Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 Investigations of the behavior of the geomagnetic field on geological timescales rely on globally distributed data sets from dated lava flows. We present the first suitable data from the Arctic region, comprising 37 paleomagnetic directions from Jan Mayen (71°N, 0.2–461 ka) and Spitsbergen (79°N, 1–9.2 Ma) and five paleointensity results. Dispersion of the Arctic virtual geomagnetic poles over the last 2 Ma (27.3 ± 4.0°) is significantly lower than that from published Antarctic data sets (32.1 ± 5.0°). Arctic average virtual axial dipole moment (76.8 ± 24.3 ZAm2) is high in comparison to Antarctica over the same time interval (34.8 ± 8.2 ZAm2), although the data are still too sparse in the Arctic to be definitive. These data support a long-lived hemispheric asymmetry of the magnetic field, contrasting higher, more stable fields in the north with lower average strength and more variable field directions in the south. Such features require significant non-axial-dipole contributions over 105−106 years.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The fundamental assumption used for plate reconstructions is that when averaged over geological timescales, the Earth's magnetic field can be approximated by a geocentric axial dipole (GAD). This simplicity is reflected in many geomagnetic field models derived from observations and numerical geodynamo simulations of convection in the Earth's liquid outer core. However, significant departures from a GAD field are observed in the present field (SA in Figure 1), and measurements over historical and archaeological timescales [e.g., Korte et al., 2011] indicate that non-axial-dipole effects persist at least over the last 10 kyr. In this paper, we explore the geomagnetic field over geological timescales at high latitude and demonstrate long lived asymmetries in field behavior between the northern and southern hemispheres. This asymmetry has consequences for models explaining the origin of the Earth's magnetic field, particularly those that invoke variations in heat flux at the core-mantle boundary (CMB) to explain changes in reversal rate, long-term departures from GAD in the time-averaged field, and geographical variations in paleosecular variation (PSV; e.g., Glatzmaier et al., 1999; Bloxham, 2000).
 The GAD hypothesis predicts that geomagnetic intensity should be twice as strong at the poles as at the equator and that inclination will vary with latitude according to the following equation:
where I is inclination and λ is latitude. Early compilations of paleointensity data [e.g., Tanaka et al., 1995] and directional data [e.g., Opdyke and Henry, 1969] appeared to support the GAD hypothesis with only minor deviations [e.g., Wilson, 1971]. Moreover, models of secular variation of an essentially GAD field [e.g., Constable and Parker, 1988; McElhinny and McFadden, 1997; Tauxe and Kent, 2004; Linder and Gilder, 2012] produce directional scatter that is symmetric about the equator.
 The modern geomagnetic field (Figure 1) exhibits hemispheric asymmetries that deviate significantly from an axial-dipole field. Northern hemisphere field structure is controlled by two high-intensity flux lobes at high latitudes, while the southern hemisphere is dominated by the low-intensity South Atlantic anomaly and a single high latitude flux lobe. The South Atlantic anomaly is a regional, non-zonal structure that historically is associated with low average field strengths and high scatter [Korte and Constable, 2005]. If such features are persistent over million year timescales we would expect to see northern high-latitude fields with low directional scatter and high paleointensities, while observing southern hemisphere paleofields similar to those observed in Antarctica [Lawrence et al., 2009].
Lawrence et al.  collected paleointensity (47) and directional (131) estimates from Antarctica spanning the last 13 Ma. Surprisingly, their paleointensity data showed no high latitude enhancement of field strength, and the directional data were significantly more scattered than predicted by standard PSV models [e.g., Constable and Parker, 1988; McElhinny and McFadden, 1997; Constable and Johnson, 1999; Tauxe and Kent, 2004]. To date there are no suitable Arctic (> 66°N) data sets with which to compare to Antarctica and establish whether the observed Antarctic field behavior is symmetrical at high latitudes, or if it is unique to the southern hemisphere. We present here paleomagnetic data from expeditions to the Arctic islands of Jan Mayen and Spitsbergen that are consistent with the prediction of asymmetrical field behavior at high latitudes.
