Studies of the recent history of Earth's magnetic field have revealed a rich spatial and temporal structure, but face limitations by a lack of Southern Hemisphere archeomagnetic data. Studies of Iron Age (200-1850 AD) peoples of southern Africa have revealed a potentially rich source of archeomagnetic information in the form of ceramics (specifically pottery). Additionally, contemporary pottery made with traditional techniques and materials can still be found. Reported here is the first step in addressing whether ancient pottery from southern Africa might faithfully record the geomagnetic field. We analyze contemporary pottery made with traditional techniques and methods. Rock magnetic measurements, including magnetic susceptibility as a function of temperature and magnetic hysteresis behavior, are discussed. Intensity results generated by three common paleointensity methods: Thellier- Coe double heating experiments, the multi-specimen method of Dekkers and Böhnel, and Shaw's method (with and without the corrections of Kono) are compared to the known field at the time of firing. The Thellier-Coe method reproduces the field (with an accuracy of 1.3 μT), the Shaw technique with the correction approach of Kono overestimates the field by 3.7%. The multispecimen method overestimates the field by 20%, however improvement upon this could be expected given recent improvements to the technique. These values bound the accuracies we can expect when applying the methods to ideal samples, representing a best-case for dealing with archeological ceramics from southern Africa
Studies of the recent history of Earth's magnetic field have revealed a rich spatial and temporal structure. Global datasets covering several millennia have been assembled by various authors in an attempt to learn about the short timescale behavior of the geomagnetic field. Jackson et al.  and later Jonkers et al.  collected navigational records from shipping logs to build a global database of field directions from the years 1590-1840 AD. Gubbins et al.  added absolute intensity data and concluded that the global dipole moment was relatively stable during this period and began decaying at its present rate near the onset of global monitoring in 1850. Later analyses, using larger datasets and going back further in time, suggest the dipole decay started as early as 1000 AD [Valet et al., 2008] or even 500 AD [Genevey et al., 2008]. A common deficiency in all of these models is a large hemispherical bias arising from the fact that an overwhelmingly large fraction of the data comes from the Northern Hemisphere.
Recent compilations of global archeomagnetic datasets [Donadini et al., 2009] have allowed field reconstructions of high spatial and temporal resolution going back several thousand years [Korte et al., 2009]. These models suffer from a similarly large hemispherical bias resulting from a relative lack of reliable intensity and directional data from the Southern Hemisphere. The availability of suitable archeomagnetic carriers in Eurasia and North America has allowed very detailed modeling of the Northern Hemisphere dating back almost ten thousand years, while spatial and temporal sampling in the Southern Hemisphere lags significantly. Apart from a few lake sediments near the equator, there are no data from the southern half of Africa included in the most recent global models of Korte et al. .
These limitations are brought into perspective when addressing features such as the South Atlantic Anomaly (SAA). This region of relatively low field intensity centered over the southern Atlantic Ocean has been attributed to a reversed flux patch near the coremantle boundary off the southern tip of South Africa [Hulot et al., 2002]. The anomaly is thought to be greatly contributing to the current decay of the dipole field. In many geodynamo models, such reversed patches can signal an oncoming field reversal [Glatzmaier et al., 1999]. The history of the anomaly is not well constrained before geomagnetic field monitoring started around 1850 AD. In the analysis of Gubbins et al. , the onset of dipole decay was related to growth of reversed flux at the core-mantle boundary in the southern hemisphere. They also noted that the presence of these features could be deduced from directional data alone. This, combined with the lack of evidence of such patches [Jackson et al., 2000, Jonkers et al., 2003], has led to the belief that the SAA is a relatively new feature. Paleointensity information from the Southern Hemisphere, particularly in Africa, would add powerful constraints to the nature and history of the SAA.
