Two thousand years of archaeointensity from West Africa

Two thousand years of archeointensity from West Africa Ritayan Mitra a Lisa Tauxe a Susan Keech Mcintosh b a Scripps Institution of Oceanography, La Jolla CA 92093-0220 b Rice University, Houston TX Abstract This study presents 17 archeointensity estimates from Senegal and Mali, two neighboring countries in West Africa, for the period 1000 BCE to 1000 CE. The archeological artifacts used in this study were collected during the course of two separate projects, together spanning 22 years and across 8 separate excavations. A primary objective of this study was to get accurate dates, hence, only samples with independent age constraints from pottery style, detailed stratigraphy and 14 C dates were used. A total of 236 specimens from 63 samples were subjected to a double heating paleointensity experiment (IZZI method) from which 95 specimens were selected using a set of very strict selection criteria. The paleointensity results were corrected for differential cooling rate effects and remanence anisotropy. Additionally, we demonstrate the equivalence of using tensors derived from anhysteretic and thermal remanences for correcting remanent anisotropy of the specimens and use the former for the anisotropy correction. Our data show good agreement with the most recent paleosecular variation model but are lower than the pre-existing data, which are mostly from Egypt and Morocco. The presence of substantial non-axialdipolar contributions in the region is evident when virtual axial dipole moments (VADMs) from the published literature are calculated for 20◦ latitudinal bands and compared with our data - the average VADM values show a distinct latitudinal gradient. A prominent feature of this dataset is an intensity high observed prior to 700 CE in both Senegal and Mali. Comparing this peak with existing records from regions further to the north suggests a small but significant temporal offset and is interpreted to be additional evidence for a geomagnetic field with a significant and rapidly changing non-axial-dipolar contribution. Key words: Preprint submitted to epsl 8 December 2012 1 5 10 15 20 25 Introduction Archeomagnetism provides a unique window onto the changes in the Earth’s magnetic field over the last few millennia. Compared to lavas, archeological artifacts are numerous, especially in more recent times. They can be used to construct regional reference curves, which can then, in principle, be used for the dating of other archeological artifacts from the region. The magnetic elements obtained from these items are also inputs for global field models such as the CALSxK generation of models (Korte and Constable, 2003, 2005, 2011), which often form the basis for answering fundamental geodynamo questions (e.g., Amit et al., 2011). In spite of its potential, obtaining the full vector information from an archeological, or for that matter any kind of sample, is not a trivial task. While obtaining the direction of the ancient field is relatively simple, intensity measurements are more complicated because of the sensitivity of the experimental protocol to mineralogical alterations and domain states of the remanence carrier. Furthermore, there is no consensus yet on the experimental protocol and data reduction criteria and workers often use very different methods (Biggin, 2010). These could add to the scatter observed in an already impoverished dataset. Experimental difficulties and lack of access to suitable materials make the global coverage of data extremely heterogeneous with respect to space as well as time. As a result, the paleointensity data for the last 10 millenia are skewed, with the last two millennia contributing more than 50% of data. Moreover, the data are heavily biased to the northern hemisphere and even there, there is a high concentration of data from Europe (see Fig. S1 of the Supplementary Material). 35 In an effort to contribute to the critical need for high quality data from the under-represented parts of the globe, we turn to the archeological findings from two areas of West Africa with rich Iron Age sites, the Middle Senegal Valley in Senegal and the Niger Valley in Mali (Fig. 1). West Africa, a geographically extensive region, comparable in size with the contiguous United States has sparse coverage in terms of archeomagnetic data (Kovacheva, 1984; Gomez-Paccard et al., 2012). The archeological samples in the present study are obtained from two groups of sites separated by 1200 km. These archeological artifacts were collected during two multi-phase, decade-long archeological projects by one of the authors. 40 In this study, we reconstruct the evolution of the geomagnetic field intensity in the region, between 1000 BCE and 1000 CE. The location of the sites lies to the south of a pronounced geomagnetic flux patch over Europe during the time that might have influenced the samples in our sites (Korte and Constable, 2011). The passage of such a flux patch under a region would result 30 2 45 in large non-axial-dipolar variations in the recorded paleointensity. Recently it has been proposed that the ‘archeomagnetic jerks’ of Gallet et al. (2003), or the rapid changes in the field strength coupled with long term directional changes in the field, are caused by the waxing and waning of geomagnetic flux patches in a region (Dumberry and Finlay, 2007; Amit et al., 2011). In this study, we find evidence for large non-axial-dipolar intensity variations, which may or may not be associated with the long term directional changes of the geomagnetic field and are more likely to be the outcome of rapid paleosecular variation (PSV) in a region influenced by geomagnetic flux patches. 50 2 55 Archeological Background The artifacts used in this study are sherds of domestic, locally-manufactured pottery that were collected over a course of two separate multi-year projects designed to gain insights into the development of iron-using societies along the Middle Niger Valley, Mali and the Middle Senegal Valley, Senegal (Fig. 1). In the rest of this section, we discuss briefly the archeological context of these two neighboring localities. 2.1 The Middle Senegal Valley samples 60 65 70 75 Between 1990 and 1999, five seasons of NSF-funded archaeological research were conducted at a variety of sites in the central region of the Senegal Valley, where the historical polities of Takrur and Sila – first mentioned in an eleventh century Arab chronicle – were thought to be located. The research aimed to investigate the origins and development of these early Sudanic polities. Prior archaeological work in the Senegal Valley was sparse and of variable quality, so chronology-building was a primary goal of the research. The numerous settlement mounds, up to six meters in height, along the river between Cubalel and Walaldé proved to be well-suited for this purpose, representing in each excavated instance between three to ten centuries of domestic deposits that accumulated between 2600-1000 BP(Deme and McIntosh, 2006; McIntosh et al., 2012). The pottery samples and radiocarbon dates used in this article come from five of these mounds, viz., Walaldé, Siwré, Cubalel C-1,C-3 and C-6, and one non-mound site, Sincu Bara, located 80 kilometers downriver (Table 1). During excavation, individual deposition contexts were identified and excavated separately. This maximized the likelihood that cultural materials and charred organics recovered from within each distinct context were deposited contemporaneously. All recovered materials from each discrete context were labeled to preserve the data on association and context. 3 80 85 90 95 100 For each excavation unit, stratigraphy is an important source of information on chronology, allowing reconstruction of the sequence of deposition for the different contexts. These events may range from very brief (such as the scatter of ashes from a small hearth) to lengthy (e.g., the slow decay and wall melt of an abandoned wattle-and-daub house). The stratigraphic sequence was complemented by the establishment of a ceramic sequence that recognized time-sensitive changes in the locally-produced pottery assemblage. Petrographic analysis confirmed that the pottery was produced using local clay sources. Detailed analysis of ceramics from all excavated contexts resulted in identification of characteristic changes in pottery style through time (Fig. 2). This permitted the recognition of time-successive pottery phases within the mound deposits. Separate Phase sequences were defined for Walaldé (Deme and McIntosh, 2006), the Cubalel/Siwré site group (McIntosh, 2012) and Sincu Bara (McIntosh and Bocoum, 2000). As a third, independent source of chronological information, radiocarbon dating of charcoal samples from reliable (uncontaminated) contexts in all the units verified both the stratigraphic and ceramic phase interpretations. Charcoals from hearths were abundant at these sites; only 1 of every 6 samples collected was submitted for radiocarbon analysis. This permitted great selectivity and let us focus on charcoal from the most secure contexts. All charred organic samples were collected during excavation at the time the pottery samples were collected. They were collected with tweezers and placed in aluminum foil packets that were stored for potential radiocarbon dating. Seven of the stored samples, from levels that produced pottery samples that yielded successful paleointensity estimates in this study, were dated recently as a part of this study and agreed well with the rest of the dates, which were obtained fifteen or more years ago, contemporaneous with the excavations. 