-velocity anomaly (HV in Figures 6-8), extending continuously from the uppermost mantle down to ~300 km depth (E–E′ profile). Beneath the central Songliao Basin, our model shows weak low velocity (dlnVs ≤ 1%) between 150 and 330 km depths.

Other than the “LV1” beneath the Changbaishan volcano, two additional low-velocity anomalies are observed in the upper mantle roughly beneath the Halaha and Abaga volcanoes, labeled as “LV2” and “LV3” in Figures 7 and 8. Both of these low-velocity anomalies extend from ~60 km to a maximum depth of ~300 km (A-A′, B-B′, C-C′, and D-D′ profiles). These low-velocity anomalies are separated from the “LV1” beneath the CBM by the high velocities in the upper mantle beneath the Songliao basin, that is, the “HV” shown in the profile A-A′ and B-B′ of Figure 7. Furthermore, “LV2” and “LV3” are also separated from each other by the relatively high velocities beneath the Xingmeng belt (profile D-D' in Figure 8).

Figure 9 compares the joint inversion model with models either from surface wave inversion or body wave inversion alone. In general, the first-order velocity features from our joint inversion is very similar with the surface wave tomography of Guo, Chen, et al. (2016) at depths from the surface to 150 km, and similar with teleseismic body wave tomography of Tang et al. (2014) at 300 to 700 km depths (Figure S1 in the supporting information). For example, significantly low velocities beneath the Changbaishan volcano, Abaga volcanoes, and Halaha volcanoes, and high velocities beneath the Songliao basin are all well imaged by three independent inversions at depths above 300 km. However, at depths between 150 and 300 km depths, some differences are observed between the joint inversion and surface/body wave tomography alone. As expected, surface wave loses resolution at depths below 250 km (Figure 9a); and body wave tomography has poor constraints at the depths shallower than 100 km (Figure 9b), resulting from nearly vertical ray paths in the shallow depths, which leads to smeared structures from greater depths to the crust (Figure 9a). Therefore, the actual depth range of low-/high-velocity bodies cannot be accurately estimated from the body wave or surface wave alone. Our joint inversion with improved resolution provides a more complete image that helps to reveal the origins of intraplate volcanoes in NE China.

Figure 9

Comparison of S wave velocity along A-A′ profile from (a) surface wave inversion, (b) body wave tomography, and (c) joint inversion.

Resolution Test

To further illustrate the ability of our joint inversion in imaging the crustal and mantle structure of NE China, we perform four checkerboard resolution tests. The horizontal dimension of input checkerboard anomalies in the four tests is set to 200 km × 200 km (Figures 10a–10d) and 400 km × 400 km (Figures 10e–10f), respectively. The vertical scale of these checkerboard anomalies vary from 40, 100, 200 km, to 300 km in different tests (Figure 11). The amplitudes of checkerboard velocity anomalies are set to ±4% relative to the 1-D regional reference velocity. Recovered models from joint inversion are shown in Figures 10 and 11 for depth slices and vertical slices, respectively. In general, the sizes of input checkerboard anomalies can be recovered reasonably well, although the vertical resolution decreases with depth. Vertical smearing effects are pronounced at the margins of the study region, where the ray path coverage is limited. In addition, the amplitude of recovered anomaly is slightly damped relative to the input model.

Figure 10

Results of checkerboard resolution tests shown at different depths with different sizes of input anomalies. (a)–(d) Recovered velocity anomalies for the input anomalies with a horizontal dimension of 200 km × 200 km; (e, f) Recovered velocity anomalies for the input anomalies with a horizontal dimension of 400 km × 400 km.

Figure 11

Cross sections of checkerboard resolution tests along the A-A′ profile with its location shown in Figure 7a. The left column = (a) the input anomalies with a dimension of 200 × 40 km, (b) 200 × 100 km, (c) 200 × 200 km, and (d) 400 × 300 km. The right column = recovered velocity anomalies for the input models shown in the left.

