Corresponding author: C.-W. Oh, Department of Earth and Environmental Sciences, Chonbuk University, Chonju 561-756, South Korea. (email@example.com)
 Here we use the global gravity field data set EGM2008 for 3-D crustal density modeling of the Mount Paekdu stratovolcano and surrounding area located on the border between North Korea and China. Curvature analysis and Euler deconvolution are used to assist interpretation, and the 3-D model is constrained by multiple geological and geophysical data sets. Mount Paekdu is characterized by a low Bouguer anomaly of −110 × 10−5 m/s2, which is caused by the combined gravity effects of (1) a depth to the Moho of about 40 km, (2) a zone with lower P wave velocity and density than the surrounding, (3) low density volcanic rocks on the surface, and (4) the presence of a magma chamber that has not previously been identified. The modeled magma chamber has a mean thickness of 5 km and a density of about 2350 kg/m3 and is located <10 km from the surface. Magma chambers are also modeled beneath Mount Wangtian and Mount Nampotae. However, the results of the 3-D density modeling do not confirm the existence of a previously proposed midcrustal low-velocity zone in the area 70 km to the north of Mount Paekdu. Since the Pliocene, volcanic activity in the Mount Paekdu region has migrated from the east coast of North Korea to the northwest, following the path of NW-SE trending faults.
 Mount Paekdu (42°00′N, 128°03′E) with a height of 2750m is located at the border between Northern Korean Peninsula and China (Figure 1). Mount Paekdu (known as Mount Changbai or Baitoushan or Tianchi volcano in Chinese) is one of the world's largest stratovolcanoes. The volcano possesses a caldera lake (named Cheonji in Korean and Tianchi in Chinese), which was formed by a rhyolitic eruption that took place at circa. A.D. 1000 [Yun et al., 1993; Horn and Schmincke, 2000; Wu et al., 2005; Stone, 2010; Xu et al., 2012]. During the eruption, volcanic ash was blown over to the Japanese islands, forming a 5 cm thick layer [Machida and Arai, 1983; Horn and Schmincke, 2000]. Since the last eruption of Mount Paekdu, in 1903, the volcano has been quiet. However, as recently discussed [Stone, 2010; Xu et al., 2012], there are several signs indicating possible seismic unrest around Mount Paekdu, suggesting that the volcano may erupt in the near future. About 1000 tremors were recorded around Mount Paekdu in 2003, while about 100 small tremors per year were registered in 1904–2002. The majority of the hypocenters of these events were clustered below the Cheonji caldera lake within the uppermost 5 km of the crust. The height of the volcano increased by 6.8 cm between 2003 and 2004 and the 3He/4He ratio during that time changed dramatically, indicating that volcanic activity may occur in the near future [Wu et al., 2005; Stone, 2010; Liu et al., 2011; Xu et al., 2012].
 Results of seismic and receiver function analysis along two intersecting seismic profiles in the research area (Figure 1) have confirmed the existence of a low-velocity zone (LVZ) in the continental crust under Mount Paekdu, indicating the presence of a high-temperature magma chamber system whose movement caused the tremors [Zhang et al., 2002a, 2002b; Hetland et al., 2004; Liu et al., 2004; Duan et al., 2005; Yang et al., 2005; Song et al., 2007; Xu et al., 2012]. However, there was a notable discrepancy among the results regarding estimates of the location of the LVZ. Recent seismic interpretations [Song et al., 2007] were used to infer the presence of a midcrustal LVZ under the region 70 km to the north of Mount Paekdu, whereas most results prior to that study [e.g., Tang et al., 1998; Zhang et al., 2002a, 2002b] indicated the LVZ as being directly beneath Mount Paekdu. The receiver function study of Hetland et al.  suggested an alternative in which the LVZ is thought to be located under both areas.
 Although the volcanic conditions and the reasons for seismic activity in and around the volcano between 2003 and 2005 are keenly debated [Wu et al., 2005; Stone, 2010; Liu et al., 2011; Xu et al., 2012], no terrestrially measured potential field data with sufficient quality are available to image the magmatic structures. This is partly due to Mount Paekdu's position at the border between China and North Korea and the associated geopolitical problems in the area. The global gravity field model provides a basis for three-dimensional (3-D) gravity modeling to constrain the location of the LVZ and to analyze the magmatic system. The EGM2008-derived gravity field has an average resolution of 10 km [Pavlis et al., 2012]. In this study, the gravity model is used to resolve the volcanic structures at local to regional scales. We model the crustal density of the Mount Paekdu area by using data from two intersecting seismic profiles and all available geophysical data sets in an interactive 3-D modeling software environment. Additional constraints for the modeling include previous seismic results, receiver function analysis, geological interpretations, and petrophysical information. Curvature analysis and the Euler deconvolution method are also applied to the interpretation of the gravity field. The density modeling provides detailed insights into the spatial distribution of crustal density structures in the research area to identify the likely locations of magma chambers as well as to anticipate the possible evolution of volcanic ruptures.
