Transitional continental–oceanic structure beneath the Norwegian Sea from inversion of surface wave group velocity data

We have analysed the fundamental mode of Love and Rayleigh waves generated by 12 earthquakes located in the mid-Atlantic ridge and Jan Mayen fracture zone. Using the multiple ﬁlter analysis technique, we isolated the Rayleigh and Love wave group velocities for periods between 10 and 50 s. The surface wave propagation paths were divided into ﬁve groups, and average group velocities calculated for each group. The average group velocities were inverted and produced shear wave velocity models that correspond to a quasi-continental oceanic structure in the Greenland–Norwegian Sea region. Although resolution is poor at shallow depth, we obtained crustal thickness values of about 18 km in the Norwegian Sea area and 9 km in the region between Svalbard and Iceland. The abnormally thick crust in the Norwegian Sea area is ascribed to magmatic underplating and the thermal blanketing e V ect of sedimentary layers. Maximum crustal shear velocities vary between 3.5 and 3.9 km s − 1 for most paths. An average lithospheric thickness of 60 km was observed, which is lower than expected for oceanic-type structure of similar age. We also observed low shear wave velocities in the lower crust and upper mantle. We suggest that high heat ﬂow extending to depths of about 30 km beneath the surface can account for the thin lithosphere and observed low velocities. Anisotropy coe Y cients of 1–5 per cent in the shallow layers and > 7 per cent in the upper mantle point to the existence of polarization anisotropy in the region.

INTRODUCTION the crustal structure of the region. Deep sea drilling projects The Norwegian-Greenland Sea region ( Fig. 1) is tectonically have found that the reflections correlate with the top of basalts complex. Whilst the seafloor spreading process along the and that these basalts were extruded at very high rates during ridges north of the Jan Mayen fracture zone proceeded symthe initial stage of spreading in the area (Eldholm & Sundvor metrically, the process south of the zone has been characterized 1980). by westward shifts of the spreading axis (Eldholm et al. 1990).
Although extensive studies have been carried out to deter-This has resulted in extinct ridges scattered through the mine the crustal thickness and velocity structure of the southern part of the study area. The Jan Mayen, Greenland Norwegian continental margin (e.g. Chroston & Brooks 1989; and Senja fracture zones intersect the area (Fig. 1), and the Jan Planke et al. 1991;Mjelde et al. 1992;Kodaira et al. 1995, Mayen Ridge micro-continent further complicates the tectonics. 1998), very little work appears to have been undertaken on Oceanic basins containing thick layers of sedimentary rocks the lithospheric structure (especially the upper mantle) of the are found east and west of the mid-Atlantic ridge. Those east margin and Norwegian Sea area. Until now, surface wave of the mid-Atlantic ridge are bounded on their eastern margins studies have been impractical since most of the earthquake by the Vøring plateau and Faeroe-Shetland escarpments. The events that occur in the vicinity of this region are small Vøring and Greenland-Faeroe plateaux were formed by the (M<5.0). Published studies concentrating on the lithospheric Iceland hot spot ( Vink 1984).
structure are of regions outside our area of interest (e.g. Chan Reflections observed beneath the smooth acoustic basement & Mitchell 1985;Clark 1983;Evans & Sacks 1979Lévêque 1980;Calcagnile & Panza 1978). This study attempts of the Vøring plateau and in the Lofoten basin (Eldholm &  to estimate the crustal and upper mantle structure of this associated with fracture zones (e.g. the Jan Mayen fracture zone). High-quality long-period and broad-band records of region using group velocities of the fundamental mode of these earthquakes are available from seismic stations operated Love and Rayleigh waves. This is achieved by simultaneously by the University of Bergen (UiB) and the Norwegian Seismic inverting Rayleigh and Love wave group velocities. Owing to Array (NORSAR). its simplicity, this method has been widely used to estimate We selected 12 earthquakes (Table 1) recorded at five seismic crust and upper mantle structure. We did not take into account stations (Fig. 2). Of the five stations, Kongsberg (KONO) and the effect of anisotropy, but performed a simple test to assess Kings Bay (KBS) are operated by the seismology group at the extent to which polarization anisotropy exists in the region UiB, whilst the other three are part of the NORESS (NRS), of study. We believe that the results obtained in this study ARCESS (ARC) and Svalbard (SPB) arrays operated by the constitute a valuable basis for future velocity structure studies Norwegian Seismic Array group (NORSAR). in this region.
