close

Вход

Забыли?

вход по аккаунту

?

Magnetic mapping and dating of prehistoric and medieval iron-working sites in northwest Wales.

код для вставкиСкачать
Archaeological Prospection
Archaeol. Prospect. 9, 163–182 (2002)
Published online 2 August 2002 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/arp.191
Magnetic Mapping and Dating of
Prehistoric and Medieval Iron-working
Sites in Northwest Wales
PETER CREW*
Plas Tan y Bwlch, Snowdonia National Park Study Centre, Maentwrog, Blaenau
Ffestiniog, Gwynedd LL41 3YU
ABSTRACT
As part of a long-term research project on prehistoric and medieval iron-working sites in northwest
Wales, a technique has been developed for processing magnetic survey data to improve the
presentation and recognition of the high-amplitude dipolar signals that are characteristic of ironsmelting furnaces. This technique has now been used successfully on some 37 British iron-working
sites. High-resolution surveys, on a 10-cm grid, have been made both by caesium magnetometer
and fluxgate gradiometer over four prehistoric furnaces and one medieval furnace giving detailed
maps of their magnetic signals. Mathematical models of these maps, using multiple dipoles, has
given estimates of the directions of total magnetization of the furnaces. Three of the furnaces were
also surveyed at subsequent stages of excavation and after the removal of the furnace, giving
the background signal. The background signals can then be subtracted from the survey data to
give clean residual maps. These are easier to model, giving more reliable results, and they give
useful information on the contribution of the different furnace materials to the overall magnetic
signals. The corrections required for the influence of induced magnetism and for viscous remanent
magnetism are examined. The results are compared with the British archaeomagnetic curve, and
with the archaeomagnetic date estimates from three of the furnaces, to assess the value of the
modelling technique in giving usable estimates of the date of last firing of the furnaces. Copyright
 2002 John Wiley & Sons, Ltd.
Key words: mapping; dipolar signals; modelling; dating
Introduction
For the past 12 years magnetic surveys have
been used as an integral part of two long-term
projects in northwest Wales, at the prehistoric
iron-working settlement at Crawcwellt and at a
group of 14th century AD bloomeries in Coed y
Brenin.
The settlement at Crawcwellt is over 300 m
above sea-level and the terrain is now rough
grazing with frequent areas of peat bog. This area
has never been ploughed and so the archaeology
*
Correspondence to: Peter Crew, Plas Tan y Bwlch, Snowdonia National Park Study Centre, Maentwrog, Blaenau
Ffestiniog, Gwynedd LL41 3YU. E-mail: peter.crew@eryriupa.gov.uk
Copyright  2002 John Wiley & Sons, Ltd.
is in good condition. The site consists of some
4 ha of enclosures and in the northeastern zone
there are stone-founded round houses (Figure 1).
Excavations from 1986 to 1999 have revealed
earlier phases of settlement with circular stakewall buildings. These contain the remains of
furnaces and smithing hearths and outside them
are dumps of iron-working slag. There is now an
estimated 6.5 t of slag from Crawcwellt, which
is one of the largest quantities known from
prehistoric Britain (Crew, 1989, 1998).
The slag dumps are not more than 30 cm
thick and were not visible at the surface. They
were first detected and mapped in 1989 when
the central 2 ha of the settlement was surveyed,
with traverses spaced at 1 m, using a fluxgate
gradiometer. This indicated three large anomalies
Received 8 June 2001
Accepted 24 May 2002
164
Figure 1. Crawcwellt, northeast area of the settlement,
showing stone enclosure walls, round stone buildings at
J1, J2 and H, and a later rectangular building. J4, J5 and
J6 are anomalies found by magnetic survey. E2 is a stone
Bronze Age burial cairn. Contours in metres above OD. Scale
1 : 1250.
in area J, which helped to focus subsequent
excavation in this area. All these anomalies, J4, J5
and J6, initially were thought to be slag dumps.
Detailed magnetic susceptibility surveys, on a
1 m grid, were carried out over these anomalies
and were used as a basis for estimating the weight
of slag (Crew, 1990, 1995).
Before excavation of the J5 area a higher resolution fluxgate survey was carried out, which
revealed the presence of two possible furnaces.
The difficulty of interpreting the grey-scale plot
led to the development of the mapping technique described below. This showed that the main
anomalies in J5 were well-defined dipoles, which
subsequently were shown by excavation to be
furnaces within a sequence of stake-wall buildings (Crew, 1998). The strength and clarity of the
magnetic signals is a result of the furnaces being
just below the surface and to the small amount of
slags covering them.
There are three phases of structure here, with
the northern pair of furnaces belonging to the
earliest phase and the southern furnace belonging
to the middle phase. The final phase had evidence
Copyright  2002 John Wiley & Sons, Ltd.
P. Crew
Figure 2. Crawcwellt J5 excavation plan, with post-holes,
stake-holes and other subsoil features. Possible wall lines
of the three circular buildings are shown. Furnace lining is
black and burnt clay is stippled. J5A, the largest and earliest
building, contained the northern furnaces. J5B, the mid-size
building, contained the central furnace. Note the way in which
all the furnace arches are orientated towards the entrance
of the buildings. The smallest and latest building was only
used for smithing, but no hearth survived. All the slags, which
covered the west half of the buildings, had been removed for
the furnace surveys. Boxes are the survey areas, 3 m square.
The site grid is 29.5° W of magnetic north and 34.7° W of true
north. Scale 1 : 125.
only for smithing, with characteristic slags, but
no structures (Figure 2). As there are no ceramics
or closely datable artefacts, the chronology of the
settlement depends entirely upon radiocarbon
and archaeomagnetic dating. The data available
so far suggests that the earliest furnaces from the
timber phase could be as early as 300 BC and that
iron-working continued in the later stone phase,
probably until the Roman conquest of this region
in AD 79.
All the Crawcwellt furnaces survived as shallow pits lined with clay, which had been burnt to
a bright red colour and vitrified on the inside
face. These remains are of non-slag tapping
shaft furnaces, about 30 cm internal diameter,
with walls 20 cm thick and an original height of
rather less than 1 m (Crew, 1991). The gaps in
the vitrified lining are from low arches in the
furnace wall, through which slag was removed
from the furnace. Thus the truncated furnaces
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
have a C-shape around which the distribution of
fired material is irregular. Except for the western furnace at J5a, they all had varying amounts
of in situ slag remaining in the bottom of the
furnace.
It became apparent that the dipolar signals
from high-resolution surveys possibly could be
used to recover the direction of total magnetization of the furnaces, thus potentially giving an
indication of their dates. This clearly could be
of value in order to refine the chronology of the
Crawcwellt site.
Thus, for the past 4 years, considerable effort
has been expended in carrying out magnetic
surveys at very high resolution, using a caesium magnetometer and a fluxgate gradiometer.
The detailed maps of the dipolar signals have
been modelled mathematically to give estimates
of the angles of magnetization. Samples from
the furnaces also were taken for conventional
archaeomagnetic dating and the results from both
techniques are compared.
This work has been carried out to try and
understand the magnetic characteristics of these
furnaces, to refine the techniques of survey and
analysis, and to try to improve the estimates
of the dates of the furnaces. A further stimulus
was the potential of being able to estimate
the date of a furnace from magnetic maps of
unexcavated iron-working sites. As only a small
proportion of iron-working sites is ever likely to
be excavated this could be a valuable source of
chronological information, which would require
only a relatively small effort in addition to the
magnetic mapping of a site.