2 Geological Setting and Petrophysical Overview
 Mount Paekdu is a 2750 m high stratovolcanic cone with a prominent caldera and rests on a basaltic shield and plateau situated on the North Korea-China border. The basement of the basaltic shield and plateau consists of Archean to middle-late Proterozoic metamorphic rocks, Paleozoic strata, and predominantly Mesozoic granite (Figure 2) [Wang et al., 2003; Yun et al., 1993]. Mount Paekdu (“P” in Figure 2) forms the volcanic zone (green area in Figure 2) with Mount Nampotae (“N” in Figure 2) and Mount Wangtian (“W” in Figure 2). Mount Paekdu has been an active volcano since the Oligocene and is still regarded as a high-risk volcano, having provided one of the planet's largest eruptions in the past 2000 years [Yun et al., 1993; Horn and Schmincke, 2000; Wu et al., 2005; Stone, 2010; Xu et al., 2012].
 In the study area, the regional NE-SW oriented tectonic fault lines (e.g., Mishan-Dunhua and Tianchi-Guanping Faults in Figure 2) run parallel to the northwest Pacific subduction zone, where many seismic events occurred in depths up to 600 km [Wang et al., 2003]. The intersections of the NE-SW and NW-SE trending fault lines provide the loci for the volcanic activity of Mount Paekdu (Figure 2).
 The volcanic history of Mount Paekdu can be divided into four stages based on age data, eruption style, and magma compositional changes [Wei et al., 2007], as follows (from the oldest to the youngest): (i) preshield plateau-forming eruptions, (ii) basalt shield formation, (iii) construction of a trachytic composite cone, and (iv) explosive ignimbrite-forming eruptions. In the first stage, a fissure eruption produced the Maansan, Zeungbongsan, Naedusan, and Changbaeg basalts from the Oligocene to the Miocene (28–13 Ma) [Liu, 1983, 1988]. A preshield plateau was formed by these basalts, which are distributed in a zone that trends NE-SW, suggesting that the fissure eruption took place along a NE-SW oriented rift. These volcanic rocks contain upper mantle inclusions and are generally characteristic of alkaline rocks, indicating mantle-origin intraplate magmatism [Chen et al., 2003]. The basalt plateau became deeply eroded, forming valleys that were filled with lava flows from the shield-forming eruptions of the Pliocene and early Pleistocene (4.21–1.70 Ma) [Liu, 1983]. Fissure and central eruptions occurred together during the shield-forming eruptions. The Pyeongjungchon basalt was erupted along a fissure at approximately 4.21 Ma, and the Mangcheona basalt and trachyandesite formed by fissure and central eruptions at approximately 4.43–3.56 Ma. At 3.11 Ma, alkali rhyolite erupted from Mount Hongdu near Mangcheona. The Kunhamsan and Bocheon basalts formed between 2.90 and 1.70 Ma, constructing a shield volcano in the area to the north and east of Mount Paekdu and in the area around Bocheon to the south of Mount Paekdu [Liu, 1983; Wei et al., 2007]. In the third stage, the trachytic composite volcano, with a diameter of 20–30 km and a height of 650 m, formed during the Pleistocene (0.61–0.09 Ma) [Liu, 1983; Liu et al., 1989]. In this stage, there were six eruptions, at 0.611–0.551, 0.442, 0.329, 0.281–0.219, 0.101, and 0.0978–0.0876 Ma, and magma changed to an acidic melt to produce alkali trachyte, trachytic tuff, tuff breccia, alkali rhyolite, and obsidian. The latest stage has been characterized by explosive ignimbrite-forming eruptions during the Holocene, producing trachytic pumice, ash tuff, welded tuff, and tuff breccia. The ash produced in this stage has reached as far as the East Sea (Sea of Japan) and Japan. During this stage, a caldera formed at the top of Mount Paekdu. The Holocene eruptions occurred between 1690 B.P. and A.D. 1702 [Yun et al., 1993]. The largest eruption, which had a 35 km high column and produced comenditic ignimbrite around the volcano, took place at circa A.D. 1000 [Yun et al., 1993; Horn and Schmincke, 2000; Wu et al., 2005; Stone, 2010; Xu et al., 2012].
 The India-Eurasia collision at approximately 30 Ma caused E-W extension and clockwise rotation at the margins of the western Pacific Plate and eastern Eurasian Plate, leading to the reactivation of NE-SW trending faults in northeast Asia (including the Mount Paekdu area) and to the opening of the East Sea. The rollback of the subduction zone in the Pacific side can be another possible reason for E-W extension and the opening of the East Sea. The volcanism of the early basaltic plateau stage, which occurred along a NE-SW trend, may be related to these tectonic events. The opening of the East Sea ceased at approximately 15 Ma, and during the late Cenozoic, the extensional environment became compressional at the margins of the western Pacific Plate and the eastern Eurasian Plate [Wang et al., 2003]. As a result, since the late Pliocene, the Mount Paekdu area has been uplifted due to intracontinental compression related to the westward subduction of the western Pacific Plate. Volcanic rocks of the Mount Paekdu area, from the Pliocene basalt shield stage to the Holocene pantelleritic volcanic eruption stage, are distributed mainly along the NW-SE trending zone. These data suggest that the NW-SE trending faults may have formed in response to late Cenozoic compression and that the volcanic activity since the Pliocene may have occurred along the NW-SE trending faults. Yun et al.  suggested that volcanism has migrated northwestward along the NW-SE trending faults from the Pliocene to the present.
