Paleomagnetism of Paleocene‐Maastrichtian (60–70 Ma) Lava Flows From Tian Shan (Central Asia): Directional Analysis and Paleointensities

The Tuoyun volcanics from the Tian Shan range (Central Asia) give an opportunity to investigate the variability of the Earth's magnetic field during the Cretaceous and Early Paleogene. In the paper we focus on Maastrichtian‐Paleocene basalts and report new paleomagnetic results from 70 lava flows (respectively 45 directional groups) sampled near Tuoyun village (75.33°E; 40.18°N) from three distinct sections. Combined with previous results, our new data set yields a virtual geomagnetic pole at φS = 180.2°E, ϑS = 49.5°N with an angular standard deviation S=20.8∘|18.8∘23.1∘ (for N = 93 directional groups). The mean inclination Is = 36.2° is indicative of an emplacement of the Tuoyun volcanics at a paleolatitude around 20°N, consistent with the high dispersion k = 14.6 of the directions. Such an inclination value is 10–15° lower than the predictions from the apparent polar wander paths for stable Europe and East Asia. This observation can be partly explained by crustal shortening within the East Asian plate during the India‐Asia convergence, but also needs to invoke a local field anomaly in Central Asia as previously proposed. Our absolute paleointensity experiments conducted on two lava flows and two baked sedimentary layers yield a mean virtual dipole moment of 58.9 ± 6.9 ZA·m2, consistent with the values at 60–70 Ma in the global paleointensity database. Compared to relative paleointensity results during the Cretaceous Normal Superchron and during the Plio‐Pleistocene, the relative variability in intensity—supposed to be a proxy for the geodynamo's activity—estimated here at 31 ± 5% (N = 15) from pseudo‐Thellier experiments possibly corroborates a correlation between geomagnetic reversal frequency and paleosecular variation rate during the past 120 Myr.


Introduction
The Earth's magnetic field, generated by a dynamo process in the fluid outer core, stochastically reversed its polarity approximately four times per Myr during the Plio-Pleistocene (e.g., Ogg, 2012). The field's reversal frequency, thought to be mainly controlled by changes in the thermal boundary conditions imposed by the mantle on the core (e.g., McFadden & Merrill, 1984, nevertheless varied over geological time, from prolonged periods (>20 Myr) with stable polarity termed superchrons documented both during the Phanerozoic and the Proterozoic (e.g., Pavlov & Gallet, 2005, 2010 to episodes of hyperactivity with a reversal frequency greater than 10 per Myr documented in particular during the Jurassic and the late Ediacaran (e.g., Gallet & Pavlov, 2016). For the last 160 Myr, the geomagnetic polarity time scale can be divided into four segments lasting approximately 40 Myr each, yielding an average reversal frequency of 4 Myr −1 for 0-40 Ma, 1.3 Myr −1 for 40-80 Ma, naught during the Cretaceous Normal Superchron (CNS,, and 2.4 Myr −1 for 120-160 (e.g., Gallet & Hulot, 1997;Lowrie & Kent, 2004). This can be hypothetically explained by abrupt changes in boundary conditions at the core-mantle boundary (e.g., Courtillot & Olson, 2007) or by spontaneous nonlinear transitions of the geodynamo (e.g., Hulot & Gallet, 2003).
An open question is whether the changes in the frequency of reversals correlate with the gradual temporal changes of the magnetic field during stable periods, referred to as paleosecular variation (PSV). An increase of PSV rate with reversal frequency has often been proposed (e.g., Biggin et al., 2008). Identifying such a correlation however depends on our ability to accurately quantify PSV through geological time. This is traditionally made using the angular standard deviation S of the virtual geomagnetic poles (also known as VGP scatter) calculated from a population of individual poles derived from several lava flows. This quantity depends on site latitude (e.g., Cromwell et al., 2018;Johnson et al., 2008) and its ability to discriminate different geodynamo states has been recently questioned (e.g., Doubrovine et al., 2019;Linder & Gilder, 2011). Another PSV proxy, the relative variability in paleointensity-the standard deviation normalized by the average value for a population of individual intensities derived from several sites-is thought to be almost independent of latitude (e.g., Lhuillier & Gilder, 2013). Albeit more favorable than the VGP scatter for comparing PSV over time, its determination is challenging due to the low success rate of absolute paleointensity (API) and relative paleointensity (RPI) methods (e.g., Tauxe & Yamazaki, 2015).
The Tuoyun basin is an optimal location in the Tian Shan range (Central Asia, Figure 1a) to investigate PSV, with several tens of lava flows emplaced both during the Albian-Aptian and Maastrichtian-Paleocene periods. Lhuillier et al. (2016) found a VGP scatter 20-25% lower during the CNS than 50-60 Myr later, consistent with the hypothesis of a greater stability of the field during superchrons. This study also documented a bimodal distribution of inclinations depicting presumably dipolar and multipolar components. Such a behavior was more pronounced at 112-115 Ma than at 60-70 Ma, suggesting more temporal variability than previously expected during the CNS, in agreement with Granot et al. (2012). This bimodal distribution of directions was interpreted in terms of a two-state dynamo, namely, a bistable dynamo that can spontaneously fluctuate between two distinct stable regimes (e.g., Simitev & Busse, 2009, 2012. Cretaceous-Cenozoic paleoinclinations in Central Asia were often found to be more than 10°lower than those predicted by the Eurasian apparent polar wander path (APWP, e.g., Cogné et al., 2013;Dupont-Nivet et al., 2010;Gilder et al., 2003;Tan et al., 2003). While undetected inclination flattening in sediments or insufficient sampling of PSV in volcanics can be invoked in particular cases, this inclination anomaly is possibly ascribed to an anomalous behavior of the geomagnetic field (e.g., Chauvin et al., 1996;Lhuillier et al., 2016;Si & Van der Voo, 2001;Westphal, 1993) and/or continental deformation within the Eurasian plate (e.g., Cogné et al., 1999Cogné et al., , 2013Hankard et al., 2007). Identifying the primary cause of this inclination anomaly could help to better quantify the amount of crustal shortening that occurred during the India-Asia collision and/or to appraise the validity of using paleomagnetic data for tectonic reconstructions in Central Asia (e.g., Dupont-Nivet et al., 2010;Meng et al., 2017Meng et al., , 2018Meng et al., -2020van Hinsbergen et al., 2012).
The aim of this paper is thus to scrutinize the field behavior around 60 Ma with a new sampling of the top of the Maastrichtian-Paleocene section in order to (i) identify the source of inclination anomaly in Central Asia; (ii) question the robustness of the two-state dynamo hypothesis; (iii) conduct an API and RPI survey in addition to the paleodirectional analysis. To this end, we first delineate the geological setting and sampling procedure (section 2). We then report our directional results (section 3) with a focus on the detection of serially correlated and transitional flows, as well as our intensity results (section 4) with an emphasis placed on the preselection and postselection of the samples. Finally, we discuss the implications of our results in terms of geomagnetic field behavior and Central Asian tectonics (section 5).