The 1700 year period prior to European colonization in southern Africa was known as the Iron Age. During this time several migrations brought new Bantu-speaking peoples south from areas in the Nigeria-Cameroon-Congo region [Huffman, 2007]. Each migration resulted in subtle changes in culture and traditions, and these changes are recorded in the stylistic evolution of the pottery made by these groups. These ceramics hold great potential for archeomagnetic analyses, and will significantly improve global data coverage in the Southern Hemisphere. Contemporary pottery made using traditional techniques and raw materials can still be found, and thus should be analogous to our archeological samples. Reported here is a first step in addressing whether ancient pottery from southern Africa might faithfully record the geomagnetic field. We analyze contemporary pottery made with traditional techniques and methods. The sampling of materials that acquired a TRM in a recent, known field as a first step in understanding the ancient record has a long history in paleointensity studies [e.g. Coe and Gromme, 1973; Cottrell and Tarduno, 1999; Genevey and Gallet, 2002; Spassov et al., 2010; Morales et al., 2011].
2. Sample Origin and Rock Magnetic Properties
During a field excursion in June of 2008, J.A.T. and M.K.W. obtained two pots, one baked and one unbaked, from a potter in the small village of Mukhondweni (S 23°15.242', E 30°6.519'), in northern South Africa. The pots were made by hand using wet clay as can be seen in figure 1a. The clay came from a nearby riverbed shown in figure 1b. It is most likely derived from weathered granitic gneisses spanning the entire Limpopo region, and contains a natural grit, which apparently acts as temper. No wheel was used and the pots were simply formed by hand. After the pots are formed, they are placed at the bottom of a shallow pit dug into the soil. The pit is then filled with wood and set on fire. The temperatures measured in pit fires often reach 800°C [Livingstone Smith, 2001], but there is a range (typically ~700-900°C) due to a large number of variables related to the local firing. The firing and cooling process for our sample pot took about twelve hours. The unbaked pot (seen in figure 1c) was obtained for comparison studies. The baked pot (figure 1d) was selected because it had been fired during the weeks prior to the date of purchase (June 19, 2008).
All magnetic measurements were performed in the paleomagnetic laboratory at the University of Rochester. Magnetic hysteresis measurements made with a Princeton Measurements Corporation MicroMag 2900 Alternating Gradient Force Magnetometer (AGFM) were performed on bulk samples to infer domain state distributions. Figure 2a shows a typical hysteresis loop from the fired pot. Some loops show wasp-waisted shapes, which are attributed to mixed domain states, or possibly the presence of hematite along with a softer magnetite phase. Figure 2b shows a Day plot summarizing magnetic hysteresis properties [Day et al., 1977] for the fired and unfired pots. In general, the samples showed pseudo-single domain behavior, with the tendency to trend above the magnetite mixing lines of Dunlop , likely due to the presence of impure magentite, superparamagnetic grains, and possibly hematite.
In addition to magnetic hysteresis measurements, measurements of susceptibility versus temperature were performed on bulk samples using an AGICO KLY-4CS Kappabridge. This instrument has a excitation frequency of 875 Hz and a field of 300 A/m was used. For these measurements, samples were crushed and then vacuum desiccated to remove moisture. Sample plots for measurements made in an argon atmosphere are shown in Figure 2c and 2d. The fired pot was fairly stable during heating in both argon and air. Both the heating and cooling curves show the same 565°C Curie temperature (determined by intersecting tangents). This result is consistent with a slightly oxidized or impure magnetite carrier. Additionally, there is no evidence for the Verwey transition, precluding the presence of pure magnetite. As expected, the unfired pot showed drastic changes between the heating and cooling curves regardless of atmosphere, suggesting major alteration during heating. These results suggest that the normal firing process stabilizes the magnetic carriers in the clay against alteration during subsequent heatings.
