Bowles et al., 2002

1. Archaeomagnetic intensity results from California and
Ecuador: evaluation of regional data
Julie Bowles Ã
, Je¡ Gee, John Hildebrand, Lisa Tauxe
Scripps Institution of Oceanography, 9500 Gilman Drive, MC 0208, La Jolla, CA 92093, USA
Received 5 April 2002; received in revised form 19 August 2002; accepted 19 August 2002
Abstract
We present new archaeointensity data for southeastern California (V33‡N, V115‡W, 50^1500 yr BP) and
northwestern South America (Ecuador, 2.4‡S, 80.7‡W, 4000^5000 yr BP). These results represent the only data from
California, as well as the oldest archaeointensity data now available in northwestern South America. In comparing
our results to previously published data for the southwestern United States and northwestern South America, we note
that significant scatter in the existing data makes comparisons and interpretations difficult. We undertake an analysis
of the sources of data scatter (including age uncertainty, experimental errors, cooling rate differences, magnetic
anisotropy, and field distortion) and evaluate the effects of scatter and error on the smoothed archaeointensity record.
By making corrections where possible and eliminating questionable data, scatter is significantly reduced, especially in
South America, but is far from eliminated. However, we believe the long-period fluctuations in intensity can be
resolved, and differences between the Southwestern and South American records can be identified. The Southwest
data are distinguished from the South American data by much higher virtual axial dipole moment values from V0^
600 yr BP and by a broad low between V1000^1500 yr BP. Comparisons to global paleofield models reveal
disagreements between the models and the archaeointensity data in these two regions, underscoring the need for
additional intensity data to constrain the models in much of the world.
ß 2002 Elsevier Science B.V. All rights reserved.
Keywords: archaeointensity; paleomagnetism; California; United States; Ecuador; South America
1. Introduction
Archaeological materials have long been recog-
nized as a possible means of recovering a closely
spaced record of paleointensity variations over the
last 5000^10 000 yr (see [1], and references there-
in]. They can potentially provide greater temporal
and spatial resolution than volcanic materials.
While extensive archaeointensity data now exist,
coverage is far from uniform. In particular, large
parts of the western and southern hemispheres
lack intensity data. Furthermore, much of the ex-
isting data are several decades old, of question-
able reliability by today’s standards, and exhibit
signi¢cant scatter.
This dearth of intensity data in particular re-
0012-821X / 02 / $ ^ see front matter ß 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 1 2 - 8 2 1 X ( 0 2 ) 0 0 9 2 7 - 5
* Corresponding author. Tel.: +1-858-822-4879.
E-mail address: jbowles@ucsd.edu (J. Bowles).
EPSL 6408 21-10-02
Earth and Planetary Science Letters 203 (2002) 967^981
www.elsevier.com/locate/epsl

2. gions can have important consequences in the de-
velopment of any global geomagnetic model. Be-
cause the virtual axial dipole moment (VADM) at
a given time varies dramatically with location,
good spatial coverage is essential in these global
models. For example, in approximating the aver-
age global dipole moment, as in McElhinny ad
Senanayake [1] or Yang et al. [2], the resulting
value will be biased towards regions with large
quantities of data.
In this paper, we examine archaeointensity data
from the southwestern United States and north-
western South America, two regions that could
bene¢t from additional intensity data. Studies
from both regions show that signi¢cant di¡eren-
ces exist between the European paleointensity
curve and the Western Hemisphere data. How-
ever, considerable scatter among the data sets
has made comparisons and conclusions di⁄cult.
We ¢rst examine the possible sources of scatter
in archaeomagnetic data. We make corrections
where possible and apply minimum reliability cri-
teria to extract a more reliable data set. We then
compare this data set, along with our new data
from southwestern California and southern Ecua-
dor, to global paleointensity models.
2. Examination of existing data
We have assembled all published archaeointen-
sity data from the southwestern United States and
northwestern South America, plus the volcanic
paleointensities of Champion [3] (Table 1; see
also Background Data Set1
). The materials used
for the archaeointensity studies are mostly ce-
ramics, with some baked clays (Table 1). The ex-
isting southwestern United States data span a re-
gion approximately 1000 km across, covering
parts of Arizona, New Mexico, Colorado and
Utah (Fig. 1). The ¢rst archaeomagnetic study
in this region was undertaken by Bucha et al.
[4], and signi¢cant amounts of data were added
by Lee [5], Hsue [6], and Sternberg [7]. The South
American data span a region roughly 2000 km
across, covering parts of Ecuador, Peru and Bo-
livia (Fig. 1). Data are contributed by Nagata et
al. [8], Kitazawa and Kobayashi [9], Gunn and
Murray [10], Kono et al. [11], and Yang et al.
[12].
Data from the southwestern United States are
concentrated in the last 2000 yr (Fig. 2), because
earlier materials are rare in this region. The lone
point at 4800 yr BP is a volcanic sample. The
majority of the South American data are also
concentrated in the last 2000 yr, but a signi¢cant
amount of data covers the period from 2000 to
4000 yr BP.
Both sets of data exhibit substantial scatter
(Fig. 2) that may obscure any real trends. Poten-
tial sources of scatter and uncertainty in the data
include dating errors, experimental inaccuracy,
cooling rate di¡erences, magnetic anisotropy,
and distortion of the ancient ¢eld during ¢ring.
We examine these age and experimental uncer-
tainties below, using data from the southwestern
United States and northwestern South America to
illustrate the magnitude of the errors.
2.1. Sources of scatter
2.1.1. Age uncertainty
Age uncertainty is perhaps the biggest concern
in archaeointensity data sets and the most di⁄cult
to eliminate. Dates are assigned to archaeological
materials through a variety of methods, including
stratigraphic association, dendrochronology, pot-
tery style, archaeomagnetic dating, association
with radiocarbon dates, and thermoluminescence
dating. Each of these methods has its drawbacks,
and the best dates are usually derived from some
combination of methods.
With the exception of thermoluminescence ^
and occasionally radiocarbon ^ dating the age
of archaeomagnetic materials invariably involves
association with other independently dated mate-
rials. The most common method of association
relies on the stratigraphy of excavations. Strati-
graphic association can provide accurate dates if
the stratigraphy is properly identi¢ed, including
non-horizontal layering, cultural deposits, intru-
sions and erosional channels that cut through
stratigraphy [13]. Unfortunately, in instances1
http://www.elsevier.com/locate/epsl.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981968