Geology and Sampling
The Tian Shan (or Tengri Tagh) range in Central Asia was principally formed by convergent tectonism during the Late Paleozoic (e.g., Windley et al., 1990). Although this region was then mostly amagmatic, some igneous units were emplaced within Mesozoic-Paleogene sediments (e.g., Sobel & Arnaud, 2000). With similar lithologies suggesting a common origin, those units were found both in the central portion of the Kyrgyz Tian Shan as well as in the Xinjiang autonomous region of China. The Xinjiang volcanics are mostly present within the Tuoyun intermontane basin, north of Kashgar and west of the Talas-Ferghana (TF) strike-slip fault. Volcanism was documented at 110-120 Ma during the Albian-Aptian as well as at 50-70 Ma across the Paleocene, with the oldest ages being found in the vicinity of the TF fault. Sobel and Arnaud (2000) associated the Tuoyun basalts to a deep-mantle component possibly mixed with a lithospheric component. In the absence of geological support for continental rifting, they speculated that Tuoyun volcanism may originate from a small plume rooted at a shallow level in the asthenosphere.
In the vicinity of Tuoyun village, both Early Cretaceous and Paleocene basalts outcrop within a distance of a few kilometers (Figure 1b, see Huang et al., 2005;Lhuillier et al., 2016;Li et al., 1995;Sobel & Arnaud, 2000, for radiometric ages). In the most recent paleomagnetic study in this area, Lhuillier et al. (2016) sampled a 550-m-thick sequence (P section, 59-70 Ma) spanning the Cretaceous-Paleocene boundary and consisting of 76 lava flows and 5 intercalated layers of baked sediments. They also sampled two possibly distinct Early Cretaceous sections dated 112-115 Ma: the 100-m-thick K-section (22 lava flows), and the 200-m-thick L-section (35 lava flows).
In this study, we sampled three additional basaltic sections of Paleocene age. As illustrated in Figure 1b, section A (75.25°E; 40.21°N) composed of 32 lava flows and section B (75.23°E; 40.23°N) composed of 12 lava flows are located in the vicinity of Sites ty1-34 (75.25°E; 40.23°N) from Huang et al. (2005), with two 40 Ar/ 39 Ar isochron ages of 58.5 ± 1.3 (2σ) Ma and 60.4 ± 1.3 (2σ) Ma for neighboring outcrops (Huang et al., 2005). Based on repeated readings for 12 distinct horizons, the flows from the 320-m thick A-section have an average bedding attitude of 28°/41°(dip-direction/dip convention). Based on repeated readings for two distinct horizons, the flows from the 45-m-thick B-section have an average bedding attitude of 14°/60°. From field observations, the A and B sections are likely distinct due to the possible presence of a fault following the Suyue River. We evaluated that the bottom of section A is 50-m stratigraphically higher than the top of section B, but note that the exact offset is uncertain. Section C (75.32°E; 40.26°N) composed of 26 lava flows is located in the vicinity of Sites c904-c907 from Gilder et al. (2003), with an 40 Ar/ 39 Ar isochron age of 60.82 ± 0.56 (2σ) Ma for Site c905 (Lhuillier et al., 2016). Based on repeated readings for six distinct horizons, the flows from the 170-m-thick C-section have an average bedding attitude of 140°/47°.
We drilled and oriented a total of 750 paleomagnetic cores, with an average of ∼9 cores per site (Table 1). Each site presumably corresponds to an independent lava flow and sampling was done within an area of several square meters to integrate the intraflow geomagnetic variability. The magnetic anomaly between magnetic azimuths and Sun azimuths (the latter having been measured in ∼50% of the cases) has a median of 4.47 ± 2.48°(2σ), which is consistent with the declination value of 4.33°predicted at Tuoyun in September 2017. In addition to the three basaltic sections, we also sampled two layers of baked sediments (Figures 1f and 1g) stratigraphically equivalent to Sites P06 and P17 from Lhuillier et al. (2016) to conduct API experiments.

Paleomagnetic Experiments
All experiments in this study were carried out in the paleomagnetic laboratory of the Ludwig-Maximilian-University, Munich: alternating field (AF) demagnetization for all samples using the automated Sushibar system based on a three-axis 2G Enterprises superconducting magnetometer and a custom-made coil (Wack & Gilder, 2012), thermal demagnetization for pilot samples using an ASC TD-48 furnace together with the Sushibar. We determined the specimen-level characteristic remanent magnetization (ChRM) directions using principal component analysis (PCA, Kirschvink, 1980) with the help of the paleomagnetism.org environment (Koymans et al., 2016). To determine likely remanence carriers, we computed the median destructive field (MDF)-the value for which half of the natural remanent magnetization (NRM) is randomized-for each sample from the AF decay curve reconstructed by modular subtraction, and the unblocking temperature T u -the value for which 95% of the NRM is demagnetized-for each pilot sample from the thermal decay curves reconstructed by modular subtraction.

Demagnetization Behavior
For the three new sections as well as section P from Lhuillier et al. (2016), the NRM values range from 0.02 to 39.8 A/m with a median of 1.62 A/m (Figure 2a). The unblocking temperature has a median of 567°C with an interquartile range of 55°C, which indicates the presence of low-titanium titanomagnetite as the main remanence carrier (Figure 2b). The MDF has a median value of 23.6 mT with an interquartile range of 35.2 mT (Figure 2c). Its peak at ∼12.5 mT is mostly ascribable to overprint components rather than low coercivities of remanence.
In order to get a fully consistent data set, we first reinterpreted the demagnetization diagrams from section P (Lhuillier et al., 2016) and chose to only retain the most robust sites for which the vector endpoints of the ChRM component unambiguously went to the origin. A small secondary component was usually removed by 10-20 mT or 200-300°C. The ChRM directions determined by origin-anchored PCA fits led to a maximum angular deviation (MAD) lower than 1.4°in 50% of the cases, only exceeding 5°in 5% of the cases.  For the new samples from sections A-C, the secondary component was a little stronger than for section P but easily removed by 20-30 mT or 300-400°C. With an unidirectional decay of the ChRM vector endpoints to the origin, all the directions were fitted by origin-anchored PCA, leading to a MAD lower than 1.7°in 50% of the cases, exceeding 5°in 9% of the cases. Figures 3g-3i (respectively Figure 3j-l) show the demagnetization behavior for a site with low (respectively high) coercivity samples. In both cases, the individual directions led to a well-defined site mean direction with a 95% confidence radius (α 95 ) lower than 5°.

