Introduction

The geomagnetic field occasionally deviates in a direction far beyond the range of secular variation. Such events are termed geomagnetic excursions, usually defined as a temporary deviation of a field direction whose virtual geomagnetic pole (VGP) deviates more than 45° from the geographical pole1. Geomagnetic excursions, together with reversals, represent the dynamic nature of the geodynamo at times when the dipole field strength decreases to less than half of the present value2. Geomagnetic excursions typically occur at intervals of 104 to 105 yr3, but their internal structure has rarely been observed in sediment-magnetisations, time-averaged geomagnetic field records, due to their short duration (typically less than ca. 2 kyr). Therefore, palaeomagnetic data of excursions are still insufficient and geographically biased, making it difficult to analyse such an event on a global scale. Moreover, very few datasets4 resolve subcentennial structure of excursions, with some excursions terminating within just a few hundred years.

The Laschamp Excursion was first discovered in the magnetisations of lava flows from the Chaîne des Puys, France5, and further investigations confirm a deviation in the magnetic field6,7. This excursion was subsequently recognised in other lavas in Iceland8,9 and New Zealand10,11, as well as in deep-sea sediments12,13,14,15, Chinese loess sequences16, a speleothem17, and lake sediments18. The lavas that record the excursion have been 40Ar/39Ar dated to between 40 and 46 kyr BP11,19. The dates are consistent with the estimated mid-point ages of between 40 and 43 kyr BP identified in sediments based on δ18O stratigraphy, climatostratigraphy, and/or 14C-, optically-stimulated luminescence- and U–Th-dating4,13,16,17. The estimated duration for an excursional-direction interval, on the other hand, vary across a wide range, grouping at 200‒300, 400‒500 and >900 yr (Supplementary Table 1). In spite of this difference, the sedimentary records exhibit many common features, such as the precursor immediately (<500 yr) predating the main excursion4,14 and the clockwise open loop of the VGP-pass4,15. Therefore, the variability of the estimated duration possibly arises from inaccuracy of chronology and the filtering effects of low-resolution data series, other than regional variabilities20,21. Precisely dated, high-resolution records of the Laschamp Excursion, therefore, have been highly anticipated to reconcile these ambiguities. Well-dated Laschamp Excursion records that comprise multiple excursional directions (Supplementary Table 1) have seldom been reported from sedimentary archives with a sedimentation rate (s.r.) lower than ca. 20 cm/kyr, which is thought to be smoothing the post-depositional remanent magnetisation (PDRM) signal22,23. High-resolution records are also anticipated from cosmic ray studies associated with excursions24,25.

The varved sediment of Lake Suigetsu (Supplementary Fig. 1) has the potential to provide a well-dated, high-resolution temporal dataset through the Laschamp Excursion. It is one of the most precisely age-controlled sedimentary archives in the world26,27, with annual layers (varves) counted by dual methods28,29,30 as well as >800 radiocarbon dates obtained on terrestrial leaf fossils27,31. The dataset has been used as a central component of the IntCal13 and IntCal20 radiocarbon calibration models32,33. The age uncertainties (1σ) given to the Suigetsu sediment at around the expected Laschamp Excursion are only ±60‒100-yr, and the mean s.r. of 90‒100 cm/kyr during this period is very high compared with many other sedimentary records (Supplementary Table 1). The preservation of varves provides support that a good palaeomagnetic record can be expected from the sediments, which should have acquired magnetisations unaffected by bioturbation. In this article, we report the high (21-yr) resolution palaeomagnetic record from Lake Suigetsu varved sediments that reveal decadal structures of the Laschamp Excursion with a duration of 790 varve years and a new excursion with a duration of 550 varve years about 2600-yr postdating the Laschamp Excursion. These excursions comprise multiple subcentennial directional-swings caused by four stationary non-axial dipolar sources intermittently dominated one after another in clockwise order.

Results and discussion

The Laschamp and new post-Laschamp excursions

High-resolution samples (every 2-cm depth around excursions) for magnetic analyses were taken from the varved “SG14” Lake Suigetsu sediment core (Methods). We use a characteristic remanent magnetisation (ChRM) isolated between 200 and 450‒590 °C from each sample (Supplementary Discussion, and Supplementary Figs. 2 and 3). It is regarded as a primary natural remanent magnetisation (NRM) carried mainly by detrital magnetites (Supplementary Discussion and Supplementary Fig. 2). The ChRMs reveal two intervals of excursion comprising multiple swings of directions whose VGPs deviate >45° from the North Pole (Fig. 1, Supplementary Figs. 4, 5, and Supplementary Data 1). Based on the Bayesian age model31,34, the lower excursion ranges from 42,400‒41,700 IntCal20 yr BP and the upper one from 39,090‒38,580 IntCal20 yr BP (Figs. 1 and 2). The lower excursion consists of five swings, and the upper one consists of two. We label these swings A, B, C, D, E, F and G in ascending order (Fig. 1). The non-excursional field represented by a single sample between swings D and C exhibits full normal polarity with a VGP latitude of 80.7°N, hence we separate D and C with confidence. On the other hand, the non-excursional fields within swings A and G have near excursional VGPs with latitudes 47.4 and 57.2°N, respectively, and thus we regard them as a single episode. However, the possibility remains that there were shorter episodes of full normal polarity that were not detected due to the analytical resolution. The duration of each swing, including the thickness of a discrete sample (2 cm), ranges from 18 to 160-yr (based on varve counts; Table 1).