Mapping of furnaces
The data from fluxgate gradiometer surveys
normally is presented in the form of grey-scale
print-outs. The data usually are clipped to a
narrow range, so that low-amplitude linear or
spatial anomalies can be identified more clearly.
Iron-working sites, however, typically produce
data that vary rapidly over a wide range and
which can include high-amplitude anomalies
from in situ furnaces or hearths. Such data are
not well represented in a grey-scale format, as
illustrated in Figure 3.
Copyright  2002 John Wiley & Sons, Ltd.
165
Figure 3. Crawcwellt, 1995 fluxgate gradiometer survey of J5,
before excavation. Parallel traverses, readings on 0.5 m grid.
Grey-scale plot, data clipped to 1 sigma (18 to C22 nT).
Area 16 m square. Scale 1 : 250.
A technique has been developed to process
the magnetic survey data from iron-working
sites to clarify the location and nature of these
strong anomalies (Crew, 1997). The raw data are
imported into Surfer to produce a filled contour
plot. The main element of the technique is the
use of a semi-logarithmic scale to emphasize
the zones of positive and negative readings.
The scale is selected, according to the range of
readings, to show in situ features in the clearest
way possible and usually this is achieved with
a scale that increases or decreases by a factor of
two at each step. The data are not manipulated
or processed in any way, except to smooth the
contours in order to reduce the inevitable linearity
of detected features caused by eccentric survey
grids. Interpolation has been avoided, because
this can easily introduce spurious data. The
results are much more easily appreciated when
plotted in colour, with the high positive readings
in shades of red, low negative readings in shades
of blue and the mid-range readings as grey tones.
Figure 4 shows extracts from a series of surveys
of one of the circular buildings at Crawcwellt, at
increasing resolution. In the 1989 survey, with 1 m
spaced traverses, the furnaces are represented
by only a small number of positive readings
Archaeol. Prospect. 9, 163–182 (2002)
166
P. Crew
and have weakly defined negative zones. As the
furnaces are only 25 cm internal diameter, with
an overall fired zone of some 50 cm, a maximum
traverse spacing of 0.5 m is necessary to show
them in any detail, as in the 1995 survey. The
improvement in resolution with a 0.25 m grid
is shown by the 1996 survey, with the northern
distorted anomaly now clearly from a pair of
furnaces.
Because the furnaces are in situ and have relatively strong remanent magnetism, they give
north–south orientated dipolar signals. The key
to their recognition is the halo of negative
readings on the northern side. However, this negative zone is relatively weak and can be distorted
easily or masked completely by overlying slags
and other features. Dipolar signals with other
orientations can occur, caused by large pieces of
magnetic material that are no longer in position.
Randomly oriented material, such as slag spreads
or dumps, give generally high positive readings,
but without any clear dipolar patterning.
Caesium magnetometer surveys and
modelling
Figure 4. Crawcwellt building J5 surveys 1989–1996, 10 m
square, at increasing resolution. Edge ticks correspond to
readings grids. Positive areas outlined, scale in nT. Grid
orientation 35° west of true north.
Copyright  2002 John Wiley & Sons, Ltd.
It became clear at an early stage of this project
that these dipolar signals had different orientations and could preserve the archaeomagnetic
orientation of the furnaces, but that this could be
recovered only from higher resolution data. Work
on this aspect had been carried out in the Snorup
area of southwest Jutland on the large blocks
of slag left from smelting in slag-pit furnaces
(Bevan and Smekalova, 1996, in press; Abrahamsen et al., 1998). As a result of the co-operation of
Bruce Bevan, Tatyana Smekalova and Olfert Voss
a detailed study was carried out in 1998 on the
furnaces at Crawcwellt and Gelli Goch.
Each of the furnaces was surveyed at very
high resolution with a caesium magnetometer.
Both 5 cm and 10 cm reading grids were used,
but the latter was shown to give adequate data
for modelling. Surveys were also carried out
at different heights. For the small prehistoric
furnaces a 30 cm sensor height gives the best
results and the larger furnace at Gelli Goch gave
good results at a height of 55 cm. A wooden
H-frame was used, both to control the precise
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
position of the sensor, in three dimensions, and
to ensure that repeat surveys were on exactly the
same grid (Crew et al., 2002, Figures 34-2, 34-3,
34-4).
The resulting magnetic maps were then analysed by Bruce Bevan, using the modelling technique described below. Examples of the surveys
and models are given in Figure 5. Although single
dipoles gave reasonable models of the magnetic
maps, it was found that pairs of close-set dipoles
modelled the elliptical shape of the dipolar signals
more closely and with lower root mean square
(RMS) errors. However, the inclination angles of
the two dipole models are some 10° shallower
than would be expected.
In contrast to the Danish slag blocks, these
furnaces are not compact nor nearly spherical
features and have an irregular distribution of
fired material. Maps of the residual errors show
that there are small anomalies which have not
167
been fully modelled, suggesting that more dipoles
would be required to provide more acceptable
results.
Anomaly J5B gave a clear dipolar signal,
which could be modelled with confidence.
However, the survey maps of the furnaces at
J5A and site H show the strong influence of
the background anomalies. Although these background anomalies could be partly modelled by
dipoles set at Earth’s field, the results can be
regarded only as a coarse approximation and
it was clear that the models would be far
better without the complication of the background field.
Fluxgate gradiometer differential
surveys
Because of the encouraging results of modelling
of the maps from the caesium magnetometer
Figure 5. (Top) Crawcwellt 1998 high-resolution caesium magnetometer surveys of furnaces in buildings H and J5, 3 m
and 4 m squares. Positive contours solid at 100 nT, negative contours dashed at 10 nT. Resolution and edge ticks at
10 cm intervals. (Bottom) Calculated maps of multiple dipole models by Bruce Bevan. Stars mark the centre of each dipole.
Each furnace has been modelled with two dipoles constrained to the same directions. Other dipoles to model background
anomalies are set at Earth’s field. The RMS errors are the root mean square differences between the survey and model data.
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
168
surveys, it was decided to continue with further
surveys of the furnaces. These were all carried out
with a Geoscan FM36 fluxgate gradiometer partly
because this instrument is more easily available
in Britain and partly to assess if good quality
results could be obtained using only the vertical
gradient data.
The main aim of this work was to carry
out surveys at different stages of excavation,
both to assess the contribution of the different
furnace materials and to obtain the background
signal after the removal of the furnace, so that
clean residual maps could be prepared for easier
modelling. This also provided the opportunity
for conventional archaeomagnetic sampling, so
that the results of both techniques could be
compared.
The first stage was to resurvey each furnace
in exactly the same state as when surveyed by
the caesium magnetometer, using the same sensor spacing and height, controlled by the wooden
H-frame (AS1, BS1 and HS1 in Figures 6, 7 and 8).
Modelling of the vertical field data using pairs of
P. Crew
dipoles gave very similar inclination/declination
(I/D) directions to the total field data, but
with systematically lower moments. This initially was thought most probably to result
from the non-linearity of the fluxgate gradiometer in high gradient fields, although
detailed analysis has shown this not to be the
case. Further work is necessary to clarify this
problem.