3. where soil layering and stratigraphy are not visi-
ble (and even in some cases where they are), ex-
cavation by metric level is used instead. This in-
troduces an ‘arti¢cial stratigraphy’ by arbitrarily
assigning a common age to all artifacts at a given
depth [13].
Pottery style and archaeomagnetic dating pro-
vide indirect age constraints through comparison
with existing records of temporal changes in pot-
tery style or the regional magnetic ¢eld. Associa-
tion of a unique pottery style with a particular
cultural period may yield excellent dates (within
tens of years) if the cultural period is short and its
age is well-constrained by other methods.
In some instances, artifacts can be sorted into
proper age order, but the details of the absolute
cultural chronology may be in dispute. For exam-
ple, many of the southwestern samples excavated
near Snaketown, Arizona, were produced by the
Hohokam culture. Until recently, con£icting
chronologies for this culture resulted in age di¡er-
ences of up to 700 yr. In 1989, Sternberg [7] ac-
knowledged this problem and discussed both the
‘long-count’ chronology of Haury [15], as well as
the ‘short-count’ chronology of Schi¡er [16],
which is much compressed and results in consis-
tently younger dates. Further work suggests that a
short-count chronology is more likely [17^19]. We
therefore use a recent chronology [18] to adjust all
Hohokam data in our ¢nal compilation (Fig. 3).
It should be pointed out, however, that details of
the Hohokam chronology before approximately
1300 yr BP remain in dispute.
As with dating by pottery style, the accuracy of
archaeomagnetic dating depends directly on the
quality of existing records used for comparison.
Two early studies which provide a signi¢cant
amount of data for the Southwest [5,6] cite ar-
chaeomagnetic dating for some samples. In the
case of oriented samples, archaeomagnetic dating
involves comparing the remanence with a refer-
ence curve of directional secular variation. How-
ever, all of the samples from Hsue [6] and many
of the samples from Lee [5] are unoriented, mean-
ing the sherds must be compared to a reference
paleointensity curve. Because these were the ¢rst
two extensive paleointensity studies in the South-
west, it is di⁄cult to accept that even an approx-
imate date could be assigned in this manner.
In addition to possible uncertainties from the
indirect associations outlined above, the absolute
ages of datable material may also have signi¢cant
errors. For example, radiocarbon dating has nu-
merous pitfalls that are important to recognize
[20]. Laboratory errors, resulting primarily from
errors in counting the radioactive disintegrations,
Table 1
Previous paleomagnetic studies of the southwestern United States and northwestern South America
Author Location Material used Method pTRM
checks?
Anisotropy
correction?
Dating
method
Champion [3] American West lava and ash £ows Thellier yes n/a C, or D and
AM
Bucha et al. [4] American Southwest ceramics and baked
clay
Thellier no no C, AM, PS,
D, S or M
Lee [5] American Southwest,
Peru, Bolivia
ceramics and baked
clay
Thellier no no A, C or AM
Hsue [6] American Southwest ceramics and baked
clay
Thellier no no A, C, or AM
Sternberg [7] American Southwest ceramics Thellier yes yes A, PS, D, HD
or M
Nagata et al. [8] Peru, Bolivia ceramics Thellier no no A
Kitazawa and Kobayashi [9] Ecuador, Bolivia ceramics Thellier no no A or C
Gunn and Murray [10] Peru ceramics Shaw n/a no A, T
Kono et al. [11] Peru ceramics Thellier no no C, S, PS
Yang et al. [12] Peru ceramics Thellier no ‘not needed’ A, T
C = 14
C, S = stratigraphy, PS = ceramic (pottery) style evolution, T = thermoluminescence, M = modern, AM = archaeomagnetic,
D = dendrochronology, HD = historic documents and material culture, A = other archaeological and historical methods.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 969

4. can become greatly ampli¢ed in the calibration
from radiocarbon years before present to calendar
dates. Even if the laboratory’s reported standard
deviation is small (20 yr) and the data fall on a
‘steep’ part of the calibration curve, the estimated
calendar age range can be as great as 100 yr
(68.3% con¢dence level) [20]. Calibration across
a relatively £at portion of the curve can lead to
estimated calendar date ranges of 250 yr or more.
Unfortunately, many of the 14
C dates used in our
compilation studies have much larger laboratory
standard deviations, leading to much larger calen-
dar date ranges (500 or even 700 yr at the 68.3%
con¢dence level). Finally, because the calibration
curve is not monotonic, it is possible for a single
radiocarbon date to have more than one possible
calendar date.
Where original references were provided, we
have calibrated or recalibrated 14
C dates with
the most recent CALIB v4.3 program of Stuiver
and Reimer [21] and associated data sets [22,23].
Except in a few instances, this recalibration had
very little e¡ect on placement of age mid-points,
although it generally increased the age uncer-
tainty.
Thermoluminescence (TL) dating, used in two
studies in our compilation, was developed in the
1960s speci¢cally as a means to directly date pot-
tery (see e.g. [24]). Since then, it has come to be
accepted as a reliable method of dating baked
archaeological materials. The technique is based
on the accumulation of trapped electrons, primar-
ily in quartz and feldspar, from natural radioac-
tive decay. These trapped electrons are released
when the pot is ¢red, resetting the luminescence
‘clock’ [24]. While TL dating will never approach
the precision of radiocarbon dating, it does have
certain advantages over that method. Because
potsherds can be directly dated, errors related to
artifact association are avoided. Calibration er-
rors also do not come into play. Uncertainties
in TL dating are usually about þ 5^10% of the
estimated age and result from errors in properly
estimating the natural level of radiation to which
the sample was exposed [24].
Perhaps the most accurate ages may be ob-
tained from the stratigraphic association of wood-
en artifacts and ceramics. Dendrochronology may
Fig. 1. Map of study areas. Stars indicate sampling locations
for present study. Circles and diamonds indicate sampling lo-
cations for other studies; diamonds show sites remaining
after applying data selection criteria (see text). Note di¡erent
scales on the two enlarged maps.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981970