 Because there were no determinations of rock properties obtained from the study area, densities and P wave velocities for the gravity modeling are taken from petrophysical data for rock samples collected from Mount Halla on Jeju Island and from the southern part of the Korean Peninsula. Located in the South Sea off the Korean Peninsula, Mount Halla has been formed by multiple eruptions of basaltic magma since the late Cenozoic, similar to Mount Paekdu. The bulk densities of the alkali basalt and trachyte sampled from Mount Halla range from 1700 to 2460 kg/m3 and from 1860 to 2120 kg/m3, respectively. The mean density of the volcanic rock samples from Mount Halla is <2300 kg/m3 [Kwon et al., 1993; Lee et al., 2007]. Park  and Park et al.  carried out more than 2500 property measurements on igneous, metamorphic, and sedimentary rock samples from the southern part of the Korean Peninsula. The mean bulk densities of granites of Paleoproterozoic to Mesozoic age, gneiss of Archean to Paleoproterozoic age, and sedimentary rocks are 2580, 2670, and 2600 kg/m3, respectively [Park, 1999; Park et al., 2009]. The physical rock properties of density and P wave velocity (Table 1) are used as input parameters for the gravity modeling described in section 6.
Table 1. Physical Properties of the Different Rock Types Sampled From Mount Halla and From the Korean Peninsula
A: Lee et al. ; B: Kwon et al. ; C: Park  and Park et al. .
3 Seismic Results Constraining the Model
 To obtain direct evidence about the Moho depth and subsurface crustal structures within the volcanic area, two wide-angle reflection and refraction seismic survey profiles were conducted: one along a south-north line including Mount Paekdu (A-P-A′-A″ in Figure 1) and another along a west-east line (B-A′-B′ in Figure 1) crossing the northern part of the volcanic area. The surveys were performed by the Research Center of Exploration Geophysics (now the Geophysical Exploration Center) of the China Seismological Bureau (now the China Earthquake Administration) and the State University of New York (SUNY) at Binghamton [Tang et al., 1998; Zhang et al., 2002a, 2002b; Yang et al., 2005]. Previous results of deep seismic soundings along the two seismic lines allowed four velocity layers (C1, C2, C3, and Moho in Figure 3a) to be identified. The crustal thickness varies from 31 km in the area north of the caldera lake to about 40 km directly under Mount Paekdu. The upper crust (from the surface to layer C2 in Figure 3a) is about 22 km thick, while the lower crust varies from 13 to 17 km in thickness. E-W lateral variations in crustal thickness are not expected.
 The most significant feature in the two seismic profiles is a series of LVZs in the crust directly beneath Mount Paekdu between 10 km depth and the Moho [Zhang et al., 2002a, 2002b]. Here the P wave velocity is approximately 2%–4% slower than that in the surrounding crust (Figure 3a). It is also interpreted that the shallowest LVZ below Mount Paekdu as extends from the stratovolcano to the NNE at a depth range of about 9–15 km. The lateral extents of the other LVZs decrease with increasing depth [Tang et al., 1998; Zhang et al., 2002a, 2002b]. In contrast, the most recent seismic study [Song et al., 2007] concluded that midcrustal LVZs do not exist directly under Mount Paekdu but that one LVZ exists in the crust under the place about 70 km north of the caldera lake (Figure 3b). A receiver function analysis [Hetland et al., 2004] resolved LVZs both beneath Mount Paekdu and in the area to the north. We evaluate these different findings using gravity modeling in section 6.
 A third seismic profile (P-N-D-C in Figure 1) extends from Mount Paekdu to the east coast across the North Korean territory (Figure 4) [Pak et al., 1993]. Due to the simplified interpretation of this profile, only general information is available: (1) the densities of the upper and lower crusts are about 2710 and 2950 kg/m3, respectively, and (2) the depth of the Moho varies from about 40 km under Mount Paekdu to <30 km at the margin of the East Sea. This information can be used to modify crustal density structures of the lithosphere in the vicinity of the profile.