Site Mean Directions and Directional Groups
For the four investigated sections, we computed the site mean directions using Fisher (1953) statistics (Tables 1 and 2 and Figure 4). As reported in Lhuillier et al. (2016), section P recorded both normal and  Note. Underlined sites represent baked sedimentary layers. a Flows/layers that were merged into directional groups highlighted in italics. b Flows/layers that were chosen for absolute paleointensity experiments.
reversed polarities: 31 m of reversed polarity (R1), 18 m of normal polarity (N1), 375 m of reversed polarity (R2), and 101 m of normal polarity (N2). With an optimal cutoff angle of 41.2°for this data set (Vandamme, 1994), no transitional directions were detected. With an angle θ ¼ 1.3°between the normal The equal-area projections (right column) show the intrasite dispersion in stratigraphic coordinates with the mean direction (D s ,I s ) in black. The parameter k (respectively α 95 ) is the Fisher (1953) precision parameter (respectively 95% confidence radius), whereas δ TH AF is the average angular distance between couples of thermally and AF demagnetized specimens.

Table 2
Paleomagnetic Results of Section A (28°/41°, Dip-Direction/Dip), Section B (14°/60°, Dip-Direction/Dip) and Section C (140°/47°, Dip-Direction/Dip) Using the Same Convections as in Table 1 hstrat Journal of Geophysical Research: Solid Earth and reversed directions lower than the critical angle θ c ¼ 11.0°for which the two populations would be different at the 95% confidence level, the reversal test is positive (Rc classification of McFadden & McElhinny, 1990). If only the youngest part of section P is considered (stratigraphic height above 300 m), we note that the reversal test is also positive (θ ¼ 7.3°, θ c ¼ 11.0°, Rc classification of McFadden & McElhinny, 1990).
Section A recorded both polarities: 61 m of reversed polarity, 124 m of normal polarity, and 146 m of reversed polarity. Relying on radiometric ages and polarity changes, we propose that the magnetozone of normal polarity is associated with magnetozone N2 of section P, so that we added an offset of 372 m for the stratigraphic height. With an optimal cutoff angle of 47.0°, no transitional directions were detected. With an angle θ ¼ 4.1°between the normal and reversed directions lower than the critical angle θ c ¼ 13.6°, the reversal test is positive for section A (Rc classification of McFadden & McElhinny, 1990). We note that reversal tests, separately applied to two consecutive polarity intervals of section A, are also positive (Rc classification of McFadden & McElhinny, 1990). Sections B and C only recorded stable reversed polarities. To match the directional hump at ∼375 m in section P, we added an offset of 272 m to the stratigraphic height of section C. We relied on our field observations to link the stratigraphic height of sections A and B, with the bottom of section A being ∼50 m higher than the top of section B.
To detect the potential existence of volcanic pulses in the four investigated sections, we considered the concept of directional groups extensively used for large igneous provinces (e.g., Mankinen et al., 1985). A directional group (DG) was inferred if the angular distance δ between the mean directions of two consecutive sites is lower than the threshold value δ c defined as the quadratic sum of the corresponding α 95 values (e.g., Chenet et al., 2008;Lhuillier & Gilder, 2019;Moulin et al., 2017). Excluding Sites P16, P48, P49, and P52 from the procedure due to their high dispersion (α 95 ≥ 10°), we detected 10 DGs (from 23 serially correlated sites) in section P, 6 DGs (from 13 serially correlated sites) in section A, 1 DG (from 5 serially correlated sites) in section B, and 5 DGs (from 13 serially correlated sites) in section C (Tables 1 and 2). The DGs are subsequently treated as individual paleomagnetic directions.
We note that sections A and P unequally sample normal and reversed polarities. Whereas the abundantly represented population of reversed polarity can be considered roughly unimodal, the poorly represented population of normal polarity shows two modes that can be distinguished by their declination values: D s ¼10-40°E on the one hand, D s ¼ 60-70°E on the other. The second population only corresponds to five  Journal of Geophysical Research: Solid Earth sites (A12, A13, A14, A16, and A15) from section A and four sites (P68, P69, P72, and P73) from section P. These sites occur at comparable stratigraphic heights (465-501 m for section A, 450-485 m for section P), which corroborates the stratigraphic match that we previously made between sections A and P. Their higher declination values may be the result of enhanced PSV after a reversal. Nevertheless, the apparent dichotomy in declination is more likely ascribable to an insufficient sampling of the directions of normal polarity.

Average Pole Compared to Reference Poles
To get a robust pole, we combined the results from sections A-C and section P. Statistical boostrap experiments reveal a maximum of the eigenvalue τ 1 associated with the principal axis of the orientation matrix (Tauxe & Watson, 1994;Tauxe et al., 1991) for an unfolding degree ranging between 88 and 111% at the 95% confidence level, so that the fold test is positive ( Figure S1 in the supporting information). The combined data set also passes the reversal test (classification Rb of McFadden & McElhinny, 1990) with the angle θ ¼ 5.6°not exceeding the critical value θ c ¼ 9.8°for which the normal and reversed populations would be different at the 95% confidence level. A more robust reversal test-that does not assume Fisherian distributions and compares the principal axes from bootstrapped distributions (e.g., Tauxe, 2010)-is also positive with the intersection of the Cartesian components of the normal and reversed populations at 95% confidence level ( Figure S2). The coordinates of the mean pole of the combined data set are φ S ¼ 180.2°E and

10.1029/2019JB018631
Journal of Geophysical Research: Solid Earth As illustrated in Table 3, this average pole is largely discordant with the predictions from the APWPs for stable Eurasia at 65 Ma (Besse & Courtillot, 2002Torsvik et al., 2008Torsvik et al., , 2012. The vertical axis rotation R (required to align an observed declination with an expected declination) and the flattening F of inclination (expected inclination minus observed inclination) yield values: R ¼ 34.4 ± 6.6°and F ¼ 15.9 ± 5.5°for the APWP of Besse and Courtillot (2002); R ¼ 29.0 ± 5.8°and F ¼ 20.8 ± 4.6°for the APWP of Torsvik et al. (2012). This suggests that the investigated lava flows, located in the vicinity of the major TF fault, were clockwise rotated by ∼32°with respect to stable Eurasia, consistent with previous paleomagnetic results in the Tuoyun basin (Gilder et al., 2003;Huang et al., 2005;Meng et al., 1998). If such a rotation is assumed, the new coordinates of the mean pole are φ S ¼ 226.6°E and ϑ S ¼ 69.0°E. We note that a vertical-axis rotation does not change the geometry (dispersion, elongation) of the distribution of directions (or VGPs), so that the PSV analysis can equally be done on the original or rotation-corrected data sets. In both cases, the possible interpretations of the ∼18°inclination flattening will be discussed in section 5.2.