We obtained three-component long-period data from all five stations. In addition, KONO station and the three arrays SEISMOLOGICAL DATA operate three-component broad-band stations. The magnitude The region shows moderate seismicity, with the highest concenof the earthquakes selected is such that the surface waves produced at the recording stations are not very large. However, tration of epicentres along the mid-Atlantic ridge (MAR) and  ( Table 1) recorded at KONO. The signal-to-noise ratio is good between 0.01 and 0.2 Hz. This record is typical of those used in the dispersion analysis.

DISPERSION ANALYSIS
Fundamental mode Love and Rayleigh waves were extracted using the multiple filter technique (MFT) to isolate the group velocities (Dziewonski et al. 1969;Herrmann 1973). The Rayleigh wave group velocities were obtained from the radial and vertical components, and the Love wave velocities from the transverse component. Instrument responses were removed using the  software package (Havskov 1997). The MFT was then applied to the corrected amplitude spectra yielding a plot of amplitude as a function of group velocity and they are recorded with a good signal-to-noise ratio, and show dispersion. period (Fig. 4). Fig. 4 shows a typical example obtained from the long-  Because of the tectonic complexity of the region and to reduce the effects of lateral structural differences, we grouped together paths passing through similar structures and travelling in approximately the same direction. Thus, the propagation paths shown in Fig. 2 are divided into five groups: W, W1, W2, W3 and W4 ( Table 2). The propagation paths of group W traverse the region between Iceland and Svalbard, crossing the Iceland plateau. Group W1 paths cover a region including the Lofoten and Greenland basins, and also cross the Senja fracture zone and the Mohns and Knipovich ridges. W2 propagation paths cross the region between Iceland and the Norwegian mainland including the Iceland-Faeroe ridge and Faeroe-Shetland escarpment. The paths of groups W3 and W4 traverse regions with similar structure, the difference between the groups being in the orientation of the paths as shown in Fig. 2. W3 paths cross the Lofoten basin as well as Table 2. The propagation paths between epicentres and stations, divided into the five groups W, W1, W2, W3 and W4. (The numbers correspond to the epicentre order as in Table 1 (Table 1) recorded at the NRS array. Black squares represent the maximum spectral amplitude at each period and correspond to the fundamental mode. The trace to the far right is that obtained after band-pass filtering the original signal using a Gaussian filter, and, immediately to the left, the same trace is plotted as amplitude versus apparent velocity (Herrmann 1987).
NORESS array. Dispersion errors at short periods are due to effects such as horizontal reflection in inhomogeneous media, Table 3. Layer parameters of the three initial models used in the multipathing and high-frequency interference, but were reduced inversion for the shear wave velocities.