2. Geology and Sampling
2.1. Jan Mayen
 Jan Mayen is a volcanic island dominated by the Earth's northernmost active volcano, Beerenberg (2277 m above sea level). Jan Mayen (Figure 2a) sits in the North Atlantic Ocean, just south of the Jan Mayen Fracture Zone between the Mohns and Kolbeinsey Ridges. Imsland  estimated surface volcanism to be approximately 400 ka or younger based on a set of 40K/40Ar dates by Fitch et al.  and Duncan, R. [unpublished]. Fitch et al.  collected paleomagnetic samples from 10 cooling units on the flanks of Beerenberg. These were of uniformly normal polarity. The oldest surface flows on Jan Mayen are part of the Havhestberget formation, a group of submarine hyaloclastites which can be found in the southwestern (Sör-Jan) and midsections (Midt-Jan) of the island. Overlying the Havhestberget formation is the subaerially erupted Nordvestkapp formation and the more recent Inndalen formation. These constitute the majority of surface volcanics and are found throughout the island. Historical lava flows date back to 1732 [Imsland, 1978] and are found in the northwest of the island (Nord-Jan), proximal to Beerenberg. The geology, geochemistry, and petrology of the island are described extensively by Imsland  and Imsland .
 We collected oriented samples from 23 sites in August 2009. Our sampling strategy was to target each of the geologic formations described above (Havhestberget, Nordvestkapp and Inndalen) and our individual site selection was guided by several unpublished 40K/40Ar dates [Duncan, R., unpublished] and by outcrop accessibility. Two Havhestberget exposures were sampled, one in the southern sea cliffs of Sör-Jan (JM012) and the other in sea cliffs along the north coast of Midt-Jan (JM006–009). Sör-Jan is dominated by a low-lying plateau of Inndalen lavas to the north (JM003 and JM013–015), while the southern highland regions contain a mixture of Inndalen scoria and trachyte cones and older Nordvestkapp lavas (JM002, JM004, JM011, and JM021–023). Midt-Jan consists predominantly of Nordvestkapp lavas and unconsolidated beach and lagoonal sediment. Nord-Jan lava flows radiate out from Beerenberg and consist of interfingered Inndalen (JM024, JM019–020, and JM025–027) and Nordvestkapp lavas (JM016–017). Sampling beyond the southwestern slope of Beerenberg was inhibited by steep cliffs along the shoreline and glacial ice.
 All samples taken in Jan Mayen were oriented with magnetic compass and either sun compass or a ProMark3 differential GPS system (Figure S1, supporting information).1 Certain samples were oriented with all three methods. Care was taken avoid oversampling the same geomagnetic field state by only sampling nearby/overlying lava flows if geologic evidence for temporal gaps could be determined.
 Spitsbergen is an island in the northwest of the Svalbard archipelago, on the northwestern edge of the Barents shelf in the North Atlantic (Figure 2b). The geology of Svalbard is dominated by fold-thrust belts associated with Paleozoic tectonics [Lyberis and Manby, 1999; Dewey and Strachan, 2003; Torsvik et al., 2001] and the Cenozoic opening of the Norwegian-Greenland Sea [Talwani and Eldholm, 1977; Maher et al., 1995]; for a detailed description of the geology of the island see, for example, Harland .
 There are two phases of Neogene volcanic activity in the Woodfjord region [Vagnes and Amundsen, 1993] where our sampling is concentrated. A thick sequence of tholeiitic plateau basalts have an age of 9–12 Ma [Prestvik, 1977] and lie to the east of Woodfjord. A second pulse of volcanism to the west of Woodfjord is exposed in three eruptive centers [Skjelkvale et al., 1988]: Sverrefjell, Halvdanpiggen, and Sigurdfjell. Preliminary paleomagnetic data [Halvorsen, 1972] suggest that Sverrefjell is normally magnetized, consistent with 40K 39Ar dates giving an upper age limit of <1 Ma [Burov and Zagruzina, 1976]. Halvorsen  also reported reversely magnetized lavas from nearby Halvdanpiggen, but the directions were anomalously shallow and could be Brunhes age excursions. The three centers are of nearly identical geochemistry and are arrayed along the same N-S fault line. Sverrefjell erupted during and prior to the last major glaciation in Bockfjord, believed to have occurred between about 100 and 250 ka [Skjelkvale et al., 1988]. Geomorphological and geochemical arguments suggest that Sigurdfjell is approximately the same age as Sverrefjell while Halvdanpiggen may be somewhat older.
 In July 2001, we collected oriented samples from 14 independent lava flows. Two sites are from Sverrefjell (SP100–101), two from Halvdanpiggen (SP102–103) and one site from Sigurdfjell (SP104) (all sites <1 Ma). The remaining sites (SP105–113) derive from the earlier eruptive phase (9–12 Ma) to the east of Woodfjord (see Figure 2b). Various lithologies were sampled including pahoehoe lava flows, palagonite tuffs, and basaltic dikes. Care was taken to collect samples where there was no evidence of tilting, slumping, or any frost heaving effects. Samples taken in Spitsbergen were oriented by magnetic compass and either sun compass or a Beeline differential GPS system (Figure S2, supporting information, described by Lawrence et al.  and Tauxe ).