J5A furnaces (Figure 6)
At J5A the next stage was to excavate the
eastern and later furnace and then to resurvey
the isolated western furnace (AS2). Subtraction
of the two sets of survey data, AS1 less AS2,
then gives a residual for the eastern furnace
(AR2), which is a rather weak but well-defined
dipolar signal with an eastern orientation. The
negative residuals in the area of the western
furnace probably are the result of small positional
errors of the sensor in the areas of highest
gradient, as the AS1 and AS2 surveys were
Figure 6. (Top) Crawcwellt 1998–1999 high-resolution fluxgate gradiometer surveys of the northern furnaces in building J5,
at different stages of excavation. Surveys 3 m square, resolution and edge ticks at 10 cm. (Bottom) AR1 and AR3, residual
maps calculated from the survey data less background data; AR2 and AR4, intermediate residuals. Positive contours solid at
100 nT, negative contours dashed at 10 nT, except AS4 (5 and C5 nT) and AR4 (10 and C20 nT).
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
169
Figure 7. (Top) Crawcwellt 1998–1999 high-resolution fluxgate gradiometer surveys of the central furnace in building J5,
at different stages of excavation (BS3 not shown). Surveys 3 m square, resolution and edge ticks at 10 cm. (Bottom) BR1
and BR5 residual maps calculated from survey data less background data; BR3 and BR4, intermediate residuals. Positive
contours solid at 100 nT, negative contours dashed at 10 nT, except BS4 and BR5 (5 and C10 nT), BS5 (5 and C5 nT) and
BR4 (5 and C20 nT).
Figure 8. (Top) Crawcwellt 1998–1999 high-resolution fluxgate gradiometer surveys of the furnace in building H, at different
stages of excavation. Surveys 3 m square, resolution and edge ticks at 10 cm. (Bottom) HR1, HR2 and HR4 residual maps
calculated from survey data less background data; HR3, intermediate residual. Positive contours solid at 100 nT, negative
contours dashed at 10 nT, except HS4 (5 and C10 nT).
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
170
carried out on different occasions and the wooden
H-frame may not have been reset in exactly the
same position. The third survey, AS3, was after
the removal of the ’basal black’ deposit in the
bowl of the furnace. This is a mixture of small
fragments of slag, ore and charcoal stained clay,
characteristic of the basal deposit in all furnaces.
An intermediate residual can be calculated (AR4,
being AS2 less AS3) indicating the complex,
non-dipolar nature of the contribution of the
basal black deposit to the overall magnetic
signal.
The furnace lining of the western furnace was
then sampled for conventional archaeomagnetic
dating, after which the remains of the lining
and the relatively small amount of external fired
clay were removed for the final survey of the
background, AS4. This shows a low positive
anomaly in the northeast corner and a gradient
on the southern side of the survey area, owing
to the negative zone of the dipolar signal from
the central furnace. In the central zone is a
weak positive anomaly, which corresponds to
the tenuous traces of an unexpected third furnace,
stratigraphically earlier than the other two.
The AS4 background signal can then be subtracted from each survey to give clean residuals,
AR1 for both furnaces and AR3 for the western
furnace. The latter is a notably simpler dipolar signal, with a markedly eastern orientation,
although it is rather elongated in shape, reflecting
the typical C-shape of the furnace lining.
J5B furnace (Figure 7)
For the central furnace J5B the procedure was
rather different. After the initial resurvey BS1, the
furnace lining was sampled for archaeomagnetic
dating. This revealed an earlier phase of lining
on a more circular plan and the second survey
(BS2) gave a less distorted dipolar signal. The
earlier furnace proved to have an unusually
large and regular furnace bottom, weighing some
30 kg. An intermediate survey (BS3, not shown)
was carried out after the removal of the furnace
bottom, which seems to have provided the major
contribution to the overall magnetic signal (BR3).
The next survey was after the removal of the
basal black deposit, giving the signal from the
burnt clay (BS4) and a residual for the basal
Copyright  2002 John Wiley & Sons, Ltd.
P. Crew
black deposit (BR4). This is a surprisingly clear
dipolar signal, suggesting that this deposit may
have been refired in situ and had then been
protected from subsequent disturbance by the
furnace bottom. The final survey, BS5, gave the
background signal which shows a slight residual
from the burnt clay and small but weak positive
anomalies for pre-furnace features not excavated
at this stage.
Subtraction of the background signal gives
a series of particularly well-defined dipolar
residuals, which all seem to have a similar
orientation. The C-shape of the BR5 residual
accurately reflects the distribution of the burnt
clay. The southward shift of the residuals for
the basal black and the burnt clay is notable,
reflecting the generally greater depth of these
deposits.
H furnace (Figure 8)
In building H the procedure was similar to that
for J5B, with archaeomagnetic sampling followed
by a series of only three surveys, as the basal
black deposit was very slight and was removed
with the furnace bottom slag (HS2, HS3 and
HS4). The final survey revealed a small positive
anomaly, the cause of the distortion of the
negative zone. This was the result of a small patch
of slag, sealed by redeposited natural, which
was thus not recognized until the final stages of
excavation. The subtraction of this background
signal gives a series of residuals that are all
relatively well-defined dipolar signals and all
seemingly with a similar orientation. The HR4
residual is rather strong, reflecting the amount
of intensely burnt clay surviving from the two
earliest phases of the furnace. As at J5B the
HR3 residual indicates that the furnace bottom
slag provides a major contribution to the overall
magnetic signal.
Analysis of the magnetic maps
The program used for the modelling of the
magnetic maps has been developed by Bruce
Bevan. This uses a simple iteration procedure to
find a least squares fit between the measured field
and the magnetic field of a dipole, or a series of
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
dipoles. Six parameters are used to define each
dipole, three to locate its centre and three to
define the vector of magnetization. In addition
a topographical map of the sensor elevation is
required, if this is not horizontal.
The parameters are set to initial values. For
a first model with a single dipole, these parameters can be program generated. The program
cycles repeatedly through each dipole parameter, changing the parameters slightly until small
changes no longer decrease the RMS error. Up to
nine dipoles can be analysed simultaneously and
the dipoles can be constrained to the same I/D
angles or they can be allowed to have independent directions. Although this program is slow,
it is very stable and has always found a single set
of parameters giving the minimum RMS value.
On the completion of a model its calculated
map can be subtracted from the initial data to
give a residual map, which shows areas of the
data not fully modelled. This map can be used to
suggest the location of further dipoles for a more
complex analysis. This process continues until the
residual map shows no obvious patterning.
This is illustrated by a sequence of models
for the western furnace at J5A (Figure 9). As
the number of dipoles increases, the RMS error
171
decreases and the residual error map becomes
simpler. The final model with five dipoles gives a
very close fit to the survey data. In such a sequence
each of the models can give quite different
I/D angles, depending on the orientation of the
furnace and the relative positions of the dipoles.
It is only when a model has a relatively low RMS
error and a simple residual map that it can be
regarded as reliable.
Results of modelling of the surveys
and residuals
Considerable effort has been expended on modelling the surveys and the residuals. The removal
of the background readings makes a significant
difference to the clarity of the dipolar signals
and most of the residuals could be modelled
accurately with five dipoles, with RMS errors
generally less than 5. The results are listed in
Table 1 and plotted on Figure 10, with the most
reliable models emphasized in bold text. The
results for the different structural elements of the
furnaces give some insight into the way in which
they contribute to the overall magnetic signal and
to the modelled directions of magnetization.