To further demonstrate that the main velocity structures we interpret in this study (Figure 12) are robust features, we carry out a “restoring” synthetic test using the final 3-D model from our joint inversion as the testing model in the resolution test. Synthetic surface wave dispersion curves and body wave traveltimes are calculated based on the test model. Random noises are added to synthetic data as real data. Then, “restoring” joint inversion is conducted using the same regularization and weighting parameters as the real inversion using real observations. Figure 12 shows the comparison between the testing model and the restored model, which illustrates that the joint inversion can resolve main velocity features of the input model and provide better vertical resolution and magnitude recovery compared with the body/surface wave tomography alone.

Figure 12

The “restoring” resolution test. (a) The input model from joint inversion using real data sets. The profile is along 42°N. (b) The “restoring” 3-D model from inversion of synthetic surface wave dispersion curves. (c) The “restoring” 3-D model from synthetic body wave traveltime. (d) The “restoring” 3-D model from joint inversion of synthetic surface wave and body wave data.

5 Discussions

Deep Origin of Changbaishan Volcano and Asthenosphere-Lithosphere Interactions

Beneath the Changbaishan volcano, our 3-D image clearly exhibits a low-velocity anomaly dipping westward with depth and continuously extending into the mantle transition zone or deeper. This low-velocity anomaly has the greatest velocity reduction up to ~6% at depths between 60 and 200 km. The notation of the deep origin of mantle upwelling beneath the Changbaishan volcano is not new (Lei & Zhao, 2005; Tang et al., 2014). However, the actual depth extent of the upwelling, particularly the top extent, is controversial as most previous tomography studies in this area are based on teleseismic body waves alone, which lack resolution in the uppermost mantle (e.g., Tang et al., 2014). Our joint inversion with good resolution in the uppermost mantle, thanks to the constraints from surface wave, clearly reveals that the upwelling reaches the depth around 60 km.

The mantle upwelling beneath the Changbaishan volcano does not represent a classic mantle plume that originates from the core-mantle boundary. There is lack of clear age progression volcanic track; no signatures of high values of He³/He⁴ materials (Chen et al., 2007) have been found; and more importantly, the global or continental-scale seismic tomography studies do not reveal a continuous plume tail in the lower mantle that is extending down to the core-mantle boundary (French & Romanowicz, 2015). Using teleseismic body wave traveltime tomography, Lei and Zhao (2005) observed high-velocity anomalies in the mantle transition zone and a columnar low-velocity anomaly in the upper mantle extending down to 400 km beneath the volcano. Based on these imaged velocity features, they suggested that the intraplate volcanism in Changbaishan is caused by the upwelling of hot and wet asthenospheric materials, related to the deep subduction and stagnation of the Pacific slab in the transition zone (Lei & Zhao, 2005). Recently, by using teleseismic body wave traveltime recorded by the NECESSArray, Tang et al. (2014) observed that the low-velocity anomalies beneath the Changbaishan volcano extend through the transition zone from the uppermost mantle to just below 660 km. Based on the observed velocity features, they suggested that the mantle upwelling beneath the Changbaishan volcano is more likely triggered by the upwelling of the hot materials that were entrained by the subducting slab and escaped later from a slab gap in the mantle transition zone rather than by the deep dehydration of the Pacific slab (Tang et al., 2014). Our joint inversion uses almost the same body wave data as Tang et al. (2014) and generates very similar velocity features at depths below 300 km as Tang et al. (2014), thus tending to more support the interpretation of Tang et al. (2014).