3.1 Relationships Between P wave Velocity and Crustal Density
 Knowledge of the distribution of density in the lithosphere is crucial for gravity modeling. Duan et al.  proposed a relationship between density (Rho) and velocity (Vp) using the following equation: Rho = 0.32 Vp + 0.77 (black line in Figure 5). The relationship applies to the NE China lithosphere including our research area. In addition, we used another equation, Rho = 0.34 Vp + 0.47 (red line in Figure 5) [DeNosaquo et al., 2009], for poorly constrained densities of partial melts in the volcanic area. The analysis of densities in the volcanic area by DeNosaquo et al.  indicated that the melted magma chamber (blue ellipse in Figure 5) characterized by a LVZ in the seismic interpretation has a density of 100–200 kg /m3 or 5%–10% lower than the surrounding. This is due to the conditions in the magma chamber (including percent melt, composition, water content, and temperature). The crustal densities determinations calculated from the P wave velocity distribution along seismic lines (Figures 3a and 3b) and derived using the equations of Duan et al.  and DeNosaquo et al.  (Table 2 and Figure 5) are used for our initial modeled density distribution.
Table 2. Relationships Between Crustal Velocities (Vp in km/s), Densities (Rho in kg/m3), and Geology (see also Table 1)a
Rho1 and Rho2 are calculated by applying the equations of Duan et al.  and DeNosaquo et al. , respectively. Rho12 is an approximate average of Rho1 and Rho2.
4 EGM2008 Gravity Field
 Free-air gravity field data were obtained from EGM2008 (Figure 6a), compiled by the National Geospatial Intelligence Agency (NGA), to compute station complete Bouguer and isostatic anomalies. EGM2008 consists of about 4.8 million spherical harmonic coefficients complete to degree and order 2159, and provides global coverage of gravity data with the highest available resolution of 5′ [Pavlis et al., 2012]. EGM2008 is a gravity field model (ITG-GRACE03S [Mayer-Gürr et al., 2007]) that incorporates almost 6 years of GRACE satellite gravity field observations and provides a highly accurate description of the long- and medium-wavelength gravity field spectrum up to degree and order 180. The second input data set for EGM2008 consists of terrestrial- and ship-measured gravity data and satellite altimetry data for gravity anomalies over the oceans where no measured data are available [Pavlis et al., 2012].
 The NGA is able to use most data sets directly to calculate a global gravity field, but in some regions, termed “fill-in” regions (e.g., the Korean Peninsula and NE China), the agency used data with a resolution of 15′, while the higher resolution of 5′ was calculated using topographic data [Pavlis et al., 2012]. The NGA merged the 15′-resolution data set with the data set obtained from topographic estimations to derive a high-resolution gravity map for the fill-in regions [Pavlis et al., 2012]. The quality and accuracy of EGM2008 have been verified in some regions: the maximum standard deviations between EGM2008 and the terrestrial gravity field for Taiwan and China have values of about 19 × 10−5 and 18 × 10−5 m/s2, respectively, and Taiwan and China are both fill-in regions, as is our study area [Li et al., 2009; Arabelos and Tscherning, 2010]. The accuracy of EGM2008 and the maximum standard deviation between EGM2008 and the terrestrial gravity field for the Korean Peninsula and northeast Asia have not been determined until now.
 The free-air anomaly field derived from EGM2008 in the research area varies from −70 × 10−5 to 120 × 10−5 m/s2 (Figure 6a). The field shows a good correlation with the elevation data portrayed in Figure 1. In general, positive free-air anomaly values are found in mountainous regions with values up to 100 × 10−5 m/s2 (e.g., close to Mount Paekdu with an elevation of 2744 m and a free-air anomaly of 120 × 10−5 m/s2). For the calculation of the complete Bouguer anomaly from the free-air anomaly, the effect of the Bouguer plate has to be removed (assuming a constant density of 2670 kg/m3), as does the effect of topography. For both computations, elevation data from the SRTM (Shuttle Radar Topography Mission) data set with an average resolution of 90 m were used [Rodriguez et al., 2005].
 Comparing the digitized Bouguer anomaly data set (Figure 7a) to the EGM2008-derived simple Bouguer anomalies, we found remarkable anomaly difference between both data sets mainly in regions of high topography (e.g., 30 mGal in Mount Paekdu). We speculated first that the difference is caused by the difference of the reference height between ellipsoid and geoid. However, the average geoid of the study area is about 25 m, of which the maximum Bouguer gravity effect is less than 4.0 × 10−5 m/s2. The remaining difference is due to the terrain effect. Therefore, we made a theoretical station data set for the study area with a distance of 10 km and performed terrain corrections until the Hayford zone O2 (outer radius of 167 km). Station heights are interpolated from the SRTM elevation data with an average resolution of 90 m. The terrain effect was calculated using in-house software and the values vary between 0 and 30 × 10−5 m/s2. Most parts of the region have terrain correction values of <10 × 10−5 m/s2. Only the highest regions in the area (e.g., Mount Paekdu) have values of >20 × 10−5 m/s2. The terrain effects of the flatlands are negligible.
 The range of the complete Bouguer anomalies mapped over the study area is from −110 × 105 to ~60 × 10−5 m/s2 (Figure 6b). The lowest regional Bouguer anomalies are found over the Ge-Ma Plateau, to the southwest of the volcanic zone. The lower Bouguer anomalies over the volcanic zone, including Mount Paekdu, Mount Wangtian, and Mount Nampotae, are consistent with the distribution of the Cenozoic volcanic rocks (hatched area in Figure 6b). The highest Bouguer gravity values of about 60 × 10−5 m/s2 are distributed along the coastline and may be caused by the crustal thinning, as shown in Figure 4 [Pak et al., 1993].