Statistics of Paleosecular Variation
We first compared the data sets from sections A-C and P after taking the antipodes of the reversed directions (Table 3). The two populations are statistically indistinguishable with their angular distance θ ¼ 2.3°not exceeding the critical value θ c ¼ 7.7°for which the two populations would be different at the 95% confidence level (Watson, 1983). The elongation of the directions from section P (e ¼ 2.0) is compatible with the prediction of the TK03.GAD and BCE19.GAD statistical field models (  Note. N, number of flows/determinations; mean directions (D s ,I s ) and VGPs (Plon s ,Plat s ) in stratigraphic coordinates; k (or K), precision parameter; α 95 (or A 95 ), 95% confidence radius; e (or E), elongation parameter; S, angular dispersion of the VGPs with 95% confidence interval. The last column indicates the classification of the reversal test according to McFadden and McElhinny (1990). BC02 refers to the APWP by Courtillot (2002, 2003) and Tor12 to the APWP by Torsvik et al. (2012). a Vandamme et al. (1991); b Vandamme and Courtillot (1992), c Chenet et al. (2008) d Riisager et al. (2002; e Ganerød et al. (2008), f Ganerød et al. (2010) paleolatitude of 20°N (e ≈ 2.0 in both cases), whereas the distribution of the directions from sections A-C is much more elongated (e ¼ 3.7) and even exceeds the prediction of statistical field models at the equator (e ≈ 2.9). The combined data set yields an intermediate elongation (e ¼ 2.5) that would be compatible with statistical field models if the flows had been emplaced at a paleolatitude of 10-15°. The between-site dispersion S B of the stable VGPs corrected for intrasite dispersion (e.g., Johnson et al., 2008) yields 20:8 ∘ j 23:1 ∘ 18:8 ∘ . Such a value would be expected for a paleolatitude of 40-50°according to statistical field models for Journal of Geophysical Research: Solid Earth the past 10 Myr, whereas the mean inclination (I S ¼ 36.2°) of our combined data set suggests a paleolatitude of ∼20°. As the four investigated sections present some stratigraphic overlap, we cannot exclude that we sampled the same flows several times. However, the VGP scatters from sections P and A-C are almost indistinguishable to within the uncertainties (Table 3), which indicates that, even if we oversampled some flows, we probably did it in an unbiased way. For eight key samples chosen for API experiments, ∼1-cm-high slices were prepared for reflected light microscopy. They were first manually polished with silicon carbide foils (FEPA P180 to P800 with average particle diameter from 80 to 20 μm) to flatten and smooth down the surface. They were then mechanically polished during 90-120 min with a 1 μm polycrystalline diamond paste, and 30 min with a 1/4 μm polycrystalline diamond paste. The polished sections were observed with 50X and 125X oil-immersion objectives mounted on a Leitz microscope.

Results
As illustrated in Figure 5a, bulk hysteresis properties shown in a Day diagram mostly lie within the area usually associated with pseudo-single domain behavior and coincide with the single domain and multidomain mixing curves of magnetite (Dunlop, 2002). Thermomagnetic curves (Figure 5b) reveal in the majority of cases the coexistence of low-T C (350-400°) and high-T C (500-600°) ferrimagnetic phases, consistent with titanomagnetite. Heating and cooling branches are moreover largely irreversible with an average departure of 83% between the two branches at 100°C.
We only identified two basaltic sites potentially suitable for API experiments. Sites B08 (at a stratigraphic height of ∼308 m) and A22 (at a stratigraphic height of ∼585 m) are both characterized by a single high-T C (565-575°) phase and a reversible behavior to within 10%. We additionally selected two sites of baked sediments for API determinations. Samples RB1152-1165 correspond to Site P06 (at a stratigraphic height of ∼41 m) and samples RB561-573 to Site P17 (at a stratigraphic height of ∼131 m). Both sites show a perfectly reversible behavior of the thermomagnetic curves to within the uncertainties of the instrument, with a Curie temperature of ∼625°C for P06 and ∼600°C for P17.
The polished sections from basaltic sites B08 and A22 revealed an advanced stage of high-temperature oxidation. In most titanomagnetite grains, a wavery intergrowth of pseudo-brookite and hemoilmenite can be unambiguously identified (Figure 5d Coe (1967) with partial thermoremanent magnetization (pTRM) checks: a first batch of 50 specimens using a laboratory field of 50 μT, followed by a second batch of 20 specimens using a laboratory field of 25 μT. Measurements were made using an ASCTD48 furnace and the automated Sushibar system. To assess the robustness of the API determinations, the following statistics were computed: n, the number of temperature steps used for the best fit line on the Arai-Nagata diagram (Nagata et al., 1963); f, the NRM fraction used for the best fit line on the Arai-Nagata diagram; β, the standard value of the slope normalized by its absolute value; q, the quality factor (Coe et al., 1978); MAD, the maximum angular deviation of the free-floating directional fits (Kirschvink, 1980); α, the angular difference between the origin-anchored and free-floating best fit directions; DRAT, the maximum difference ratio measured from pTRM checks (Selkin & Tauxe, 2000); and CDRAT, the absolute value of the sum of the pTRM differences (Kissel & Laj, 2004). Their values were compared to PICRIT-03 selection criteria (Kissel & Laj, 2004). Stricter selection criteria such as those applied to glassy volcanics (e.g., Cromwell et al., 2015) or archeomagnetic materials (e.g., Cai et al., 2017) were unfortunately not applicable to our samples because the NRM fraction did not exceed 60%. To further assess the thermal stability of the investigated samples, successive heating cycles M i S ðTÞ for peak temperature T i max ¼ 200, 300, 400, 500, and 600°were measured on 5-mm-diameter sister cores using a VFTB. Figures 6a-6d show typical Arai-Nagata diagrams for the four investigated sites. The fits were computed from 200°C, temperature at which the weak secondary component was reasonably removed (see associated vector endpoint diagrams), to 500°C for Site P06, 525°C for Sites P17 and A22, and 575°C for Site B08. The basalts from Sites A22 and B08 experienced thermal stability up to 500°C (Figures 6e and 6f), whereas the baked sediments from Sites P17 and P06 seemed, to within instrumental uncertainties, to be perfectly stable up to 600°C (Figures 6g and 6h). The second slope above 525°C (respectively 575°C) in the Arai-Nagata diagrams for Site A22 (respectively B08) is likely ascribable to chemical alteration, as indicated by the negative pTRM checks (Figures 6a and 6b). The full API determinations with their statistics are reported in Table 4. Figure 7a reveals a higher dispersion of the API estimates converted into virtual dipole moments (VDM, e.g., Tauxe, 2010) for the basalts than for the baked sediments. Applying the PICRIT-03 selection criteria (Kissel & Laj, 2004) slightly reduces the dispersion of the estimates for the baked sediments and improve the consistency between the two sites. Contrastingly, the selection process for the basalts is inefficient to improve the consistency of the estimates.