Fig. 1: High-resolution palaeomagnetic data from Lake Suigetsu (a–d) and comparison with other data.
figure 1

Time variations of VGP latitudes (a), inclinations (b) and declinations (c) of ChRMs, and the relative palaeointensity (J200 °C/χ200 °C) normalised to a total mean (d). The successive swings are labelled 'A‒G' (see text). e Weighted five-point moving averaged Δ14C data from Lake Suigetsu31, 34. The dashed lines represent depositional ages of the key chrono-stratigraphically relevant volcanic ash (tephra) fall events identified in the SG14 core57. f Δ14C data spanning 30,000 to 50,000 cal yr BP from Lake Suigetsu31, 34, New Zealand kauri trees25 and Hulu cave speleothems36. The graphs of (a–e) and the Lake Suigetsu Δ14C of (f) are plotted against IntCal20 yr BP. The age values in cal yr BP in (f) are based on 14C-dating and tree-ring counting for kauri trees and U–Th dating for speleothems. g 10Be flux data from Greenland ice cores plotted against the GICC05 age model60. h Palaeointensity data from Black Sea sediments plotted against an age model climatostratigraphically correlated to the GICC05 age model4. In the magnetic polarity based on the VGP latitude, the black/hatched zones show normal-polarity/excursional intervals.

Full size image
Fig. 2: Age model for core SG14.
figure 2

a The age versus composite-depth plot for mid-depths of all palaeomagnetic samples, based on the correlation model: version 06 Apr. 2020. The error bars show ±1σ errors. The blue bars represent the intervals of the Laschamp (lower) and the post-Laschamp (upper) excursions. b The time variation of a basic resolution for each sample (a minimum time estimate for field-averaging) is calculated by dividing the thickness of a sample (2 cm) by a s.r. between adjacent sample levels. The low basic-resolution values are from core segments from parallel boreholes (3189 cm) and event (flood) layers (3350, 3358–3360 and 3384 cm). c Cross-section photographs of dry cubic samples after progressive THDs within Swing C of the Laschamp (i) and just below the Laschamp Excursion (ii). The original wet thickness of the sample is 2 cm (dry thickness is ca. 1.8 cm).

Full size image
Table 1 Depth ranges, mid-ages and durations of excursions and swings from Lake Suigetsu varved sediments.
Full size table

The lower excursion comprising swings A‒E is correlated with the 40‒46 kyr Laschamp Excursion including the precursor4,11,19. The mid-age of 42,050 ± 120 IntCal20 yr BP from the Lake Suigetsu record and a varve-counted duration of 790 yr are the best estimates yet for the Laschamp Excursion, from the high-resolution and precisely age-controlled palaeomagnetic data (Supplementary Table 1). In addition, the upper excursion, with a mid-age of 38,830 ± 140 IntCal20 yr BP and a varve-counted duration of 550 yr (Table 1), has not been previously recognised. The Mono Lake Excursion, dated to between 32 and 35 kyr BP13,18, is too young to be a counterpart of it, and no other excursions based on high-precision data have been reported from elsewhere between 35 and 40 kyr BP. However, some lava-flow excursion records in France7 may correlate with the upper excursion (swings F and G) as well as swing E.