Figure 9. Crawcwellt J5A, western furnace. (Top) Calculated maps of models for AR3 residual with increasing numbers of
dipoles. Positive contours solid at 100 nT, negative contours dashed at 10 nT intervals. 3 m squares, edge ticks at 10 cm.
(Bottom) Residual error maps, calculated from survey less model data, contours at 20 and C20 (left) and 10 and C10 nT.
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
172
P. Crew
Table 1. Results of multiple dipole models of the residual
signals from Crawcwellt furnaces A, B and H (see figures 6,
7, 8)
Code Model for
A1
A2
A3
B2
B6
B3
B4
B5
H1
H2
H3
H4
Both
furnaces
East
furnace
West
furnace
Early
phases
Bottom C
clay
Bottom
only
Basal
black
Clay only
All
phases
Earliest
phase
Bottom
only
Clay only
Moment Declination Inclination
(A m2 )
RMS
error
0.71
15.3
63.5
5.79
0.30
12.7
63.5
4.86
0.36
11.4
66.2
4.59
0.65
9.6
71.1
5.20
0.54
4.1
71.7
5.21
0.49
2.0
71.9
4.27
0.12
12.3
67.6
1.49
0.06
0.53
9.7
2.6
69.5
66.2
1.20
6.06
0.39
7.1
64.9
3.67
0.19
10.4
63.0
3.58
0.22
10.9
65.3
3.76
Not all of the residuals calculated give dipolar signals, especially those for the later phases
of furnace lining. This is mainly because of the
typically irregular survival of upstanding lining material, usually exacerbated by differential
cleaning for the archaeomagnetic sampling. As
shown in Figures 6, 7 and 8, the clearest dipolar
signals are from the AR3, BR3 and HR3 residuals,
without the basal black and clay deposits, corresponding to the furnace lining and slag bottoms.
This material tends to have the most regular distribution laterally and with depth and usually it
is very clear whether or not it is strictly in situ. The
models of these residuals have both the lowest
RMS errors and the simplest error maps, making
them the most reliable for each furnace.
J5A furnace
The AR3 residual was the easiest to model as the
magnetic signal derives almost entirely from an
even distribution of the surviving lower part of
the vitrified lining. Remarkably the five dipoles
seem to map with some accuracy the C-shape
of the furnace lining (Figure 9). The model of
the AR2 residual was carried out on a subset
of the data, to avoid the influence of the false
negative from the western furnace. Despite the
apparent difference in the magnetic maps the
calculated moments are similar, reflecting the
amount of magnetic material surviving at each
furnace. The A1 model is of both furnaces and
as they are essentially contemporary they were
modelled with their I/D angles constrained to the
same directions. The total calculated moment of
Figure 10. Plot of the angles of total magnetization from multiple dipole models of the Crawcwellt furnaces (see Table 1). The
centre of each symbol gives the I/D angles. The most reliable models are indicated in bold type. CA, CB and CH are the mean
remanence vectors for the furnaces, after partial demagnetization, plotted with 63% confidence error bars (see Table 4). Part
of the British archaeomagnetic curve from 550 BC to AD 150 is shown (after Clark et al., 1988).
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
the A1 model is slightly higher than the sum of A2
and A3, because the data include the signal from
the basal black deposit. One might expect the
I/D angles of the western furnace to have been
affected by having cooled with the influence from
the earlier eastern furnace, but this seems not to
have been the case. Despite these complications
the A1 and A2 models have similar declinations
to A3, although their inclinations are slightly
shallower.
J5B furnace
The residual BR2 is effectively a summation of
the BR3, BR4 and BR5 residuals, which represent
its constituent parts. The plot of the model results
shows clearly the influence of the basal black
and clay components, despite their relatively
small contribution to the overall moment. The
clay of this furnace is a mixture of material
with varied degrees of firing and its model B5
is the least reliable. The large furnace bottom
provides the most significant contribution from
this furnace and B3 is the most reliable model. An
artificial residual BR6 was also calculated, being
a summation of BR3 and BR5, to demonstrate the
weak influence of the clay, with only a slight shift
from the B3 model directions.
173
with different weights and different magnetic
characteristics. The moments for each of these
elements can be calculated directly by the models
or can be estimated from the differences between
models (Figures 6, 7 and 8). Table 3 shows the
significant variation in the relative moments
of the different materials for the J5B furnace,
which give an indication of their mean intensity
of magnetization. These data indicate why the
residuals for the furnace lining and bottom
tend to give more reliable models. Despite these
variations, the overall relative moments for each
furnace are fairly consistent and are comparable
to the limited data available (Bevan, 1999).
This provides a useful statistic for estimating
the surviving quantity of fired material from
modelled surveys of unexcavated furnaces.
Table 2. Relative magnetic moments for each furnace
Furnace
J5A, east
J5A, west
J5B
H
H furnace
Relative magnetic moments
The weight of fired material removed at each
stage of excavation was recorded, so that the
relative magnetic moments for each furnace could
be calculated (Table 2). However, each furnace
comprises three or four distinct types of material,
Copyright  2002 John Wiley & Sons, Ltd.
Weight of
material
(kg)
Relative
moment mA
(m2 kg1 )
0.30
0.36
0.76
0.53
22.03
28.64
104.90
47.49
13.6
12.5
7.2
11.2
Table 3. Relative moments for J5B furnace (Figure 7)
Material
In the case of the H furnace the H1, H2 and H3
models show a systematic shift in direction as
the more irregularly distributed later phases of
lining are removed. Both the H1 and H2 models
also show the influence of the highly fired clay,
modelled in H4. The five dipoles of the H3 model
are arranged in a C-shape, as those in A3, but they
do not map the distribution of the lining quite so
accurately, probably because of the large furnace
bottom.
Total
moment
(A m2 )
Late Lining
BR3 Bottom
BR4 Basal black
BR5 Clay
BR1 Total
Total
moment
(A m2 )
Weight of
material
(kg)
Relative
moment
(mA m2
kg1 )
0.09
0.49
0.12
0.06
0.76
8.9
30.0
27.5
38.5
104.9
10.1
16.3
4.4
1.6
7.2
Corrections to the angles of
magnetization
There are two main sources of potential error
that need to be considered before the results
of the modelling can be compared with the
archaeomagnetic curve.
The most significant error is likely to be because
the I/D angles given by the models are of the total
magnetization, which is the vector sum of both
the remanent and induced components. As the
induced component becomes more important,
Archaeol. Prospect. 9, 163–182 (2002)
174
the necessary correction progressively moves the
angles away from those of Earth’s field.
Measurements of the magnetic susceptibility
and NRM of a range of different furnace materials from the Crawcwellt furnaces have been
made, from which the Q ratio, of the remanent
to induced magnetism, can be calculated. For
the clays and lightly vitrified lining this ratio is
low, between 1 and 4, showing that the induced
component is significant. For the more heavily
vitrified lining and furnace bottom slags, however, the ratio is 5 to 10 and 15 to 80 respectively,
so that the induced magnetism contributes only
a small proportion of the total.