 To verify the quality and accuracy of the EGM2008-derived Bouguer anomalies, we compare these anomalies with the digitized Bouguer anomalies reported by the Jilin Bureau of Geology and Mineral Resources [Jin and Zhang, 1994]. The digitized Bouguer anomalies range from −110 to −10 × 10−5 m/s2 (Figure 7a). In general, the volcanic zone is characterized by low Bouguer anomalies, outside the volcanic zone anomaly value more positive of about −10 × 10−5 m/s2. The comparison (Figure 7b) indicates that the two data sets are consistent, with a mean difference of <10 × 10−5 m/s2 and a standard deviation of about 8.0 × 10−5 m/s2 (Figure 7c). More than 90% of the Bouguer anomalies derived from EGM2008 (blue line in Figure 7d) along seismic profile A-A′ (dotted line in Figures 7a and 7b) coincide with the digitized Bouguer anomalies (red line in Figure 7d).
 To calculate the isostatic residual field, the gravity effect caused by the variation in depth of the Moho was subtracted from the complete Bouguer anomaly (Figure 6b). The gravity effect of the Moho was calculated from an isostatic gravity field and topography by using an Airy-Heiskanen model with a density contrast of 350 kg/m3 between the lower crust and the upper mantle. The deepest Moho (42 km) is calculated for the Ge-Ma Plateau and the shallowest (<30 km) for the coastal area (Figure 8a). The depth of the Moho in the volcanic zone under the Mount Paekdu varies from 35 to about 40 km. These gravity-computed Moho depths are consistent with previous findings [Tang et al., 1998; Shin and Baag, 2000; Zhang et al., 2002a, 2002b; Hetland et al., 2004; Song et al., 2007]. The gravity effect due to the variation in depth of the Moho is then calculated by applying the Parker-Oldenburg algorithm [Parker, 1972; Oldenburg, 1974] assuming a crust-mantle density contrast of 350 kg/m3.
 The isostatic anomaly map (Figure 8b), which is derived from the Bouguer anomaly map (Figure 6b) by subtracting the gravity effect of the Moho, portrays the gravity effect from the density distribution in the crust only and probably reflects the gravity effect of volcanic rocks such as alkali basalt much better than do the Bouguer anomalies. Furthermore, the most negative local anomaly of about −40 × 10−5 m/s2 coincides with the location of the Cheonji caldera lake, under which a deep-reaching magma chamber evidently exists [Tang et al., 1998; Zhang et al., 2002a, 2002b; Hetland et al., 2004]. Positive anomalies of >60 × 10−5 m/s2 mapped along the coast can be explained by mantle material uplifted into the lower crust [Pak et al., 1993].
5 Interpretation and Analysis of the Gravity Field
5.1 Curvature Analysis
 Originally, curvature analysis was used to help interpret 3-D seismic arrays [Roberts, 2001], and the method has been already tested with gravity data [e.g., Choi et al., 2011]. Curvature analysis is particularly useful for enhancing linear elements (e.g., gravity lineations in an isostatic gravity field) and can help identify the orientations of the tectonic elements that cause gravity anomalies.
 The “dip curvature” (Figure 9) is calculated in the direction of the largest dip of the combined gravity surface [Roberts, 2001]. The parallel pairs of positive values (red colors in Figure 9) and negative values (blue colors in Figure 9) of the dip curvature indicate areas where strong density contrasts occur in the crust: (1) the parallel pair line (dotted line numbered with 1 in Figure 9) is correlated with the boundary of the vicinity of the volcanic rocks. We interpret that the dotted line is caused by the prominent density contrast between the basement (average density of 2600 kg/m3) and the alkali volcanic rocks (average density of 2300 kg/m3) (Table 1). The resulting area of Cenozoic volcanic rocks is calculated to amount to 32,000 km2, almost 50% greater than previously estimated (18,350 km2 [Yun et al., 1993]). (2) The most prominent parallel pair line (dashed line numbered with 2 in Figure 9) indicates an oval subsurface structure directed northeastward from Mount Wangtian (“W” in Figure 9) to Mount Paekdu (“P” in Figure 9) and is roughly consistent with the distribution of recent volcanic rocks, such as pumice with a mean density of <2200 kg/m3 (Table 1). The circle numbered with 2 can be regarded to represent the most recent volcanic eruption in A.D. 1000.