The relative palaeointensity (Methods, Supplementary Data 2) provides minima around both excursions and a gentle rise between them with a peak closer to the younger excursion (Fig. 1d). The palaeointensity variation resembles that of the Black Sea record between ca. 41.5 and 39 ka on its climatostratigraphic age scale4 (Fig. 1h), although no excursional-direction is recorded or resolved at the younger minimum. The younger excursion appears to be a rebound of the Laschamp, displaying similar characteristics. The large decrease of palaeointensity during excursions causes an increase in the cosmic ray flux, therefore, an increase in the production rates of cosmogenic nuclides in the stratosphere24,25. The production rates of nuclides 10Be and 14C would have elevated during the Laschamp Excursion. However, the Δ14C measured in various archives is delayed relative to the 14C production as it is filtered and attenuated in amplitude as it makes its way through the global carbon cycle between atmosphere, biosphere, and the ocean, whereas 10Be directly reflects cosmic ray intensity variations with almost no attenuation and a delay of only 1‒2 yr35. Concomitant Δ14C in Lake Suigetsu is very high (>400‰) during the intervals of the Laschamp and the post-Laschamp excursions, supported by other Δ14C data from other sites24,25 (Fig. 1e, f). In addition, the extremely high (>600‰) Δ14C peaks just postdate swings A‒E, F and G, by 300‒400 yr, ca. 50 yr and ca. 250 yr, respectively. These time lags, likely dependent upon frequencies of variations, are comparable with simulation estimates from a 12-box model37. Thus, the cosmic ray flux probably increased in phase with the excursions even on decadal-centennial time scales. The high cosmic ray flux during the Laschamp Excursion is also evidenced by the high 10Be flux in Greenland ice cores compared with the Black Sea palaeomagnetic data4 (Fig. 1g, h). As discussed below, the main excursion and the precursor in the Black Sea record can be correlated with a combined swing of C and D, and B (or A + B) in Lake Suigetsu, respectively. These correlations would be useful for connecting the ice core age model (GICC05) and the IntCal20 age model, which is beyond the scope of this study.

The post-Laschamp excursion record from Lake Suigetsu sediments provides a chronologically important geomagnetic-directional episode that postdates the widespread 'U-Ym' tephra by ca. 1700 years (see Methods for further details), and is separated from the main Laschamp excursion by ca. 2600 years. Here, we propose to name this the 'Post-Laschamp (Suigetsu) Excursion' for the convenience of chronological application.

Corroboration from existing lower resolution records

The detailed structure of the Laschamp Excursion that we recognise in the Suigetsu core, comprising multiple subcentennial episodes, is unique. The simpler behaviour shown by previous, lower s.r. (20‒44 cm/kyr) records (Supplementary Table 1), may be the consequence of the filtering effects of a greater PDRM process associated with lower sedimentation rates, as well as the direct loss of sampling resolution. First, we simulated how the present magnetisation data would be smoothed in records with lower s.r., without filtering by a PDRM process. The time-series vectorial data (Methods) were condensed for sediments with various s.r. assuming that magnetisations were measured on 2-cm thick (Δt-yr spanning) samples taken at 2-cm (Δt-yr) intervals. Here, we define Δt (=2 cm/s.r.) as a basic resolution (a minimum time estimate for field-averaging). Δt is 21-yr for the Lake Suigetsu sediments, consistent with the visual varve counts with dried samples after thermal demagnetisations (THDs) (Fig. 2). The condensation is equivalent with convolution integration (or filtering) with a box function with a width of Δt, with every-Δt outputs. For simulations, we used the time-series data of the orthogonal three magnetisation components converted from declination, inclination and relative palaeointensity (Fig. 3, Methods). Figure 3a shows curves of VGP latitude and relative palaeointensity calculated from the outputs for a s.r. of 48, 40, 20, 12 and 9 cm/kyr. When the s.r. is halved (48 cm/kyr), swing E disappears, and swings C and D are combined (tentatively named 'CD'). The large swing CD still remains at 12 cm/kyr, but disappears at 9 cm/kyr. The presence of excursional signals even at 12 cm/kyr is inconsistent with real data from sediments with s.r. <20 cm/kyr as almost none exhibit any Laschamp-related excursional-direction signals. Such sediments are most likely to have undergone filtering of magnetisations by a PDRM process.

Fig. 3: Simulated results to illustrate the filtering effect if our high (21-yr) resolution Lake Suigetsu magnetisation data had instead been recorded in lower s.r. sediments.
figure 3

a VGP latitude and palaeointensity curves from the Lake Suigetsu magnetisation vector (x, y, z) data condensed to every Δt (2 cm/s.r.)-yr data (equivalent with filtering with a box function with a width of Δt) for various s.r. b VGP latitude and palaeointensity curves from the Lake Suigetsu magnetisation vector (x, y, z) data convolution-integrated with an exponential lock-in rate function defined by a half lock-in depth/time (Z1/2/T1/2), and condensed to every Δt-yr data (equivalent to the dual filtering with an exponential lock-in rate function and a box function). The calculations are made with various Z1/2/T1/2 values for s.r. of 44 cm/kyr, the highest value for the previous sedimentary records of the Laschamp Excursion (Supplementary Table 1). The dark blue areas represent obvious artefacts of palaeointensity decreases <0.4, produced by the convolution integration. c VGP latitude and longitude curves of the Lake Suigetsu data before and after filtering, and the Black Sea data4. The light cyan bar represents the interval of the clockwise open-loop VGP-pass. 'A‒E' and 'CD' on the VGP latitude curves represent excursional swing episodes.