5. yield dates accurate to a calendar year [14]. While
a time lag between the death of the tree and its
association with other cultural deposits must be
considered, this method has the potential to pro-
vide very tight age constraints.
This brief review of dating methods highlights
the uncertainties inherent in the techniques. An
ideal date might come from a recently measured
14
C date (last 20 yr), combined with one or two
other methods, reducing uncertainty to less than
þ 50 yr. However, this kind of information is gen-
erally not available, and the fact remains that age
Fig. 2. All previously published data for (a) the southwestern United States and (b) northwestern South America. Vertical error
bars represent one standard deviation of repeat measurements on the same sample. Where none are shown, only one measure-
ment was made. Horizontal error bars represent age uncertainties as given in original sources (see Table 1). Where none are
shown, no age uncertainty was reported.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 971

6. control on most of the data in our compilation is
poor. The average age uncertainty in the compi-
lation is nearly þ 140 yr (1c for radiocarbon and
thermoluminescence, or the length of the assigned
ceramic phase or other archaeological context).
Thus, short-period ( 6 200 yr) £uctuations in geo-
magnetic intensity are unlikely to be reliably de-
termined.
Fig. 3. Data selected from the literature based on the selection criteria described in the text. New data from this study are shown
as squares. The light solid line is a cubic spline ¢t to the binned and averaged data. For comparison, the Hongre et al. [47] mod-
el (dotted line) and the gufm1 model of Jackson et al. [48] (bold line) have been evaluated for these two regions. The inset in (b)
is a close-up of the last 700 years showing the discrepancy between gufm1 and the archaeomagnetic data.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981972

7. 2.1.2. Experimental uncertainties
Two main techniques are currently in use for
determination of paleointensity: Thellier^Thellier
[25] and Shaw [26]. The stepwise double-heating
method of Thellier and Thellier [25], modi¢ed by
Coe [27], is commonly believed to be the most
reliable method of determining paleointensity,
and most of the studies in the compilation use a
variation of this method. The method involves
heating samples twice at each temperature step ^
once in zero ¢eld to remove a portion of the nat-
ural remanent magnetization (NRM) and once in
a controlled ¢eld to determine the partial ther-
moremanence (pTRM) gained. The ratio of
NRM lost to pTRM gained is proportional to
the ancient ¢eld. Errors associated with the Thel-
lier method include uncertainty in oven temper-
ature and lab ¢eld, errors in the measurement of
sample intensity, and sample thermal alteration.
Measurement errors should average out if multi-
ple specimens are taken from each sherd. It is
important that temperature between in-¢eld and
zero-¢eld steps is accurately reproduced. Thermal
alteration of the sample can also lead to signi¢-
cant error. The so-called ‘pTRM checks’ are de-
signed to reveal such alteration by back-tracking
to repeat a lower-temperature in-¢eld step. Most
of the early studies in the compilation do not use
these checks.
The Shaw method [26], modi¢ed by Kono [28]
and Rolph and Shaw [29], has been used in one
study in the compilation. Designed to avoid or
correct for thermal alteration, the Shaw method
uses AF demagnetization, but still requires heat-
ing the sample above the Curie temperature to
impart a total TRM. In theory, some thermal
alteration from this heating can be corrected for
[29,30]. In practice, results generally tend to agree
with those of the Thellier method, but in several
cases have been reported to show more scatter
[31,32].
2.1.3. Anisotropy of TRM
Rogers et al. [33] report strong magnetic aniso-
tropy in pottery samples that could lead to errors
of 30^40% (up to 60% for wheel thrown pottery)
if left uncorrected. They suggest that this aniso-
tropy stems from preferential alignment of mag-
netic grains during shaping of the pot which re-
sults in an easy plane of magnetization in the
plane of the pot. Scatter from remanence aniso-
tropy can be corrected by measuring the TRM
anisotropy tensor [36].
2.1.4. Cooling rate
Another potential source of data scatter is in-
troduced through di¡erences between cooling
rates in antiquity and in the lab [34,35]. A slower
cooling rate produces a larger TRM, as magnetic
moments have more time to come into equilibri-
um with the ¢eld. In general, lab cooling rates are
faster than original cooling rates, leading to an
overestimation of the ancient ¢eld, the degree of
which depends on the actual cooling rate di¡er-
ence, as well as the blocking temperature. Lab
cooling times reported in the compilation studies
ranged from 10 min to 2 h.
For ceramics and bricks ¢red in brick kilns,
original cooling can take over 24 h, leading to
an overestimation of the ¢eld by approximately
10% (based on a lab cooling time of 25 min)
[36]. Most of the ceramic samples from the com-
pilation studies were likely ¢red instead in the
open or in small pit kilns, with cooling times on
the order of 1^12 h, based on observations of
Native American potters [37,38]. Pots may be re-
moved from the ¢re while still at high temperature
[37], or they may be left over the ¢re until cool
[38]. How long it takes pots left over the ¢re to
cool depends on the kind and amount of fuel
used, as well as to what degree the fuel surrounds
the pots and how well they are insulated [38].
Actual lab overestimation of ancient ¢eld could
therefore range from V2 to 8%.
Given the variation among modern potters, it is
di⁄cult to comment on precise techniques in a
given location thousands of years ago. It is likely
that the maximum cooling rate error is 10%, and
in many cases it is probably less than 5%. If no
correction is made, paleointensity will always be
overestimated, leading to an upward bias of
the entire averaged curve. For this reason, we
choose to make a 5% correction to all the data.
While some of the resulting paleointensity esti-
mates will now be too low and others still too
high, this blanket correction should have the ef-
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 973