Our 3-D joint model confirms the deep origin of Changbaishan volcano and also reveals some new features compared to previous body wave tomography (Tang et al., 2014). Our model shows that the upwelling of upper mantle may reach the shallow depth of ~60 km, and the upwelling asthenosphere may spread at the base of the lithosphere along the CBM with its head ~5 times larger than the tail in the lower upper mantle. Therefore, the ~1.5 km topography of the southern CBM could be partially dynamically supported by such large-scale mantle upwelling (Tao et al., 2014). The large head of the upwelling mantle may trigger significant basal erosion to the overlying lithospheric mantle, leading to a generally thinned lithosphere beneath the CBM observed by our model. Our observations also agree well with previous receiver function studies (Guo et al., 2014; Zhang, Wu, et al., 2014), which reveal a ~60–80 km thick lithosphere beneath CMB. The preexisting weak zones in the lithosphere, such as the Tanlu fault zone, which is believed to cut through the entire lithosphere (Chen et al., 2006), may provide the pathway for the transportation of hot and basaltic melt to the surface.

Local Asthenospheric Upwelling Beneath Volcanic Groups in the Xingmeng Belt

In contrast to the Changbaishan volcano, low-velocity anomalies beneath the Abaga and Halaha volcanoes in the Xingmeng belt are relatively smaller and their depth is shallower. These low-velocity anomalies extend to a maximum depth of ~200 km and are separated by a relatively thick lithosphere in between 140 and 160 km depths (Zhang, Wu, et al., 2014). Moreover, unlike that there is an eruption center of the Changbaishan volcano (the Tianchi crater), volcanoes in the Halaha and Abaga areas are characterized by isolated eruptions with diffusively distributed volcanic cones of a small volume of melt (Ho et al., 2008; Ho et al., 2013).

We therefore suggest that the magmatism of the Halaha and Abaga volcanoes is more likely due to isolated and localized asthenospheric upwelling processes, rather than originating from the common deep source as the Changbaishan volcano with the mantle upwelling related to the deep subduction of the Pacific slab. The geochemical and petrological studies (Chen et al., 2015; Ho et al., 2013) also corroborate this interpretation, which suggest that basalts in the Halaha and Abaga regions derive from relatively low-degree partial melting of garnet peridotite source in the asthenosphere, and furthermore, the strong chemical heterogeneities of these basalts reflect that they originate from different mantle sources with strong compositional heterogeneities.

Small-Scale Sublithospheric Convection and Mantle Dynamics for Intraplate Volcanism in NE China

Another important feature of our 3-D model is the prominent high velocity (HV) beneath the southern Songliao basin, which extends to ~300 km depth and is confined to the area roughly between 42°N and 45°N, rather than occupying the whole basin as suggested before (Guo, Chen, et al., 2016). This high-velocity anomaly extending to ~300 km unlikely represents the thick lithosphere of the south Songliao basin, as previous receiver function studies indicate the thickness of lithosphere beneath the Songliao basin is only about 100 km (Guo et al., 2014; Zhang, Wu, et al., 2014). We consider that the drip-shaped high velocity may indicate a sublithospheric downwelling in response to the large-scale mantle upwelling (LV1) beneath the Changbaishan volcano as previously proposed by Guo, Chen, et al (2016). This interpretation is reasonable given the facts that (1) the volume of “HV” is similar to that of the “LV1”, and (2) the size of the convection cell is around 500 km, estimated from the distance between the center of the downwelling and that of the upwelling, and (3) the convection cell vertically extends to a depth of ~300 km.

The presence of the sublithospheric downwelling beneath the southern Songliao basin is also in agreement with the complex patterns of anisotropy across the Songliao basin revealed by shear wave splitting measurements, which are shown in Figure 13 on top of averaged S wave velocity between 100 and 300 km depths. The fast polarizations of splitting waves change abruptly from dominant NNW-SSE direction in the northern Songliao basin (46°N–48°N), which roughly parallels to the subduction direction, to the nearly W-E direction in the basin interior (45–46°N). Beneath the southern basin, there is no preferred fast polarization alignment. Furthermore, delay time reduces from 0.8 to 1.0 s outside the southern basin to less than 0.5 s inside the southern basin, where the downwelling could occur and a large number of null measurements are found. Such striking variations of the splitting pattern could be the consequence of the rapid change of mantle flow direction from the horizontal direction outside the downwelling to the vertical direction inside the downwelling. The similar abrupt change of anisotropy feature has been reported in the Great Basin, western USA, where West et al. (2009) suggested that the sublithospheric downwelling occurring between 200 and 800 km controls the mantle flow pattern beneath the western USA.