Full size image

A PDRM is a time-averaged geomagnetic field filtered with a magnetisation lock-in rate function38,39,40. Therefore, dually filtered sediment-magnetisations can be calculated using the lock-in rate function and a box function. We simulated the filtering effect for a fixed s.r. with various values of Z1/2 (T1/2), a half lock-in depth (time) where 50% of a magnetisation is locked-in38 (Methods, Supplementary Fig. 6). In the VGP latitude curves for a s.r. of 44 cm/kyr (the highest value for the previous sedimentary records spanning the Laschamp Excursion), at 0.5 cm for Z1/2, swing E disappears, and a combined swing 'CD' appears (Fig. 3b). Swing A just disappears at 1 cm, B at 2 cm and CD at 4 cm. The excursion comprising swings B and CD at Z1/2 of 1.0 and 1.5 cm spans about 450 yr, which correlates well with the double-episode Laschamp Excursion spanning 400‒500 yr recorded in the North Atlantic12,14. The excursion with only CD at Z1/2 of 2.0 cm spans about 140-yr, which can be correlated with the single-episode Laschamp Excursion spanning 200‒300 yr in China and the North Atlantic16,41. The swing CD at Z1/2 of 3 cm spanning <50 yr would be difficult to detect. A Z1/2 of 1.0–1.5 cm for natural sediments is realistic, whereas that of >2 cm is not (Methods)23,42,43. The output VGP latitude and longitude variations for CD correlate well with those for the main excursion seen in the Black Sea record4 (Fig. 3c), which indicates the dominance of non-axial dipolar fields. For the remaining interval, they disagree with each other, except the output swing B that can be correlated with the precursor in the Black Sea; the VGP variation has a comparable signal in latitude but does not in longitude, which may be caused by a filtering effect due to the low s.r. (20‒30 cm/kyr) or reflect regional fields.

The relative palaeointensity curves in any simulations have remarkable drops, by 70‒80% at their maximum, around swings B, C and D (Fig. 3a, b). These swings include largely deviated fields with VGPs in the southern high-latitudes. The vector sum of such nearly-reversed and normal-polarity magnetisation components forges an apparent large decrease of palaeointensity. We should note that sediment-magnetisation intensity more or less contains such forged decrease components, which sometimes largely lowers the palaeointensity value and may even shift a minimum in position (time), especially around geomagnetic excursions and reversals38 (Supplementary Fig. 6). Relative palaeointensity artefacts are introduced even by NRM acquisition efficiency variations in terms of remanence carrier alignment and magnetic components with different intrinsic ratios between their NRM contributions and a magnetisation normaliser40,44. In the Black Sea data (Fig. 1h), the palaeointensity takes a flat value of ca. 12 μT, about 1/4 of the present field intensity, around the centre of the main excursion that coincided with the cosmic ray peak. However, the wide interval of palaeointensity values much lower than 12 μT with a minimum of 3 μT, predates the flat interval, centred between the main excursion and the precursor (Fig. 1h). This extremely low palaeointensity interval, without a corresponding much higher cosmic ray increase (Fig. 1g), may be an artefact. Careful interpretations of sediments’ palaeointensity records are always necessary, including even the present Lake Suigetsu data. The minor peaks of Δ14C between ca. 41,000 and 40,000 IntCal20 yr BP (Fig. 1d, e) are not reflected in the palaeointensity curve, which indicates the collapse of inverse correlations between palaeointensity and Δ14C changes. We need more datasets to assess the correlation during this extremely low palaeointensity period.

VGPs longitudinally constrained pass and clockwise shifting cluster

The VGP-path is constrained to longitudinal bands through the Pacific Ocean for swings A and C, and eastern Africa‒the Indian Ocean for swings D, E and F (Fig. 4a). The VGPs of swings B and G arrived in eastern Africa‒Indian Ocean band via the Pacific band, returning to northern high-latitudes through different bands. Swing C reached the highest southern latitude (71.3°S) within 43 vyr. Such a rapid near-reversal can only occur in the field in the Earth’s liquid outer core, which reverses in direction on time scales of 500-yr or less45. During the Laschamp Excursion, rapid near-reversals occurred within 20–45 vyr in swings B, C and D. It is noteworthy that the excursional VGPs of lava flows from France, Iceland, New Zealand, and Japan related to the Laschamp Excursion are in or near those bands recognised in Lake Suigetsu (Fig. 4a).