through filtering during the MFT. Scatter observed in the group velocities is a reflection of both systematic and random errors. These include errors in the traveltime due to source finiteness and rise time. We did not correct for these errors since earthquakes used in this study are of so small a magnitude that the error term is insignificant. For example, earthquakes of magnitude M b =5.6, would result in rise time errors of about 1 s, and source finiteness errors of about 4 s (Chave 1979;Clark 1983). This would result in a typical traveltime error of 5 s. An average path length of about 1500 km and Love wave velocity of 4.5 km s−1 gives rise to an error of 0.079 km s−1 in group velocity. Hence, since the majority of earthquake events used in this study are of magnitude less than 5.0 ( Table 1), most of the velocity errors from this study are even less than 0.079. Further systematic errors are introduced by epicentre mislocations and in data processing. The error due to epicentre mislocation increases as the epicentral distance decreases. Other workers have shown that these errors also have a negligible effect on the group velocity. For example, Clark (1983) showed that similar earthquake sizes and epicentral distances produce errors of ±0.075 km s−1. Singh (1987) has shown that instrumental errors, origin time errors, interference due to wave propagation at contact boundaries of distinct lithological structures, and mode conversion result in an error of about ±0.04 km s−1 in the group velocity estimates. On the strength of the results from the above-mentioned studies, we did not correct for any of the errors mentioned. the sheared ridges in the region, whilst those from group W4 As a result of the different structures across which all these paths travel, the estimated group velocities are an average cross the Norway basin, Jan Mayen fracture zone and the Iceland-Faeroe ridge. Henceforth, the regions will be referred of the velocities along each path. For each group of paths (W-W4), we calculated the average of the group velocities for to by the name of the group of paths crossing them. For example, the region traversed by paths making up group W each period. For each mean velocity, we calculated the standard deviation using the standard relation. will be referred to as region W. where W i is the component of the weighting matrix, inversely proportional to the layer thicknesses in the model (Mitchell 1976), l i is the damping factor, s2 is an adjustable parameter called the problem variance, and V ij are the eigenvectors associated with the rows of A, where A is the matrix containing the partial derivatives of group velocities with respect to the shear wave velocity (Rodi et al. 1975). The reliability of the estimated models is shown by resolution kernels obtained from the rows of the resolution matrix (Jackson 1972;Wiggins 1972).

RESULTS AND DISCUSSION
Computed and observed group velocities for both Love and Rayleigh waves are shown in Fig. 5. The computed group velocities generally fall within the standard deviation of the observed velocities. Since anisotropic effects are not accounted for, standard deviation values of the observed velocities in all the groups except group W are large. Even in group W, the standard deviations of the velocities at short periods are large. These, however, decrease with increasing period, reflecting the homogeneity of the velocity structure with depth. Because of the small size of the earthquakes used (Table 1), the longer periods are less reliable since the amplitudes at these periods are small. Christensen et al. (1980) observed a similar effect when they used events of similar magnitudes to study the deep structure beneath the Atlantic Ocean. Therefore, to reduce the impact of this effect, only periods up to 50 s were used in this study.

INVERSION OF GROUP VELOCITY DATA
The group velocities observed over the whole region are The shear wave velocity models were estimated by simulgenerally lower than the North Atlantic Ocean values obtained taneously inverting Love and Rayleigh wave group velocities by Christensen et al. (1980), but are similar to those obtained using the Herrmann (1987) software package. The inversion for the Iceland plateau by Evans & Sacks (1979). The exception subroutines given by Lawson & Hanson (1974) and used in is W1 group velocities, which are very low compared with this package are based on a general form of the stochastic velocities observed in the other four groups. The maximum inversion procedure which is equivalent to the damped leastvelocities are less than 3.5 km s−1. The low velocities, especially squares method (Herrmann 1987). Details of this method are at short periods, may be a reflection of the thick sedimentary discussed in Jackson (1972) and Wiggins (1972). A basic layers (up to 10 km) in this part of the region. outline is given below.
The group velocities were inverted to obtain the shear wave A starting model is selected and is used to compute the velocity models shown in Fig. 6. The top layer, which represents dispersion curves and partial derivatives of the group velocities the water depth, is given an average thickness of 1.5 km for with respect to the unknowns. The dispersion curves are the whole region. Although we also inverted for depth, most compared with the observed data, and model corrections are of the layer thicknesses do not show much change from those of computed. After applying first-order corrections to the starting the initial models. Most of the observed changes are in the model, the whole process is repeated until the changes in shear velocities. An examination of the models along with their the model are small. In our calculations, the inversion was resolving kernels (Fig. 7) enables one to judge where features terminated once the model rms error was small (e.g. 0.0001) in the models are artefacts of the inversion process. The and had stopped improving. resolution kernels plotted in Fig. 7 correspond to the velocity A simple oceanic model was selected as the principal starting models for the regions traversed by the paths making up model. To test for stability, two further models (Table 3) were groups W, W2, W3 and W4. The kernels shown in the plots also used as initial models. We inverted the group velocities are quite similar, implying that resolution is uniform across for both shear wave velocity and depth of interface. Initially, the whole region. The maxima of the curves coincide with the the layer thicknesses were held constant whilst inverting for layer depths up to about 100 km below the surface. The width velocity. Inversion for interface depth was only undertaken at the of the resolving kernels at shallow depths (<20 km) and at final stage once the computed dispersion curves for both Love great depths (>100 km) is wider than the layer thicknesses at and Rayleigh waves fitted well within the data uncertainties of the same depths, implying that the resolution of layer structure the observed dispersion values (Fig. 5).