2.2 The Middle Niger Valley samples 105 110 115 Most of the Mali samples come from two adjacent settlement mounds, JennéJeno and Hambarketolo, that were built up largely contemporaneously in the first and early second millennia CE. Together they cover an area of over 110 acres. The samples come from four units excavated in 1977 (M1, see McIntosh and McIntosh, 1980), 1981( LX-N, HAMB, see McIntosh, 1995) and 1997 (DT, unpublished) (Table 1). While several other units have additionally been excavated at these mounds (see Supplemental Material), reinforcing the ceramic phase chronology and the radiocarbon dating of the deposits, Table 1 includes only the units from which pottery samples were analyzed for archeomagnetism. At their highest points, both mounds revealed between 4.5 - 5.5 meters of accumulated occupation deposits. The methodology employed for 4 excavation and for the establishment of a ceramic phase chronology was identical to that described for the Senegal Valley sites. Detailed stratigraphy and chronological information for the sites discussed so far are available in the Supplemental Material (Fig. S2-6). 120 125 130 135 140 The remaining Mali samples were recovered from excavations in 1999 at two occupation mound clusters ∼100 km north-north-west of Jenné-jeno in a nowdesiccated alluvial region known as the Méma: Kolima SE and Akumbu. At the low mound of Kolima SE, two units, 1 and 3, were situated close together and yielded a similar pottery assemblage throughout the 1.5 meters of deposits. The material from both units can be considered as part of the same ceramic phase. This material has not yet been fully studied but the radiocarbon dates obtained in this study are in agreement with the stratigraphy. Akumbu comprises several occupation mounds up to 4 m in height. In 2000, Unit 1 was excavated at one of these mounds to a depth of 1.5 m, when the excavation was ended before reaching sterile soil. The pottery assemblage in the upper levels 1-9 belongs to the Middle ceramic phase identified from prior excavations at another mound in the Akumbu cluster and radiocarbon dated between the sixth and fourteenth centuries CE (Togola, 2008). The pottery below level 9 is identical to the Early Phase ceramics identified by Togola, which date to the early first millennium CE. The sherd samples used in the analysis come from levels 6 and 13, separated by over 50 centimeters of deposits. The new radiocarbon dates obtained during the course of this study, 271±58 CE for Level 13 and 899±56 CE for Level 6 were found to be in excellent agreement with the previously established chronology of Togola (2008). 5 Table 1 List of archeological site with dimensions of excavated units and pottery phases excavated at each unit. Site acronyms in brackets are the paleomagnetic sites corresponding to the archeological sites (see text). Archeological Sites Excavation Units (size) Depth Phases Walalde (WA) WAL 1 (4x3m) 4.5 m WAL I, II WAL 2 (3x2m) 4.5 m WAL I, II 1S (4.3m) 5.5 m CUB/S I C-1 C-1 (4x3m) 3.0 m CUB/S II,III C-3 C3-A (5x6m) 5.5 m CUB/S I,II,III C3-B (5x6m) 5.5 m CUB/S I,II,III C-6 C-6 (4x3m) 3.8 m CUB/S II,III,IV Sincu Bara(SB) SB1 (3x3m) 3.2 m SB 1, II, III1 Jenné -jeno(JJ) LXN (5x6m) 5.5 m JJ I, II, III, IV DT (2x3m) 2.0 m JJ I M1 (3x3m) 5.0 m JJ I, II, III, IV Hambarketolo (JJ) HAMB (2x2m) 4.5 m JJ II, III, IV Kolima SE (KS) UNIT 1 (2x3m) 1.5 m NA UNIT 3 (2x3m) 1.5 m NA UNIT 1 (3x3m) 1.5 m NA Siwré (SW) Cubalel (CB) Akumbu (AK) 1 SB Phase III had pottery identical to CUB/S Phase III. SB Phases I and II were similar but not identical to CUB/S I and II, and contemporaneous with them, according to radiocarbon dates. 6 Table 2 Table showing list of samples and specimens from the successful sites. Sample(Specimen) column shows number of samples (specimens) selected for site level estimates/number of samples(specimens) measured. Age calibrated according to the OxCal program (Ramsey et al., 2010). Latitude Longitude Samples Specimens Calibrated Age (CE) Intensity (µT) VADM (x1022 Am2 ) SB1 15.8◦ N 13.3◦ W 2/3 4/8 502±72 35.32±3.34 8.26±0.78 SB2 15.8◦ N 13.3◦ W 2/2 6/8 502-624 38.75±1.27 9.06±0.3 SB3 15.8◦ N 13.3◦ W 1/1 5/5 624±52 42.23±1.54 9.88±0.36 SB4 15.8◦ N 13.3◦ W 2/4 6/14 815±81 36.96±4.08 8.64±0.95 WA1 16.5◦ N 14.3◦ W 2/4 7/12 -628±92 40.66±2.04 9.43±0.47 CB1 16.0◦ N 13.6◦ W 2/2 5/6 124±73 38.89±3.08 9.07±0.72 CB2 16.0◦ N 13.6◦ W 2/4 4/7 221±70 36.17±1.61 8.43±0.37 CB5 16.0◦ N 13.6◦ W 1/3 4/8 485±75 36.40±2.70 8.49±0.57 CB6 16.0◦ N 13.6◦ W 1/2 4/8 221-410 34.93±1.23 8.15±2.86 JJ5 13.7◦ N 4.5◦ W 3/3 8/12 27±47 40.06±2.49 9.58±0.6 JJ6 13.7◦ N 4.5◦ W 1/2 3/8 125±48 41.88±0.38 10.0±0.09 JJ7 13.7◦ N 4.5◦ W 1/1 3/4 664±31 43.37±3.51 10.4±0.84 KS1 15.4◦ N 5.5◦ W 1/1 4/4 -662±72 40.82±1.11 9.59±0.26 KS2 15.4◦ N 5.5◦ W 1/1 4/4 -812±12 40.09±0.66 9.41±0.15 KS3 15.4◦ N 5.5◦ W 1/1 4/4 -865±25 42.71±2.25 10.03±0.53 AK1 15.3◦ N 5.5◦ W 1/2 4/8 899±56 39.62±1.01 9.32±0.24 AK2 15.3◦ N 5.5◦ W 1/2 4/8 271±58 35.87±1.45 8.44±0.34 7 Sites 2.3 Chronology and Nomenclature 145 150 155 160 165 170 175 As evident from the foregoing discussion, the chronology of the excavated sites has been iteratively constrained by the archeologists with independent evidence from pottery style, stratigraphy and 14 C dating. Within a single unit, stratigraphy and pottery style served to corroborate the available 14 C dates. Often in archeological excavations, we have events such as pit digging or other disturbances that can complicate the stratigraphic interpretation. Therefore, care was taken so that all the pottery and charcoal samples were obtained only from the most unambiguous strata, maximizing the likelihood of their contemporaneity. Across the units, pottery/occupation phase offered an independent check for the 14 C dates. Samples for which the pottery style, stratigraphy and 14 C dating did not agree were deemed unfit for the study and were rejected (See Fig. S3 and S5 in the Supplemental Material). Table 2 shows the list of all the successful sites. For two sites, viz., SB2 and CB6, the age was constrained using stratigraphic constraints and 14 C dates from the adjoining strata. The sense of the term paleomagnetic site as used in this study is different from its archeological equivalent, i.e., the archeological site. While an archeological site comprises a number of excavation units, a paleomagnetic site is defined as being made up of a number of samples from a particular horizon within an single excavation unit of an archeological site. For example, the paleomagnetic sites starting with CB (Table 2) refer to a distinct layer in one of the Cubalel mound site excavation units (Table 1). The detailed stratigraphic and chronological contexts of the paleomagnetic sites and their respective archeological sites are presented in the Supplementary Material. Samples, as used in this study, refer to individual sherds (eg., cb07), whereas specimens refer to the multiple pieces, ranging between three to six, sub-sampled from the sherds (eg., cb07a. See Table S1 in the Supplementary Material for the full list of specimens). 3 Rock Magnetism To identify the major magnetic mineral phases in the samples we carried out a three-axis IRM demagnetization experiment (Lowrie, 1990). Three DC pulse fields of 1.2 T, 0.4 T and 0.2 T were used. The samples were subsequently thermally demagnetized up to 600◦ C. Finally, hysteresis loops (peak field of 1T) were measured for these specimens. Together, these reveal varying contributions of different phases of magnetite and hematite. The single most important remanence contributor in most of the samples had a 8 180 185 190 195 200 205 210 low coercivity (<0.2 T) and demonstrated distributed unblocking to maximum temperatures of 580◦ C. This behavior is characteristic of titanomagnetite with varying Ti content as the remanence carrier (Fig. 3a). A typical hysteresis plot displays a narrow, saturated loop with hysteresis parameters (0.5 > Mr /Ms > 0.05 and 1.5< Hcr /Hc <5 falling well within the PSD domain in Day plots (Day et al., 1977) (Fig. 3e). In some samples, the presence of a harder component (>1.2T) with maximum unblocking above 600◦C is indicative of a somewhat higher proportion of hematite (Fig. 3b). The remanence contribution of this higher coercivity mineral in such samples is still secondary to the lower coercivity mineral. The hysteresis loop is constricted in the middle confirming the presence of two phases with different coercivities (Fig. 3b), presumably magnetite and hematite. A totally different mineralogical makeup was evident in some samples (Fig. 3c,d,g,h). The goose-necked (sensu Tauxe et al., 1996) hysteresis loops indicate nonsaturation at 1 T with a heavily constricted middle, which suggests an equal, or in some cases, greater contribution of higher coercivity minerals (Fig. 3g,h). The IRM demagnetization experiments supported these observations (Fig. 3c,d). The medium and high field components showed a marked unblocking between 200 and 250◦ C. This component carried the bulk of the remanence in these specimens. Goethite, despite a very high coercivity of > 5T (Roberts et al., 1995), has a maximum unblocking temperature of 120◦ C (Strangway et al., 1968) and is therefore not a suitable candidate. Even dehydration of goethite to form a more stable phase of hematite as suggested by Roberts et al. (1995) is unlikely because of the remarkable stability shown by other specimens from the same samples during the paleointensity experiments. We therefore believe that these minerals are similar to those found by McIntosh et al. (2011, 2007), wherein the authors demonstrated widespread occurrence of minerals with similar rock magnetic properties in archeological artifacts collected from 12 European countries. They called this mineral phase HCSLT, an acronym for High Coercivity, thermally Stable, Low unblocking Temperature phase. Ubiquitous in most archeological samples (Hartmann et al., 2010; McIntosh et al., 2011), these are yet to be identified unambiguously but possibilities such as hemoilmenite, ferri-cristobalite or a substituted hematite are currently under consideration (McIntosh et al., 2011). 