5.2 Euler Deconvolution
 To obtain information regarding the distribution of 3-D source points in the volcanic area prior to forward modeling, we use Euler deconvolution. The Euler deconvolution technique has been applied to potential field data for more than two decades [e.g., Thomson, 1982; Reid et al., 1990; Pašteka et al., 2009] and assists the interpretation of potential fields in areas of simple density distributions (e.g., dike-like or isolated structures). The degree of homogeneity N in the Euler equation can be interpreted using the structural index (SI), which is a measure of the rate of change of a field with distance [Pašteka et al., 2009], including the center of mass point (SI = 2), the center of a dike-like feature (SI = 1), and the position of the top of a dike-like feature or tectonic boundary (SI = 0). Euler source points were calculated from the isostatic residual field with the in-house program REGDER. Pašteka et al.  proposed using SI = 2 for calculations of the center of mass source points. The mass points located between 4 and 11 km depth (red-, blue-, and green-filled circles in Figure 10) in the volcanic zone (grey hatched area in Figure 10a) are clustered mainly around Mount Paekdu. Comparing our results to the seismic velocity structures (Figure 10b) along profile A-A′, the distribution of shallower and deeper source points around Mount Paekdu refers to existence of a vertical, isolated cylinder-like structure with a maximum thickness of about 10 km, which extends between the Baishan-Tianchi Fault (2 in Figure 10) and the Mi'anshan-Sanbaobaihe Fault (6 in Figure 10). The diameter of the structure is estimated to be about 20 km.
6 Three-Dimensional Gravity Modeling
6.1 Method and Constraining Data
 The software used for the gravity modeling was IGMAS+, an extension of the original 3-D modeling package IGMAS developed by Götze ; IGMAS+ is a modern interpretation system with geographic information system functionality [Götze and Schmidt, 2002; Schmidt et al., 2004, 2010]. The system uses polyhedrons with triangulated surfaces to approximate bodies of density and/or magnetic susceptibility within the Earth's crust and mantle, the geometries of which are defined by a number of parallel vertical modeling sections [Götze and Lahmeyer, 1988]. The positions of the 35 vertical model sections used here are chosen to be parallel to the north-south seismic profile (A-P-A′-A″ in Figure 1; see Figure 3 for the P wave velocity distribution). The distances between adjacent vertical sections varied between 2 and 20 km depending on the complexity of near-surface geology.
6.2 Constraints on Geometry and Densities
 The interpretation of potential field data requires interdisciplinary knowledge and the integration of information from various types of independent data. Several independent constraints, obtained from geophysical, geological, and tectonic interpretations, are taken into account:
 An SRTM elevation model with a 2 km grid (Figure 1) was used to define the most superficial geometries.
 The modeled structures in the topography and the subsurface were defined on the basis of the geological and tectonic map (Figure 2).
 To obtain direct evidence on the thicknesses, structures, and densities of features between the surface and the Moho, seismic wide-angle refraction/reflection profiles (Figures 3 and 4) were used, together with density-velocity relationships (Figure 5). The A-P-A′-A″ seismic profile (Figure 3a) defined velocity discontinuities at depths of about 12 km (6.1 to 6.26 km/s, C1 in Figure 3a), 20 km (6.27 to 6.4 km/s, C2 in Figure 3a), 30 km (6.4 to 6.6 km/s, C3 in Figure 3a), and 35 km (6.8 to 8.0 km/s, Moho in Figure 3a). From these velocity discontinuities, we calculate the densities of modeled crustal layers (Table 2 and Figure 11b) using velocity-density relationships (Figure 5). The most significant feature revealed along the seismic lines is the LVZ distributed from a depth of 10 km to the Moho directly under Mount Paekdu (Figure 3a).
 Density determinations obtained from existing geological and petrophysical information are used to estimate model densities (Tables 1 and 2).
 The complete Bouguer gravity field (Figure 6b) is used to fit the density contrasts in the crust.
 The spatial distribution of the Moho, computed using the Airy-Heiskanen model, is used to define a regional Moho gravity effect in the 3-D model.
 Dip curvature calculations (Figure 9) are used to interpret the existing anomalies and density sources within the subsurface, such as alkali basalt.
 The results of Euler calculation (Figure 10) are used to obtain preliminary information on the spatial distribution of source points.
6.3 Interpretation and Results
 A modeled section (Figure 11) along the A-P-A′-A″ seismic profile [Tang et al., 1998; Zhang et al., 2002a, 2002b] reveals that the gravity field is characterized by a very low gravity anomaly of about −100 × 10−5 m/s2 over the Mount Paekdu area (red line in Figure 11a). The density model has been modified by (i) using Bouguer anomalies, (ii) calculating the density distribution from seismic velocities (Table 2 and Figure 5), (iii) computing the variation in the depth of the Moho using the Airy-Heiskanen model (red dotted line in Figure 11b), and (iv) incorporating petrophysical analysis for the three LVZs (pink, bright red, and pale brown polygons in Figure 11b), which have a velocity 3% lower and a density 10% lower than that of the surrounding material (Table 2). Because of technical aspects by modifying magma chambers, we made one polygon for two magma chambers located between lower crust and the Moho as shown in Figure 11b. The joined magma chamber in the model, however, has the same thickness as two magma chambers have, and the density is averaged from two densities of both chambers. Therefore, the modified three magma chambers have the same gravity effect as the four magma chambers as shown in the seismic profile (Figure 11c). The ascending regional anomalies (grey line marked by “M” in Figure 11a) from south to north can be explained by the variation in the Moho depth (red dotted line in Figure 11b) from about 40 km in the Mount Paekdu area to about 30 km near the city of Dunhua to the north. The depth of the Moho calculated by the Airy-Heiskanen isostatic model agrees well with the Moho digitized from the velocity model.