Fig. 4: Mollweide projection of VGPs centred on 180° longitude.
figure 4

a VGP-path for the Laschamp to the post-Laschamp excursions, as recorded in Lake Suigetsu, and excursional VGPs from lava flows related to the Laschamp Excursion. The lava-flow VGPs are from France6, 7, 19, Iceland9, 61, New Zealand10, 11, 62 and Japan63. The small black numbered dots show the sites of samples. b VGP clusters for the Laschamp Excursion centred in the North Pacific (NP), the South Pacific (SP), the Southern Indian Ocean (SI) and eastern Africa (EA). The dotted circle indicates a possible cluster (see text) around Australia. The symbols are the same as (a). c Comparison of VGP paths. Aqua colour; the Lake Suigetsu data filtered with an exponential lock-in rate function with a Z1/2 of 1.5 cm and condensed to a resolution of 45-yr per data point (Fig. 3b). Black colour; the Black Sea Laschamp data4, 48. The solid/open circles represent excursional/non-excursional VGPs. The red/brown open symbols represent the peak of the precursor (triangle), the start (circle) and the end (square) of the main excursion for the Lake Suigetsu/Black Sea data. d Time-averaged radial field component at the core-mantle boundary of the time-varying spherical harmonic model (CALS3k.4b) spanning 1000 BC to AD 1990 (adapted from ref. 50).

Full size image

The excursional VGPs make clusters centred in the North Pacific (NP), the South Pacific (SP), the Southern Indian Ocean (SI) and eastern Africa (EA) (Fig. 4b). In addition, most VGPs of lava flows from various locations fall into one of these clusters, except for the SI cluster. The clustering of VGPs from distant (>9000 km) sites indicates that the excursions are dominated by dipolar fields possibly originating from near-radial negative flux concentrations under each cluster. The dipolar source under the NP cluster, whose magnetic north pole locates under the NP, dominated swings A and B, followed by the dominance of the source under the SP during swing C. The source under the SI mainly dominated swing D, and partly swings B and F. The source under the EA dominated swing E. This source recovered ca. 2,600-yr later, to dominate swings F and G. Those sources became dominant generally in clockwise order from the NP to EA, with sporadic extra dominances by the source under the SI. In addition, each dominance interval is relatively short (<ca. 100 yr). The non-excursional intervals dominated by geocentric axial-dipole fields between swings are rather long (240 yr maximum). The VGPs of swings B, D and F dropped in the SI cluster and left it very swiftly. Thus, dipolar-source dominances in the SI cluster were intermittent and short in duration, which might make instantaneous spot-recordings of excursions unlikely for lava flows. The dipolar sources may correlate with the negative flux concentrations at the core-mantle boundary (CMB) in the spherical harmonic (SH) field models of the Laschamp Excursion46,47, whose positions agree well with the four VGP clusters. The models are mainly based on low-resolution deep-sea palaeomagnetic data. Therefore, the flux concentrations at the CMB reflect the features of relatively long-time-averaged fields during the excursion. Due to the low temporal resolution, the SH models do not resolve the frequent normal polarity (non-excursional) field dominances between swings that are observed in the Lake Suigetsu data. In addition, the SH models have an extra concentration under the low-latitude North Atlantic, which may reflect the uneven geographical distribution of the contributing palaeomagnetic data.

The filtered data represent a large open clockwise loop of VGP, consistent with the data from Black Sea sediments4,48, although the southernmost VGPs somewhat shift northward by filtering (Fig. 4c). The two datasets should have captured the same high-cut filtered field-behaviours of the main Laschamp Excursion. However, the absolute age ranges are somewhat discordant, probably ascribed to the application of different dating methods; i.e., the chronology of the Black Sea data is based on climatic wiggle matching. The multi-core data stack from the Black Sea has an analytical interval of ca. 10-yr, but the basic resolution is 50‒80-yr (Supplementary Table 1), with which it is difficult to resolve subcentennial swings. Many sedimentary records of the Laschamp Excursion show similar open clockwise VGP-loops from high northern latitudes through the Pacific Ocean to high southern latitudes, with a return through the Indian Ocean and Africa back to high northern latitudes. The records of the Iceland Basin Excursion at about 190 kyr3 show an open anticlockwise VGP-loop through the same pass15,49. The consistent open clockwise VGP-passes show evidence for the dipolar-field dominance during the Laschamp Excursion. In addition, the consistent VGP-passes shown by different excursions (Laschamp and Iceland Basin) indicate the stationary dipolar-sources recurrently dominated in the same manner.

The four dipolar-field sources (under the NP, the SP, the SI and the EA), correlated with the negative flux concentrations at the CMB of the SH model46,47, are similar to the four stationary high-latitude radial-flux concentrations at the CMB (over Canada, the South Pacific, the Southern Indian Ocean, and Eurasia) that have repeatedly waxed and waned on decadal‒centennial time scales throughout the past 3-kyr50 (Fig. 4d). The hemispherically-symmetric pattern of four stationary flux concentrations may have persisted even during the excursions, provided that the latitudes somewhat lowered and the southern-latitude concentrations reversed in sign. The latter behaviour (sign reversal) is probably common with geomagnetic reversals, and has not occurred in the past 3-kyr of secular variation. The persistence of stationary radial-flux concentrations can be caused by the influence of mantle heterogeneity in thermal core-mantle coupling51,52. The SH models represent no clockwise-order waxing and waning of the stationary radial-flux concentrations at the CMB, due to the low-resolution of data used46,47. However, the flux concentration under the NP predominated in the early stage of the excursion, and that under the EA slightly dominated in the final stage, which possibly concerned swings A/B and E, respectively. Thus the SH models may partly support the clockwise order dominance of dipolar sources.