The magnitude of the correction also depends
on the proximity of the model I/D angles to those
of Earth’s field. This is illustrated in Figures 11
and 14, which indicate the theoretical corrections
for Q ratios of 10 and 5 for a selection of models. It
is, of course, impossible to extrapolate accurately
from the Q ratios of small samples, especially for
those residuals from combinations of different
furnace materials. However, the Q ratios for the
heavily vitrified lining and furnace bottom, which
produced the AR3, BR3 and HR3 residuals, would
all be relatively high, probably greater than 10,
P. Crew
so that the corrections needed for their models
would be small enough to be of little significance.
There also is a potential error caused by
the contribution from any viscous remanent
magnetism (VRM) that the furnaces might have
acquired when lying in the ambient magnetic
field. Correction for VRM also should result in a
shift of the I/D angles away from Earth’s field, or
more probably from its average direction during
some period since the last firing of the furnaces
(Tarling, 1983, pp. 130–131).
It generally is assumed that it is VRM which
is the main component removed during partial demagnetization. This can be examined
by comparison of data from small samples
removed for conventional archaeomagnetic dating (Figure 11). If the low coercivity components
removed were the result of primarily VRM then
one might expect there to be a systematic shift
from the mean NRM vector to the most stable
vector, which clearly is not the case. In any case
the heavily vitrified lining and the furnace bottoms, which give the most reliable models, are
so strongly magnetized that the VRM component
is not likely to be significant. This is confirmed
by the relatively slow decrease of the intensity of
Figure 11. A3, B6 and H2 are selected models of the Crawcwellt furnaces. If the magnetization is entirely remanent the square
marked R is the best estimate of the I/D angles. The other squares show the theoretical shift away from Earth’s field (marked
by a cross), for Q ratios of 10 and 5, owing to the induced magnetic component. B28ff and C04ff are the mean remanence
vectors of selected furnaces from Bryn y Castell and Crawcwellt. These are calculated from the NRM data (N) and the most
stable data (S), after partial demagnetization, to show the effect of the removal of the low coercivity elements of the remanent
magnetization. Part of the British archaeomagnetic curve from 550 BC to AD 150 is shown (after Clark et al., 1988).
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
furnace-lining samples on partial demagnetization. It seems then that the components removed
during partial demagnetization are more probably the result of small-scale inhomogeneities in
the magnetic fabric (Tarling and Davis, 2001).
However, comparison of the results of the models with different combinations of furnace materials, for example B6 to B3 and H2 to H3, might indicate I/D angle changes owing to the influence of
VRM on the less strongly magnetized fabric. This
emphasizes the need to assess carefully the merits
of each model and the value of using residuals of
furnace materials that are least likely to be influenced by other components of magnetization.
It should be noted that the examples given in
Figure 11 for the changes in the mean remanence
vectors cannot be used directly to indicate the
likely magnitude of errors owing to the contribution of low coercivity components to the magnetic
maps. These are calculated from a number of
small samples used for conventional archaeomagnetic dating, on each of which the influence
of low coercivity components would be relatively
strong. The shifts in I/D angles for the individual
samples are generally greater and more random
than the statistically smoothed calculation of the
mean vectors. As the magnetic maps are made
at some distance from the source, the effect of
any local and relatively weak inhomogeneities
would be much diminished and smoothed out. It
is probable that this will prove to be one of the
major advantages of this technique for estimating
the dates of iron smelting furnaces.
Crawcwellt archaeomagnetic
determinations
Archaeomagnetic dating of iron-working furnaces has been carried out routinely since 1979
at Bryn y Castell and Crawcwellt, with a total of
20 features being sampled. The results from Bryn
y Castell have been published (Clark et al., 1988,
table 1; Crew 1986), showing a very tight cluster
of dates in the mid 1st century AD and a small
group of 2nd to 3rd century AD. The results from
Crawcwellt, which are much less precise, are
summarized in Table 4. This includes the data
from the J5A, J5B and H furnaces, which were
Copyright  2002 John Wiley & Sons, Ltd.
175
Table 4. Archaeomagnetic results from Crawcwellt
Year
Furnace
1986 C41
C10
1987 C04
C62
1990 C207
1993 C292
C285
1999 CA
CB
CH
Declination Inclination
353.1
352.5
359.9
5.9
355.4
354.5
351.6
7.6
352.8
355.8
75.8
70.4
71.0
72.6
69.7
71.6
66.3
68.5
75.9
63.4
n
k
˛95
10
6
12
4
9
10
7
11
6
8
165
226
77
160
58
68
94
34
102
34
3.8
4.5
5.0
7.3
6.8
4.0
4.1
5.4
4.3
6.5
sampled in 1999 to provide comparative data for
the modelling results (Tarling and Davis, 2001).
Fourteen samples were taken from each furnace, evenly distributed around the C-shaped
vitrified lining. Each sample was weighed, allowing detailed comparison of their intensity of
magnetization. Excluding a small number of samples with anomalous values, J5A was the most
strongly magnetized, averaging 18 š 12 mA m2
kg1 . The other furnaces had similar initial intensities of 9 š 8 at J5B and 10 š 6 at H. These figures
are broadly comparable to the relative moments
calculated from the models of residuals (Table 2).
Although the individual sample directions
showed high consistency and linearity on partial
demagnetization, the final directions are widely
scattered and the means are poorly defined
(Table 4). Some samples had very large deviations
and were excluded from the final calculations.
The inclinations are generally high, which means
that the declinations are large and therefore nondiagnostic (Figure 10).
These rather poor quality results are not
unusual from iron smelting furnaces and the
reasons are not yet fully understood. The samples
are all from solid and heavily vitrified lining, so
post-firing movement is not a source of error.
The differences in sample intensity might suggest
considerable variation in the local geomagnetic
field direction. However, there is no indication
from any systematic deviation in the individual
sample directions to suggest that this is the result
of differential cooling of the furnace structure
causing distortions in the ambient geomagnetic
field.
Orientation of the all of the samples was carried
out both by magnetic and sun compass. This
showed significant discrepancies between the
Archaeol. Prospect. 9, 163–182 (2002)
176
two methods, with deviations up to 40° , owing
to highly localized magnetic anomalies. The very
high initial intensities of some samples indicate
that the anomalies probably result from small iron
prills trapped in the slagged furnace lining. This
was also indicated by high-resolution magnetic
susceptibility measurements of the in situ furnace
lining. Other inhomogeneities are probably the
result of grain size and uneven distribution of
different oxides and silicates in the glassy matrix
of the lining.
Normally small inhomogeneities would be
cancelled out during partial demagnetization in
the spinning magnetometer, but this clearly is
not the case with furnace lining material. It
would seem, therefore, that the scatter of the
results may be caused primarily by relatively
strong and very localized changes in the magnetic
fabric of the furnace lining, which can have a
marked effect on the small samples used, thus
introducing an inherent uncertainty in the quality
of archaeomagnetic determinations from ironworking furnaces.
Comparison to the archaeomagnetic
curve
Part of the British archaeomagnetic curve is
shown on Figures 10 and 11, against which the
results of these analyses can be compared. The
curve used is the well-known 1988 version, which
is a subjective evaluation of the data collected
from archaeomagnetic measurements over many
years. The prehistoric section of the curve has very
few independently dated points of good quality
and it relies almost exclusively on a calibration of
lake sediment data (Clark et al., 1988).
Despite the uncertainties of this section of the
curve it is still of potential use for dating 1st millennium sites, partly owing to its rapid east–west
movement and because of the difficulties of dendrochronological correction of 14 C dates during
this period. There is some uncertainty between
about 100 BC and AD 50, as the curve kinks back on
itself in a way not yet fully defined, but thereafter
the Roman section has relatively rapid changes
in inclination to a minimum at about AD 250.