3.1. Thellier-Coe Analysis
Three different types of paleointensity experiments were conducted on the fired pot. Six specimens were treated with the Thellier [Thellier and Thellier, 1959] double heating method as modified by Coe . For these experiments, roughly 1 cm3 cubical specimens were cut from the pot using a bronze-bladed rock saw. Cooling water was applied as needed with a squeeze bottle to limit absorption and sample softening. Because the material is fairly porous, specimens were vacuum desiccated overnight in a magnetically shielded environment after cutting. During experiments, specimens were treated with paired field-off and field-on heatings. These started at 100°C and continued through 580°C in steps of 25°C (with the final step jumping from 550°C to 580°C). An applied field of 30 μT was used for field-on steps. Starting at 250°C, pTRM back-checks were performed every 50°C to monitor for alteration [Coe et al., 1978], with a change in TRM carrying capability greater than 10% resulting in sample rejection. Experiments were conducted entirely within a magnetically shielded room, with maximum background fields of approximately 100 nT. Heatings were conducted in an ASC TD-48 thermal demagnetizing oven with a background field less than 5 nT, and temperature repeatability of ±1°C. The oven took between 30-45 minutes to reach temperature, and specimens were held at temperature for 10-15 minutes. Cooling took approximately one hour. Moment measurements were performed using a 2G Enterprises 755R DC SQUID magnetometer.
All specimens show a noisy low temperature component that is removed by 200–300°C. Once removed, specimens show single component decay to the origin. This decay occurs in a temperature range that coincides with linear behavior in Arai plots [Nagata et al., 1963]. Figure 3 shows Arai plots generated from all six specimens, with associated orthogonal vector plots of field-off steps inset. At 500°C all six specimens show an increase in slope and a pTRM check showing increased TRM carrying capability (ranging from 2.9% for B9 to 10.3% for B12). Furthermore, the next pTRM check at 550°C shows diminished TRM carrying capability for all six specimens. These sample-wide effects indicate that temperatures above 450°C should not be considered reliable for paleointensity analysis. One specimen (B12) was not fit due to failed pTRM checks. The remaining five were fit from 200°C to 450°C. Results from all specimens are summarized in Table 1, with fit parameters following Coe et al. . The resulting paleointensity values ranged from 26.8 ± 0.5 μT to 30.3 ± 0.9 μT. A weighted mean of the five results [Coe et al., 1978] yields a field value of 28.3 ± 1.3 μT.
Table 1. Thellier-Coe Analysis. Hp, b, σb, f, g, and q were calculated as in Coe et al. .
Specimen B8 had the highest weight factor during averaging due to the small variance in its fitted slope; however, a close examination of this specimen reveals that it has the largest scatter in the orthogonal vector plot of the field off steps with a mean-angular dispersion of 6.9° over the range fit. In this case the lower apparent paleointensity uncertainty of specimen B8, and its associated higher weight in the average, might be misleading. A simple arithmetic mean of the five specimens results in our preferred paleofield estimate of 29.3 ± 1.3 μT. This reproduces the International Geomagnetic Reference Field value at the site (IGRF Model 11, 29.3 μT).
3.2. Multi-specimen Analysis
Multispecimen paleointensity experiments [Dekkers and Böhnel, 2006] were also performed. For these, eight 1 cm3 specimens were cut. Sample preparation and treatment were as described for the Thellier-Coe analysis. Each specimen was heated once, with its NRM direction aligned with a laboratory field. The heating was done at a sufficiently high temperature (300°C in this case) to allow the activation of stable remanence carriers, but low enough to avoid significant alteration. Each specimen was heated in a different applied field, the first at 10 μT, the second at 20 μT, and so on to 80 μT. The change in magnetic moment after heating is normalized to the NRM value, and the resulting percent change is plotted versus applied field intensity for a complete set. For applied fields less than the firing field, the percent change will be negative, and for fields higher than the firing field the change will be positive. The points should thus fall along a line which crosses the x-axis at the original firing field.
An experiment was conducted on both the fired and unfired pot for comparative purposes. Results are shown in figure 4. Vertical error bars were generated by propagating standard deviations in the measured NRM and pTRM values. These standard deviations are the result of averaging three measurements of both the NRM and pTRM values. For each measurement the specimen was removed from the magnetometer and repositioned in the sampler handler. The unfired pot showed no trend when analyzed, which was expected since it would not have acquired a natural TRM. It also shows high measurement uncertainties. In contrast, the fired pot showed a well defined linear trend, and reduced moment uncertainties, indicating the grains may be more stable in the fired pot. The data for the fired pot were fit using a delete-1 jackknife method to yield a paleofield of 35.1 ± 3.0 μT. This represents an overestimate of the known IGRF field of roughly 20%.