The VGPs exhibited by three discrete samples dropped in a minor cluster around eastern Australia (Fig. 4b) may represent a transit cluster or a part of an expanded SP or SI. The cluster is near the region where a minor, negative radial-flux concentration has sometimes appeared at the CMB during the late Holocene50. This region overlaps with a cluster of transitional VGPs of lava flows and sediments with high s.r. associated with the Matuyama–Brunhes polarity reversal53,54. This may be evidence of a persistent stationary radial-flux concentration around the region, coupled with the core-mantle dynamics55,56. Thus, this minor cluster is a feature common with the Matuyama–Brunhes reversal, as well as the geomagnetic secular variation.

Summary

The palaeomagnetic record from the varved sediments of Lake Suigetsu (SG14) reveals a sub-structure associated with the Laschamp Excursion at a high chronological resolution, as well as a new post-Laschamp excursion, tentatively named here the 'Post-Laschamp (Suigetsu) Excursion'. The mid-ages and the varve-counted durations are 42,050 ± 120 IntCal20 yr BP and 790 SG062018 vyr for the Laschamp Excursion, and 38,830 ± 140 IntCal20 yr BP and 550 SG062018 vyr for the Post-Laschamp (Suigetsu) Excursion, respectively, coincident with the cosmic ray flux proxy (Δ14C) maxima, with time lags induced by the global carbon cycle. Our data reveal that both excursions are characterised by multiple subcentennial field-directional swings containing rapid near-reversals within 20–45 yr. Simulations of filtering effects on sediment-magnetisations show that our high-resolution record replicates most of the features seen in previous Laschamp Excursion records with lower s.r. The VGPs of the high-resolution Laschamp Excursion dataset make four clusters centred in the North Pacific, the South Pacific, the southern Indian Ocean, and eastern Africa, three of which include the VGP clusters seen in the snapshot records of lavas from various regions. This agreement of the VGP clusters, together with the consistent clockwise open loops of the VGP-pass among the previous low s.r. records and the filtered data of the high-resolution record of this study, demonstrate that the Laschamp Excursion was dominated by non-axial dipolar fields. The dominance of non-axial dipolar fields was intermittent, generated in clockwise order by the four stationary sources, and possibly linked to lower mantle heterogeneity.

Methods

Collection of the SG14 core from Lake Suigetsu

Lake Suigetsu is a tectonic lake with an area of about 4.3 km2 and a maximum depth of 34 m, located close to the Sea of Japan coast, central Japan (Supplementary Fig. 1). The lake has no direct river influx and thus is supplied with no coarse detrital materials. Lake Suigetsu is connected to the neighbouring Lake Mikata by a shallow and narrow sill (ca. 4 m deep, ca. 45 m wide), through which freshwater from the Hasu River, the only freshwater source to the multiple lake system, is supplied together with fine-grained detrital materials which pass through the natural filter of Lake Mikata (Supplementary Fig. 1). Lake Suigetsu has a calm water surface with virtually no high-energy flows, partly due to the protection from winds by surrounding hills, and thus forms a stable stratified water column. The absence of major vertical mixing of water results in less oxygen supply to the lake bottom, creating anaerobic environments and greatly reduced biological activity at the lake bottom. Seasonally different types of material are deposited, comprising a layer of Aulacoseira spp. diatoms with a small amount of siderite overlain by an aeolian dust layer in spring, a light amorphous organic material layer in summer, an Encyonema diatom layer overlain by a siderite layer in autumn, and a clay layer in autumn to winter26. Identification of these seasonal layers, comprising varves, was the key for counting and producing an independent chronology. More than 50,000 annual layers were identified in the top 45 m of the sequence of SG06 (an earlier core retrieved from Lake Suigetsu), and sporadic event layers relating to floods, earthquakes and tephras, along with regular varve accumulation, facilitated the correlation of sections obtained from parallel boreholes26.