A revised version of this curve has been
published which gives a critical re-assessment
Copyright  2002 John Wiley & Sons, Ltd.
P. Crew
of the Roman and medieval sections (Tarling
and Dobson, 1995). This is based primarily on
the best data available, taking into account the
quality of the archaeological age assessments,
as well as that of the magnetic measurements.
Although the prehistoric section of the curve
was not examined, because there is too little
reliable data, it was suggested that it may have a
rather more shallow inclination than in the 1988
version.
Another version of the curve results from a
statistical smoothing of the data using a moving
window technique (Batt, 1997). However, this
treatment necessarily treats all the archaeological
data as if they are of equal quality and assumes
that the errors have a Gaussian distribution. The
result of this approach is a smoothing of the
amplitude of the swings of the earlier curves.
Because of the limited data from the prehistoric
period there are large uncertainties in the I/D
angles and the curve becomes so contorted that it
is almost unusable.
So, for the purposes of this discussion it is
simplest for the moment to use the familiar 1988
version. Bearing in mind all the uncertainties,
of the archaeomagnetic curve, of the archaeomagnetic dates and of the calculated angles of
magnetization, a comparison of the results is still
of some value. Figure 10 shows clearly the clustering of the results of the models, which gives
some confidence in the apparent difference in
their directions of magnetization. As discussed
above, the A3, B3 and H3 models are the most
reliable. In the case of the J5A and H furnaces, the
model results fall within the 63% error bounds
of the archaeomagnetic dates. For the J5B furnace the archaeomagnetic inclination is rather
too steep, although the B3 model plots almost
exactly on the curve. This result is partly supported by several of the archaeomagnetic results
from other furnaces (Table 4), which seem to
imply a major phase of activity between about
200 and 50 BC. Although the archaeomagnetic
result for the H furnace could indicate a date
in the second century AD, it seems most likely that
this furnace is earlier. The model result from H3
suggests that it could be related to the 50 BC turning point of the curve, which is also supported
by the archaeomagnetic results from two other
furnaces.
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
177
The results from the J5A furnace are of particular interest. A radiocarbon date from the J1 building suggests that iron-working may have started
at Crawcwellt as early as 300 BC. This would fit
with the declinations of both the model results
and the archaeomagnetic date. Although the inclinations seem to be rather too shallow, the archaeomagnetic curve is not well defined in this region.
Thus, although there are significant uncertainties with all the elements in this discussion, the
results would fit reasonably well both with the
archaeological evidence and the limited independent dating evidence from Crawcwellt (Crew,
1998, p. 32).
Gelli Goch, archaeological
background
Some 5 km to the southwest of Crawcwellt there is
a group of large bloomeries, with slag heaps up to
200 t, now in modern conifer plantations. These
sites can be dated by historical records to the
late 14th century (Smith, 1995; Crew and Crew,
1995). The first fluxgate gradiometer surveys were
carried out in 1989. At Gelli Goch a well-defined
dipolar signal was revealed, although its full
significance was not recognized at the time (Gater
and Gaffney, 1989).
All of these sites have since been surveyed by
fluxgate gradiometer at higher resolutions and at
both Llwyn Du and Gelli Goch furnaces identified
by the mapping technique have been excavated
(Crew, 1999). The furnaces are considerably larger
and more complex than those from the prehistoric
sites, with internal diameters of 50 cm and thick
clay walls. Each furnace has a wide arch for
tapping the slag into a working pit, where the
clays are much more intensely fired than at the
prehistoric furnaces.
The dipolar signal from Gelli Goch was particularly clear as the slags have been ploughed
downhill to the southwest (Figure 12). This complex signal originally was interpreted as being
from a pair of furnaces, but excavation showed
the westerly positive zone to be from a 1 m deep
working pit filled with slag and collapsed furnace
structure. In every case where there has been subsequent excavation the original interpretation has
required modification.
Copyright  2002 John Wiley & Sons, Ltd.
Figure 12. Gelli Goch, 1995 survey before excavation, 10 m
square. Resolution and edge ticks at 0.5 ð 0.25 m. Positive
areas outlined, scale in nT. Grid orientation 11 ° E of true north.
Gelli Goch surveys and modelling
With the experience gained from the surveys of
the prehistoric furnaces at Crawcwellt, the procedures adopted for the high-resolution surveys of
the furnace at Gelli Goch were more comprehensive. This was partly because of the much more
complicated nature of the furnace and associated
deposits, but also because of the relatively steep
slope and the very irregular topography of the
site, caused by the deep working pit immediately
west of the furnace. A series of detailed tests
were also made to check the repeatability of the
survey data and to check the effects of positional
errors on the gradiometer readings. Using the
more precise control measures adopted for these
surveys any measurement errors are insignificant and the resulting maps are as good as they
possibly can be.
The surveys and excavation were only completed in May 2001 but some preliminary
results are available. A total of eight surveys have been carried out at two different
heights, 32 and 55 cm, from which a large
number of residuals can be generated. As at
Crawcwellt, the aim was to assess the best
stages of excavation at which to survey, to
give the clearest possible residual for modelling.
As the Gelli Goch furnace is of known date
and the archaeomagnetic curve for the 14th
Archaeol. Prospect. 9, 163–182 (2002)
178
century is reasonably well defined, any results
obtained from this site could be useful as a
control.
The earlier surveys were made to isolate the
contributions of the last run of tap slag, still in
situ in the working pit, and of a large quantity
of very burnt clay that had collapsed into the
tapping arch. A selection of the later surveys at
a height of 55 cm, is shown in Figure 13. Survey GS5 still includes a considerable quantity
of the basal black deposit, both in the furnace
and the working pit. Survey GS6 is after the
removal of this deposit, giving a simpler though
still distorted dipole. The GR5 residual shows the
contribution from the basal black deposit, which
gives a weak dipolar signal, perhaps indicating
that the deposit had been heated or reheated in
situ. The final survey, GS7, was made after the
P. Crew
removal of the large quantity of heavily vitrified
and slagged furnace lining and a thin furnace
bottom. The eccentric background signal results
from the remains of the earliest phase of lining,
a considerable quantity of burnt clay structure
with a very uneven distribution and a veneer
of heavily burnt clay in the tapping arch and
working pit.
The GR6 residual gives a clean dipolar signal
for the lining, slag and furnace bottom and
modelling this with only four dipoles gave a
low RMS error (Table 5). It is of great interest
that the four dipole centres are in an inverted Lshape, which maps the surviving lining and slag
with some accuracy. This seems to confirm the
earlier results from Crawcwellt which show that
the multiple dipole models may give a reasonable
indication of the distribution of the fired material.
Figure 13. Gelli Goch. (Top) GS5, GS6 and GS7 4 m square surveys at 55 cm height, at later stages of excavation. (Bottom)
GR5 residual for the basal black deposit, GR6 residual for the furnace lining, etc. GR6d, four dipole model of GR6, with the
dipole centres marked by a star, proportional to their size. Moments calculated from models or by subtraction. Contours at
10 and C50 nT.
Copyright  2002 John Wiley & Sons, Ltd.