 The gravity field (brown line marked with “Z” in Figure 11a) calculated by incorporating the gravity effects of crustal layers and LVZs is roughly correlated with the distribution of density structures; however, very low anomalies over the Mount Paekdu area do not fit the calculated anomalies. To enhance the match between the observed and computed Bouguer anomalies in the Mount Paekdu area, it is necessary to modify the density structures for this area. As mentioned in section 2, the volcanic zone including Mount Paekdu is covered by alkali basalt with a mean density of 2200 kg/m3 (Table 1 and Figure 2) and a maximum thickness of around 1.5 km [Yun et al., 1993; Wang et al., 2003]. When considering the spatial distribution of the volcanic rocks in the subsurface (Figure 11b), we found a better fit between the observed and computed anomalies in the area (black line marked by “V” in Figure 11a) but an anomaly difference of about 25 × 10−5 m/s2 remained. For the best fit (blue line marked by “C” in Figure 11a), we additionally require a cylinder-like density structure directly under Mount Paekdu with a thickness of around 5 km, a width of around 20 km, and a density of about 2350 kg/m3, which is supported by the interpretation of curvature analysis (circles numbered “2” in Figure 9) and by Euler calculations (Figure 10). The modified cylinder-like structure (red polygon marked “MC” in Figure 11b) is assumed to be a magma chamber, which is probably related to the most recent eruption in A.D. 1000 [Yun et al., 1993; Wang et. al., 2003; Stone2010; Xu et al., 2012].
 Therefore, on the basis of the modeling results for this area, we conclude that the very low Bouguer anomalies found over the Mount Paekdu area are caused by the combined gravity effects of the deepest Moho in the area, an LVZ with a density 10% lower than the surrounding material, volcanic rocks with lower densities than the adjacent area, and the predicted magma chamber with a density of about 2350 kg/m3. The predicted magma chamber (red polygon in Figure 11c) sits above LVZ1 (dark yellow polygon in Figure 11c), which extends from the Mount Paekdu area to the northern part of the volcanic area and ranges in depth from 15 km in the south to 9 km in the north. The volumes of the magma chamber and LVZ1 are estimated to be about 800 and 40,000 km3, respectively.
 A cylinder-like magma chamber is predicted to exist not only under Mount Paekdu but also under both Mount Wangtian and Mount Nampotae (Figure 12). These chambers have caused the most negative isostatic anomalies (< −40 × 10−5 m/s2) (Figure 12a) and the prominent high elevations (>2 km) of the three volcanoes (Figure 12b). The density of the magma chamber under Mount Nampotae (2450 kg/m3) is higher than the densities of the chambers under Mount Paekdu and Mount Wangtian (2350 kg/m3) (Figure 12c) suggesting that Mount Nampotae is an inactive volcano. The active volcanoes, Mount Paekdu and Mount Wangtian (“P” and “W” in Figure 12a), are separated from the inactive volcano Mount Nampotae (“N” in Figure 12a) by the Changbaishan-Zhenfeng Fault line (line marked “7” in Figures 12a and 12b).
 The gravity field along the NW-SE seismic profile (line P-N-D-C in Figure 1) (Figure 13a) varies from a very low anomaly of about −100 × 10−5 m/s2 observed over Mount Paekdu to the most positive anomalies of >80 × 10−5 m/s2 on the east coast. For structural modifications of the crust, we assumed that Mount Nampotae and Mount Duru both contain magma chambers at depths of about 5, 10, and 20 km (as for Mount Paekdu). However, the structural model in which the densities of magma chambers under Mount Nampotae and Mount Duru are the same as those under Mount Paekdu (Figure 13c) generated a notable anomaly difference of about 40 × 10−5 m/s2 in the area of Mount Duru (Figure 13a). Therefore, we modified the petrophysical properties of the magma chambers (Figure 13b) to optimize the model and obtain the best match between the observed and computed Bouguer anomalies (black and red lines in Figure 13a). As a result, we find that the local positive anomalies over Mount Nampotae (70 to 90 km along the profile in Figure 13a) and Mount Duru (100 to 120 km along the profile) can be explained by magma chambers with higher densities than the chamber under Mount Paekdu. The final density modification (Figure 13b) allows us to distinguish the petrophysical characteristics of the three volcanoes. The density of the magma chamber under Mount Nampotae is higher than that under Mount Paekdu, possibly reflecting the lower temperature of Mount Nampotae chamber as proposed by Denosaquo et al. . Mount Duru, probably the oldest volcano found along the profile and suspected to be inactive, has a magma chamber system with the highest density and lowest temperature among the three volcanoes.