3. Reliability of Orientation Methods at High Latitude
 Two thirds (22/36) of our northern high-latitude sample sites in this study were oriented, at least in part, by differential GPS systems, either the Beeline baseline system or the ProMark3 described in the previous sections. The recent Antarctic study by Lawrence et al.  also used the Beeline system to orient 28 out of 129 sites. These three study areas (Jan Mayen, Spitsbergen, and Antarctica) represent the bulk of published high latitude paleomagnetic data and all known sites oriented by differential GPS systems, therefore an evaluation of the reliability of the differential GPS orientation method is in order.
 We use sun compass azimuths as the standard for evaluating the differential GPS method because of its widespread use in the paleomagnetic community as a reliable orientation method. Uncertainties in sun compass orientations are estimated to be about 3° [Tauxe, 2010] and derive from errors in the field orientation process (e.g., improper insertion of the Pomeroy into the drill hole, deviations of the Pomeroy from horizontal, the width of the gnomon, and uncertainties in time and location of the measurement).
 A total of 241 samples from 37 sites were oriented with magnetic compass in addition to either sun compass or differential GPS (or both). Magnetic azimuths were corrected to “true north” using the IGRF predicted declinations for each location. Because all samples were oriented with either sun compass or differential GPS methods, we did not routinely check the magnetic azimuths by backsighting, however there are a total of 63 samples whose orientations were estimated with backsighted azimuths of which 23 were also oriented with a sun compass.
 Azimuths obtained from the differential GPS systems were evaluated by (1) comparing GPS and sun compass directions on samples where both were measured (total of 198 samples, Figure 3a) and (2) by comparing the site level directional scatter obtained from the two methods on sites where both were measured (total of 24 sites, Figure 3b). The first approach gives us a measure of the agreement between the two methods at the sample level and the second tells us which method is the most effective at reducing site level scatter, hence, which is the most precise.
 In Figure 3a, we compare the deviations of Beeline (dark blue) and ProMark3 (red) GPS samples, magnetic compass (purple) and backsighted (light blue) azimuths to sun compass azimuths. These are plotted as cumulative distribution functions (CDFs).
 Magnetic compasses are very precise instruments, however the range of azimuth deviations in Figure 3a confirm the widely held belief that their orientations cannot be trusted when used on lava flows, at least at high latitudes. Of the magnetic compass azimuths, 48% fall within 5° of sun compass estimates. Backsighting in the field is generally thought to minimize, and even remove, the effects of magnetization from lava flows, however, our backsighted azimuths are very asymmetric about 0° and only 17% are within 5° of sun compass azimuths. Backsight and magnetic compass MAD* values are equivalent, indicating that backsighting does not improve overall orientation accuracy and should not be relied upon to correct magnetic deviations in the field.
 Overall, both differential GPS methods perform significantly better than magnetic compasses (MAD* values of 1.6° and 6.3°, respectively) even with the presence of a few large outliers. Of all GPS orientations, 95% deviate less than 13.9° from their respective sun compass azimuths and approximately 73% of GPS azimuths deviate less than 5°. Two deviant samples oriented with the ProMark3 (JM021i and JM021j, marked with white “+”s in Figure 3a) are determined to be caused by insufficient satellite coverage and are not included in GPS statistics. Four samples from the Beeline system have deviations greater than 30°, and although the source of uncertainty cannot be determined it is certain that these are erroneous measurements. GPS statistics change considerably if these four samples are excluded (All GPS: ; Beeline GPS: ).
 Significant GPS orientation errors are rare (6/296 samples) but if undetected could affect site mean directions. By implementing appropriate minimum criteria for site level precision, such as α95 or the Fisher precision statistic kw, sites with inaccurately oriented samples will be removed from PSV analysis.