9 4 Experimental Techniques 4.1 Paleointensity Experiments 215 220 225 230 235 240 245 250 A total of 236 specimens from 63 samples were prepared for the archeointensity experiments. From each pottery sample, 3-6 specimens were chosen for the paleointensity experiments. Care was taken to sand off the slip (possibly a low temperature addition) that is a common feature in potteries. Thick edges and rims of potteries showing an even red coloration signify that high temperatures were reached and were therefore sampled wherever possible. The specimens were uneven in shape and approximately 1 mm3 in dimension. All the specimens were wrapped in glass microfibres and completely immersed in potassium silicate, which dried over time. The archeointensity experiments were carried out using the in-field, zero-field, zero-field, in-field (IZZI) protocol of Tauxe and Staudigel (2004) in the magnetically shielded paleomagnetic laboratory of the Scripps Institution of Oceanography. The measurements were made in a 2G cryogenic magnetometer. Two water-cooled ovens were used for the experiments. The IZZI experiments consisted of double heating at steps from 100◦ C to 585◦ C (or 600◦C in some cases). We used coarser steps of 100◦ C up to 200◦C, 50◦ C steps up to 500◦C and smaller steps at higher temperatures. After every zero field heating during the zero-field and in-field pair, an additional in-field step was inserted, which involved heating to a temperature two steps lower than the current. This treatment demonstrated whether the specimen had altered during the intervening steps and served as an alteration (pTRM) check. Acceptable paleointensity estimates should have a linear Arai plot (Nagata et al., 1963) and a single natural remanent magnetization (NRM) component that decays to the origin. Such behavior is expected for single domain grains that have been heated once to very high temperatures. Larger grain sizes, alteration during laboratory experiments, multiple heating of household potteries during their period of use and experimental errors invariably lead to significant departures from the aforementioned criteria. The acceptability limits for this departure are still a matter of discussion and are critically dependent on the choice of a set of parameters that can quantitatively classify the quality of the data. Therefore, selection of data remains subjective because of the wide array of parameters to choose from (Tauxe, 2010) and also the broad range of acceptance thresholds for each of those parameters. In this study, we have used a fairly strict set of criteria (see Fig. S6 of the Supplementary Material). For a specimen to be selected, at least 60% of the remanence, as estimated by the fvds (fraction of the total remanence used for 10 255 260 265 270 275 280 the slope calculation as estimated by vector difference sum as formulated in Tauxe and Staudigel (2004)), spanning at least 6 double heating steps were used to define the stable component. The parameter β is the error of the best fit slope in an Arai plot normalized to the slope (Coe et al., 1978) and we constrain it to be less than 0.06. Furthermore, the deviation angle (DANG) (Tauxe and Staudigel, 2004), the deviation of the best fit line from the line joining the center of mass to the origin in the Zijderveld plot, and the maximum angle of deviation (MAD) (Kirschvink, 1980), which is a measure of the scatter of the points about the best- fit line, were both restricted to a maximum value of 5. To assess the alteration we used the difference ratio sum (DRATS), which is the sum of the difference between the original partial TRM (pTRM) at a given temperature step and the pTRM check, normalized by the length of the best-fit slope through the NRM/pTRM data (Tauxe and Staudigel, 2004). DRATS was constrained to less than 10 for the selected specimens. A successful sample with a mean of µ had to have at least two successful specimens with a within-sample scatter, measured by the percent standard deviation (100σ/µ), of less than 10%. For a site level average to be included in the study it should have at least one sample and a minimum of three specimens. Therefore, a site having only one successful sample with two successful specimen level estimates would be rejected, but a site with two samples with two specimen level estimates each or a site with one sample and three specimen level estimates would be included in the study. The choice of the bounds involves some amount of discretion on the part of the worker but such strict criteria help to ensure unambiguous interpretations such as those shown in Fig. 4 a-c. For example, in cb07a, if the temperature bounds between 200◦ C and 585◦ C were chosen, the sample would be excluded on the basis of low fvds , whereas bounds from 100-585◦C would also discard the specimen because of low β and high MAD. Looser criteria, say β ≤ 0.1 and, MAD ≤ 8 or fvds > 0.4 would have lead to an acceptable intensity estimate with either of the aforementioned selection criteria. However, the obtained value would hardly be unambiguous; 62.4 µT for the former and 49.2 µT for the latter. The low β used in this study ensured that Arai plots with pronounced concavity were rejected. Specimens similar to cb11b were discarded because of the high MAD and/or DANG values. 285 4.2 Anisotropy 290 Thermoremanent magnetization (TRM) anisotropy in pottery is largely believed to be the result of alignment of the remanence carrying fraction during the creation of the pot when the minerals get aligned at a tangent to the surface of the pottery. Bricks, which do not undergo any flattening processes, are largely found to be isotropic lending credence to this hypothesis. As a result of 11 295 300 305 310 315 320 325 330 partial alignment of the magnetic particles, the strength of the TRM depends on the direction of the applied field. One way to partially correct for this is to align the laboratory field along the NRM direction (e.g., Aitken et al., 1984). However, this is not foolproof because the NRM in an anisotropic material is unlikely to be exactly parallel to the ancient field. A better way would be to find the anisotropy tensor of each specimen and use it to correct the ancient field estimate obtained from a Thellier-type experiment. It is largely agreed that the anisotropy of magnetic susceptibility (AMS) is carried by grains that are larger than the remanence carrier and hence should not be used to correct for the anisotropy (Selkin et al., 2000). Anisotropy of anhysteretic remanent magnetization (AARM) and anisotropy of TRM (ATRM) have long been known to display very similar properties for single domain magnetites and are carried by the same grain fraction (Levi and Banerjee, 1976). While some authors eschew using AARM because of the widely held belief that AARM tensors are different from the ATRM tensors (Chauvin et al., 2000), others (e.g., Selkin et al., 2000; Ben-Yosef et al., 2008) have continued using AARM because of the purported similarity in the grain sizes of the remanence carrying population and the lack of additional alteration while heating at high temperatures. In order to resolve which of the two is more suitable for our sample set we decided to correct the data with tensors obtained from both the methods. The specimens were given an ARM (180 mT alternating field; 0.05 mT bias field) along 9 directions. A zero-field demagnetization step was inserted between each position and was used for baseline subtraction. Subsequently, we conducted partial TRM anisotropy experiments at 520◦ C in six positions with an initial zero step that was taken as the baseline. An additional step in the end was used to detect alteration. It is worth noting that all the specimens had at least 60% of the remanence fraction removed by 520 ◦ C. The specimen intensities corrected with the AARM and ATRM tensors showed a marked correlation (r=0.95) (Fig. 5a). Ideally, correction for anisotropy should be able to reduce the scatter at the sample level. Site level reduction in scatter should only be a natural consequence of this correction but is not inevitable because a single site may have potteries from considerably different periods of time. Therefore, in order to test which method is better, we plot the cumulative distribution of the sample level scatter (Fig. 5b). The withinsample scatter in the raw data (black squares) improved with both the ARM tensors (red solid circles) and the TRM tensor (blue open circles), although the AARM demonstrates a lower median of 4.03% as opposed to 4.98% for the ATRM. The two were otherwise indistinguishable as revealed by a two sample Kolmogorov-Smirnov test with the test statistic, D=0.17, being less than the critical value of D; Dcrit = 0.35 at 95% confidence level. We believe that the slightly (but insignificantly) higher scatter found in the ATRM corrected intensities could be a result of the blocking temperature dependent anisotropy as was originally suggested by Aitken et al. (1981). If so, and if ATRM tensors are used to correct paleointensity, we recommend doing the experiments close 12 335 to the maximum blocking temperature of the specimens. 