8. fect of centering the scatter closer about the true
value.
2.1.5. Field distortion during ¢ring
A possibility exists that the ¢eld seen by the
pots during ¢ring was not equivalent to the an-
cient ambient ¢eld. Field distortion from a self-
demagnetizing ¢eld and/or proximity to other
pots, sherds or metal during ¢ring can also in-
crease scatter in paleointensity data. A self-de-
magnetizing ¢eld will always act to subtract
from the main ¢eld, but little can be done to
reproduce or correct for this, although Aitken et
al. [39] attempted to measure the possible degree
of self-demagnetization. Distortion from external
sources is a distinct possibility, as open pit kiln
construction can include placing broken sherds
both under and over the pots to be ¢red [37].
After introduction of iron by Europeans roughly
500 years ago, there is also the possibility that
iron objects were used to support or shield pots
during ¢ring [39], although this is probably more
common only in modern times [38,40]. Aitken et
al. [39] suggest that one way to partially mitigate
¢eld distortion e¡ects is simply to take multiple
specimens from di¡erent parts of the pot and mul-
tiple pots from each time period.
2.1.6. Temporal and spatial ¢eld variation
Two ¢nal sources of scatter in the data may be
geomagnetic in origin. Temporal variations in in-
tensity may appear as scatter if samples cannot be
well dated or if sampling density is insu⁄cient.
Spatial variation in the ¢eld may also introduce
apparent scatter. Based on the present ¢eld, re-
gions the size of our study areas (V1500 km)
may have intensities that di¡er by as much as
15%. As more data become available worldwide,
it will be possible to average over smaller regions,
eliminating some of this spatial scatter.
2.2. Data selection
While some of the potential sources of scatter
mentioned above cannot be eliminated, many can
be compensated for or minimized. By selecting
only the most reliable data, we should produce
a much less scattered record that more accurately
represents the true paleo¢eld behavior. We can
use updated ceramic chronologies, calibrate or re-
calibrate 14
C dates, and make cooling-rate correc-
tions to existing data. Corrections for other fac-
tors cannot be made in retrospect, however, and
some data must be excluded. Because data repro-
ducibility is essential, we eliminate over 40% of
the data from the compilation because only one
specimen per sample was tested. We further ex-
clude any samples that show excessive intra-sam-
ple scatter as measured by the standard deviation
of the paleointensity estimates divided by the
mean (cB/Bavg).
Ideal data points would have at least two sub-
samples, cB/Bavg 9 0.10, use the method of Thel-
lier and Thellier with pTRM checks, have an ani-
sotropy correction, and an age uncertainty of less
than þ 100 yr (1c). Sadly, this would leave only
one study from each region. Instead, we apply
somewhat less stringent acceptance criteria: sam-
ples must have at least two specimens; the Thel-
lier technique should be utilized; and cB/Bavg
should be less than 0.20 for each sample to ensure
at least a reasonable degree of within-sample scat-
ter. The Lee [5] and Hsue [6] studies are excluded
because of extreme uncertainty in dating. They
state that they use radiocarbon dating, historical,
or archaeomagnetic dating. As discussed above,
archaeomagnetic dating in this context is highly
questionable. They also provide no information
on their radiocarbon dates so the calibration sta-
tus is uncertain. Because neither study speci¢es on
a sherd by sherd basis the method of dating, we
reject all of these samples from the compilation.
We have also not included a study on Peruvian
ceramics [41] because site locations and individual
results were not presented.
After applying these selection criteria, the scat-
ter is indeed reduced (Fig. 3), especially in South
America. However, very few points remain to de-
¢ne changes in the geomagnetic ¢eld through
time, and we must recognize that even these
points are not ideal. Many of these points are
from studies which did not use pTRM checks or
made no anisotropy correction. It is clear that
new data adhering to strict reliability standards
will go far in resolving a more dependable paleo-
intensity record.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981974

9. 3. New results
In an attempt to increase the amount of high-
quality data available for the southwestern United
States and northwestern South America, we car-
ried out paleointensity experiments on six pot-
sherds from the southern California desert
(V33.3‡N, V115.5‡W) and 14 from Ecuador
(Table 2). The surface-collected California sherds
are ceramics of the Lowland Patayan tradition
[42] and were obtained from the San Diego Mu-
seum of Man where they were typed by Michael
Waters. These sherds have been classi¢ed into
three time periods spanning the past 1500 years
based on pottery style. Unfortunately, more pre-
cise dating in southern California is not possible
due to very plain ceramic styles and the lack of
dendrochronology data. The South American
sherds were excavated from the Real Alto (Valdi-
via) site at 2.37‡S, 80.72‡W and span the period
of roughly 4000^5000 yr BP. These sherds have
also been classi¢ed based on style, and the dates
of these styles are based on extensive radiocarbon
assays as well as some TL dating [43]. Several
individual sherds have also been directly associ-
ated with radiocarbon-dated material. The Real
Alto sherds were obtained from the University
of Illinois Laboratory of Anthropology collec-
tions and were typed by James Zeidler.
We trimmed four specimens from each sherd
and pressed them into salt pellets for ease of han-
dling and to ensure consistent orientation
throughout the experiments. We used the stepwise
double-heating method of Thellier and Thellier
[25], modi¢ed by Coe [27], to recover an estimate
of the ancient ¢eld. Thermal alteration of the
sample was monitored by pTRM checks after ap-
proximately every third temperature step. Two
salt pellet ‘blanks’ were also included in the Thel-
lier experiments and showed no remanence acqui-
sition.
The reliability of the ancient ¢eld determina-
tions was assessed using the criteria of Selkin
and Tauxe [44], which are much more stringent
than the selection criteria applied above. These
criteria are: (1) The temperature interval selected
for paleointensity interpretation should corre-
spond to the ¢nal remanence component of mag-
netization ^ what we hope is the sample’s original
thermoremanence. This is illustrated by the decay
of the zero-¢eld steps in the selected interval to
the origin of a vector endpoint diagram (e.g. Fig.
4); the angle (K) between the principal component
of the selected interval and the vector average of
the data should be less than 15‡. (2) The maxi-
mum angular deviation of this principal compo-
nent must also be less than 15‡. (3) The slope
calculated from the NRM^pTRM pairs (e.g.
Fig. 4) must have the ratio (L) of the standard
error of the slope to the absolute value of the
Table 2
Sherd locations, ages and types
Sample Lat. Lon. Site Sherd provenance Style Historical period Age range
(N) (W) (yr BP)
sic182 33.10 114.87 C85, trail Surface collected Black Mesa Bu¡ Patayan I 1000^1300
sic188 32.76 114.72 C86N, trail Surface collected Colorado Red Patayan I 1000^1300
sic194 33.12 115.36 C14, village Surface collected Salton Bu¡ Patayan II 500^1000
sic195 33.10 114.87 C85, trail Surface collected Tumco Bu¡ Patayan II 500^1000
sic198 33.02 116.28 C160, village Surface collected Colorado Bu¡ Patayan III 50^500
sic201 32.26 115.79 LC54, village Surface collected Colorado Bu¡ Patayan III 50^500
sio001 32.37 80.72 Real Alto Trench A, Structure 1,
£oor deposit, Unit 59/52,
Level 1
Valdivia Polished Red Valdivia Phase 3 4400^4800
sio004 32.37 80.72 Real Alto Trench A, Structure 13,
wall trench, 0^10 cm
Valdivia Nicked Rim Valdivia Phase 6 3950^4100
sio013 32.37 80.72 Real Alto unprovenanceda
Valdivia Red Engraved Valdivia Phase 2 4800^5300
sio014 32.37 80.72 Real Alto unprovenanceda
Valdivia Red Engraved Valdivia Phase 2 4800^5300
a
Assigned to Valdivia Phase 2 based on ceramic style.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 975