In summer 2014, for the purpose of exhibiting varved sediments at the ‘Fukui Prefectural Varve Museum’ (http://varve-museum.pref.fukui.lg.jp/en/), the local government funded the collection of the ‘SG14’ sediment core (officially named ‘Fukui-SG14’) from the central flat plain of Lake Suigetsu, about 350 m east of the SG06 coring point (Supplementary Fig. 1). Like SG06, sections comprising SG14 core were recovered from four parallel boreholes with overlaps, ensuring that there were no gaps in the retrieved composite stratigraphy. Most of the sections are ca. 1-m long. The coring method was similar to that for the SG06 core. The total length of the composite profile of SG14 is 98.2 m, reaching the basement gravel layer. A series of core splitting and high-resolution photography of the exposed sediment cross-sections was performed in the temporary lakeshore laboratory, in parallel with the coring work on the lake. Owing to such parallel works, we could chronologically monitor each section by stratigraphic correlation with the SG06 data including high-resolution cross-section photographs, which enabled prediction of the depths of the target sections for oriented coring. The half-cut core sections were sealed immediately after they were photographed, and kept in plastic bags containing oxygen absorbent material until magnetic sampling.

Magnetic analyses

We preliminarily collected cubic sediment samples of 10 cm3 by pushing plastic boxes into the exposed core sections, followed up with retrieval of 2 cm3 × 2 cm3 × 100 cm3LL-channel samples from each 1 m-section for the main analysis26. From the LL-channel samples, 2 cm3 × 2 cm3 × 2 cm3 samples were cut using a knife at contiguous 2 cm depth intervals for excursional zones and 4‒30 cm depth intervals for the remaining sections. We primarily took samples from the oriented core sections, but also took supplemental samples from unoriented sections. A total of 131 samples, 119 for palaeomagnetic and 12 for rock magnetic analyses, were collected from an age interval of 35 to 45 kyr (most intensively sampled from 37.9 to 42.7 kyr).

All the samples were subjected to progressive THDs at intervals of 30–50 °C up to 590–680 °C. An NRM of a sample was measured using a 2 G cryogenic magnetometer. At each step of THD, low-field magnetic susceptibility (χ) was measured using a magnetic susceptibility metre (MS2, Bartington Measurements, UK) to monitor magnetic mineral alterations. To estimate magnetic carriers, we performed thermomagnetic analyses in an air atmosphere for selected samples using an NMB-89 magnetic balance (Natsuhara Giken Corporation, Osaka, Japan). A sample was heated from 50 °C to 700 °C and then cooled to 50 °C at a rate of 10 °C/min. Isothermal remanent magnetisation (IRM) acquisition experiments and hysteresis measurements were performed for both non-heated and heated (350 °C) samples using an AGM microMag 2900 (Princeton Measurements Corporation, Westerville, OH, USA). The IRM was measured at 100 levels from 0.1 to 1000 mT. Magnetic carriers and primary NRM components based on these experiments’ results are discussed in the Supplementary Discussion.

ChRMs were calculated by principal component analysis. Declinations of ChRMs were corrected by rotating the core segment frame to adjust the mean declination of viscous remanent magnetisations to the present geomagnetic north (Supplementary Discussion, Supplementary Figs. 4,  5 and Supplementary Data 1). For relative palaeointensity, the NRM intensities of samples after 200 °C- and 350 °C-THDs (J200 °C and J350 °C) were normalised by the magnetic susceptibility values of samples after 200 °C- and 350 °C-THDs (χ200 °C and χ350 °C), respectively. The ratios of J200 °C /χ200 °C and J350 °C /χ350 °C further normalised by total means were used for discussion (Supplementary Discussion, Supplementary Fig. 5 and Supplementary Data 2).

Chronology

Construction of the age-depth model for SG06 involved a three-stage process. Firstly, a pure varve (SG062018 vyr) chronology was developed, based upon thin section microscopic counts and an objective varve interpolation programme30. Secondly, the cumulative uncertainty inherent to such layer counting methodologies was reduced through utilising the 639 radiocarbon dates from the varved section of the core and applying a Bayesian Poisson-process deposition model to fit the Suigetsu data onto the independently U–Th dated Hulu Cave radiocarbon calibration dataset36, thus combining the relative dating strength of varves, and the absolute dating precision of U-Th methods34. Finally, the Lake Suigetsu radiocarbon data on this ‘preliminary timescale’, along with all of the other contributory datasets, were incorporated into the construction of the IntCal20 consensus radiocarbon calibration curve itself33. A useful aspect of this methodology is that it directly produces marginal posterior dates (i.e. IntCal20 ages) for the Lake Suigetsu stratigraphy. Accordingly, the IntCal20 chronology for the palaeomagnetic data presented herein is thoroughly robust. The Bayesian age model is transferred to the SG14 core, based on detailed (1 mm-precision) stratigraphic correlations using event layers (e.g. tephra layers) and micro facies captured in high-resolution photographs before oxidisation26. The tephra layers can also be used to correlate Suigetsu to other regional records. In particular, there are tephra layers that are stratigraphically located very close to the two excursions and if they are found in other records they could be used as a proxy for the position of the excursions. One, potentially widespread tephra from Changbaishan volcano, on the border of North Korea and China, is identified just below the Laschamp Excursion. This tephra, B-Sg-42, is found at a composite depth of 3378.9–3379.9 cm57, which corresponds to a date of 42,580 ± 120 IntCal20 yrs BP, and only 16 cm (~180 years) below the onset of the Laschamp Excursion. Another, the U-Ym tephra (SG14-3216), sits between the Laschamp (90 cm after it ends; ~870 years) and the Post-Laschamp (Suigetsu) Excursion (143 cm below the onset; ~1700 years). It is from an eruption of Ulleungdo Island, South Korea, which is dated to 40,830 ± 120 IntCal20 yrs BP based on its composite depth of 3216.2 m57.