4.3 Cooling-rate 340 345 350 355 360 365 370 Dodson and McClelland-Brown (1980) and Halgedahl et al. (1980) first pointed out that if during cooling, an assemblage of single domain grains is allowed to equilibrate at a particular temperature for a longer period of time, it will acquire a higher magnetization. In other words, a magnetic assemblage in a slowly cooled system can equilibriate with the field to a lower temperature before the magnetization gets blocked. In traditional paleointensity experiments, the sample is cooled in the lab for typically less than an hour. Archeological artifacts could have taken anywhere from a few hours to a few days to cool. Because of the different cooling rates in the laboratory and the original environment, the same field could have induced a higher remanent magnetization when the artifact cooled in its ancient environment than when it cooled in the laboratory. Therefore, in a paleointensity experiment, where the laboratory field is multiplied by the ratio of ancient magnetization to the laboratory magnetization to get an estimate of the ancient field, the ancient field strength would be overestimated. Theoretical calculations have shown that for single domain magnetite, an order of magnitude of cooling rate difference produces a 7% (Dodson and McClelland-Brown, 1980) to 10% (Halgedahl et al., 1980) overestimation. In other words, although for most ancient artifacts a cooling rate correction is mandatory, some uncertainty in the precise knowledge of ancient cooling time can be accommodated. This fact is especially useful because often the ancient cooling time depends on factors like the kind of oven (open hearth, pit etc.) and fuel (twigs, branches, leaves, coal etc.) and is often a matter of speculation. For example, in Fig. 6a, we see that even if an extreme misjudgment of cooling time was to occur and an ancient cooling time of two days was mistaken as six hours, the difference in the overestimate is unlikely to be more than 5% for grains with low blocking temperatures. For grains with higher blocking temperatures this difference is expected to be even lower. This is also borne out by experimental studies where multiple cooling rates were used to account for the uncertainty in the ancient cooling time (Hartmann et al., 2010). To find the possible overestimates in our samples we have designed a cooling rate experiment very similar to Chauvin et al. (2000) and Genevey and Gallet (2002). A sister sample was used from every potsherd. Three partial TRMs were measured at a high temperature. The temperature was high enough to cover the blocking temperatures of grains carrying at least 60 % of the remanence. In our experiments 525◦ C is an appropriate temperature for this purpose. The first experiment involved cooling for the normal laboratory time 13 375 380 of 30 minutes (TRMf ast1 ). The second involved cooling at a slower rate. For this purpose, we turned off the cooling fan which generated an exponential cooling from 520◦ C to 40◦ C in 10 hrs (TRMslow ). The third experiment was exactly like the first (TRMf ast2 ) and the difference between the TRMf ast1 and TRMf ast2 was used to detect alteration. The correction factor is given by the ratio of the average of TRMf ast1 and TRMf ast2 to TRMslow . The alteration cutoff was fixed at 5%. 5 385 390 395 400 405 Results The fairly strict paleointensity selection criteria we have used for selecting the specimens ensured that only the most robust estimates were retained. Approximately half of the selected specimens decayed straight to the origin with little to no viscous component (eg., Fig. 4d), whereas most of the rest (∼ 40%) had a sharp low temperature component which was removed by ∼ 200◦ C and the remanence decayed linearly to the origin thereafter (Fig. 4e). Only a few (∼ 10%) had some curvature in the vector plots, but not significant enough to warrant exclusion (Fig. 4f). Of the 236 specimens, 103 passed the selection criteria at the specimen level. Of these specimens, 98 passed the sample level criteria and 95 passed the site level criteria (Table 2 and Table S1 in the Supplementary Material). However, sites JJ2 and WA1 were discarded because of contradicting archeological evidence (see Supplementary Material). For this study, we have shown the effectiveness of using either AARM or ATRM tensors for correcting the anisotropy in the specimens. Therefore, we report here the specimen intensities that were corrected using the tensor obtained from the AARM correction. The specimens in our study were found to be moderately anisotropic with ARM anisotropy degree (τ1 /τ3 ) varying between 1.05 to 1.50, with 78% of the specimens yielding values less than 1.20. Anisotropy correction markedly reduced the site level scatter in the data as evident from the convergence of the specimen-level intensity values (examples shown in Fig. 5c). The cooling rate correction of the samples varied between 0.85 to 1.05 and displayed a median of 0.95 (Fig. 6b). A factor greater than unity could suggest the presence of interacting single domain or multi domain grains (Brown, 1984). 14 6 410 415 420 425 430 435 440 445 Discussion Africa is a large continent with limited archeointensity data. In the northern hemisphere it has a latitudinal extent of 37◦ from the equator and a longitudinal extent from 17◦ W to 51◦ E. In the last three millenia we have only 74 paleointensity estimates from the region (http://geomagia.ucsd.edu). Most of these estimates come from Egypt and Morocco and all of these estimates are from latitudes greater than 25◦ N latitude. Our data from West Africa are the first to be reported from a lower latitude (∼15◦ N). Coupled with other studies from Africa, mostly from Egypt and a few from Morocco, the data help us to reconstruct the geomagnetic field evolution in Africa between 1000 BCE and 1000 CE. The data from Aitken et al. (1984) show that the field increased from the present day value to 1.6 times that value during 1500 BCE to 1000 BCE (Fig. 7). Our data demonstrate that the field could have dropped after that, between 1000 BCE and 600 BCE. We had no successful samples between 500 and 0 BCE. The results from Hussain (1987) demonstrate that the field might have reached a value as high as twice that of the present day field at ∼ 500 BCE. The rapid decline of the field during the first half of the first millennium CE is quite evident in our results. An interesting feature of our data is the marked cusp observed at 300-500 CE, when the field was as low as the present field. Leonhardt et al. (2010) also found a decisive low during this period from Egyptian ceramics. Walton (1984, 1990) and Walton and Balhatchet (1988) found similar low values from Greece. Western European data also suggest a low around 250 CE (Chauvin et al., 2000). Similar low values are also observed in the Levantine curve of Ben-Yosef et al. (2008). The broad picture arising out of these disparate datasets is that of a pronounced low over Europe and the northern part of continental Africa between 300 and 500 CE. In the global context, we find no evidence for this (see Fig. 9 in (Genevey et al., 2008)), which suggests that the marked low around this period is likely to be a regional feature. The high scatter observed in the Egyptian data (black symbols in Fig. 7) and the paucity of data from West Africa (blue symbols) make it difficult to draw strong conclusions. The scatter observed in the Egyptian data has been previously discussed in Leonhardt et al. (2010). They attributed this to experimental techniques in general, which were ill-equipped to detect errors arising from differences in cooling rates, anisotropy, multi-domain grains and undetected alteration of the samples. The data from Morocco provides an interesting perspective - these plot higher than our data for all contemporaneous periods. This is unlikely to be a result of systematic and undetected experimental error because three different studies with very different experimental controls (Thellier, 1959; Kovacheva, 1984; Gomez-Paccard et al., 2012) report 15 450 455 460 465 470 475 480 485 490 similar results. Morocco is located 2000 km north of Mali and Senegal. Therefore, it is possible that there is a significant latitudinal gradient in the field intensities that cannot be explained by an axial-dipole. This line of reasoning would also suggest that the very high values of Hussain (1987) could well be a true geomagnetic signal because the sites were located east of Cairo (30◦N, 31◦ E) at a latitude similar to the sites in Morocco. Despite the above reasoning, the scatter observed in the site estimates could indeed be as speculated in Leonhardt et al. (2010). Our data generally agree well with the global CALS3K.4 model (Fig. 8) with a few exceptions (JJ6, SB3, and JJ7). The predicted range from 35 µT to 43 µT for the region is in close agreement with our data. It is important to note that although the sites from Mali and Senegal are separated by 1000 km, neither the CALS3K.4 predictions nor the data show significant differences in the two regions. The close agreement between the model and our data in general allows us to infer that during the period where we have no data, viz., 500 BCE to 0 CE, the field likely increased but not to the extent observed in Hussain (1987). This observation suggests that the field could have been different in Northern Egypt during this time. To estimate the magnitude of the difference in intensity of the field that cannot be explained by an axial dipole we make use of the VADM predictions from the CALS3K.4 model (Fig. 9a). Model results for the four different locations separated either longitudinally or latitudinally were used to estimate the median differences expected in the VADM