This is quite poor compared to the Thellier-Coe technique, and unexpected based on the original presentation of the method [Dekkers and Böhnel, 2006]. However, subsequent to our measurements, Fabian and Leonhardt  predicted overestimates from the multispecimen technique when pseudo-single or multi-domain carriers are present. Fabian and Leonhardt  go further and offer a correction scheme based on a series of subsequent heatings. We are uncomfortable applying such large corrections here. However, we note that in analyses of very large data sets, multispecimen approaches with PSD/MD corrections may have merit in helping to detect paleointensity trends.
3.3. Shaw Analysis
The third paleointensity technique investigated was the Shaw method [Shaw, 1974]. In this technique a laboratory induced TRM is compared to a specimen's NRM, and anhysteretic remanent magnetizations (ARM) are used to check for alteration before and after the heating step. For these experiments samples were again cut into 1 cm3 specimens. Their NRM's were stepwise alternating-field (AF) demagnetized. The AF intensity ranged from 0 to 100 mT, in increments of 10 mT. After the NRM is demagnetized, an ARM (henceforth ARM1) is imparted using a 30 μT DC bias field. This ARM is then demagnetized in the same fashion as the NRM. Next, a total TRM is imparted by heating the specimen to 600°C in a 30 μT bias field, and subsequently AF demagnetized. Finally, a second ARM (ARM2) is imparted under identical conditions to ARM1, and demagnetized. Paleointensity is calculated by plotting NRM versus TRM at each AF step, and fitting the expected linear trend. The product of this slope with the field applied during the TRM heating step yields the paleofield.
The ARM demagnetizations are used to identify steps in the coercivity spectrum which have potentially altered. In an ideal experiment, plotting ARM1 as a function of ARM2 at each AF step would yield a line through the origin with a slope of unity. Assuming the ARMs were imparted under identical conditions, deviations from this trend indicate a change in the sample's coercivity spectrum, presumably during heating. Shaw  proposed the criterion that coercivity values that do not fall along the linear trend in this plot should be rejected from consideration when fitting the NRM/TRM plot. Furthermore, samples whose ARM plot had a slope greater than 1.05 or less than 0.95 were rejected outright. Kono  modified this protocol by dividing the NRM/TRM slope by the ARM1/ARM2 slope. This, he justified, corrected the paleofield calculation for uniform increases or decreases in the amount of magnetic material present across the coercivity spectrum. It also removes the slope constraint on the ARM plot, and allows more samples to be fit.
As with the Thellier-Coe experiments, six specimens were measured with this method. All specimens were stable after heating and none were rejected. Specimens were fit using all points except for B5, whose initial TRM value was corrupted by a measurement error. The NRM/TRM plot for B5 was therefore fit from 10-100 mT, while all points were fit in its ARM1/ARM2 plot. The results of all six measurements are shown in Table 2. All ARM1/ARM2 slopes were greater than unity, with specimens B1, B3, B4, and B5 failing Shaw's 1.05 cutoff. Specimen B1 had the maximum slope at 1.07. Figure 5 shows the plots resulting from the Shaw experiments performed on specimens B1 and B6; the remaining four showed similar features. The NRM/TRM and ARM1/ARM2 plots are nearly linear, though each shows slight concavity. The minimum R2 value from all twelve fits was 0.992 (shown in Figure 5a), indicating that the deviations from linearity, though systematic, were small.
Table 2. Shaw Analysis.Paleofields were calculated with and without the ARM slope correctionof Kono . TRM was imparted with a 40 μT lab field for Sample B1. A 30 μT field was used for all others.
When specimens B2 and B6 are averaged together the mean field value is 31.5 ± 0.3 μT using Shaw's original method. Applying the correction of Kono  to all six specimens yields a paleofield estimate of 30.4 ± 0.8 μT. These two mean values represent overestimates of the reference field by 7.4% and 3.7% respectively. These overestimates are small and potentially due to the presence of nearly-SD pigmentary hematite grains. Such grains would likely relax after the pot's original firing, but may retain their laboratory induced TRM (acquired immediately prior to AF demagnetization).