 Figure 3b shows the scatter in directions observed at the site level as a second measure of precision in orientation. We compare the circular standard deviation (CSD) of site level mean directions on 14 sites oriented with sun compass, differential GPS, and magnetic compass. For comparison, we also plot equivalent values of the estimated precision statistic, kw. Interestingly, magnetic compass orientations perform as well or better than both sun compass and GPS in terms of CSD. This observation can be attributed to the high precision of magnetic compasses. GPS orientations with poor precision (relative to sun compass measurements) should result in site means with large angular deviations and sites oriented with GPS do appear on average to be slightly less precise than when oriented with a sun compass. However, the kw for all GPS-oriented sites is still greater than 50, a selection criterion on a recent study of PSV in lava flows [Johnson et al., 2008], and all but one site have kws greater than 183 or a CSD of less than 6°. As there does not appear to be a systematic bias in directions, the uncertainty in GPS orientation at high latitude does not appear to be a serious problem. GPS orientation systems are therefore suitable alternatives to the sun compass, especially at high latitudes where sunshine may be scarce and the accuracy exceeds magnetic compass measurements alone.
4. 40Ar/39Ar Geochronology
40Ar/39Ar incremental heating experiments were successful on 10 sites from Jan Mayen and 3 from Spitsbergen (Figure S3 and Table S1, supporting information). Ages from Jan Mayen ranged from 6 ka to 460.9 ka. We attempted to date sites believed to be old enough for successful 40Ar/39Ar dating experiments, however one site, JM022, returned an exceptionally low age that should be considered with some caution, 6.0±14.5 ka. Site JM020 is thought to be the 1732 CE historical eruption [Imsland, 1978] and site JM019 is a surface flow in proximity to JM020. With no other age information, we designate JM019 as “historical” (<2000 years). All other Jan Mayen sites not dated are believed to be too young for 40Ar/39Ar dating.
 Sites SP100–104 were taken from a younger volcanic episode of 100–250 ka. Successful ages from Spitsbergen reported here ranged from 8.32 to 9.15 Ma, (SP109, SP111) consistent with the age estimates for earlier Neogene volcanic episodes. Sites SP106–108, SP110, SP112, and SP113 are part of the same volcanic province as SP109 and SP111 and are therefore assigned equivalent ages of 9 Ma.
 We used the IZZI modified Thellier-Thellier paleointensity experiment [Tauxe and Staudigel, 2004] to estimate intensity for Spitsbergen and Jan Mayen specimens (see supporting information). Figure 5 shows examples of representative IZZI experiments as Arai plots [Nagata et al., 1963]. Insets are the behavior of the natural remanant magnetization (NRM) (zero-field steps) plotted as Zijderveld diagrams [Zijderveld, 1967].
 There is general consensus in the paleointensity community that specimens with straight lines in the Arai plots, with no evidence of alteration (partial thermal remanant magnetization (pTRM) checks plot on top of initial pTRM measurements) or multicomponent behavior in the Zijderveld diagrams, can be used to estimate the ancient magnetic field strength (e.g., Figure 4a). However, such data are rather rare and there is little agreement on how to properly select or reject data that depart from ideal behavior (e.g., Figures 4b–4d). Recent paleointensity studies tend to require more stringent specimen level selection criteria with the goal of removing data that significantly depart from ideal behavior [e.g., Shaar et al., 2011].
 In our experiments, some specimens acquired a remanence parallel to the laboratory field which was not demagnetized by later heating steps (e.g., Figure 4b). In some cases, this behavior is less pronounced, only manifesting itself after 95% of the NRM has been removed (e.g., Figure 4c). Moreover, many specimens exhibit concave-up curves in the Arai plots, (Figure 4d). Such behavior is usually interpreted as characteristic of multidomain remanences [e.g., Dunlop and Özdemir, 2001]. To screen out multicomponent remanences or experiments that exhibited alteration, we have used the following strategy (see Table S2, supporting information, and Tauxe  for definitions): (1) each specimen must have a relatively linear component of magnetization representing a majority of the thermal remanant magnetization (TRM) applied to the specimen with a maximum angle of deviation (MAD) ≤5; (2) no data can be used at temperature steps greater than a failed pTRM check as defined by a difference ration sum (DRAT) of 20%; (3) the temperature range of the interpreted intensity component must be related to the characteristic remanent magnetization of the specimen (deviation of the angle (DANG) ≤13); (4) no temperature step may be used where the pTRM gained is lower than the gain of the previous temperature step.
 At the site level, we require at least two intensity estimates per site, NB ≥ 2, and evaluate the percent standard deviation of the mean intensity, which is . We use a dσB cutoff of 15% in order to ensure that outliers do not adversely effect our final paleointensity estimation.