at those depths is poor. Thus the results obtained here are The standard deviation for the shear velocity in each layer most reliable for the sub-Moho velocity structure. The model of the model was calculated using the relation given by Mitchell layer standard deviations ( horizontal bars on model layers in (1976) for the stochastic least-squares approach to inversion: Fig. 7) seem to reflect the standard deviations observed in the group velocity data. This is illustrated by the low group velocity standard deviation for group W, which corresponds to the low model layer standard deviation shown in Fig. 7. We Figure 6. Shear wave velocity models resulting from the simultaneous inversion of Love and Rayleigh wave group velocity data for the four different regions traversed by paths in groups W, W2, W3, W4. The thin lines represent the three models obtained by inverting from the three initial models for each region, and the thick line is an average of the three models in each group of paths.
believe that the scatter is due to azimuthal anisotropy, which (1980). Since this is an average value depending on the structures crossed by paths making up group W, the thick is not accounted for here. The effect is low in the region traversed by paths making up group W, as shown by the low lithosphere beneath the Iceland plateau (Evans & Sacks 1979) may well have influenced the results for this group. The final standard deviations. The shear velocity model for region W is very similar to model for region W1 appears to be quite unrealistic and thus it was not included in the final results. The very low group young oceanic-type structure, especially as regards the shear velocity values in the lithosphere. The thickness of the crust is velocity values obtained are reflected in the unrealistically low shear wave velocities in the model. approximately 10 km. This value must be regarded as approximate since resolution is poor at this depth. The lithospheric The models for regions W2, W3, and W4 differ significantly in upper lithospheric structure from a typical oceanic structure. thickness value of about 60 km is, however, larger than expected when compared to results obtained by Evans & Sacks The maximum shear wave velocities in the crust vary between Figure 7. Resolving kernels (right side) for the best-fitting models (left side) for regions traversed by paths in groups W, W2, W3 and W4.
3.5 and 3.9 km s−1. These are quite low compared with ness values. Vink (1984) proposed the hot-spot-fed rise model, with the Iceland hot spot producing vast quantities of basalt, expected oceanic values (i.e. greater than 4.0 km s−1). The crustal thicknesses for regions W2, W3 and W4 were estimated which cooled to form areas of thick crust and plateaux in the region. Faerseth et al. (1995) and Talwani & Eldholm (1977), at about 20 km. This value is higher than typical oceanic-type structure, but comparable to values obtained by other studies. among others, attribute the thicker than normal crust to extension processes which resulted in the creation of the By inverting surface wave group velocities, Clark (1983) found crustal thickness values of about 20-24 km in the Iceland-Norwegian Sea. The presence of deep post-Jurassic sedimentary basins and Faeroe region. Other geophysical studies by Chroston & Brooks (1989), Planke et al. (1991 and Mjelde et al. (1992) thick sedimentary layers on highs as well as in the Barents Sea area is reflected in the low crustal velocities and can also explain suggest values of 25-30 km in the Lofoten basin, whilst Evans & Sacks (1979)  Results from several studies on the evolution of the layers result in the metamorphism of existing sedimentary rocks, thereby increasing the thickness of the crust (Singh 1988). Norwegian Sea seem to explain these abnormal crustal thick- Lithospheric thickness and upper mantle shear wave and Singh's models were derived using surface wave dispersion data. Crustal thicknesses beneath the Arabian Sea vary between velocities are much better resolved than crustal values. All the models exhibit a lithospheric thickness of about 60 km. 16 and 28 km, similar to values obtained in this study. Singh (1988) concluded that thermal blanketing effects played Between approximately 10 and 30 km below the surface, the shear velocities observed in the models for regions W2, W3 a major role in generating the abnormally high thicknesses. Thus, considering the tectonic similarities of the two environ-and W4 are low, varying between 3.5 and 4.5 km s−1. Below the lid is a low-velocity zone, the bottom of which is not ments, our crustal thicknesses are reasonable, despite the poor resolution of the upper part of the lithosphere in our resolved accurately.