10. slope less than 0.1. (4) To ensure su⁄cient repro-
ducibility between two in-¢eld measurements at a
given temperature, the di¡erence between repeat
in-¢eld steps normalized by the length of the se-
lected NRM^pTRM segment must be less than
0.10. (5) q, the quality factor as de¢ned by Coe
et al. [46], must be greater than 1. (6) For each
sample, the standard deviation of the ¢eld esti-
mates divided by the mean (cB/Bavg) must be
6 0.20. Application of these criteria produced ac-
ceptable paleointensity interpretations for 39
specimens representing 10 di¡erent sherds. Repre-
sentative data are shown in Fig. 4, and data from
all interpreted specimens are presented in Table 3,
along with sample averages.
The California sherds all showed either unidir-
ectional demagnetization behavior (for the zero-
¢eld steps) or the presence of a small, secondary,
low-temperature component that was removed by
250‡C (Fig. 4a). Paleointensity interpretations
were made for all of the California specimens,
but one sample (sic182) was rejected because of
excessive intra-sherd scatter. The Ecuadorian
samples yielded fewer successful ancient ¢eld esti-
mates. Many sub-samples exhibited vector end-
point plots that did not decay to the origin, did
not pass pTRM checks or showed other non-ideal
behavior. We believe this is caused by incomplete
oxidation during ancient ¢ring or ¢ring under re-
ducing conditions. As observed by others (e.g.
[4,7]) the well-oxidized sherds that display a red
to orange color in cross-section typically provide
better results than sherds that are gray to black in
color. Of the Ecuadorian samples that produced
interpretable results, most showed two or more
components, but the characteristic component
was usually isolated by 350‡C (Fig. 4b). We spec-
ulate that the lower-temperature components re-
sult from a second ¢ring at a lower temperature,
removal of the pots from the ¢re mid-way
through cooling [37], or use of the pots as cooking
vessels. Of the four Valdivia sherds that gave
good results, one is rejected because of excessive
intra-sherd scatter (sio014).
Fig. 4. Typical results from Thellier^Thellier experiments. (a) Vector endpoint plot of zero-¢eld NRM steps (left) and NRM^
pTRM (right) plots for California sample sic188-1a. (b) Same for Ecuador sample sio1-1a. x^y projection on the vector endpoint
plot is in circles; x^z projection is in squares. Closed circles on the NRM^pTRM plot indicate temperature steps chosen for
paleointensity determination; open triangles are pTRM checks. Numbers on both plots refer to temperature steps.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981976

11. Finally, we made a correction to the paleointen-
sity values for TRM anisotropy and cooling rate
as described by Selkin et al. [36]. The anisotropy
tensor for each specimen was determined by heat-
ing above Tc and cooling in lab ¢eld in six di¡er-
ent positions ( þ x, þ y, þ z). The degree of aniso-
tropy as measured by the ratio of the maximum
to minimum eigenvalues [45] ranges from 1.09 to
1.35. Thirty out of 39 specimens show weakly to
moderately developed oblate anisotropy ellipses,
Table 3
Results of Thellier^Thellier experiments on Californian and Ecuadorian ceramic pot sherds
Specimen c B B* B*avg cB cB/B*avg f g q VADM VADM2
(U1022
A m2
) (U1022
A m2
)
sic182-1 0.0689 83 92 70.6 14.83 0.21 0.53 0.73 13.0 13.30 12.63
sic182-2a 0.0312 66 70 0.64 0.81 29.3
sic182-3a 0.0157 58 59 0.71 0.81 54.5
sic182-3b 0.0214 59 61 0.71 0.81 41.4
sic188-1a 0.0548 66 61 64.3 2.75 0.04 0.57 0.78 12.4 12.17 11.56
sic188-2a 0.0224 69 67 0.64 0.81 39.1
sic188-3a 0.0186 58 63 0.68 0.83 47.0
sic188-4a 0.0640 64 66 0.61 0.78 12.3
sic194-1a 0.0199 63 64 62.7 2.64 0.04 0.49 0.72 28.6 11.82 11.23
sic194-2a 0.0458 66 64 0.52 0.78 14.2
sic194-3a 0.0256 53 59 0.66 0.84 31.5
sic194-4a 0.0260 62 64 0.61 0.81 30.7
sic195-1a 0.0242 45 46 44.2 6.81 0.15 0.53 0.73 18.0 8.33 7.91
sic195-2a 0.0271 52 53 0.38 0.70 12.8
sic195-3a 0.0589 40 43 0.36 0.71 4.6
sic195-4a 0.0311 34 36 0.35 0.71 7.1
sic198-1 0.0284 70 60 63.6 3.02 0.05 0.40 0.74 15.6 12.00 11.40
sic198-2a 0.0195 64 67 0.69 0.87 51.1
sic198-3a 0.0211 65 65 0.69 0.86 45.9
sic198-4a 0.0150 66 63 0.70 0.86 63.1
sic201-1 0.0600 64 61 60.7 1.75 0.03 0.58 0.77 11.4 11.57 10.99
sic201-2 0.0680 65 63 0.59 0.79 10.7
sic201-3a 0.0439 55 60 0.57 0.80 15.7
sic201-4a 0.0397 54 59 0.60 0.80 17.8
sio001-1a 0.0234 22 24 19.7 2.64 0.13 0.47 0.74 9.1 6.81 6.47
sio001-2a 0.0236 25 26 0.49 0.78 10.4
sio001-3a 0.0304 26 29 0.43 0.76 7.8
sio004-1a 0.0161 30 31 33.3 1.90 0.06 0.43 0.74 15.6 8.60 8.17
sio004-2a 0.0462 36 36 0.49 0.75 7.2
sio004-3a 0.0063 26 33 0.68 0.81 71.0
sio004-4a 0.0211 28 33 0.67 0.80 20.9
sio013-1a 0.0194 36 36 39.3 2.58 0.07 0.74 0.77 26.2 10.17 9.66
sio013-2a 0.0205 38 40 0.77 0.83 31.0
sio013-3a 0.0103 41 42 0.76 0.83 64.2
sio013-3b 0.0164 39 40 0.73 0.84 37.0
sio014-1 0.1020 51 59 40.3 14.95 0.37 0.57 0.42 3.5 10.43 9.91
sio014-2a 0.0694 27 32 0.20 0.65 1.5
sio014-3a 0.0807 36 44 0.12 0.71 1.2
sio014-4a 0.0440 21 26 0.18 0.73 1.9
Specimens beginning with ‘sio’ represent sherds from Ecuador, while ‘sic’ denotes Californian sherds. c, standard error of slope;
B, ancient ¢eld (WT); B*, ancient ¢eld (WT) corrected for magnetic anisotropy of sample; B*avg, site mean; cB, standard devia-
tion of site values; f (NRM fraction), g (gap factor), and q (quality index) as de¢ned by Coe et al. [46]; VADM, virtual axial di-
pole moment (U1022
A m2
); VADM2
, cooling rate-adjusted VADM (5%). Note that samples sic182 and sio014 were rejected
based on excessive intra-sample scatter.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 977