10.1029/2019JB018631
Journal of Geophysical Research: Solid Earth As we expected the Tuoyun volcanics to be emplaced at a midlatitude, our first batch of experiments was made with a 50 μT laboratory field. When we realized that the API estimates were twice lower, we conducted a second batch of experiments with a 25 μT laboratory field (save for Site A22 for which we had no available specimens anymore) to check the potential dependency of API estimates on laboratory field (e.g., Selkin et al., 2007). Before the application of the selection criteria, the API estimates are on average 7% higher for a 50 μT laboratory field than for a 25 μT laboratory field. After the application of the selection criteria, the API estimates obtained with a 50 μT laboratory field are still a few percents higher than the API estimates obtained with a 25 μT laboratory field. However, we did not apply any correction for the API estimates obtained with a 50 μT laboratory field as the bias with respect to a 25 μT laboratory field is poorly constrained-due to the limited number of successful API estimates obtained with both 25 and 50 μT laboratory fields-and lies within the intrinsic uncertainties of API estimates.

Pseudo-Thellier Experiments 4.3.1. Methods
We conducted RPI experiments using the pseudo-Thellier method (Tauxe et al., 1995) on a selection of 90 previously AF demagnetized specimens (among 19 cooling units) chosen for their homogeneous demagnetization behavior. While the use of RPI is common in sedimentary rocks, its usage in volcanic rocks was promoted by Yu et al. (2003). The purpose of the pseudo-Thellier experiments in this study is not to provide calibrated paleointensities (e.g., Bogue, 2018;de Groot et al., 2013) but to investigate the relative variability in intensity as a PSV proxy (e.g., Lhuillier et al., 2016Lhuillier et al., , 2017. More specifically, the pseudo-Thellier method implies the stepwise acquisition of an anhysteric remanent magnetization (ARM) in a laboratory field B lab (30 μT in our case), followed by the stepwise demagnetization of the last imparted ARM. In this study, the analysis was performed using the automated Sushibar system in 5 mT steps up 30 mT, then in 10 mT steps up to 90 mT. To account for the variable amounts of NRM remaining after the 90 mT step, we removed by vector subtraction the residual magnetization from all remanence data (Lhuillier et al., 2017). The results were analyzed in the framework of a pseudo-Arai diagram that compares the remaining NRM with the acquired ARM. The slope b between these two quantities over the chosen AF segment-computed in the sense of Coe et al. (1978) with its standard deviation σ b , NRM fraction f and relative standard error β-yields the RPI estimate.
To check the reliability of the estimates, the pseudo-Arai diagram can be supplemented with a demag-demag plot (ARM remaining versus NRM remaining) as well as an ARM-ARM plot (ARM remaining versus ARM gained) according to the terminology of Paterson et al. (2016). Both diagrams are expected to yield straight lines over the chosen segment, with the additional constraints that the best-fit line on the demag-demag plot should intersect the origin and the slope on the ARM-ARM plot should tend toward one. The linearity (respectively the trend toward the origin) can be quantified by the coefficient R of correlation (respectively the fraction f res of residual ARM at the origin). To circumvent the problem of the dependency of pseudo-Thellier estimates on the size distribution of the remanence-carrying grains (Yu et al., 2003), selection can also be done using the grain-size indicator B ARM 1=2 -the AF field for which half of the maximum ARM is imparted-as suggested by de Groot et al. (2013).
We note that the choice of the selection criteria is widely debated. In addition to the linearity criteria, de Groot et al. (2013) recommended to only retain the measurements for which 23mT<B ARM 1=2 <63 mT and suggested that a general calibration law can be employed to produce API estimates. de Groot et al. (2015) then recognized that the pseudo-Thellier slopes may not only depend on the grain size (Yu et al., 2003) but also on the chemical composition of the remanence carriers. For a heterogenous set of samples, Paterson et al. (2016) questioned the efficiency of the criterion based on B ARM 1=2 (de Groot et al., 2013) and emphasized the difficulty in finding universal selection criteria that reduce the dispersion of the calibration law. To produce reliable RPI estimates, we thus adopt in this study a pragmatic approach and choose the range of B ARM 1=2 and f that tend to minimize the dependency on the pseudo-Thellier slope. Figure 8 shows typical examples of pseudo-Thellier results for a specimen yielding a RPI on the order of unit (panels a-c) and for a specimen yielding a RPI approximately three times higher (panels d-f). In both cases, the pseudo-Arai diagrams (Figures 8c and 8f) show a high degree of linearity of the slopes that can be unambiguously determined over a NRM fraction exceeding 60%.

Results
Before applying any selection criteria, the site mean pseudo-Thellier slopes range from 0.3 to 2.3, with a relative standard error from 15% to 90% (Figure 7b). As shown in Figure 7c, the pseudo-Thellier slope clearly depends on the grain-size indicator B ARM 1=2 and, to a lesser extent, to the NRM fraction f. To limit the dependency of the pseudo-Thellier slope on B ARM 1=2 and f, we chose to only retain specimens satisfying 15mT<B ARM 1=2 <25mT and f > 0.40, in addition to the linearity criteria (R > 0.95 for the three diagrams) and f res < 0.15. It leads to a set of 37 (out of 88) robust RPI estimates, the site means of which range from 0.8 to 2.5, with a relative standard error from 2% to 46% (Figure 7b).
From these results, we derived the interflow relative variability ε F known to be a proxy for PSV (Lhuillier & Gilder, 2013) and found a value of 31 ± 5% estimated with bootstrap experiments on 1000 draws. Due to the low number of considered cooling units (N ¼ 15 after the selection process) with sometimes only one accepted determination per site, this value must be of course treated with caution. It was unfortunately not possible to conduct pseudo-Thellier experiments on a larger number of sites because of the very heterogeneous behavior of the AF decay curves.