The smooth composite depth‒age relationship for the discrete samples used for magnetic analyses confirms the highly precise stratigraphic correlations (Fig. 2), except four levels where a small basic resolution (proportional to a reciprocal of s.r.) is estimated between adjacent sample depth/age data, from different-hole core segments or event layers. We could visually count ca. 20 to 25 varves in dry cubic samples around the expected Laschamp horizon after THDs, which is consistent with the mean s.r. of 96 cm/kyr (=ca. 21 yr in 2 cm). For example, we count 22 and 25 annual layers in the dry sample photographs in Fig. 2c, consistent with the basic resolution (Fig. 2b). The durations of excursional episodes estimated by the varve counts of the varve chronology model 201830 are close to those estimated by the IntCal20 age differences (Table 1). In this study, we use the IntCal20 age model for absolute age estimates and the varve count data for duration estimates for excursions and short episodes.

Filtering of sediment-magnetisation data

Consideration of the filtering (smoothing) effects upon sedimentary magnetic records are essential when comparing the high-resolution Lake Suigetsu palaeomagnetic data with existing lower-resolution datasets. Magnetisation data of sediments measured with a discrete sample suffer two kinds of filters. One is a filter with a box function whose width equals the thickness of a sample. The other is a filter with a lock-in rate function as a result of PDRM processes due to progressive sediment consolidation and dewatering, for which various kinds of functions have been proposed23,38,39,40. As s.r. decreases, time widths of both functions increase to cause higher filtering effects. We estimate how the high-resolution magnetisation data from Lake Suigetsu sediments, with a mean s.r. of 96 cm/kyr, would be modified in sedimentary archives with a lower s.r. The magnetisation data of Lake Suigetsu are already filtered with a box function with a width of 21-yr (i.e. the mean time span of our 2-cm thick sample), but the estimation of filtering effects upon the data is useful for variations with a wavelength longer than 21-yr. We prepared time-series vector (x, y, z) data using the ChRM directions and the relative palaeointensity values from 42,600 to 41,400 cal BP, a dense-sampling interval. For relative palaeointensity values, we used the value of NRM intensity normalised by magnetic susceptibility of a sample demagnetised at 200 °C (J200 °C/χ200 °C). Equally spaced data were generated by condensing densely spaced data and/or linear interpolations for a data blank. The obtained 21-yr resolution time-series data have almost the same structure as the original data except for swing A, in which three densely spaced data were condensed into one. For a filter of a PDRM process, we used an exponential lock-in rate function based on the laboratory-experiment results, r (z); r (z) = A exp (−Az) (A = ln2/Z1/2) for z ≥ 0, and r (z)  0 for z < 0 (Supplementary Fig. 6)38. Z1/2 is a half lock-in depth where 50% of magnetisation is locked-in (95% of magnetisation is locked-in at 4.3Z1/2, and 99% at 6.6Z1/2). Z1/2 values smaller than about 2 cm, excluding the thickness of the bioturbation zone, have been estimated for natural fine sediments. A Z1/2 value of ≤1 cm has been estimated for the fine marine clays of Osaka Bay42. Tauxe et al. 43 estimated a value of about 2 cm for the so-called lock-in depth, a downward shift of the magnetic polarity reversal boundary, which equals Z1/2. Roberts and Winklhofer23 estimated that 95% of the PDRM is locked in 5 cm below the surface mixed layer, which is equivalent to a value of 1.2 cm for Z1/2. These estimates are inconsistent with much larger lock-in depths, e.g. 15 cm58 and 21–34 cm59. However, the former includes the thickness of the bioturbation (no lock-in) zone, and the latter is probably a lock-in depth of magnetosomal origin magnetites, respectively.

For filtering with a box function with a width of Δt, we condensed the original data to that averaged over Δt every Δt. For filtering with an exponential lock-in rate function, we integrated the original data convolved with the function whose value was set to zero beyond 6.6T1/2 (Supplementary Fig. 6).