 As illustrated in Figures 4c and 4d, some specimens display concave-up Arai plots. Selecting the low-temperature component results in a steeper slope and therefore a high-ancient field estimate, while selecting the high temperature component results in a shallower slope and a smaller ancient field. While either of these selections could meet the criteria outlined above, they are incompatible with each other and it is difficult to justify choosing one over the other. Dunlop and Özdemir  suggest that the appropriate measure of paleointensity for multidomain specimens is normalizing by the full TRM, that is, by taking the first and last temperature steps of the IZZI experiment. To ensure that the full vector is used in the paleointensity calculation of specimens with sagged Arai plots, we required the value of fvds to be at least 0.95. If the implications of Dunlop and Özdemir  are correct, all specimens from a given site should yield the same answer and pass our site selection criterion.
 Our attempts to define the intensity component for concave-up experiments using the total TRM resulted in highly variable within-site field estimates which exceeded our 15% cutoff for dσB. Interestingly, all the estimates derived from curved Arai plots were lower than those derived from straight Arai plots from the same site. Moreover, when given a fresh laboratory TRM and treated to a second IZZI experiment, all of our specimens behaved in an ideal fashion so the curvature was not reproducible on laboratory time scales.
Shaar et al.  conducted a series of experiments on synthetic multidomain specimens and observed a decay of NRM intensity in multidomain specimens, without the acquisition of secondary directional components, over a time span of several years. Specimens showing this behavior would not fail pTRM checks or DANG criteria but will yield paleointensity estimates that are too low. All aberrant experiments observed by Shaar et al.  were markedly concave-up. The evident variability of total and partial TRM estimates for concave-up specimens has lead us to reject all multidomain paleointensity estimates until some constraint can be placed on the reliability of concave-up Arai plots.
 Despite our reluctance to include specimens displaying concave-up Arai Plots, a few exceptions were made for specimens from three sites where we determined that the observed behavior was not actually multidomain (Figure 4c). Specimens from sites JM002, JM019, and JM020 are included because the concavity of the Arai plot begins after a failed pTRM check, indicating that alteration is a plausible cause of this behavior. Arai plots in each of these cases are relatively straight prior to the failed pTRM check and pass all specimen level criteria (1–4). It can be assumed that if alteration did not occur then the slope of the intensity component would have continued through the remainder of the IZZI experiment. We therefore assume that the straight, lower-temperature intensity component, before the failed pTRM check, is an accurate estimate of the ancient magnetic field in each specimen. If our interpretation is incorrect and the observed behavior in these specimens is not caused by alteration, then the low-temperature components we selected would likely overestimate the ancient magnetic field strength (e.g., Figure 4d, purple component).
 Five sites from Jan Mayen (Table 1, Figure 7) meet our selection criteria while no samples from Spitsbergen are deemed reliable. The majority of those specimens that failed our selection criteria altered during the multiple heatings of the IZZI experiment. The average intensity of our Jan Mayen data set (n = 5) is 56.9 ± 18.0 μT (σ) with an equivalent virtual axial dipole moment (VADM) of 76.8±24.3 ZAm2 (σ). This data set exhibits the high latitude enhancement of field strength expected from a GAD field, and is nearly a factor of two higher than comparably aged data from Antarctica (n = 7), 26.8±6.0µT (σ) and 35.3±7.9 ZAm2 (σ) [Lawrence et al., 2009].
NB is the number of specimens used in paleointensity calculation; B (μT) is the average field strength at each site. σB is the standard deviation of the specimens used to calculate the site mean. dσB is the standard deviation of the specimen intensities as a percent fraction of the mean intensity. VADM (ZAm2) is the virtual axial dipole moment. σVADM is the standard deviation of specimens used to calculate the mean VADM. VDM is the virtual dipole moment, along with its standard deviation, σVDM.
 We demagnetized at least five specimens per site using alternating field (AF) and thermal demagnetization techniques (see supporting information). Principal component analysis (PCA) [Kirschvink, 1980] was used to determine characteristic directions or best fit planes for each specimen. Generally, uni-vectorial decay was fit with a line while data affected by laboratory overprints like pTRM acquisition were fit by planes. The assumption here is that the original characteristic direction is constrained to lie within the best-fit plane. Specimen lines and planes were deemed acceptable if they had at least four consecutive demagnetization steps and MAD ≤5°.
 Two types (Type I and II) of site level demagnetization behavior were observed. Type I sites (7 total, e.g., Figures 5a–5c) decayed linearly to the origin in both AF and thermal demagnetization experiments for all specimens. Type II sites (30 total, e.g., Figures 5d–5i) were categorized by some specimens exhibiting laboratory overprints. Site means for Type I sites (Figure 5c) were calculated using Fisher statistics [Fisher, 1953] and the combined lines and planes method of McFadden and McElhinny  was used for Type II sites (Figure 5f). We required all sites to have a minimum of five specimens (n ≥ 5), a Fisher precision parameter, κw ≥50 and a demagnetization code (DC) of 4 or 5 [McElhinny and McFadden, 2000] (complete demagnetization, use of PCA). See Tables 2 and 3 for all acceptable site mean directions and virtual geographic poles (VGPs).