estimates within Africa. The latitudinal gradient of the field as predicted by CALS3K.4 is ∼3.5% for every 10◦ , but the longitudinal gradient is much lower at ∼0.4% for a similar distance; more toward the east, this gradient increases to ∼2%. To evaluate how the raw data from Africa compare with those from Europe, we compiled average VADMs from different latitudinal bands (Fig. 9b). We have used the GEOMAGIA database (Korhonen et al., 2008; Donadini et al., 2009) and included only those data that were obtained with a double heating experimental protocol spanning the longitudinal band of 20◦ W to 60◦ E. An obvious problem with such a comparison is the extreme heterogeneity in the sampling of the respective latitudinal bands. While the region between 40◦ N and 60◦ N has 690 intensity estimates (blue squares), the region between 20◦ N and 40◦ N has only 293 estimates (yellow triangles). This also explains the considerably smaller standard errors in the former. The southern-most 20◦ band, from 0◦ to 20◦ N, happens to be the least studied and this study is the only source of data (red circles). In spite of this heterogeneity, some interesting observations can be made. Post 0 CE, we observe an increasing divergence in the VADMs from the different latitudinal bands. African intensities are consistently lower than those from the more northerly and largely European data. Such a systematic variation in the VADMs indicates a strong non-axial-dipolar contribution. Prior to 0 CE, we observe little evidence of such a difference, at least up to 500 BCE. Between 1000 BCE and 500 BCE, the VADMs from 16 495 500 505 510 515 520 525 530 0◦ N to 20◦ N plot considerably lower than the other two latitudinal bands. A possible reason for the very high VADMs of 20◦ N to 40◦ N (yellow triangles) could be the sampling bias introduced by the Ben-Yosef et al. (2008) dataset. Ben-Yosef et al. (2008) sampled a very high resolution section, viz., a slag mound. Multiple samples from the same slag mounds yielded 17 intensity estimates ranging from 11.02 to 25.08x1022 Am2 . When the same calculation was done with an average of these intensity estimates (14.83x1022 Am2 ), instead of all the 17 estimates, the VADMs for 900 and 800 BCE dropped lower to 11.34x1022 Am2 and 11.75x1022 Am2 , respectively. Therefore, even after accounting for the possible bias, we do not observe any significant difference in the VADMs from 40◦ N to 60◦ N (blue squares) and 20◦ to 40◦ N (yellow triangles), but both are considerably higher than the VADMs from 0◦ to 20◦ N (red circles). We can therefore conclude that with the currently available data, significant non-axial-dipolar contributions are fairly evident from 0 CE to 1000 CE and also possibly existed from 1000 BCE to 500 BCE as well. The evidence for significant non-axial-dipolar contributions in this region also comes from the radial field simulation at the core mantle boundary by the CALS3K.4 model. The simulations show the presence of a persistent but highly mobile flux patch in this region, between 1000 BCE and 1400 CE (http://earthref.org/ERDA/1478/). The rate of drift in these flux patches could be as high as ∼ 0.2◦ per year (Amit et al., 2011). Consequently, the field at the surface of the earth would also display significant departure from the axial-dipolar structure. For example, at 800 BCE and 400 CE, we observe that the field was considerably non-axial-dipolar in nature. Compared to that, the structure of the field at 0 CE was more axial-dipolar (Fig. 9, insets). A region influenced by such an oscillating flux patch can be argued to undergo rapid field changes on a spatial scale smaller than expected otherwise. A possible manifestation of such a drifting flux patch could be offset intensity highs from neighboring regions. For example, Fig. 8 indicates the epochs where Gallet et al. (2003) found rapid changes in field intensities, the so-called “archeomagnetic jerks”. In our data, we see a distinct offset of the peak defined in both Senegal and Mali by the highs SB3 and JJ7 as well as the flanking lows SB4 and AK1. One way of reconciling the two highs would be to ascribe the observed differences in the data to age uncertainties; either our samples suffer from undetected old wood effects, i.e., the dated charcoals originate from extraordinarily thick branches, or the archeological material used in Gallet et al. (2003) had to be from an archeological context older than that of its surroundings. Neither of these explanations seem plausible. Exactly the same age offsets from the sites SB3 and JJ7 separated by 1200 km (see Fig. 8) indicate that old wood effect is unlikely to be a serious impediment. Furthermore, the radiocarbon dates coupled with stratigraphy and detailed pottery phase associations of our samples make the dating more robust than would have been possible from using any of them independently. The pottery fragment 17 535 used in Genevey and Gallet (2002) was from a well-dated production site. Such associations by their very nature limit the possibility of the pot sherds being from an older context. We therefore argue that the observed offset is likely to be a geomagnetic feature. 7 540 545 550 555 560 565 570 Conclusion We have provided well constrained archeointensity results from West Africa. The field varies between 35 µT and 43 µT and defines a decisive low between 300 and 500 CE. The data provide the first set of paleointensity estimates from Africa between 0 and 20◦ N. We do not see a marked difference between the data from Senegal and Mali, with a longitudinal separation of 9◦ . The global CALS3K.4 model predicts a stronger gradient across the latitudes. This study provides experimental support for the model prediction - our data are significantly lower than those from Morocco (20◦ farther north). The CALS3K.4 model predicts a ∼3.5% difference for every 10◦ latitudinal separation. In the western part of Africa, the longitudinal differences in the intensities are minimal (also found in our study) and amounts to ∼ 0.4% for every 10◦ of longitudinal separation. Further eastward, this difference increases to ∼ 2% for the same longitudinal span. Significant non-axial-dipolar components are also evident when we compare the African and European data. The CALS3k.4 model, which is based on a global compilation of all the available data, shows the presence of a persistent and highly mobile flux patch at the core-mantle boundary between 1000 BCE and 1400 CE. We find further evidence of the flux patch in a pair of intensity highs observable in the dataset; a noticeable peak in intensity, observed in both Senegal and Mali, is found considerably offset from the ’archeomagnetic jerk’ at 800 CE (Gallet et al., 2003). We speculate that such asynchronous peaks should be more common in these regions and their discovery is critically dependent on the continuity of the archeomagnetic dataset and proper age control. Archeomagnetic data by their very nature are discrete and discontinuous. Uncertainty in the age is usually a few hundreds of years. Together, these factors can be very effective in hiding true offsets of intensity highs from neighboring regions. It is therefore desirable to pay very close attention to the age control in any archeomagnetic study. The substantial difference in VADMs observed in Africa and the offset intensity highs are a reminder of how restricted the reach of regional archeomagnetic curves for the purpose of dating could be. A corollary of that would be to not attribute offset intensity peaks/troughs in regional curves from adjacent regions to age uncertainties. The possibility of offset peaks/troughs being true geomagnetic features, controlled largely by the growth and decay of flux 18 575 patches, should also be investigated. Additionally this study also shows that the scatter observed in the compilations of archeomagnetic records from a region could be, in part, a telltale signature of true geomagnetic field variability. 8 We thank the NSF Arizona AMS laboratory for the fast processing of seven radiocarbon samples. The 14 C dates corroborated and supplemented the preexisiting archeological database. We thank Jason Steindorf for his help in the laboratory. RM thanks Ron Shaar for his thoughtful comments on many aspects this study. The authors thank Jeff Gee and Cathy Constable for their comments which greatly improved the quality of the manuscript. Part of this work has been funded by the NSF grants EAR 0944137,EAR 1013192 and EAR 1141840 to LT. 20°N Mali Timbucktu r ge Ni Walalde Siwre Cubalel Saint Louis Sincu Bara Kolima SE Akumbu 15°N Dakar Senegal Dia Jenne-Jeno Tambacounda Bamako Se ne ga l 580 Acknowledgement 10°N 5°N 0° 20°W 15°W 10°W 5°W 0° Fig. 1. Location of archeological sites in Senegal and Mali (triangles). Important cities and ports are shown with red dots. 19 Fig. 2. Assorted ceramic fragments demonstrating various surficial ornamentation, which determined the pottery style. sb04 a) -5 12 x10 8 10 Moment (Am 2 ) b) Hard (1.2T) Medium (0.4T) Soft (0.2T) sb03 x10 -5 7 6 8 10 4 4 3 0 200 400 600 0 25 20 15 4 10 2 200 o T ( C) e) 2.0 30 6 1 400 o T ( C) 5 0 600 200 400 o T ( C) 0 600 200 400 600 o T ( C) g) f) 3.0 Hard (1.2T) Medium (0.4T) Soft (0.2T) 35 8 2 2 ak01 d) x10 -5 40 Hard (1.2T) Medium (0.4T) Soft (0.2T) 12 5 6 jj08 c) -5 14x10 Hard (1.2T) Medium (0.4T) Soft (0.2T) h) 2.0 2.0 1.0 1.0 1.0 M/Ms 1.0 0.0 0.0 0.0 0.0 -1.0 -1.0 -1.0 -2.0 -2.0 -1.0 -0.5 0.0 B (T) 0.5 1.0 -3.0 -1.0 -1.0 -0.5 0.0 B (T) 0.5 1.0 -2.0 -1.0 -0.5 0.0 B (T) 0.5 1.0 -1.0 -0.5 0.0 B (T) 0.5 1.0 Fig. 3. Rock magnetic measurements on sister specimens from samples that yielded successful paleointensities. a-d)Three