In Fig. 8, the shear velocity models obtained in this study study.
The results from this study were also compared with a are compared with three other shear velocity models, namely The Arabian Sea model (Singh 1988), a 45 Myr old oceanic 45 Myr old oceanic crust model (Evans & Sykes 1980) and with the Baltic Shield structure as observed by Calcagnile & crust (Evans & Sykes 1980) and the Baltic Shield (Calcagnile & Panza 1978) model. The Arabian Sea area studied by Singh Panza (1978). As expected, the average lithospheric depth is smaller than the Shield value. Kanamori & Press (1970) found (1988) is a continent-ocean transition zone, with the oceanic part of similar age to the Norwegian Sea area studied here, an ocean basin lithospheric thickness value of 70 km, which compares well to the value of 65 km obtained by Evans & The Love wave model has larger velocities at a depth of about 30 km below the surface. There is a clear mismatch between the Sacks (1980) for oceanic structure of age 45 Myr. Compared with these, our value of 60 km is low. models in the other three regions. This difference in the models points to the existence of polarization anisotropy in the whole The problem of crust and upper mantle elastic anisotropy is not addressed in this study. However, the large standard region. Having attributed the difference to anisotropy, the anisotropy coefficient x for all regions is estimated by deviations shown in Fig. 5 point to the existence of azimuthal anisotropy. We investigated the presence of anisotropy by performing a simple test for polarization anisotropy. Love and x= SH−SV S mean .
( 2 ) Rayleigh wave group velocities were inverted separately for each of the regions W, W2, W3 and W4. The results are The shallow layer coefficients are 1-5 per cent for all regions except region W2, where the value is extremely large (>10 per shown in Fig. 9. Rayleigh and Love wave models for region W3 are not very different from the model obtained after the cent). The coefficients between the crust and lithosphere lid are also large (>7 per cent). joint inversion of Love and Rayleigh wave group velocities. continent-ocean transition covered by the paths making up CONCLUSIONS groups W2, W3 and W4 has been shown to have a crustal thickness of between 18 and 20 km, which compares well with We have presented the results of a study of Rayleigh and Love wave dispersion across the Norwegian-Greenland sea previous geophysical studies in the region. However, lower crustal and upper mantle shear wave velocities are much region. The models obtained for the region covered by the paths making up group W ( between Svalbard and Iceland) lower than expected, possibly reflecting high heat flow values extending to depths of about 30 km beneath the surface. are similar to those of typical young oceanic-type structures; however, the observed lithospheric thickness of about 60 km Average lithospheric thicknesses of 60 km are observed. Evans & Sacks (1980) found that ocean of crustal age 45 Myr has a is greater than expected for young oceanic-type structure. This can be attributed to the contribution of the Iceland plateau lithospheric thickness of about 65 km. Thus our value is lower than expected, especially considering that the region under velocity structure to the final average structure. The low crustal thickness value of 9 km is reasonable for this area and is study is a continent-ocean transition zone. It is possible that the thick sedimentary layers in the region have reduced the probably due to the influence of the young oceanic region near the ridge segments crossed by the group W paths. The normal cooling rate of the lithosphere, resulting in lower shear