Age (ka) is calculated 40Ar/39Ar radiometric age with 2σ error or estimated age (see section 2). All other flows are determined to be younger than the oldest surface flow, JM008. Dec and Inc are mean site declination and inclination, respectively. nl/np is the number of best-fit lines and planes, respectively, used in site mean calculations. N is the combined number of best-fit lines and planes. k is an estimate of the Fisher  precision parameter, R is the resultant vector of N unit vectors, and α95 is the Fisher  circle of 95% confidence. VGP Lat/Lon are the virtual geomagnetic poles calculated for each site. NB is the number of specimens used in paleointensity calculation. Lat*, Lon*, PLat*, and PLon* are site latitude, longitude, and recalculated VGP lat and lon, respectively, after adjusting for plate motion using the NNR-MORVEL model Argus et al. . (Plate corrected locations are not listed for Jan Mayen because there is essentially no change.)
Age (Ma) is calculated 40Ar/39Ar radiometric age with 2σ error or estimated age (see section 2). See Table 2 for a description of column definitions.
 Several lava flows have overlapping directions (e.g., JM009 and JM013; JM011 and JM020; SP106 and SP107) but there is no indication of preferentially oversampling any temporal field since all the sites in question are from geographically distinct units and/or show evidence of temporal gaps between units (e.g., paleosols).
 Figure 6 shows equal area plots of site mean directions (a and b) and orthographic plots of site virtual geomagnetic poles (VGPs) (c and d) for the two locations. Combined site mean directions for each of the Jan Mayen and Spitsbergen locations (green triangles) are consistent with those expected from a GAD field (stars). Reverse polarity sites from Spitsbergen are antipodal to normal sites and the data pass both Watson's Vw and bootstrap reversals tests [Tauxe, 2010]. Combining the antipodes of the reverse directions produces distributions with estimated κF values of 36 and 27 for Jan Mayen and Spitsbergen, respectively.
 It has become traditional to use the statistic S of Cox  to quantify VGP dispersion. We modify S and use the statistic SF [McElhinny and McFadden, 1997] in order to correct for within-site directional scatter.
where N is the number of sites, Δi is the angle between the ith VGP and the spin axis, Sw is within-site scatter (defined as , where kw is the Fisherian precision statistic), and nni is the number of specimens in the ith site. We list SF for the Jan Mayen and Spitsbergen data sets in Table 4.
Table 4. VGP Dispersion, SF, with Bootstrapped 95% Confidence Bounds, Adjusted for Plate Motion Using the NNR-MORVEL Model of Argus et al. a
SF is calculated for Jan Mayen and Spitsbergen individually, combined sites from both locations, JM/SP, Iceland [Udagawa et al., 1999], and Antarctica [Lawrence et al., 2009]. Lat* is average location latitude corrected for plate motion [Argus et al., 2011], N is number of sites used to calculate VGP scatter using no latitudinal cutoff, SF, the iterative cutoff of Vandamme , SFv, and a 45° cutoff, SF45·λv is the Vandamme colatitudinal VGP cutoff.
 Jan Mayen and Spitsbergen site VGPs (Figures 6c and 6d) are highly scattered (κF values of 11 and 10, respectively) as a consequence of the VGP mapping at high latitudes. The dotted black circles represent the 45° colatitudinal VGP cutoff often applied in PSV studies. Four sites have VGP latitudes close to this arbitrary cutoff and two (JM016 and SP102) exceed the cutoff. The use of VGP cutoffs (e.g., 45° or that of Vandamme, 1994), has been shown to bias PSV estimates [e.g., Lawrence et al., 2006] by arbitrarily defining some directions as transitional. This same bias is exaggerated at high latitudes.