10.1029/2019JB018631
Journal of Geophysical Research: Solid Earth

Discussion
In this discussion, we first compare in section 5.1 our directional results with those from coeval volcanic rocks in other regions (Deccan Traps and North Atlantic Igneous Province). We then focus in section 5.2 on the interpretation of the low inclination values at Tuoyun, in terms of crustal shortening and/or two-state geodynamo. We finally compare in section 5.3 our paleointensity results with the API database and other RPI results over the past 150 Myr.

Directional Behavior During the Paleocene 5.1.1. Considered Data Sets
Our paleodirectional results from the 60-70 Ma lava flows from Tuoyun can be compared to those from the Deccan Traps emplaced around 65 Ma (e.g., Chenet et al., 2007;Knight et al., 2003) and those from the North Atlantic Igneous Province (NAIP) emplaced between 55 and 62 Ma (e.g., Ganerød et al., 2010;Sinton et al., 1998). For the Deccan Traps, we combined the 21 independent directions from the Nagpur-Bombay traverse (Courtillot et al., 1986;Vandamme et al., 1991), the 13 independent directions from the Mandla region (Vandamme & Courtillot, 1992), and the 25 independent directions from the Mahabaleshwar escarpment (Chenet et al., 2008) to produce a collection of 59 directions, two of which are transitional according to the Vandamme (1994) cutoff criterion. For the Eastern part of the NAIP, we combined the 43 independent directions from the Faroe Islands (Riisager et al., 2002) dated 55-58 Ma, the 30 independent directions from the Isle of Mull in Scotland (Ganerød et al., 2008) dated 58-60 Ma, and the 36 independent directions from the Antrim Lava Group in Northern Island (Ganerød et al., 2010) dated 60-62 Ma to produce a collection of 109 directions, two of which are transitional according to the Vandamme (1994) cutoff criterion. According to the APWPs (Besse & Courtillot, 2002;Torsvik et al., 2012) for India (resp. Europe), the data set for the Deccan Traps (respectively the NAIP) is representative of PSV at a paleolatitude of 24-27°S (respectively 45-52°N).

General Trends During the Paleocene
Instead of working with VGP distributions, we based our analysis on the distribution of directions (recentered about their principal axis for each data set) to avoid any assumption about the field geometry. As reported in Figure 9 and Table 3, the elongation parameter is 25% lower for the NAIP (e ¼ 1.4) than for the Deccan traps (e ¼ 1.9), with values close to those predicted by the BCE19.GAD field model for the past 10 Myr (e ≈ 1.3 @ 50°N, e ≈ 1.8 @ 25°S). This observation points to a similar geometry of the Earth's magnetic field at 55-65 and 0-5 Ma. The precision parameter of the directions is 30% higher for the NAIP (k ¼ 27.2) than for the Deccan traps (k ¼ 20.7), with a trend compatible with BCE19.GAD (k ≈ 17.1 @ 25°S, k ≈ 23.5 @ 50°N) although the values are in both cases 20-30% higher than the predictions. This lower dispersion of the directions coincides with the lower rate of reversals during the Paleocene-Maastrichtian (ca. 1 Myr −1 , Ogg, 2012) than during the past 5 Myr (∼5 Myr −1 , Ogg, 2012), corroborating a possible increase of PSV rate with geomagnetic reversal frequency (e.g., Biggin et al., 2008;Lhuillier & Gilder, 2013).

Paleodirections of Tuoyun Volcanics
In comparison, the Tuoyun volcanics, the paleolatitude of which is uncertain-around 20°N according to our data set in the framework of a geocentric axial dipole (GAD) field, 34-36°N according to the APWPs for stable Europe-show recentered directions elongated in the WS-NE direction, inconsistent with BCE19.GAD that predicts an elongation in the meridian plane. That notwithstanding, the elongation (e ¼ 2.5) together with the higher dispersion (k ¼ 14.6) of the directions make an emplacement at low latitude plausible. Were the Tuoyun volcanics emplaced at mid latitude, their directional distribution should be intermediate between those observed for the Deccan Traps and NAIP in the framework of a GAD field. Such a departure from the GAD geometry (i.e., a higher dispersion and a higher elongation than predicted by BCE19.GAD) may be the signature of a non-zonal component of the global field, a feature that was not observed for the Deccan Traps and NAIP. This departure could be more likely the signature of a local anomaly specific to Central Asia. Such an anomaly was proposed to have occurred from the Middle Eocene (e.g., Chauvin et al., 1996), from the latest Paleocene (e.g., Si & Van der Voo, 2001) or possibly as early as ∼115 Myr ago (e.g., Lhuillier et al., 2016).

Interpretation of the Low Inclinations at Tuoyun
The inclination values obtained from the Tuoyun volcanics are more than 15°lower than the predictions from the APWPs for stable Europe (Besse & Courtillot, 2002;Torsvik et al., 2012). In the previous paragraph, we however showed that the distribution of the directions may be consistent with low inclinations, unless a local field anomaly is invoked. In this section, we review the different explanations that could reconcile these apparently contradictory observations.

Structural or Rock-Magnetic Causes
Structural and rock-magnetic causes are in our opinion unlikely to explain the low inclination values. First, the investigated sections are continuous with clearly identified lava flows. Second, the inclination values do not correlate with any rock-magnetic parameter such as the Curie temperature or the anisotropy ratio  Shcherbakov & Sycheva, 2006 for the past 150 Myr. The black curves define the ±1σ envelope produced by a moving average over a 10 Myr window. (b) Relative variability ε F in intensity as a function of reversal frequency. The black curve is the prediction from numerical dynamo simulations (Lhuillier & Gilder, 2013).