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
179
Table 5. Results of multiple dipole models of the
survey and residual signals from the Gelli Goch
furnace (Figure 13)
Code
Model
for
GS5
GS6
All materials
No basal
black
Final clay
Basal black
Lining C slag
GS7
GR5
GR6
Moment Decli
Incli RMS
(A m2 ) nation nation error
4.18
3.38
1.0
7.6
50.5
43.4
4.72
6.60
2.29
0.85
1.05
8.5
13.5
6.2
56.4
45.3
49.0
5.74
3.55
4.04
Table 6. Relative moments for Gelli Goch furnace materials
Material
Tap slag
Arch clay
GR5 Basal black
GR6 Lining C slag
Total
Total
moment
(A m2 )
Weight of
material
(kg)
Relative
moment
(mA m2 kg1 )
0.15
0.15
0.85
1.05
2.20
40.47
45.56
153.90
57.35
297.28
3.7
3.3
5.5
18.3
7.5
All of the excavated furnace deposits were
weighed after excavation and Table 6 gives the
calculated relative moments of the different
materials. This is only a preliminary estimate,
as the calculated moments need more refinement
and the furnace deposits have not been fully
characterized and sorted.
These relative moments can be compared with
those from Crawcwellt J5B (Table 3). The figure
for the basal black deposit is rather similar but
the clay and furnace lining values are rather
higher, reflecting the different furnace operating
conditions at Gelli Goch. The total relative
moment of 7.5 mA m2 kg1 for the excavated
materials is very similar to that from J5B, but
this results from the different proportions of
the deposits with varying relative moments,
especially the basal black.
Gelli Goch results
The directions of total magnetization calculated
from a variety of the Gelli Goch survey and
residual data are shown on Figure 14. The models of the caesium magnetometer data show a
systematic trend as the models become more
sophisticated, getting progressively closer to the
turning point of the archaeomagnetic curve.
Copyright  2002 John Wiley & Sons, Ltd.
Figure 14. Plot of the I/D angles of total magnetization
for the Gelli Goch furnace. G23–G27 are models of the
caesium magnetometer survey data, with increasing numbers
of dipoles (after Crew et al., 2002, table 1). GS5, GS6 and GS7
are multiple dipole models of the fluxgate gradiometer survey
data and GR5, GR6 of the calculated residuals (see Table 5).
The plot for GR6 shows the estimates for entirely remanent
magnetization (R) and for the removal of the induced magnetic
component for Q ratios of 10 and 5. The cross is a multipole
model for GR6 with 95% confidence error bars. Part of the
British archaeomagnetic curve from AD 1200 to 1450 is shown
(after Clark et al., 1988).
Most of the fluxgate gradiometer data models
plot on the same trend, giving some confidence in the results. However, only the model
from the GR6 residual can be regarded as reliable. Although its declination fits well with the
late 14th century section of the archaeomagnetic curve, its inclination is some 5° shallower.
The theoretical corrections for Q ratios of 10
and 5 are also indicated, although any likely
correction for induced magnetization should
result in only a slight further shallowing of the
inclination.
Archaeol. Prospect. 9, 163–182 (2002)
180
Nevertheless, considering the topographic and
structural complexities of this site, this is
a very encouraging result. The archaeomagnetic curve is relatively well defined for the
medieval period and despite the too shallow
inclination, it is unlikely that the model result
could indicate anything other than a 14th century date.
One of the main difficulties of assessing
the validity of the model results is that the
program used only gives the final inclination
and declination values, with no estimation of
the likely errors. A new technique is being
developed using a different approach, in which
the non-dipolar elements of the data are modelled
by higher orders of multipoles. This technique
uses a sophisticated NAG optimizer that gives
confidence limit estimators for all of the modelled
parameters. These can then be used to derive
corresponding error estimates for the calculated
angles of total magnetization.
The multipole model for the GR6 residual is
shown on Figure 14, with its 95% confidence
errors, plotting remarkably close to the late 14th
century section of the archaeomagnetic curve.
This initial result is very promising, but a great
deal more work still needs to be carried out to
refine and test this technique.
Discussion
The development of these techniques for the
mapping and dating of iron-working sites has
taken considerable time and effort over the past
5 years and the work is clearly not yet complete.
The examples of mapping presented in this
paper are restricted to the illustration of furnaces
but the technique has potential for much wider
use. Data from a considerable number of ironworking sites have now been processed and the
results are of great value in showing their layout
in some detail.
An early opportunity to test the technique was
provided by the survey data from Stanley Grange,
Derbyshire, where a large medieval iron-smelting
site was being excavated in advance of open-cast
coal extraction. Because the site had been heavily
ploughed, removing most of the superficial slags,
clear dipolar signals predicted the location of
Copyright  2002 John Wiley & Sons, Ltd.
P. Crew
all the furnaces and numerous other features
(Challis, 1997).
In Cumbria some 25 bloomeries have now been
surveyed at high resolution. In most cases the
maps have been useful in identifying furnaces,
often in pairs, and a wide range of other features,
including possible hearths, ore roasting and charcoal burning areas. These surveys were carried
out as part of a long-term project by the Lake District National Park and the National Trust, both
to provide information for management purposes
and to identify sites worthy of further survey and
excavation.
As only a small proportion of sites is ever likely
to be excavated, the information recovered from
magnetic mapping is crucial to their interpretation. This will contribute to building a typology
of sites and, ultimately, to the recognition of technological, chronological and regional patterns.
However, the detailed interpretation of the magnetic maps is fraught with difficulty. This has
been shown at each of the sites subsequently excavated, emphasizing the need for closer dialogue
between archaeologists and surveyors.
One objective of the differential fluxgate gradiometer surveys was to determine the stages at
which to survey, during the excavation process,
in order to obtain the best quality magnetic signal
for analysis. It is clear now that a minimum of
two surveys is required. The first should be of
the furnace when it has been defined as fully as
possible, with an even distribution of lining and
furnace bottom and after the removal of the basal
black deposit. This is not always possible, however, as any surviving furnace bottom will seal
basal black deposits, although in such a case they
are more likely to have been fired in position.
The second survey should be after the removal of
the lining and any furnace bottom, but leaving in
position any surviving basal black deposits and
as much as possible of the surrounding burnt
clay. This survey then provides an effective background signal, from which a clean residual can
be calculated and analysed.
In practice, two surveys of this kind take about
the same time as conventional archaeomagnetic
sampling. With care it should be possible to
carry out the surveys during the normal course
of an excavation, as long as the surrounding
Archaeol. Prospect. 9, 163–182 (2002)
Iron-working Sites in Northwest Wales
archaeological features remain the same. Furnaces would not then need to be isolated and
excavated out of sequence, which usually has to
be the case for archaeomagnetic sampling. This
has a real advantage in that the furnace can be
excavated and recorded in a normal manner,
thus recovering the maximum of archaeological
information.
One of the original aims of this project was to
discover if the date of a furnace could be estimated
from the magnetic maps of unexcavated sites.
However, in view of the complications outlined
in this paper, such an estimate would be possible
only under the relatively rare conditions when
the magnetic signal from a furnace is not
unduly complicated by overlying slags or other
archaeological features. If such a circumstance
did arise in the field, it would require only a small
additional effort to resurvey a feature at a higher
resolution. Trials have shown that readings on
a 25-cm grid can give a reasonable estimate
of the moment and of the size of a furnace,
but not always of the angles of magnetization.