component IRM demagnetization experiments. e-h) Hysteresis loops of the same specimens at 1T. 20 ak03c cb07a a cb11b c 300 b 200 400 0.6 200 300 540 600 300 1.0 400 0.1 0.2 0.5 0.3 500 585 0.4 0.5 0.0 0.6 0.5 1.0 600 2.0 1.5 2.5 500 540 0.2 0.0 cb12a 585 0.4 0.6 0.8 pTRM gained wa17a 200 200 585 540 200 300 500 400 400 0.6 500 540 400 1.2 NRM remaining 1.2 1.2 300 500 1.0 0.8 400 0.6 500 200 300 0.8 400 400 0.6 540 585 0.4 0.6 0.8 1.0 pTRM gained 300 200 0.4 0.4 0.2 1.0 200 0.4 0.2 f 300 1.4 NRM remaining d NRM remaining 0.6 585 pTRM gained e 0.0 300 400 0.2 540 sb04a 0.8 0.8 500 pTRM gained 1.0 540 1.0 0.4 400 0.2 NRM remaining 500 0.8 0.4 500 1.5 NRM remaining NRM remaining 200 300 400 500 1.0 0.0 400 1.2 1.2 500 0.2 500 0.2 540 540 1.2 0.0 0.5 1.0 1.5 pTRM gained 2.0 0.0 0.2 0.4 0.6 0.8 Fig. 4. IZZI experiment results. a-c) Arai plots and Zijderveld diagrams of the specimens that failed to pass the selection criteria. d-f) Representative plots of the specimens that passed the selection criteria. The temperature interval used for isolating the characteristic remanence is marked with red line and squares. The insets are the vector components of the zero field steps with x in the abscissa and y and z in the ordinate. The circles are (x, y) pairs and the squares (the temperature steps are marked alongside) are (x, z) pairs. The directions are in the specimen coordinate system. The laboratory field was applied along the z-axis in the in-field steps. 21 1.0 pTRM gained 1.2 a) b) 60 c) 1 55 y = 2.7751 + 0.95032x R= 0.95499 55 0.8 50 45 40 35 CB2 CB05a CB05b CB06a CB06b 0.6 Field (μT) Cumulative fraction ATRM 50 0.4 45 KS02a KS02b KS02c KS02d 40 0.2 KS2 30 25 25 30 35 40 45 50 55 60 0 0 2 4 6 8 10 12 14 35 Scatter (%) AARM Fig. 5. a) Correlation between the ATRM and AARM corrected sample intensities. b) The cumulative distribution function of raw (black squares), ATRM corrected (blue open circles) and AARM corrected (red solid circles) intensities; the median values are 6.05, 4.98 and 4.03, respectively. The scatter is expressed as the percent relative standard deviation. c) Anisotropy correction using AARM tensors for two different sites. Open(solid) circles show uncorrected(corrected) specimen level intensity. Thin (thick) line and light (dark) grey area indicate the site level mean and the standard deviation of the uncorrected(corrected) specimens, respectively. b) a) 10 2 days 1.10 8 Overestimate Number of Samples 1 day 1.08 12 hrs 1.06 6 hrs 1.04 4 2 1.02 1.00 0.0 6 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 Log ( ancient/lab cooling time) 1.6 1.8 0.8 0.85 0.9 0.95 Correction Factor Fig. 6. a) Predicted overestimates for SD magnetites having five grain sizes with axial ratio=1.6. Each line represents a single grain size with laboratory blocking temperatures of 308◦ C(black), 436◦ C(blue), 486◦ C(green), 512◦ C(red) and 528◦ C(cyan). Black dots show representative time intervals. For example, a potsherd that was cooled in a furnace for 1 day is expected to overestimate the ancient field by 8%, if it unblocks at 436◦ C in the lab. The calculations are as in Halgedahl et al. [1980]. b) Histogram showing the correction factors for the specimens assuming a 10 hr cooling time. 22 1 1.05 1.1 20 18 14 VADM (x10 22 2 Am ) 16 12 10 8 6 -1500 -1000 -500 0 500 1000 Age (Years,CE) Fig. 7. Collection of data from continental Africa for the period 1500 BCE to 1000 CE. The data are from Morocco (blue symbols) and Egypt (black symbols). Symbols are for different studies: Aitken et al. (1984)(open triangle), Odah (1999)(closed diamond), Thellier (1959)(blue solid square), Odah et al. (1995)(open diamond), Kovacheva (1984) (blue solid diamond), Hussain (1983)(up solid triangle) , Hussain (1987)(down solid triangle), Leonhardt et al. (2010)(open square), Gomez-Paccard et al. (2012)(blue open circle). Horizontal line shows present day IGRF field at Mali converted to VADM. Our data are shown as solid red circles. The Geomagia database (Korhonen et al., 2008; Donadini et al., 2009) was queried and only experiments using a double-heating paleointensity technique were included. 23 Senegal Mali CALS3K.4 CALS3K.4 50 45 Intensity (μT) JJ7 KS3 JJ6 SB3 40 KS2 WA1 JJ5 KS1 CB1 SB4 35 AK2 -500 AK1 CB5 CB2 30 -1000 SB2 0 CB6 SB1 500 1000 Age (Years,CE) Fig. 8. Intensities from Senegal and Mali and the corresponding CALS3K.4 model predictions. Grey bars show the ’archeomagnetic jerk’ periods of Gallet et al. (2003). 24 14 12 15°N 5°W 15°N 15°E 15°N 35°E 35°N 5°W 11 μT 70 13 60 50 40 VADM (x1022 Am2) VADM (x1022 Am2) 12 10 9 400 CE 11 10 9 μT 75 65 25 8 8 7 55 μT 45 75 35 65 55 800 BCE 45 35 0 CE 7 -1000 -500 0 Age (Years,CE) 500 1000 6 -1200 -800 -400 0 400 800 1200 Age (Years,CE) Fig. 9. a) CALS3K.4 predictions for four different locations in Africa. b) The average VADM curves for 60◦ N to 40◦ N (blue squares), 40◦ N to 20◦ N (yellow triangles) and 20◦ N to 0 ◦ (red circles). The longitudinal extent of the data is between 20◦ W and 60◦ E. A 200 yrs moving window shifted by 100 yrs was used for the analysis. We used GEOMAGIA database (Korhonen et al., 2008; Donadini et al., 2009) and only paleointensity estimates obtained with a double-heating protocol were included in the analysis. The standard error is plotted on the y-axis and a fixed error of 100 yrs on the x-axis. The insets show the CALS3K.4 predictions at the surface for different periods in time. 70°N 50°N 30°N 10°N 10°S 26 30°S 50°S N <200 <400 <600 <800 <1000 70°S 585 160°W 120°W 80°W 40°W 0° 40°E 80°E 120°E 160°E Fig. S1: a) Spatial and temporal distribution of paleointensity data from the last 10 millennia as available in the Archeoint database (Genevey et al., 2008). The color-coded grid shows the absolute number of paleointensity estimates (N) available in each box and the histograms show its distribution with time. Each bin in the histogram represents 1000 years with left being older. The boxes that include less than 20 paleointensity estimates have no histograms. C3A C3B Phase 3b Levels 1-10 Phase 3b Levels 1-10 C-6 0m CB11 1121 +- 56 Phase 3b CB10 C-1 1S 2S Phase 3a Levels 1-19 mixed Phases 1b, 3 and 4 mixed Phases 4 and 4 Levels 1-19 Levels 1-9 3S SB4 Levels 1-12 1239 + - 45 1192 + - 60 Phase 5 1670 SB1 SB2 SB3 Phase 3b Level 1-14 Phase 3b Levels 1-6 Phases 1,2 and 3 1210 +- 70 1190 +- 50 1240 +- 50 Phase 3a Levels 7-8 +- 80 Levels 1-12,14 1.0 m Phase 3a Levels 11-26 Phase 3a Levels 11-20 1466 + - 49 1364 + - 61 + 1490 - 52 Phase 3a Levels 13-17 CB9 Phase 2 Levels 18-34 2.0 m Phase 2 Levels 27-38 Phase 2 Levels 21-31 1575 +- 60 1495 +- 55 CB8 SW1 1590 +- 50 Phase 1b Levels 12-20 Phase 1b 1610 + - 65 Deposits continue, excavation ended Phase 3a Levels 15-17 SB3 SB2 Phase 2 Levels 20-26 Phase 2 Levels 18-23 1367+- 76 SB1 27 1370 +- 95 1540 +- 45 1640 +- 70 Phase 1b Levels 20-32 3.0 m 1570 +- 60 CB4 1660 + - 40 4.0 m CB5 CB3 1570 + - 70 Some Phase 1b materials CB6 Phase 1A Levels 41-42 Phase 1A Levels 33-35 1750 5.0 m 590 CB2 1800 + - 50 CB1 1890 + - 60 5.6 m 1910 + 60 - 1840 + - 40 CB7 1410±60 +- 50 1550 +- 70 Phase 2 Level 9 Phase 1 Level 10 Phase 2 Level 13 1520 + - 65 Phase 1 Levels 15-17 Fig. S2: Stratigraphy of Cubalel, Siwré and Sincu Bara units. For simplicity, phases have been numbered without reference to their 595 respective sites; eg., Phase CUB/S II (in the main text) is simply written as Phase 2 in the respective units. Units 2S, 3S, SB2 and SB3 yielded only charcoal samples for the current study. Triangles show radiocarbon samples and the corresponding uncalibrated date (BP). Solid red (Open black) circles shows samples at their excavation depths that passed (failed) the selection criteria of the paleointensity experiments. The paleomagnetic site names are recorded beside the samples. 28 OxCal v4.1.7 Bronk Ramsey (2010); r:5 Atmospheric data from Reimer et al (2009); Phase 3A Phase 1A Unit C3A Unit C3A 1490±52 1910±60 1364±61 1890±60 1466±49 1800±50 Unit C1 Unit C3B 1575±60 1840±40 Unit SB1 Phase 1B 1410±60 Unit 1S Phase 3B 1750±50 Unit C3A 1660±40 1192±60 1590±50 1239±45 Unit 2S Unit C6 1610±65 1121±56 Unit 3S Unit SB1 29 1670±80 1240±50 Phase 2 1210±70 Unit C3A Unit C3B 1570±70 1190±50 1640±70 1540±45 1000 1370±95 Unit C6 1570±60 1367±76 Unit C1 1495±55 Unit SB1 1550±70 Unit SB3 1520±65 1000 500 BCE/CE 501 1001 1501 500 BCE/CE 501 1001 1501 600 Fig. S3: Calibrated radiocarbon dates from all the charcoal samples shown in Fig. S1. Uncalibrated ages of the charcoals are reported in the column and the corresponding output age distributions with 1- and 2-sigma brackets. Oldest dates appear at the top. Units (in blue) have dates ordered according to stratigraphic heights from which the charcoal samples were collected. The same pottery phase (in pink) could be found across multiple units hence many units can be part of the same phase. Since no information is available regarding the stratigraphic ordering of samples across the units, arrangement of units within a phase are arbitrary. For example, unit C3A has four dates from charcoals found associated with pottery showing distinct Phase 2 characteristics. Being in the same unit they could be stratigraphically ordered. But other Phase 2 potteries from units C1, SB1 and SB3 are listed under their respective units without any hierarchical significance. 30 WAL 1 WAL 2 0m 2201±39 Phase 2 Levels 1 -19 Phase 2 Levels 1 -5A 1m 2218±38 2394±43 2m 2364±43 2412±43 2497±43 Phase 1 Levels 5B -7 3m Phase 1 Levels 20 -25 WA2 2592±43 2543±43 4m 4.5 m 2470±53 2496±43 WA1 605 Fig. S3: Stratigraphy of Walalde units. WA2, which was successful in the PI experiments (see Table 8) had contradicting stratigraphic and radiocarbon information and hence was not included in our study. Symbols have the same meaning as in Fig. S1. 