12. consistent with an easy plane of magnetization
within the plane of the sherd, as predicted by
Rogers et al. [33]. This resulted in corrections to
the preliminary ancient ¢eld values of up to 30%.
A further 5% reduction to intensity values was
applied to correct for cooling rate.
4. Discussion
Our new data from California generally agree
with earlier data from the southwestern United
States that meet the minimum reliability criteria
outlined above (Fig. 3). While the dates of the
Californian sherds examined in this study remain
too poorly constrained to provide clear distinc-
tions between new and existing data, the lower
of the two points at 750 yr BP suggests that the
decrease in ¢eld intensity between 500 and 1000 yr
BP was possibly lower than previous data imply.
In general, the combined southwestern compila-
tion and new data show signi¢cant di¡erences
from both South American and global data. In
Fig. 5 we show the new and compilation data
averaged into 500 yr bins to compare with the
global archaeointensity curve of Yang et al. [2].
The southwestern data show a distinct low be-
tween V1000 and 1500 yr BP that is not present
in either the South American or the global curves.
Our new paleointensity results from Ecuador
are signi¢cant in that they extend the reliable
South American chronology back over 1500 yr
(Fig. 3b). The well-dated point at approximately
4000 yr BP is especially useful in providing an
accurate paleointensity estimate for a well-con-
strained time. These data roughly agree with
what is expected on a global basis, falling on ei-
ther side of the global curve (Fig. 5). Like the
southwestern compilation data, the South Ameri-
can data show a much narrower peak than the
broad, 2000^3000 yr high of the global curve,
which is dominated by European data. However,
after removing the European data from the curve,
we see that the remaining world data (bold,
dashed line in Fig. 5) provide a closer match to
the long-period variations in South American
data.
To examine shorter-period trends represented
by the southwestern and South American data
sets, we bin and average the selected compilation
data along with our new data in 200 yr bins over
the last 2500 yr (Fig. 3). The arithmetic mean
VADM of each bin is given an age equal to the
arithmetic mean of the sample ages within the bin.
We connect the averaged data points (represented
by asterisks in Fig. 3) with a cubic spline, merely
to guide the eye rather than to suggest a ‘true’
paleointensity curve. We must recognize that the
data scatter (especially age uncertainty) and the
method of averaging degrade any higher-fre-
quency signal that may be present in the data.
Certainly earlier than 2000 yr BP, both curves
are poorly de¢ned and we cannot expect that
the true curve is adequately represented by con-
necting the few dots.
Nonetheless, results of this averaging process
highlight potentially signi¢cant di¡erences be-
tween the two regions, including a higher
VADM in the southwestern United States from
V0 to 600 yr BP and the broad low from
V1000^1500 yr BP mentioned above. These dif-
ferences illustrate the importance of having spa-
tially well-distributed data sets for any kind of
global geomagnetic model. It is equally impor-
tant, however, not to use data of questionable
Fig. 5. Global archaeointensity curve of Yang et al. [2] (bold
solid line), and the global curve minus European data (bold
dashed line). The southwestern United States (thin solid line
and circles) and South American (thin dotted line and trian-
gles) data of Fig. 3 are shown here averaged in 500 yr bins
to match the bins used by Yang et al. Symbols shown with
no connecting line represent single data points rather than
binned averages.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981978