For example, from the 27 sites surveyed in
Cumbria there are only eight dipolar signals
that are clean enough to give indications of
the dates of the furnaces. A radiocarbon dating
programme of these sites in now in progress
and it will be of great interest to compare
the results. Nevertheless, information gained
by modelling of the dipolar signals of these
furnaces has been shown to be of some value
in refining the interpretation of low-resolution
field surveys.
The potential applications of the techniques
described are not, of course, restricted to furnaces.
Any well burnt feature of simple shape, such
as a hearth or a small oven, as survive on
many archaeological sites, could be surveyed and
analysed in this way. However, any estimates
of date would be difficult to make because of
the potentially high contributions of induced
magnetism and viscous remanent magnetism.
Careful measurements of the Q ratios and of
magnetic viscosity would be essential to ensure
that adequate corrections could be applied.
The value of analysing very high-resolution
maps to recover estimates of the dates of iron
smelting furnaces is still debatable. However, on
the basis of the evidence presented in this paper
Copyright  2002 John Wiley & Sons, Ltd.
181
it would seem that the results of modelling are
no less useful than the results from conventional
archaeomagnetic dating. Each technique clearly
has its limitations and only further studies,
with better independent dating evidence, will
reveal whether or not the mapping approach
can be refined further. If so, then a systematic
programme of magnetic survey and analysis
could result in the acquisition of new high-quality
data, which could be of particular value for the
dating of prehistoric sites and which, in the longer
term, could help to refine the archaeomagnetic
curve.
Acknowledgements
This project has been possible only with the
help of many friends and colleagues. Above all
I should like to thank Bruce Bevan, for modelling the caesium magnetometer surveys, for
allowing the use of his programs and for his
continued advice and encouragement. Thanks
also to Tatyana Smekalova for the caesium
magnetometer surveys and to Olfert Voss for
making them possible. The fluxgate gradiometer surveys have been carried out with great
care and patience by John Price and Kathy
Laws, of Engineering Archaeological Services.
Thanks especially to Don Tarling for the archaeomagnetic sampling and for his general advice
and comments. Tony Clark was involved with
this project for many years, sampling our furnaces, and he would have been delighted at
the progress made. Ian Foster and his colleagues
at Coventry University kindly arranged for the
Q-ratio measurements, which were carried out
by Tony Chapman. Particular thanks are due
to Terry Williams for his stimulating contribution in developing the multipole modelling
technique.
None of this project would have been possible
without the constant support, during the excavations and the surveys, of Sheila Rawson and Ian
Devine and, especially, my wife Susan.
References
Abrahamsen N, Koppelt U, Jacobsen B, Smekalova T,
Voss O. 1998. Iron age slags at Snorup (Denmark):
magnetic prospecting, modelling, reconstruction
and dating. Geologica Carpathica 49(1): 61–70.
Batt CM. 1997. The British archaeomagnetic calibration
curve: an objective treatment. Archaeometry 39(1):
153–168.
Archaeol. Prospect. 9, 163–182 (2002)
182
Bevan B. 1999. The Magnetic Properties of Archaeological
Materials. Geosight Technical Report No. 5, Geosight:
Virginia, USA.
Bevan BW, Smekalova TN. 1996. Magnetisation
Directions of Iron Slag at Snorup. Manuscript Report
for Olfert Voss, National Museum: Copenhagen.
Bevan BW, Smekalova TN. in press. Magnetisation
directions of iron slag in Denmark. Prospezioni
Archeologiche.
Challis K. 1997. Rescue excavation of a Medieval
monastic site at Stanley Grange, Derbyshire. In Early
Ironworking in Europe: Archaeology and Experiment,
Crew P, Crew S (eds). Plas Tan y Bwlch: Blaenau
Ffestiniog; 35–39.
Clark AJ, Tarling DH, Noël M. 1988. Developments
in archaeomagnetic dating in Britain. Journal of
Archaeological Science 15: 645–667.
Crew P. 1986. Bryn y Castell hillfort—a late prehistoric
ironworking settlement in north-west Wales. In The
Crafts of the Blacksmith, Scott BG, Cleere H (eds).
Ulster Museum: Belfast; 91–100.
Crew P. 1989. Excavations at Crawcwellt West,
Merioneth, 1986–1989. A late prehistoric upland
ironworking settlement. Archaeology in Wales 29:
11–16.
Crew P. 1990. Crawcwellt West, Trawsfynydd.
Archaeology in Wales 30: 46–47.
Crew P. 1991. The experimental production of
prehistoric bar iron. Historical Metallurgy 25(1):
25–36.
Crew P. 1995. Excavations at Crawcwellt, a prehistoric
ironworking settlement in Merioneth. In Iron for
Archaeologists, Crew P, Crew S (eds). Occasional
Paper No. 2, Plas Tan y Bwlch: Blaenau Ffestiniog;
32–35.
Crew P. 1997. Geophysical survey and furnace location
at ironworking sites. In Early Ironworking in Europe:
Archaeology and Experiment, Crew P, Crew S (eds).
Plas Tan y Bwlch: Blaenau Ffestiniog; 113–115.
Copyright  2002 John Wiley & Sons, Ltd.
P. Crew
Crew P. 1998. Excavations at Crawcwellt West,
Merioneth: a late prehistoric upland ironworking
settlement. Archaeology in Wales 38: 22–35.
Crew P. 1999. Magnetic surveys of ironworking sites
in north-west Wales. In Traditions and Innovations
in Early Medieval Iron Production, Gömöri J (ed.).
Dunafer: Sopron-Somogyfajsz, Hungary; 41–47.
Crew P, Crew S. 1995. Medieval bloomeries in northwest Wales. In The Importance of Ironmaking:
Technical Innovation and Social Change, Magnusson G
(ed.). Jernkontorets Berghistoriska Utskott H58:
Stockholm; 43–50.
Crew P, Smekalova TN, Bevan BW. 2002. High
resolution magnetic surveys of prehistoric and
Medieval iron smelting furnaces in north-west
Wales. In Prehistoric and Medieval Direct Iron Smelting
in Scandinavia and Europe: Aspects of Technology and
Society, Nørbach L, Voss O (eds). University Press:
Aarhus, 209–222, 315–321.
Gater J, Gaffney C. 1989. Geophysical Surveys of
Bloomeries in Coed y Brenin. Manuscript report for
Snowdonia National Park Study Centre: Blaenau
Ffestiniog.
Smith KES. 1995. Ironworking in north-west Wales in
the late fourteenth century. Archaeological Journal 152:
246–290.
Tarling DH, Davis P. 2001. Archaeomagnetic Results from
three Furnaces, Crawcwellt. Manuscript report for
Snowdonia National Park Study Centre: Blaenau
Ffestiniog.
Tarling DH. 1983. Palaeomagnetism: Principles and
Applications in Geology, Geophysics and Archaeology.
Chapman and Hall: London.
Tarling DH, Dobson MJ. 1995. Archaeomagnetism: an
error assessment of fired material observations in the
British directional database. Journal of Geomagnetism
and Geoelectricity 47: 5–18.
Archaeol. Prospect. 9, 163–182 (2002)
Документ
Категория
Без категории
Просмотров
0
Размер файла
531 Кб
Теги
wales, working, site, medieval, dating, mapping, prehistoric, northwest, magnetic, iron
1/--страниц
Пожаловаться на содержимое документа