31 OxCal v4.1.7 Bronk Ramsey (2010); r:5 Atmospheric data from Reimer et al (2009); Boundary BOTTOM Phase 1 Unit W1 2496±43 2470±53 2592±43 2497±43 Unit W2 2543±43 Phase 2 Unit W1 2364±43 2218±38 2201±39 Unit W2 2412±43 2394±43 Boundary TOP 2500 2000 1500 1000 500 501 1001 BCE/CE 610 Fig. S4: Calibrated radiocarbon dates from all the charcoal samples shown in Fig. S3. Convention as in Fig. S2. 32 LX-N M1 ALS 550±110 Phase 4 (late) Levels 1-5,12,13 Phase 4 Levels 1-12 Phase 4 Levels 1-3 DT HAMB WFL KAN Phase 4 Levels 1--8 Phase 4 (late) Levels 1--2 Phase 4 Levels 1--6 0m 1970±41 JJ5 790±100 1310±110 1.0 m Phase 4 Levels 14-22,24 Phase 3 Levels 4-10 Phase 3 Levels 14-20 2.0 m 940±110 Phase 4 (early) Levels 25-28 mixed 3.0 m Phases 3,4 Levels 29-31 Phase 3 (late) Levels 35-38 970±110 1300±110 Phase 3 (early) Levels 39-42 4.0 m Levels 43-45 JJ6 1887±36 Phases 1,2 Level 46,47 1776±40 Phase 1,2 (Early) Levels 48-51 2060±110 1910±110 2090±110 2018±36 5.0 m JJ1 615 Phase 1 Levels 1-20 2036±36 1430±170 JJ2 Phase 4 (early) Levels 3--6 1220±110 Phase 3 (late) Levels 9-13 JJ3 1364±35 Phase 2 Levels 21-24 Phase 3 (early) Levels 14-19 Phase 3 Levels JJ7 7-17 1310±110 1640±150 1910±50 Phase 1 Levels 25-31 Phases 1,2 Levels 20-25 JJ4 1750±100 1800±120 Phases 1,2 Levels 11-13 Fig. S5: Stratigraphy of Jenné-Jeno and Hambarketolo units in Mali. Symbols have the same meaning as in Fig. S1. JJ2, which was successful in the PI experiments (see Table 8) had contradicting stratigraphic and radiocarbon information and hence was not included in our study. Units ALS, WFL and KAN yielded only charcoal samples that were used for the chronology. 33 1210±110 b) 25 a) 20 c) 25 DANG 10 5 0 0.01 0.03 15 10 5 0 0.05 1 2 3 4 20 15 10 5 0 5 e) 20 d) 20 1 2 5 15 No. of Samples No. of Specimens 10 4 5 Scatter (%) DRATS 15 3 f) 8 Fvds No. of Specimens MAD 20 No. of Specimens 15 No. of Specimens No. of Specimens β 10 5 6 4 2 0 0.6 620 0.7 0.8 0.9 1 0 2 4 6 8 10 0 2 4 6 8 10 Fig. S6: Selection criteria for the IZZI experiments. a-e) Histograms showing the distribution of the acceptance criteria for the selected specimens (see text for description) f) Histogram showing the sample level scatter as percent relative standard deviation. 34 OxCal v4.1.7 Bronk Ramsey (2010); r:5 Atmospheric data from Reimer et al (2009); Phase 3 Boundary BOTTOM Unit ALS Phase 1/2 1310_110 Unit DT Unit WFL R_Date 2036_36 1310_110 1970_41 Unit HAMB Unit LX-N 1364_35 2018_36 1220_110 2090_36 Unit LX-N 1910_110 1300_110 2060_110 970_110 1776_40 Unit M1 35 1887_36 1430_170 Unit M1 Phase 4 1910_50 Unit KAN 1640_150 1210_110 Unit ALS Unit LX-N 1800_120 940_110 Unit HAMB 790_100 1750_100 3000 550_100 2000 1000 1001 BCE/CE 2001 Boundary TOP 3000 2000 1000 1001 2001 BCE/CE 625 Fig. S7: Calibrated radiocarbon dates from all the charcoal samples shown in Fig. S5. Convention as in Fig. S2. Site Name Specimen Z Name SB1 sb02a Fvds Intensity (T) MAD N Q T(◦ C) 0.656 0.719 3.52E-05 1.6 8 15.3 200-520 0.06 2.4 3.1 0.724 0.733 3.36E-05 3.7 8 10.2 200-520 1.5 0.029 2.5 7 0.583 0.647 4.39E-05 4.4 8 16.5 250-540 0.009 1.2 1.7 0.744 0.757 3.94E-05 1.7 9 73.7 300-585 sb05a 1.1 0.037 4.4 0.1 0.901 0.872 4.46E-05 4 10 20.8 100-540 sb05b 1.3 0.025 2.3 2.5 0.897 0.868 4.29E-05 3 10 30.6 100-540 sb03c 36 SB3 Fcoe 2.1 sb03a 1.4 0.036 DANG DRATS 0.7 sb02c SB2 β 0 0 sb05c 0 0.042 1.5 0.5 0.911 0.862 4.22E-05 2.1 10 18.7 100-540 sb05d 0 0.029 0.4 9.3 0.93 0.909 4.32E-05 3.2 10 26.7 sb06a 0 0.023 0.7 6.1 0.681 0.682 3.47E-05 2.5 9 sb06c 0 0.059 1.4 7.8 0.75 0.744 4.25E-05 4.6 11 11.1 200-585 sb04a 0 0.012 0.2 1.3 0.856 0.859 4.62E-05 1.6 11 60.7 100-560 sb04b 0 0.018 0.2 2.4 0.709 0.707 4.28E-05 3.7 9 sb04c 0 0.032 0.5 1.3 0.894 0.875 4.64E-05 4 11 24.3 100-560 sb04d 0 0.016 0.6 6.5 0.886 0.893 4.34E-05 2.6 11 46.7 100-560 sb04e 0 0.022 1.3 5 0.926 0.888 4.29E-05 3.3 11 36.7 100-560 0-520 24.7 300-585 33.5 300-585 Site Name Specimen Z Name SB4 37 WA1 WA2 sb07a β 1.2 0.018 DANG DRATS Fcoe Fvds Intensity (T) MAD N Q T(◦ C) 13 48 0-585 1.3 2.6 0.942 0.963 4.43E-05 2.1 sb07b 0 0.025 1.8 3.8 0.957 0.902 3.83E-05 5 sb07c 0 0.014 1.7 1.3 0.841 0.812 4.18E-05 2.2 11 54 200-585 sb08c 0 0.016 3.5 0.4 0.953 0.934 3.28E-05 5 12 54 100-585 sb08w 0 0.053 2.9 4.9 0.887 0.863 3.35E-05 3.8 10 14.2 200-600 sb08x 0 0.045 4.9 2.7 0.842 0.81 2.99E-05 4.9 10 15.7 200-600 wa08a 0 0.019 1.2 2.2 0.929 0.838 0.000045 4.7 12 42.7 100-585 wa08b 0 0.049 1.9 9.4 0.92 0.826 0.0000428 4.9 10 16.6 100-540 wa08x 0 0.025 1.2 1 0.888 0.829 0.0000434 4.6 10 30.9 200-600 wa12a 0 0.016 1 4.6 0.643 0.706 0.0000463 1.7 9 wa12b 0 0.018 1.5 8.3 0.627 0.678 0.000046 2.1 9 29.7 200-560 wa12c 1.5 0.03 3.1 7.5 0.563 0.604 0.0000434 5 8 15.8 200-540 wa12d 1.2 0.036 1.8 7.5 0.59 0.631 0.0000422 2.1 8 13.8 200-540 wa02w 2.9 0.057 1.5 4.7 0.868 0.83 0.0000566 2.4 8 1 9.5 0.893 0.863 0.0000549 2.3 10 41.5 200-600 wa02x 0 0.018 12 34.8 100-585 34 13 200-560 200-540 Site Name Specimen Z Name DANG DRATS Fcoe Fvds Intensity (T) MAD N Q T(◦ C) wa16a 0 0.029 2.3 2.2 0.771 0.76 0.0000452 3.2 8 wa16b 0 0.024 0.6 9.4 0.903 0.916 0.0000496 0.9 10 32.3 200-600 wa16c 0 0.027 1.7 9.1 0.882 0.869 0.0000471 2.3 9 27.1 100-540 wa16d 0 0.023 0.7 9.2 0.775 0.743 0.0000448 3.2 8 28.3 200-540 wa18a 1.1 0.021 0.1 9.9 0.835 0.895 0.0000563 0.9 10 34.4 200-600 wa18b 1.5 0.017 0.6 0.5 0.827 0.879 0.0000658 1.8 10 42.6 200-600 22.4 200-540 38 wa18c 0 0.019 0.4 8.4 0.731 0.828 0.0000617 2.7 9 32.9 250-600 wa18d 0 0.024 1.2 9.8 0.706 0.816 0.0000668 2.5 9 24.3 250-600 1.5 0.045 2.6 0.6 0.748 0.758 0.0000452 2.9 9 14.2 250-600 1 3.2 0.661 0.672 0.0000409 4.8 8 13.7 300-600 0.86 0.62 0.0000436 4.7 11 41.3 200-585 wa19a wa19b CB1 β 0 0.039 cb01a 1.3 0.018 3.9 6 cb01b 2.3 0.023 0.3 2.7 0.876 0.883 0.0000468 1.1 11 33.5 200-585 cb01c 1.3 0.025 1.1 4.2 0.838 0.805 0.0000475 2.7 11 30.1 200-585 cb02a 0 0.02 1.4 5.1 0.774 0.715 0.0000445 3.4 9 cb02c 0 0.013 0.7 2.5 0.722 0.0000381 0.9 9 0.67 32.8 300-585 48 300-585 Site Name Specimen Z Name CB2 CB5 39 CB6 DANG DRATS Fcoe Fvds Intensity (T) MAD N Q T(◦ C) cb05a 0 0.029 2.4 0.1 0.727 0.705 0.0000379 4.9 9 cb05b 0 0.031 1.9 1.5 0.815 0.816 0.0000451 4.4 10 22.7 200-560 cb06a 0 0.018 1.7 1.7 0.82 0.739 0.000041 3.2 11 40.8 200-585 cb06b 0 0.021 1.1 3.8 0.955 0.956 0.0000391 2.2 12 41.7 100-585 cb09a 1.3 0.04 1.1 4.8 0.589 0.676 0.0000458 3.2 7 12.2 400-585 cb09b 1.4 0.024 2.1 4.3 0.676 0.745 0.0000392 4.5 8 24.1 350-585 21.8 200-540 cb09w 0 0.015 1.7 2.8 0.615 0.725 0.000042 2.7 6 31.3 400-600 cb09x 0 0.021 0.2 2.1 0.606 0.711 0.0000482 2.8 6 22.1 400-600 cb12a 0 0.012 1 0.6 0.726 0.739 0.0000387 1.2 10 50.8 250-585 cb12b 0 0.017 2.8 1.2 0.829 0.615 0.0000352 5 11 41 200-585 cb12w 0 0.01 1.7 1.7 0.781 0.806 0.0000348 2.5 8 65 250-560 1.5 0.017 1.3 3 0.745 0.824 0.0000337 3.2 8 36.5 250-560 cb12x JJ2 β jj03a 0 0.032 4.5 8.8 0.622 0.624 3.48E-05 4.7 9 16.5 200-560 jj03b 0 0.041 2.5 8.9 0.605 0.77 4.07E-05 2.9 9 12.9 200-560 jj03c 0 0.032 2 6.6 0.631 0.674 4.67E-05 2.9 8 16.7 200-540 Site Name Specimen Z Name jj03d JJ5 40 JJ6 JJ7 jj08a 0 β 0.028 2.4 0.057 DANG DRATS Fcoe Fvds Intensity (T) MAD N Q T(◦ C) 2 9.7 0.623 0.704 4.12E-05 3.9 9 18.9 200-560 4.8 8.9 0.928 0.945 4.38E-05 2.4 9 11.6 100-540 jj08c 0 0.019 4.1 9.3 0.812 0.867 4.42E-05 2.5 7 26.6 100-450 jj10c 0 0.04 4.7 2.2 0.52 0.625 3.99E-05 2.9 9 11.1 250-600 jj10d 0 0.046 1.9 2.9 0.522 0.633 3.91E-05 3.5 9 9.6 jj11a 0 0.035 0.9 7 0.766 0.765 4.84E-05 2.3 8 18.2 250-560 jj11b 0 0.017 1.2 2.6 0.933 0.948 4.21E-05 1.7 11 48.8 100-600 jj11c 0 0.04 2.1 0.1 0.868 0.869 4.07E-05 3.2 10 18.8 jj11d 0 0.024 3 0.8 0.732 0.721 4.52E-05 2.4 7 24.4 350-600 jj13b 0 0.039 3 5 0.748 0.655 4.45E-05 2.9 7 15.2 350-600 jj13c 0 0.031 1.2 9.3 0.739 0.638 4.32E-05 1.6 7 18.3 350-600 jj13d 0 0.037 1.4 8.7 0.782 0.656 4.54E-05 1.9 7 16.7 350-600 jj14a 0 0.022 0.7 6.3 0.912 0.932 5.21E-05 2.2 10 36.5 200-600 2 9.3 0.905 0.832 5.74E-05 3.5 10 19.1 1.6 1 0.748 0.672 5.29E-05 4.9 7 jj14b jj14c 1.8 0.041 0 0.05 250-600 0-540 0-540 12.2 200-500 Site Name Specimen Z Name KS1 KS2 41 KS3 AK1 β DANG DRATS Fcoe Fvds Intensity (T) MAD N Q T(◦ C) 0.668 0.67 4.75E-05 3.3 6 10.3 300-540 0.756 0.684 4.24E-05 4 7 17.1 250-540 ks01a 0 0.051 2.3 8.6 ks01b 0 0.036 1.6 4 ks01c 0 0.036 2.3 4.7 0.65 0.665 4.76E-05 4.4 7 ks01d 0 0.047 3.4 2.9 0.663 0.634 4.23E-05 4.1 7 1 4.9 0.846 0.902 4.49E-05 2.9 10 28.4 100-560 ks02a 1.4 0.026 15 250-540 11.7 300-560 ks02b 0 0.012 0.6 7.1 0.796 0.803 4.73E-05 2.4 9 56.2 200-560 ks02c 0 0.013 0.4 9.5 0.826 0.808 4.89E-05 3.2 8 50.7 250-560 ks02d 0 0.024 2.3 5.1 0.763 0.81 5.43E-05 2.4 9 28.1 200-560 ks03a 0 0.013 2.5 0.8 0.714 0.683 4.61E-05 3.9 8 43.4 200-540 ks03b 0 0.029 4.3 1.6 0.825 0.801 5.08E-05 4.4 10 24.2 ks03c 0 0.029 3 3.7 0.674 0.614 4.79E-05 3.7 6 ks03d 0 0.023 2.3 6.7 0.837 0.82 4.65E-05 3 2.6 0.057 0.5 1.9 0.641 0.601 4.07E-05 1.2 9 9.6 250-600 0.059 0.4 1.3 0.647 0.625 4.68E-05 2.3 9 9.4 250-600 1.1 0.014 0.5 3.8 0.696 0.676 4.23E-05 0.7 9 41.9 250-600 ak01a ak01b ak01c 3 0-540 17.8 300-540 10 30.5 0-540 Site Name Specimen Z Name ak01d AK2 β 1.7 0.017 DANG DRATS Fcoe Fvds Intensity (T) MAD N 1.2 0.9 0.673 0.637 4.34E-05 2.3 9 Q T(◦ C) 33.3 250-600 ak04a 1 0.022 1.5 3.1 0.889 0.936 3.52E-05 2.3 10 34.6 200-600 ak04c 0 0.022 1.2 3.6 0.891 0.95 4.14E-05 2.4 10 35.3 200-600 ak04d 0 0.015 1.8 2.8 0.874 0.905 4.54E-05 3 10 51.8 200-600 ak04b 0 0.011 0.7 6.4 0.873 0.897 3.62E-05 2 10 69.9 200-600 42 Table S1: Paleointensity parameters of specimens selected for site level estimates. Z , the zig-zag parameter of Tauxe (2010). β, Fcoe and Q as in Coe et al. (1978). Fvds , DANG and DRATS as in Tauxe and Staudigel (2004). MAD as in Kirschvink (1980). N is the number of double-heating steps used to constrain the paleofield estimate and T represents the corresponding temperature range. 