Figure 13

Azimuthal anisotropy from shear wave splitting measurements across the Songliao basin. The red bars present shear wave splitting results using data from NECESSArray (Li et al., 2017) and the blue ones from stations of China Earthquake Administration (Li & Niu, 2010). The size of white circles represents the delay time relative to 1 s delay time shown in the right bottom legend. The red and blue dots represent stations with null measurements. The background map represents the average S wave velocity between 100 and 300 km. The white arrow indicates the subduction direction of the western Pacific plate relative to the Eurasia plate.

Furthermore, the small-scale sublithospheric mantle convection may play an important role in the temporal and spatial distribution of Cenozoic volcanism in NE China. First, several episodes of volcanism have been reported occurring at the southern Songliao basin from the late Cretaceous (~81 Ma) to the late Paleogene (~39 Ma) with a basaltic type shifting from predominantly tholeiitic to alkali during ~49 Ma. Since 39 Ma, no volcanism has been detected within the Songliao basin (Liu et al., 2001). The depression of the Songliao basin and the subsequent development of the sublithospheric downwelling since ~49 Ma may provide a plausible explanation for the change of basaltic type and the lack of volcanism in the Songliao basin.

Second, numerical modeling study has suggested that the downwelling may further induce localized asthenospheric upwelling in its surrounding areas (West et al., 2009). It is worth noting that the Halaha and Abaga volcanoes in the Xingmeng belt are all located near the edge of the downwelling limb beneath the Songliao Basin. We thus propose that localized upwelling beneath the Xingmeng belt could be triggered by the downwelling beneath the southern Songliao Basin, and such local upwelling of asthenosphere could lead to decompression partial melting in the uppermost mantle, feeding the Halaha and Abaga volcanoes. Here we need to acknowledge that more works in the future are needed to test our hypothesis for the origin of the Halaha and Abaga volcanoes, especially numerical modeling, considering that the specific geometry of the observed velocity features are needed to model if the local downwelling can trigger the local asthenospheric upwelling.

6 Conclusions

We present a high-resolution 3-D S wave velocity model by joint inversion of body wave traveltime and surface wave dispersion curves using NECESSArray. The joint inversion model significantly improves the resolution at shallow depths compared with body wave tomography and provides seismic evidence for the origin and mantle dynamics for the Quaternary intraplate volcanoes in NE China. Our model supports that the mantle upwelling beneath the Changbaishan volcano originates from the mantle transition zone or deeper. A high-velocity body extending down to more than 300 km at the southern Songliao basin could be a downwelling limb that is induced by the large-scale upwelling beneath the Changbaishan volcano. The upwelling and downwelling form a small-scale convection beneath NE China. Furthermore, the upwelling beneath the Halaha and Abaga volcanoes in the Xingmeng belt is more localized and can be only traced down to less than 300 km depth, which is not directly related to the subduction of the Pacific plate. The local upwelling beneath the Xingmeng belt could be induced by the downwelling beneath the Songliao Basin.

Acknowledgments

We thank all the people who had participated in the field work of the NECESSArray project. This study is supported by the NSFC (grant 41774052, 41674057). The seismic data set used in this study is recorded by the NECESSArray network during 2009 and 2011, which can be accessible at http://ds.iris.edu. We also thank constructive suggestions from two anonymous reviewers, which significantly improved the manuscript. This project is also supported by Australian Research Council Future Fellowship (FT130101220). This is contribution 1102 from the ARC Centre of Excellence for Core to Crust Fluid Systems (http://www.ccfs.mq.edu.au) and 1210 in the GEMOC Key Centre (http://www.gemoc.mq.edu.au).