13. quality merely because they exist where little else
does.
The lack of su⁄cient quality data in certain
areas of the world can place limitations on global
models of ¢eld behavior. This becomes evident in
the global archaeointensity model of Hongre et al.
[47]. In South America, where the Hongre et al.
model uses much of the same data as we show in
Fig. 3, an evaluation of their model does provide
a rough ¢t to the data, at least back to about 1700
yr BP. However, in the southwestern United
States, where Hongre et al. provide no intensity
control, the model only roughly approximates the
data and overestimates intensity (Fig. 3a). This
illustrates the danger of drawing any conclusions
from evaluations of such models in poorly con-
strained areas.
Models based on direct observations provide a
closer match to the archaeointensity data,
although even these models show some discrepan-
cies with the data. Since 1832, when Gauss devel-
oped a method to measure absolute intensity, one
might expect the ¢eld to be quite well-de¢ned. But
evaluation of the gufm1 model of Jackson et al.
[48], which is based on historical observations,
shows a discrepancy of V10% in South America
(Fig. 3b, inset). Whether this discrepancy is due to
a bias in the archaeointensity data or to the in-
adequate constraints on the gufm1 model in this
region, it is important to recognize the potential
limitations of such models.
5. Conclusions
Existing archaeointensity data covering the past
5000 years in the southwestern United States and
northwestern South America exhibit signi¢cant
scatter and contain more uncertainty than is gen-
erally recognized. Experimental inaccuracies,
cooling rate di¡erences, magnetic anisotropy, ¢eld
distortion during ¢ring, and especially dating un-
certainty serve to create scatter in the record and
obscure short-period variations in ¢eld intensity.
While we can approximate corrections for some
of the sources of scatter, only a fraction of the
existing data can be used with any con¢dence.
Much work remains to increase the quantity of
high-quality paleointensity data available and
provide more uniform spatial coverage for the
past 5000 years. While our new data start to ¢ll
in the gaps, there is still a long way to go. Com-
plete uniform global coverage may never be
achieved, but signi¢cant improvements on present
data will result in global models that are much
better constrained.
Acknowledgements
We would like to thank Jim Zeidler for select-
ing the Valdivia potsherds for our use and for
providing dates and other helpful information.
Thanks are also due to Angela Neller for facilitat-
ing the loan of the Valdivia sherds from the Uni-
versity of Illinois. Thanks to Ken Hedges and
Grace Johnson of the San Diego Museum of
Man for permission to study the California ce-
ramics. We are also grateful for the helpful com-
ments provided by Je¡rey Eighmy and Robert
Sternberg. This work was partially supported by
NSF grant EAR0099294 to L.T.[RV]
References
[1] M.W. McElhinny, W.E. Senanayake, Variations in the
geomagnetic dipole 1: The past 50,000 years, J. Geomagn.
Geoelectr. 34 (1982) 39^51.
[2] S. Yang, H. Odah, J. Shaw, Variations in the geomagnetic
dipole moment over the last 12000 years, Geophys. J. Int.
140 (2000) 158^162.
[3] D.E. Champion, Holocene geomagnetic secular variation
in the Western United States: Implications for the global
geomagnetic ¢eld, Open File Report 80-824, USGS, Men-
lo Park, CA, 1980, 314 pp.
[4] V. Bucha, R.E. Taylor, R. Berger, E.W. Haury, Geomag-
netic intensity: changes during the past 3000 years in the
Western Hemisphere, Science 168 (1970) 111^114.
[5] S.S. Lee, Secular variation of the intensity of the geomag-
netic ¢eld during the past 3,000 years in North, Central,
and South America, Ph.D., University of Oklahoma,
1975.
[6] T.S. Hsue, Archaeomagnetic intensity data for the south-
western United States 700^1900 A.D., MS, University of
Oklahoma, 1978.
[7] R.S. Sternberg, Archaeomagnetic paleointensity in the
American Southwest during the past 2000 years, Phys.
Earth Planet. Inter. 56 (1989) 1^17.
[8] T. Nagata, K. Kobayashi, E.J. Schwarz, Archaeomagnetic
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 979