10.1029/2019JB018631
Journal of Geophysical Research: Solid Earth (Lhuillier et al., 2016). One additional difficulty with lava flows is their potentially irregular eruption rate. The 40 Ar/ 39 Ar age determinations at Tuoyun suggest an enhanced volcanic activity around 60 Ma (Huang et al., 2005;Lhuillier et al., 2016;Sobel & Arnaud, 2000). As illustrated in Figure 4, our composite section shows around this age a magnetozone of normal polarity (N2) possibly associated with chron C26n. The magnetozone R3 would thus correspond to chron C25r and the magnetozone R2 partly to chron C26r. The 40 Ar/ 39 Ar age determination of Site P03 at 69.88 ± 0.63 (2σ) Ma (Lhuillier et al., 2016) suggests a time hiatus in our sequence, probably situated at the bottom of section B. In spite of that, the alternation of normal and reversed polarities concurrently with the positive reversal test suggest that a sufficient amount of time has been sampled to accurately represent the distribution of directions.

Tectonic Interpretation: A Crustal Shortening?
Under the assumption of a GAD field, the site mean paleoinclination determined in this study implies a paleolatitude of 20.1 ± 3.8°N at Tuoyun during the 60-70 Ma interval. Combining our results with the site mean directions with α 95 < 10°from Bazhenov and Mikolaichuk (2002), Gilder et al. (2003), and Huang et al. (2005) in the Tuoyun and Tekelik areas yields a slightly higher paleolatitude of 25.7 ± 3.9°N, with a pole at 177.0°E, 54.2°N (N ¼ 140,A 95 ¼ 3.9°) that is representative for the Tarim Basin at 60-70 Ma. In both cases, the obtained paleolatitudes are 10-15°lower than the value of 34-36°predicted by the APWP for stable Europe (Besse & Courtillot, 2002;Torsvik et al., 2012), which was supported by the paleolatitude of ∼39°d erived from Cretaceous inclination-corrected sediments in Tarim . However, Cogné et al. (2013) (see also Cogné et al., 1999;Hankard et al., 2007) suggested that the APWP for stable Europe may not be appropriate to interpret Asian paleomagnetic data because of continental deformation that has occurred since the Oligocene between Europe and Siberia (e.g., Puchkov, 1997). Compared to the APWP for the East Asian plate proposed by Cogné et al. (2013), the site mean directions from this study yield a paleolatitude 11.5-12°(i.e., 1,280-1,330 km) lower than the prediction (31.7-32.2°N), whereas the site mean directions from the extended Tarim data set yield a paleolatitude 6.0-6.5°(i.e., 670-720 km) lower than the prediction (31.7-32.2°N).
Under the assumption that the APWP by Cogné et al. (2013) is appropriate to interpret East Asian data, the offset between the Tarim and reference paleolatitudes can be used to quantify the amount of continental shortening that occurred within the East Asian plate north of the Tarim Basin during the India-Asia convergence. This amount is poorly constrained from observations (e.g., van Hinsbergen et al., 2011). For the Tian Shan range, at least 20-40 km of crustal shortening was documented in the Southern piedmont from geological mapping and cross-section construction (Yin et al., 1998), whereas an estimate of 203±50 km was proposed in the Northern piedmont assuming isostatic equilibrium (Avouac et al., 1993). For the Mongolian Altai range, a ∼50 km shortening was documented (Cunningham, 2005). This leads to at least 223-343 km of crustal shortening north of the Tarim Basin, to be compared with the 670-720 km latitudinal offset for the extended Tarim data set. Paleomagnetic data thus indicate that a larger amount of crustal shortening may have occurred within the East Asian plate during the India-Asia convergence and/or that the geomagnetic field was locally anomalous. Lhuillier et al. (2016) documented a bimodal distribution of the inclinations for the cospatial Albian-Aptian volcanics at Tuoyun (sections K and L), with a boundary between the two populations observed at ∼40°. Their interpreted directions during the Maastrichtian-Paleocene (section P) also supported, though in a less striking way, a bimodal distribution of the inclinations with the same boundary at ∼40°. Such a bimodality was interpreted in terms of a two-state dynamo, namely a bistable dynamo that can spontaneously fluctuate between two distinct stable regimes (e.g., Simitev & Busse, 2009, 2012. As illustrated in Figure 9e, our new data set consisting of 93 mean directions does not show such a splitting of the distribution of the inclinations at ∼40°, suggesting that the bimodality observed in Lhuillier et al. (2016) was possibly due to an insufficient sampling of the distribution.

Geomagnetic Interpretation: A Two-State Dynamo
However, the distribution of the inclinations at Tuoyun are slightly left-skewed, which is not the case or less marked for the NAIP and the Deccan Traps (Figure 9). In particular, it appears that the low inclination sites (I ≤ 30°) mostly occur in sequence (e.g., P35-P52, B05-B12, C09-C21) and could in this sense be interpreted as excursional events that were not detected by the Vandamme (1994) cutoff procedure. Low inclination values are recurrent in the Tuoyun basin (Gilder et al., 2003;Huang et al., 2005) but not visible in the more restricted volcanic sequences of the Kyrgyz Tian Shan (Bazhenov & Mikolaichuk, 2002). The slightly excessive presence of low inclination values at Tuoyun, in contradiction with the GAD geometry of the Earth's magnetic field observed at the same period in the Deccan traps and NAIP, may thus be the signature of a local anomaly of the Earth's magnetic field in the Tuoyun basin, consistent with the regional anomaly documented in Central Asia during the Cenozoic (e.g., Chauvin et al., 1996;Si & Van der Voo, 2001).

Time Constraints
Compared to the APWP for the East-Asian plate (Cogné et al., 2013), the site mean directions from this study (respectively the extended Tarim data set) yield a paleolatitude 11.5-12°(respectively 6.0-6.5°) lower than the prediction for the 60-70 Ma interval. In contrast, the 112-115 Ma volcanics from Tuoyun (I ≈ 47°N, Lhuillier et al., 2016) yield a paleolatitude 6.5-7.0°lower than the prediction. Both Albian-Aptian and Maastrichtian-Paleocene volcanics should equally detect the effects of Cenozoic crustal shortening that occurred north of the Tarim Basin during the India-Asia collision (e.g., Ji et al., 2008). If the Tarim data set is considered, the similar amount of latitudinal offset deduced from 112-115 and 60-70 Ma volcanics does not allow us to separate the effects of crustal shortening and geomagnetic field anomaly. If the the site mean directions from this study are considered, the increase of the latitudinal offset from 6.5-7.0°for 112-115 Ma volcanics to 11.5-12.0°for 60-70 Ma volcanics points to a strengthening of the geomagnetic field anomaly. However, we would like to stress that such an interpretation must be treated with caution because it critically relies on the choice and exactness of the APWP.
As a conclusion, the low inclination values observed from volcanic rocks at Tuoyun cannot be explained by structural or rock-magnetic causes. According to present geological observations, the low inclination values can be partly explained by crustal shortening within the East Asian plate during the India-Asia convergence but also needs to invoke the existence of a local anomaly of the Earth's magnetic field in Central Asia during the Maastrichtian-Paleocene.