 Combining our results with geological and petrophysical analysis (e.g., Figures 2 and 5) provides new insights into volcanic activities in the study area. We speculate that the volcanic activities have migrated toward the northwest from the east coast to Mount Paekdu, probably along the SE-NW trending Baishan-Tianchi Fault line (2 in Figure 2). The magma chamber found under Mount Duru might have been constructed during/after the volcanic eruptions that took place in the Pliocene and early Pleistocene (4.4–1.7 Ma). The locus of volcanism continued to migrate northwestward, with activity in the Mount Nampotae area occurring in the Pleistocene (0.6–0.1 Ma) while Mount Duru was cooling. Since the Holocene, volcanoes including Mount Paekdu and Mount Wangtian, located to the NW of Mount Nampotae, have been activated. Meanwhile, the intruded magma chambers under Mount Duru have completely cooled to form volcanic rocks with higher densities than the adjacent material. We also speculate that Mount Nampotae is not yet completely inactive.
 As mentioned in section 1, there are notable discrepancies in seismic interpretations regarding the location of the LVZs underlying the study area. The most recent seismic interpretation [Song et al., 2007] suggested that a midcrustal LVZ might exist under the region 70 km to the north of Mount Paekdu, while most other previous results [e.g., Tang et al., 1998; Zhang et al., 2002a, 2002b] indicated that the LVZ should lie only under Mount Paekdu. The receiver function study of Hetland et al.  resolved LVZs under both areas. In section 6, we modeled density structures along the A-P-A′-A″ seismic profile as proposed by the previous seismic interpretation of Zhang et al. [2002a, 2002b] and explained that the low Bouguer anomalies found over Mount Paekdu are caused by the combined gravity effects of the deep Moho, LVZs with lower densities than the surrounding material, volcanic rocks covered by other deposits in the area, and the predicted magma chamber underneath the volcano.
 To investigate other possible locations of LVZs as proposed by Song et al.  and Hetland et al. , we constructed two other simplified density models. The first model (Figure 14d) incorporates an LVZ located about 70 km to the north of Mount Paekdu as shown in the P wave velocity model (Figure 14e) [Song et al., 2007]. According to the velocity-density relationships (Figure 5) [Denosaquo et al., 2009], the density of the LVZ with a P wave velocity of 5.6 km/s is calculated as being about 2550 kg/m3, which is lower than that of the surroundings by about 200 kg/m3. The gravity anomalies computed from the model (green line in Figure 14a) are not consistent with the observed anomalies (black line in Figure 14a), producing a remarkable difference of >30 × 1−05 m/s2 over the LVZ, 100 to 150 km along the profile. The second model characterized two LVZs, one located under Mount Paekdu and one under the region to the north of Mount Paekdu (Figure 14c) [Hetland et al., 2004]; this result also yields a large inconsistency between computed anomalies (blue line in Figure 14a) and observed anomalies. The results of these different density models indicate that the location of the LVZ is confined to Mount Paekdu (Figure 14b), and the other possible location to the north is incompatible within the constraints of the data and modeling.
 To provide detailed insights into the spatial distribution of crustal density structures in the Mount Paekdu area, located on the border between North Korea and China, we interpreted an EGM2008-derived gravity field data set by using (1) curvature analysis and Euler deconvolution of the EGM gravity fields and (2) 3-D forward density modeling that incorporated constraints from existing geological and geophysical information. The accuracy and quality of the EGM2008-based gravity field in the research area were verified by comparing it with the terrestrially measured data set of Jin and Zhang .
 We were able to closely constrain crustal density structures and the magma chamber system in the Mount Paekdu volcanic area. The depth of the Moho varies from about 40 km under Mount Paekdu to about 30 km in the northern part of the study area. Very low Bouguer anomalies of about −100 × 10−5 m/s2 over Mount Paekdu are caused by the combined gravity effects of the deep Moho, a low-velocity zone (LVZ) with a density 10% lower than the surroundings, volcanic rocks on the surface with an average density of 2200 kg/m3, and a predicted magma chamber with a density of about 2350 kg/m3 and a mean thickness of about 5 km. Magma chambers are also predicted to lie beneath Mount Wangtian and Mount Nampotae. However, a previously proposed LVZ in the area to the north of Mount Paekdu is not confirmed by the results of the 3-D density modeling. Volcanism in the study area is proposed to have migrated northwestward from the east coast along the SE-NW trending Baishan-Tianchi Fault line.
 The study was carried out as part of the Korea-German collaborative project “Synoptic Modeling of the Lithosphere in the Korean Peninsula” and the project “Satellite-derived Gravity Modeling in Mt. Paekdu.” We appreciate funding from the Deutsche Forschungsgemeinschaft (DFG; German Research Foundation, grant GO 380/30-1) and a grant from the Korean Meteorological Administration (CATER 2012–5062). The quality of the manuscript benefits from comments of Joerg Ebbing, Tom Parsons, and an anonymous reviewer. Thank you for your efforts.