630 635 640 645 650 655 References Aitken, M. J., Allsop, A. L., Bussell, G. D., Winter, M. B., 1984. Geomagnetic intensity in Egypt and western Asia during the second millennium BC. Nature 310 (5975), 305–306. Aitken, M. J., Allsop, A. L., Bussell, G. D., Winter, M. B., 1989. Geomagnetic intensity variation during the last 4000 years. Phys. Earth Planet. Inter. 56 (1-2), 49–58. Aitken, M. J., Bussell, G., Jones, R., Shaw, C. J., 1981. Geomagnetic field intensity variations in the eastern Mediterranean within the period 5000 to 2000 BP. Geophys. J. R. Astron. Soc. 65 (1), 267–267. Amit, H., Korte, M., Aubert, J., Constable, C., Hulot, G., 2011. The timedependence of intense archeomagnetic flux patches. J. Geophys. Res. 116. Ben-Yosef, E., Ron, H., Tauxe, L., Agnon, A., Genevey, A., Levy, T. E., Avner, U., Najjar, M., 2008. Application of copper slag in geomagnetic archaeointensity research. J. Geophys. Res. 113 (B8). Biggin, A. J., 2010. Are systematic differences between thermal and microwave Thellier-type palaeointensity estimates a consequence of multidomain bias in the thermal results? Phys. Earth Planet. Inter. 182 (3-4), 199–199. Brown, E. M., 1984. Experiments on TRM intensity dependence on cooling rate. Geophys. Res. Lett. 11 (3), 205–208. Chauvin, A., Garcia, Y., Lanos, P., Laubenheimer, F., 2000. Paleointensity of the geomagnetic field recovered on archaeomagnetic sites from France. Phys. Earth Planet. Inter. 120 (1-2), 111–136. Coe, R. S., Gromme, S., Mankinen, E. A., 1978. Geomagnetic paleointensities from radiocarbon-dated lava flows on Hawaii and question of Pacific nondipole low. J. Geophys. Res. 83 (Nb4), 1740–1756. 43 660 665 670 675 680 685 690 695 700 Day, R., Fuller, M., Schmidt, V. A., 1977. Hysteresis properties of titanomagnetites : Grain-size and compositional dependence. Phys. Earth Planet. Inter. 13 (4), 260–267. Deme, A., McIntosh, S., 2006. Excavations at Walaldé: New light on the settlement of the Middle Senegal valley by iron-using people. J. African Arch. 4 (2), 317–347. Dodson, M. H., McClelland-Brown, E., 1980. Magnetic blocking temperatures of single-domain grains during slow cooling. J. Geophys. Res. 85 (Nb5), 2625–2637. Donadini, F., Korte, M., Constable, C. G., 2009. Geomagnetic field for 03 ka: 1. New data sets for global modeling. Geochem. Geophys. Geosyst. 10 (Q06007), doi:10.1029/2008GC002297. Dumberry, M., Finlay, C. C., 2007. Eastward and westward drift of the Earth’s magnetic field for the last three millennia. Earth Planet. Sci. Lett. 254 (1-2), 146–157. Gallet, Y., Genevey, A., Courtillot, V., 2003. On the possible occurrence of ’archaeomagnetic jerks’ in the geomagnetic field over the past three millennia. Earth Planet. Sci. Lett. 214 (1-2), 237–242. Genevey, A., Gallet, Y., 2002. Intensity of the geomagnetic field in western Europe over the past 2000 years: New data from ancient French pottery. J. Geophys. Res. 107 (B11). Genevey, A., Gallet, Y., Constable, C. G., Korte, M., Hulot, G., 2008. Archeoint: An upgraded compilation of geomagnetic field intensity data for the past ten millennia and its application to the recovery of the past dipole moment. Geochem. Geophys. Geosyst. 9 (Q04038), doi:10.1029/2007GC001881. Gomez-Paccard, M., McIntosh, G., Chauvin, A., Beamud, E., Pavon-Carrasco, F. J., Thiriot, J., 2012. Archaeomagnetic and rock magnetic study of six kilns from North Africa (Tunisia and Morocco). Geophys. J. Int. 189, 169– 186. Halgedahl, S. L., Day, R., Fuller, M., 1980. The effect of cooling rate on the intensity of weak-field TRM in single-domain magnetite. J. Geophys. Res. 85 (Nb7), 3690–3698. Hartmann, G. A., Genevey, A., Gallet, Y., Trindade, R. I. F., Etchevarne, C., Le Goff, M., Afonso, M. C., 2010. Archeointensity in Northeast Brazil over the past five centuries. Earth Planet. Sci. Lett. 296 (3-4), 340–352. Hussain, A. G., 1983. Archaeomagnetic investigations in Egypt - Inclination and field intensity determinations. J. Geophys. 53 (3), 131–140. Hussain, A. G., 1987. The secular variation of the geomagnetic-field in Egypt in the last 5000 years. Pure Appl. Geophys. 125 (1), 67–90. Kirschvink, J. L., 1980. The least-squares line and plane and the analysis of paleomagnetic data. Geophys. J. R. Astron. Soc. 62 (3), 699–718. Korhonen, K., Donadini, F., Riisager, P., Pesonen, L. J., 2008. GEOMAGIA50: An archeointensity database with PHP and MySQL. Geochem. Geophys. Geosyst. 9 (Q04029), doi:10.1029/2007GC001893. 44 705 710 715 720 725 730 735 740 745 Korte, M., Constable, C., 2003. Continuous global geomagnetic field models for the past 3000 years. Phys. Earth Planet. Inter. 140 (1-3), 73–89. Korte, M., Constable, C., 2011. Improving geomagnetic field reconstructions for 0-3 ka. Phys. Earth Planet. Inter. 188 (3-4), 247–259. Korte, M., Constable, C. G., 2005. Continuous geomagnetic field models for the past 7 millennia: 2. CALS7K. Geochem. Geophys. Geosyst. 6 (Q02H15), doi:10.1029/2004GC000800. Kovacheva, M. A., 1984. Some archaeomagnetic conclusions from 3 archaeological localities in Northwest Africa. Dokladi Na Bolgarskata Akademiya na Naukite 37 (2), 171–174. Leonhardt, R., Saleh, A., Ferk, A., 2010. Archaeomagnetic field intensity during the Roman period at Siwa and Bahryn oasis, Egypt: Implications for the fidelity of Egyptian archaeomagnetic data. Archaeometry 52, 502–516. Levi, S., Banerjee, S. K., 1976. On the possibility of obtaining relative paleointensities from lake sediments. Earth Planet. Sci. Lett. 29 (1), 219–226. Lowrie, W., 1990. Identification of ferromagnetic minerals in a rock by coercivity and unblocking temperature properties. Geophys. Res. Lett. 17 (2), 159–162. McIntosh, G., Kovacheva, M., Catanzariti, G., Donadini, F., Lopez, M. L. O., 2011. High coercivity remanence in baked clay materials used in archeomagnetism. Geochem. Geophys. Geosyst. 12 (Q02003), doi:10.1029/2010GC003310. McIntosh, G., Kovacheva, M., Catanzariti, G., Osete, M. L., Casas, L., 2007. Widespread occurrence of a novel high coercivity, thermally stable, low unblocking temperature magnetic phase in heated archeological material. Geophys. Res. Lett. 34 (21). McIntosh, R., McIntosh, S., Bocoum, H., 2012. Seeking the origins of Takrur : Archaeological excavations and survey in the Middle Senegal valley. Yale University Publications in Anthropology, New Haven., (In press.). McIntosh, S., 1995. Excavations at Jenné-jeno, Hambarketolo, and Kaniana (Inland Niger Delta, Mali), the 1981 Season. University of California Publications in Anthropology, 20. University of California Press, Berkeley. McIntosh, S., 2012. Ceramics. In: McIntosh, R., McIntosh, S., Bocoum, H. (Eds.), Seeking the Origins of Takrur : Archaeological excavations and survey in the Middle Senegal valley. Yale University Publications in Anthropology, New Haven., (In press.). McIntosh, S., Bocoum, H., 2000. New perspectives in Sincu Bara, a first millennium site in the Senegal valley. Afr. Arch. Rev. 77 (1), 124–178. McIntosh, S., McIntosh, R., 1980. Prehistoric Investigations in the Region of Jenné, Mali. Vol. 2 of Camb. Mono. Afr. Arch. Oxford: B.A.R. Nagata, T., Arai, Y., Momose, K., 1963. Secular variation of the geomagnetic total force during the last 5,000 years. J. Geophys. Res. 68 (18), 52775281. Odah, H., 1999. Improvement of the secular variation curve of the geomagnetic field in Egypt during the last 6000 years. Earth Planets Space 51 (12), 1325– 1329. 45 750 755 760 765 770 775 780 Odah, H., Heider, F., Hussain, A. G., Hoffmann, V., Soffel, H., Elgamili, M., 1995. Paleointensity of the geomagnetic-field in Egypt from 4000bc to 150 AD using the Thellier method. J. Geomagn. Geoelectr. 47 (1), 41–58. Ramsey, C. B., Dee, M., Lee, S., Nakagawa, T., Staff, R. A., 2010. Developments in the calibration and modeling of radiocarbon dates. Radiocarbon 52 (3), 953–961. Roberts, A. P., Cui, Y. L., Verosub, K. L., 1995. Wasp-waisted hysteresis loops - Mineral magnetic characteristics and discrimination of components in mixed magnetic systems. J. Geophys. Res. 100 (B9), 17909–17924. Selkin, P. A., Gee, J. S., Tauxe, L., Meurer, W. P., Newell, A. J., 2000. The effect of remanence anisotropy on paleointensity estimates: A case study from the Archean Stillwater Complex. Earth Planet. Sci. Lett. 183 (3-4), 403–416. Strangway, D., Honea, R. M., Mcmahon, B. E., Larson, E. E., 1968. Magnetic properties of naturally occurring goethite. Geophys. J. R. Astron. Soc. 15 (4), 345–359. Tauxe, L., 2010. Essentials of paleomagnetism. University of California Press, Berkeley, pp 380-383. Tauxe, L., Mullender, T. A. T., Pick, T., 1996. Potbellies, wasp-waists, and superparamagnetism in magnetic hysteresis. J. Geophys. Res. 101 (B1), 571– 583. Tauxe, L., Staudigel, H., 2004. Strength of the geomagnetic field in the Cretaceous Normal Superchron: New data from submarine basaltic glass of the Troodos ophiolite. Geochem. Geophys. Geosyst. 5 (Q02H06), doi:10.1029/2003GC000635. Thellier, E. & Thellier, O., 1959. Sur l’intensité du champ magnétique terrestre dans le passé historique et géologique. Ann. Géophys. 15, 285376. Togola, T., 2008. Archaeological investigations of Iron Age sites in the Mema region, Mali (West Africa). BAR IS1736 Oxford: Archaeopress. Walton, D., 1984. Re-evaluation of Greek archaeomagnitudes. Nature 310 (5980), 740–743. Walton, D., 1990. The intensity of the geomagnetic-field in the eastern Mediterranean between 1600 BC and AD 400. J. Geomagn.Geoelectr. 42 (8), 929–936. Walton, D., Balhatchet, H., 1988. Application of a new technique to Greek archaeomagnitudes. J. Geomagn. Geoelectr. 40 (12), 1503–1510. 46