14. intensity studies of South and Central America, J. Geo-
magn. Geoelectr. 17 (1965) 399^405.
[9] K. Kitazawa, K. Kobayashi, Intensity variation of the
geomagnetic ¢eld during the past 4000 years in South
America, J. Geomagn. Geoelectr. 20 (1968) 7^19.
[10] N.M. Gunn, A.S. Murray, Geomagnetic ¢eld magnitude
variations in Peru derived from archaeological ceramics
dated by thermoluminescence, Geophys. J. R. Astron.
Soc. 62 (1980) 345^366.
[11] M. Kono, N. Ueno, Y. Onuki, Paleointensities of the
geomagnetic ¢eld obtained from Pre-Inca potsherds near
Cajamarca, Northern Peru, J. Geomagn. Geoelectr. 38
(1986) 1339^1348.
[12] S. Yang, J. Shaw, T. Rolph, Archaeointensity studies of
Peruvian pottery - from 1200 B.C. to 1800 A.D., J. Geo-
magn. Geoelectr. 45 (1993) 1193^1207.
[13] E. Harris, Principles of Archaeological Stratigraphy, Aca-
demic Press, San Diego, CA, 1989, 170 pp.
[14] J.S. Dean, Dendrochronology, in: R.E. Tyalor, M.J.
Aitken (Eds.), Chronometric Dating in Archaeology, Ple-
num Press, New York, 1997, pp. 31^64.
[15] E.W. Haury, The Hohokam: Desert Farmers and Crafts-
men: Excavations at Snaketown, 1964^1965, University
of Arizona Press, Tucson, AZ, 1976, 412 pp.
[16] M.B. Schi¡er, Hohokam chronology: an essay on history
and method, in: R.M. McGuire, M.B. Schi¡er (Eds.),
Hohokam and Patayan, Prehistory of Southwestern Ari-
zona, Academic Press, New York, 1982, pp. 299^344.
[17] H.D. Wallace, J.M. Heidke, W.H. Doelle, Hohokam ori-
gins, Kiva 60 (1995) 575^619.
[18] M.D. Elson, A revised chronology and phase sequence for
the Lower Tonto Basin of Central Arizona, Kiva 62
(1996) 117^147.
[19] P.L. Crown, The Hohokam: Current views of prehistory
and the regional system, in: P.L. Crown, J. Judge (Eds.),
Chaco and Hohokam: Prehistoric Regional Systems in
the American Southwest, School of American Research
Press, Santa Fe, NM, 1991, pp. 135^157.
[20] C.E. Buck, C.D. Litton, E.M. Scott, Making the most of
radiocarbon dating: some statistical considerations,
Antiquity 68 (1994) 252^263.
[21] M. Stuiver, P.J. Reimer, Extended 14
C database and re-
vised CALIB radiocarbon calibration program, Radiocar-
bon 35 (1993) 215^230.
[22] M. Stuiver, P.J. Reimer, T.F. Braziunas, High-precision
radiocarbon age calibration for terrestrial and marine
samples, Radiocarbon 40 (1998) 1127^1151.
[23] M. Stuiver, P.J. Reimer, E. Bard, J.W. Beck, G.S. Burr,
K.A. Hughen, B. Kromer, F.G. McCormac, J. v.d. Plicht,
M. Spurk, INTCAL98 radiocarbon age calibration
24,000-0 cal BP, Radiocarbon 40 (1998) 1041^1083.
[24] M.J. Aitken, Luminescence dating, in: R.E. Taylor, M.J.
Aitken (Eds.), Chronometric Dating in Archaeology, Ple-
num Press, New York, 1997, pp. 183^216.
[25] E. Thellier, O. Thellier, Sur l’intensite¤ du champ magne¤-
tique terrestre dans le passe¤ historique et ge¤ologique, Ann.
Geophys. 15 (1959) 285^378.
[26] J. Shaw, A new method of determining the magnitude of
the palaeomagnetic ¢eld: Application to ¢ve historic lavas
and ¢ve archaeological samples, Geophys. J. R. Astron.
Soc. 39 (1974) 133^141.
[27] R.S. Coe, The determination of paleo-intensities of the
earth’s magnetic ¢eld with emphasis on mechanisms
which could cause non-ideal behavior in Thellier’s meth-
od, J. Geomagn. Geoelectr. 19 (1967) 157^179.
[28] M. Kono, Reliability of palaeointensity methods using
alternating ¢eld demagnetization and anhysteric rema-
nence, Geophys. J. R. Astron. Soc. 54 (1978) 241^261.
[29] T.C. Rolph, J. Shaw, A new method of palaeo¢eld mag-
nitude correction for thermally altered samples and its
application to Lower Carboniferous lavas, Geophys. J.
R. Astron. Soc. 80 (1985) 773^781.
[30] H. Tanaka, Theoretical background of ARM correction
in the Shaw palaeointensity method, Geophys. J. Int. 137
(1999) 261^265.
[31] S. Gonzalez, G. Sherwood, H. Bohnel, E. Schnepp,
Palaeosecular variation in Central Mexico over the last
30,000 years: the record from lavas, Geophys. J. Int.
130 (1997) 201^219.
[32] A.T. Goguitchaichvili, M. Pre¤vot, P. Camps, No evidence
for strong ¢elds during the R3-N3 Icelandic reversal,
Earth Planet. Sci. Lett. 167 (1999) 15^34.
[33] J. Rogers, J.M.W. Fox, M.J. Aitken, Magnetic anisotropy
in ancient pottery, Nature 277 (1979) 644^646.
[34] J.M.W. Fox, M.J. Aitken, Cooling-rate dependence of
thermoremanent magnetisation, Nature 283 (1980) 462^
463.
[35] S.L. Halgedahl, R. Day, M. Fuller, The e¡ect of cooling
rate on the intensity of weak-¢eld TRM in single-domain
magnetite, J. Geophys. Res. 85 (1980) 3690^3698.
[36] P.A. Selkin, J.S. Gee, L. Tauxe, W.P. Meurer, A.J. New-
ell, The e¡ect of remanence anisotropy on paleointensity
estimates: a case study from the Archean Stillwater Com-
plex, Earth Planet. Sci. Lett. 183 (2000) 403^416.
[37] H.S. Colton, Hopi pottery ¢ring temperatures, Plateau 21
(1951) 73^76.
[38] A.O. Shepard, Ceramics for the Archaeologist, Carnegie
Institution of Washington, Washington, DC, 1965, 414
pp.
[39] M.J. Aitken, A.L. Allsop, G.D. Bussell, M.B. Winter,
Determination of the intensity of the Earth’s magnetic
¢eld during archaeological times: reliability of the Thellier
technique, Rev. Geophys. 26 (1988) 3^12.
[40] C.A. Pool, Why a kiln? Firing technology in the Sierra de
los Tuxtlas, Veracruz (Mexico), Archaeometry 42 (2000)
61^76.
[41] J. Shaw, D. Walton, S. Yang, T.C. Rolph, J.A. Share,
Microwave archaeointensities from Peruvian ceramics,
Geophys. J. Int. 124 (1996) 241^244.
[42] M.R. Waters, The Lowland Patayan ceramic tradition,
in: R.H. McGuire, M.B. Schi¡er (Eds.), Hohokam and
Patayan: Prehistory of Southwestern Arizona, Academic
Press, New York, 1982, pp. 275^297.
[43] J.G. Marcos, A reassessment of the chronology of the
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981980

15. Ecuadorian formative, in: M. Guinea, J. Marcos, J.F.
Bouchard (Eds.), El Area Septentrional Andina: Arqueo-
logia y etnohistoria, Abya-Yale Editing, Quito, 1998, pp.
277^324.
[44] P. Selkin, L. Tauxe, Long-term variations in palaeointen-
sity, Phil. Trans. R. Soc. London 358 (2000) 1065^
1088.
[45] T. Nagata, Rock Magnetism, Maruzen, Tokyo, 1961.
[46] R.S. Coe, S. Gromme, E.A. Mankinen, Geomagnetic pa-
leointensities from radiocarbon-dated lava £ows on Ha-
waii and the question of the Paci¢c nondipole low,
J. Geophys. Res. 83 (1978) 1740^1756.
[47] L. Hongre, G. Gulot, A. Khokhlov, An analysis of the
geomagnetic ¢eld over the past 2000 years, Phys. Earth
Planet. Inter. 106 (1998) 311^335.
[48] A. Jackson, A.R.T. Jonkers, M.R. Walker, Four centuries
of geomagnetic secular variation from historical records,
Phil. Trans. R. Soc. London 358 (2000) 957^990.
EPSL 6408 21-10-02
J. Bowles et al. / Earth and Planetary Science Letters 203 (2002) 967^981 981