A second correction scheme proposed by Rolph and Shaw  was investigated. In this method, NRM is plotted against TRM*, which is found by multiplying the TRM value at each coercivity by the value of ARM1/ARM2 at that coercivity. For all samples this resulted in an exaggeration of the nonlinearity, and the correction was deemed unsuitable. We note that Dunlop et al.  caution against the use of corrections (specifically that of Rolph and Shaw ) that treat each point in the coercivity spectrum differently when changes aren't well characterized.
4. Discussion and Conlusions
The rock magnetic characteristics of modern pottery, made using traditional techniques and materials have been investigated as a first step in evaluating the suitability of Iron Age pottery from southern Africa as an archeomagnetic recorder. Magnetic hysteresis and susceptibility properties appear favorable, which suggests similar archeological materials may be capable of retaining meaningful remanences. Three different paleointensity techniques were used to extract field intensities, which reproduced the IGRF reference field with varied success. A simple arithmetic mean of 5 Thellier-Coe double heating experiments repreduced the reference value of 29.3 μT. An uncorrected Dekkers-Böhnel multispecimen technique yielded a result that overestimated the field by nearly 20%, but likely can be improved by recent advancements to the technique. The Shaw method produced an average value that overestimated the field by 7.4%, however applying Kono's correction reduces the overestimate to 3.7%. An uncorrected multispecimen experiment overestimated the firing field by 20%. A recent correction scheme is proposed by Fabian and Leonhardt , however we are uncomfortable applying such a large correction in cases such as that discussed here, where an accuracy < 20% is sought. Nevertheless, we feel the multispecimen approach may be useful in detecting paleointensity trends.
We chose not to apply two methods during the paleoinentensity experiments used to detect MD grains: pTRM tail checks [Riisager and Riisager, 2001] and the IZZI protocol [Yu et al., 2004], for three principal reasons. First, our magnetic hysteresis data already point toward the presence of MD grains; our aim here is not to further document their presence, but to determine which paleointensity method is least affected by these grains (an evaluation that is tenable in our case only because we know the geomagnetic field value). Second, the IZZI technique could pollute our field-off directions, which we otherwise use to assess how effectively we have removed secondary magnetizations. Unlike some studies of similar materials, we see some directional scatter at low unblocking temperatures. Conducting field-off steps first allows the removal of this scatter to be monitored. Third, our initial data indicates some thermal alteration during the paleointensity experiment. Such alteration would be exacerbated by the additional heating needed for pTRM tail checks. Similarly, no anisotropy correction was applied due to the successful reproduction of the known field, and the potential to introduce large errors due to the repeated heatings needed to detect TRM anisotropy.
According to single domain theory, laboratory cooling rates that are faster than the original can result in paleointensity overestimates [e.g. Néel, 1955; Halgedahl et al., 1980]. For multidomain grains, however, paleointensity is underestimated [McClelland, 1984]. Winklhofer et al.  showed that for pseudo-single domain grains, paleointensity would be slightly underestimated, while Yu  recently noted that for PSD grains, cooling rate effects are negligible. In grain mixtures like the ones we are dealing with, it might be that the effects of cooling rate are also negligible because the competing effects of SD and MD grains are offset, and PSD grains have little cooling rate dependence. Ultimately, because the firing timescale was similar to the laboratory timescale, and the known field was reproduced in our Thellier-Coe data, we conclude that a cooling correction is unnecessary in the material we have studied.
The Thellier-Coe method, which reproduced the value, appears to be the most reliable way to extract intensity information out of this modern pottery. The corrected Shaw method also faithfully recorded the field, though slight deviation from linearity in its characteristic plots may be cause for concern. In general, the modern pottery studied here yielded encouraging results using both both methods, and similar results may be attainable in studies of archeological samples.
The authors would like to thank Julia Voronov for help with sample preparation and rock magnetic analysis. RDC conducted the multi-specimen experiments. LPN conducted the Thellier-Coe and Shaw experiments. We thank Fabio Donandini, Elizabeth Schnepp, Associate Editor Joshua Feinberg and an anonymous reviewer for their helpful comments. This work was supported by the NSF EAR-0838185.