 Jan Mayen has 23 directional sites, all younger than 461 ka, and an unfiltered VGP dispression estimate of . Spitsbergen has 13 directional sites, unevenly distributed at <1 Ma (n = 5) and 9 Ma (n = 9), with VGP dispersion of . Jan Mayen and Spitsbergen are at different latitudes and have, two distinct age distributions, <1 Ma and ∼9 Ma, and Spitsbergen has too few data points in either age group to stand alone. However, if we combine the data from Jan Mayen and the younger Spitsbergen group together, the SF values cannot be distinguished ( (n = 28) from the older Spitsbergen group alone (n = 9)). Therefore, we combine the two data sets despite their spatial and temporal differences. When combined (JM/SP), these data represent the first estimate for directional and VGP scatter at Arctic latitudes (average of 74.2°N); SF for these data (Figure 8a) is (n = 37, 0–9 Ma) and is significantly lower than that estimated for the Antarctica data of Lawrence et al.  of (n = 131, 0–13 Ma). Note that no VGP filter was used for either calculation.
 New directional and paleointensity results suggest the possibility of hemispheric asymmetry at high latitudes. The five Jan Mayen intensity sites span the last 300 kyr, with an average VADM of 76.8±24.3 ZAm2, as do seven sites from Antarctica, 35.3±7.9 ZAm2. Only one Jan Mayen site (JM021) has a comparable age with any Antarctic sites (mc35 and mc217), too few for a comprehensive comparison. To augment the Arctic data set, we search the MagIC database (http://earthref.org/MAGIC) for published data from the last 300 kyr at high latitudes (≥ 60°) and find three studies [Pesonen et al., 1995; Donadini et al., 2007; Stanton et al., 2011] with paleointensity sites that meet our site selection criteria and do not target transitional field states. Unfortunately, these records are from the last few thousand years and do not provide comparable ages to the Antarctic data set.
 VADMs from this study and previously mentioned published data sets are plotted in Figure 7. For reference, we plot the paleomagnetic axial dipole moment model (PADM2M) of Ziegler et al. . All of the Antarctic intensity data (open diamonds) are lower than PADM2M and any contemporaneous high northerly latitude sites. Hemispheric asymmetry in average VADMs could explain a previously noted discrepancy between compilations of paleointensity from the global submarine basaltic glass data set and those from lava flows [e.g., Tauxe and Yamazaki, 2007], which until recently have been heavily biased toward northern hemisphere sampling sites. Moreover, it suggests that north-south asymmetry visible in the current geomagnetic field may persist for millions of years, as hinted at by previous studies [e.g., Kelly and Gubbins, 1997] and some numerical models with nonuniform boundary conditions [e.g., Glatzmaier et al., 1999]. Although evidence from the Arctic and Antarctic suggest that hemispheric asymmetry exists, there simply are too few Arctic paleointensity records with sufficient temporal coverage to adequately evaluate this hypothesis.
 Our directional data, however, allow us to examine high latitude northern hemisphere VGP dispersion, as quantified by SF, and compare it to dispersion from high southerly latitudes. JM/SP has considerably fewer directional sites (n = 37) than Antarctica (n = 131), so in order to provide a more robust comparison we augment our Arctic directions by drawing all high-latitude (>60° N) directional studies from the MagIC database that meet our site selection criteria: , and DC ≥4 or 5. Some paleomagnetic studies have focussed their efforts on transitional sites and we exclude such studies in the present investigation. One study from Iceland meets all requirements [Udagawa et al., 1999].
 Jan Mayen/Spitsbergen, Antarctica, and Iceland have different temporal distributions which may affect our comparisons of PSV. Antarctic paleodirections are predominantly 0–5 Ma (n = 123/131) with eight sites scattered between 5 and 13 Ma. The bulk of the combined Arctic data (n = 63/76) are 0–2 Ma with 12 sites between 2 and 9 Ma. We therefore limit our hemispheric comparison of PSV to the last 2 Ma. See Table 4 and Figure 8.
 We correct for (the very small) plate motion by adjusting the sampling latitudes of all sites using the no-net-rotation plate motion model (NNR-MORVEL) of Argus et al. . In Figure 8, we compare high latitude VGP scatter from JM/SP, Iceland [Udagawa et al., 1999] and Antarctica [Lawrence et al., 2009]. We show SF with all VGPs (Figure 8a), as well as dispersion calculated using the variable cutoff algorithm of Vandamme  (Figure 8b). Gray symbols are calculated for 0–2 Ma, black symbols are 0–0.5 Ma. Predicted VGP dispersion from the PSV model TK03 [Tauxe and Kent, 2004] is plotted with no VGP latitudinal filter (Figure 8a) and using the Vandamme cutoff (Figure 8b). The dashed line in Figure 8b is the prediction from Model G [McElhinny and McFadden, 1997]. Note that Model G was designed to fit the PSVRL database [McElhinny and McFadden, 1997] which excludes VGPs according to the Vandamme variable cutoff method.