Paleointensity Behavior Since the Cretaceous
Whereas the two previous sections were devoted to the interpretation of our directional results in terms of crustal shortening and/or two-state dynamo, we now focus on our paleointensity results and comment on the strength and variability of the geomagnetic field's intensity since the Cretaceous.

Amplitude of the Virtual Dipole Moments
We first compared our API determinations with the "world paleointensity database" (WPD) maintained by the Borok Geophysical Observatory (e.g., Shcherbakov & Sycheva, 2006Shcherbakov et al., 2002) and which includes all known published results of reconstructed dipole strength (in terms of VDM or VADM) obtained from eruptive or baked rocks. We focused on the past 150 Myr and selected the data according to the STAT criterion of Biggin and Paterson (2014), namely, at least five individual sample estimates per unit with a relative dispersion lower than 25%. As illustrated in Figure 10, our VDM average estimates for the baked sediments around 60 Ma and for the basalts around 70 Ma lie within the ±1σ envelope produced by a moving average over a 10 Myr window. Were the Tuoyun volcanics emplaced at a paleolatitude of 31.5°N as predicted by Cogné et al. (2013), the associated VADM values (52.7 ± 3.3 ZAm 2 for P17+P06, 51.4 ± 14.4 ZAm 2 for B08+A22) would be 10-20% lower than the VDM values, though still included within the ±1σ envelope of the moving average. With a conservative data selection solely based on the STAT criterion, no robust trend can be identified from the Cretaceous to present.
We then compared our API determinations with the subsets from Kulakov et al. (2019), which were analyzed using a suite of 10 paleointensity quality criteria (Q PI ). Subsets I-III yield for the late Cretaceous interval (65-84 Ma) a median dipole strength around 34 ZAm 2 , that is, twice lower than our mean dipole strength of 58.9 ± 6.9 ZAm 2 derived from Tuoyun volcanics and baked sediments for the time interval 60-70 Ma. However, subsets I-III are based on a lenient selection process that does not require the fulfilment of the STAT criterion. Only three determinations (taken from Goguitchaichvili et al., 2004;Juárez et al., 1998;Shcherbakova et al., 2007) indeed satisfy the STAT criterion and lead to the mean dipole strength 57.3 ZAm 2 , consistent with our new estimates.
As a concluding remark about API determinations, we note that caution must be exercised when analyzing the results in term of reconstructed dipole strength. First, such estimates can only be rigorously compared at a global scale if the field has a perfect GAD configuration, which may not be the case for the investigated time interval. Second, the current state of the global paleointensity database for the late Cretaceous interval 10.1029/2019JB018631 Journal of Geophysical Research: Solid Earth does not allow one to apply a stricter selection process than the fulfilment of the STAT criterion, as initially proposed by Perrin and Shcherbakov (1997).

Relative Variability of the Intensities
In addition to the dispersion of the directions, the relative variability ε F of the intensities can be used as a PSV proxy to provide insight into the geodynamo's activity. From numerical dynamo simulations, ε F was shown to be almost independent of latitude and to correlate with the reversal frequency f rev (Lhuillier & Gilder, 2013). This parameter is practically difficult to estimate from API determinations due to the low success rate of Thellier-style experiments (Thellier & Thellier, 1959) but can in some cases-when the magnetomineralogy of the investigated sequence is homogeneous enough-be more successfully estimated from RPI determinations using the pseudo-Thellier method (Tauxe et al., 1995).
Two such estimates were previously obtained on volcanic rocks. The first one (ε F ¼ 44 ± 5%, N ¼ 47) derived from eruptive units from the Boring Volcanic Field (Oregon, USA) corresponds to the 0-3 Ma interval when polarity reversals occurred in average 5 times per Myr (Lhuillier et al., 2017). The second one (ε F ¼ 31 ± 4%, N ¼ 28) derived from the Tuoyun volcanics at 112-115 Ma characterizes the behavior of the Earth's magnetic field during the CNS (Lhuillier et al., 2016). Figure 10 shows that these two estimates are, to within the uncertainties determined by the bootstrap method, marginally consistent with the empirical law between ε F and f rev predicted by Lhuillier and Gilder (2013) from numerical dynamo simulations, although the linear dependency could be in reality a little steeper.
The new estimate (ε F ¼ 31 ± 5%, N ¼ 15) for the Tuoyun volcanics at 60-70 Ma is indistinguishable from that for the Tuoyun volcanics at 112-115 Ma. Although caution must be exercised due to the low number of flows used for the determination, this result tends to confirm a distinct PSV behavior between periods with a present-day reversal rate and a low reversal rate. An open question concerns the exact dependency between ε F and f rev , possibly different from the one predicted by numerical dynamo simulations. One can wonder whether the dependency should be necessarily continuous or could undergo abrupt changes. To answer this question, more estimates of ε F are clearly needed during the past 150 Myr when reversal frequency is well documented (Lowrie & Kent, 2004).

Conclusions
We investigated the paleomagnetic directions and intensities of Maastrichtian-Paleocene basalts emplaced near Tuoyun village (75.33°E; 40.18°N) in Central Asia.
(i) The observed mean paleoinclination I S ¼ 36.2°is indicative of an emplacement of the volcanics at a paleolatitude λ ≈ 20°N, consistent with the high dispersion k ¼ 14.6 of the directions. (ii) The deduced paleolatitude λ ≈ 20°N largely differs from the predictions λ ¼ 31-32°N according to the APWP for East Asia and λ ¼ 34-36°N according to the APWP for stable Europe. (iii) Such a latitudinal offset can be partly explained by crustal shortening within the East Asian plate during the India-Asia collision. (iv) Such a latitudinal offset also needs to invoke an anomaly of the geomagnetic field, at a local rather than global scale because paleomagnetic results from Deccan traps and NAIP document a GAD geometry at the same period. (v) The relative variability in intensity-supposed to be a proxy for the geodynamo's activity-estimated here at 31 ± 5% from pseudo-Thellier experiments may corroborate a correlation between geomagnetic reversal frequency and PSV rate during the past 120 Myr.

Data Availability Statement
The raw data (in MagIC format) are available at this site (https://doi.org/10.5282/ubm/data.194).