 Unfiltered VGP dispersion for 0–2 Ma from Iceland and JM/SP are consistent with the prediction of the TK03 PSV model, although at the extreme end of the confidence bounds for JM/SP, while dispersion from Antarctica is at least 4° above TK03. When the Vandamme cutoff is used, Antarctic dispersion drops significantly and all locations have essentially the same value. This artificial suppression of VGP dispersion, coupled with our observation that the transformation from directions to VGPs makes the magnetic field appear more variable than the directions suggest, reinforces the argument that latitudinal VGP cutoffs should not be used for PSV analysis.
 Our initial analysis of the last 2 Ma spans the Brunhes and Matuyama polarity chrons. SF estimates derived from VGPs spanning reversals may differ from dispersion calculated during a single polarity interval. Paleodirections in the Matuyama chron have been shown to have a greater angular standard deviation about GAD than the Brunhes [e.g., Johnson et al., 2008], in addition some sites may sample reversals, producing transitional directions that will cause an increase in SF. To this end we analyze PSV from the last 0.5 Ma which, in theory, should produce more stable estimates of SF. Mean VGP scatter in the Arctic decreases slightly to (n = 26), while mean SF in Antarctica increases to (n = 29). Our interpretation of unfiltered dispersion for 0–0.5 Ma remains essentially the same as for 0–2 Ma, except the new Arctic estimate is in better agreement with TK03.
 We observe that unfiltered VGP dispersion for 0–2 Ma in the Arctic is in good agreement with existing PSV models while scatter in Antarctica is significantly higher. This result, coupled with systematically low VADMs in Antarctica, hints at asymmetrical PSV behavior between the two high latitude regions for at least the last 0.5 Ma.
 The comparison of high latitude directional and intensity data from both hemispheres suggests a significant zonal hemispheric asymmetry in the geomagnetic field that persists on time scales of 105−106 years. This asymmetry is not accounted for in existing secular variation models and requires long-term non-axial-dipole field contributions. A significant quadrupolar component, 20% of the dipole field, would be needed to fit the asymmetry in field strength at high latitudes (although Arctic paleointensity records are sparse (n = 5) and not yet conclusive). This high value is incompatible with observed inclination errors over the last 5 Ma [e.g., Johnson et al., 2008] indicating that the long-term geomagnetic field may contain some non-axial-dipole component or higher order terms as well.
 The statistical parameters and percent of non-axial-dipole contribution in TK03 can easily be modified to fit our observed differences in polar VGP dispersion, however, TK03 is a global model and changing parameters to fit asymmetries at high latitudes will have significant effects on PSV predictions at mid and low latitudes. Proper evaluation of the long-term hemispheric asymmetry hypothesis requires globally distributed directional and intensity data spanning 105−106 years.
 Existing numerical geodynamo simulations often explore geographic variations in thermal CMB conditions, especially potential cold regions around the Pacific rim, that could reflect seismic velocity variations near the CMB. Simulations with timescales of 100 kyrs or greater are often tested against TAF and PSV models with similar temporal resolution. The recent addition of high-quality paleomagnetic data, especially at equatorial [e.g., Kent et al., 2010; Opdyke et al., 2010] and high latitudes [e.g., Lawrence et al., 2009, this study] allows for a greater understanding of long-term geomagnetic field behavior and perhaps revisions to existing TAF and PSV models. If numerical geodynamo models can be made to reproduce the statistical behavior of these revised models, then it may be possible to understand the hemispherical asymmetries or other features seen in long-term geomagnetic field behavior.
 We thank Winfried Dallmann for significant logistical and field support in Spitsbergen and Leif-Erik Pedersen and Philip Staudigel for their hard work in the field collecting samples on Jan Mayen. We thank Brad Singer for his help with the 40Ar/39Ar experiments and for the use of his lab, and Bob Duncan for access to unpublished preliminary data from Jan Mayen. We appreciate Jason Steindorf's contributions to sample preparation and processing. Thanks to Jeff Gee for his constructive comments, and two anonymous reviewers who helped to greatly improve the manuscript. We are grateful to the crew of Jan Mayen, especially Ole Øiseth and Filip Myrvoll, for their hospitality and expertise. We thank Frank Vernon and Mert Ingraham for their assistance in adapting the Beeline differential GPS system for paleomagnetic work in Spitsbergen and Stephen Peter for training and troubleshooting with the ProMark3 GPS system. This material is based on work supported by National Science Foundation grants EAR9805164, EAR0838257, EAR0809709, and EAR1141840.