Oxford Dendrochronology Laboratory

Report 2005/23

FURTHER DENDROCHRONOLOGICAL INVESTIGATIONS

ON THE CARVED HEADS STORED AT
DEVIZES CASTLE, WILTSHIRE

Summary

A previous study of these carved ‘corbels’ in August 1989 suggested a likely felling date range of
the trees used for these heads as 1370-1405. The present study, using micro-cores extended the
chronology and enabled this range to be refined, using more modern sapwood estimates, to the
period 1398-1430. A short, previously undated series from the 1989 samples matched the new
chronology and dates its outside ring to 1407. No other information was available about this
sample, but it does suggest the felling period should be shortened to 1408-1430.

Authors: Dr M C Bridge & D W H Miles

Oxford Dendrochronology Laboratory
Mill Farm
Mapledurham
Oxfordshire
RG4 7TX September 2005
HEAD 1 HEAD 6 HEAD 11

HEAD 13 HEAD 12 HEAD 15

Photographs supplied by Richard Haddlesey

FURTHER DENDROCHRONOLOGICAL INVESTIGATIONS
ON THE CARVED HEADS STORED AT DEVIZES CASTLE, WILTSHIRE

BACKGROUND TO DENDROCHRONOLOGY

The basis of dendrochronological dating is that trees of the same species, growing at the same time,
in similar habitats, produce similar ring-width patterns. These patterns of varying ring-widths are
unique to the period of growth. Each tree naturally has its own pattern superimposed on the basic
'signal', resulting from genetic variations in the response to external stimuli, the changing
competitive regime between trees, damage, disease, management etc.

In much of Britain the major influence on the growth of a species like oak is, however, the weather
conditions experienced from season to season. By taking several contemporaneous samples from a
building or other timber structure, it is often possible to crossmatch the ring-width patterns, and by
averaging the values for the sequences, maximise the common signal between trees. The resulting
'site chronology' may then be compared with existing 'master' or 'reference' chronologies.

This process can be done by a trained dendrochronologist using plots of the ring-widths and
comparing them visually, which also serves as a check on measuring procedures. It is essentially a
statistical process, and therefore requires sufficiently long sequences for one to be confident in the
results. There is no defined minimum length of a tree-ring series that can be confidently
crossmatched, but as a working hypothesis most dendrochronologists use series longer than at least
fifty years.

The dendrochronologist also uses objective statistical comparison techniques, these having the
same constraints. The statistical comparison is based on programs by Baillie & Pilcher (1973, 1984)
and uses the Student’s t-test. The t-test compares the actual difference between two means in
relation to the variation in the data, and is an established statistical technique for looking at the
significance of matching between two datasets that has been adopted by dendrochronologists. The
values of ‘t’ which give an acceptable match have been the subject of some debate; originally
values above 3.5 being regarded as acceptable (given at least 100 years of overlapping rings) but
now 4.0 is often taken as the base value. It is possible for a random set of numbers to give an
apparently acceptable statistical match against a single reference curve - although the visual
analysis of plots of the two series usually shows the trained eye the reality of this match. When a
series of ring-widths gives strong statistical matches in the same position against a number of
independent chronologies the series becomes dated with an extremely high level of confidence.

One can develop long reference chronologies by cross-matching the innermost rings of modern
timbers with the outermost rings of older timbers successively back in time, adding data from
numerous sites. Data now exist covering many thousands of years and it is, in theory, possible to
match a sequence of unknown date to this reference material.

It follows from what has been stated above that the chances of matching a single sequence are not
as great as for matching a tree-ring series derived from many individuals, since the process of
aggregating individual series will remove variation unique to an individual tree, and reinforce the
common signal resulting from widespread influences such as the weather.

Growth characteristics vary over space and time, trees in south-eastern England generally growing
comparatively quickly and with less year-to-year variation than in many other regions (Bridge,
1988). This means that even comparatively large timbers in this region often exhibit few annual
rings and are less useful for dating by this technique.

When interpreting the information derived from the dating exercise it is important to take into
account such factors as the presence or absence of sapwood on the sample(s), which indicates the
outer margins of the tree. Where no sapwood is present it may not be possible to determine how
much wood has been removed, and one can therefore only give a date after which the original tree
must have been felled. Where the bark is still present on the timber, the year, and even the time of
year of felling can be determined. In the case of incomplete sapwood, one can estimate the number
of rings likely to have been on the timber by relating it to populations of living and historical
timbers to give a statistically valid range of years within which the tree was felled. For this region
the estimate used is that 95% of oaks will have a sapwood ring number in the range 9 - 41 (Miles
1997).

BACKGROUND TO THE STUDY

In a letter from Neil Stratford, Keeper of Medieval and Later Antiquities at the British Museum, to
Mr E. B. M. Kemp of Devizes Castle, dated 15th June 1989, it was stated that the corbels had
previously been sawn off at the back, and that the best prospects for dendrochronological dating
would be to remove a slice from the back of the corbel given to the Museum. Permission was
requested to do the same to other corbels in order to increase the likelihood of success. This
permission was given in a reply dated 22nd June 1989 written by K. H. Kemp.

A report dated August 1989 reports on the results of work on seven cross-sections from the Devizes
corbels, with an eighth sample from the Museum’s example. Five of these samples were cross-
matched to each other, and the resulting 156-year chronology was dated to the period 1213-1368.
Sapwood was noted on some timbers, but only on two of the dated samples themselves, and using
the appropriate sapwood estimates as known at that time, a suggested felling date range of 1370-
1405 was given for the carved heads. Some short series remained undated. It was also noted that
this date was much later than the suggested stylistic dating of the 12th century given by Stalley
(1970).

Examination of the mortices in the timbers clearly shows that the heads were carved on the ends of
stub-tiebeams in a principal-rafter roof. The side mortices are for a coved inner cornice plate, the
profile of which is apparent in the housings, and there is a mortice in the tops of the timbers for an
ashlar. Thus the heads would have originally been positioned at the top of a wall, at roof plate
level, and were intended to be viewed from below. One of the heads clearly belonged to an end
truss, as it is angled to one side. The roof to which these corbels were originally attached would
have been at least 7 bays in length.

SAMPLING

Sampling took place in July 2005. Clearly it was impossible to intervene on the surfaces of the
carved heads due to their intrinsic nature, and the backs had too little suitable surface to allow any
further sectioning of the cut-off ends. Any remaining areas were too abraded or damaged to allow
the rings to be accurately measured, therefore a micro-borer was used to extract the samples. This
system was initially developed for work on the medieval doors at the Tower of London,
commissioned by the Historic Royal Palaces Agency. This is accomplished by using a small 8mm
outside diameter hollow drill bit which extracts a 5mm diameter core. The drill bit is cooled and
cleared of dust with the aid of compressed air which is channelled through the inside of the cutting
tube and clears the waste from around the outside of the bit. Visually apparent holes were
afterwards plugged with an oak pellet and stained. The cores thus extracted were mounted on
grooved timber mounts and prepared by being sanded on a linisher using 60 to 1200 grit abrasive
paper, and cleaned with compressed air to allow the ring boundaries to be clearly distinguished.

Sample series from the 1989 sections were prefixed hd, whilst core samples extracted for the
present study were labelled with the prefix dev. At the time of sampling it was not made apparent
that previous work had been done on the heads, hence the difference in numbering. Cores were
mounted on wooden laths and polished with progressively finer grits down to 400 to allow the
measurement of ring-widths to the nearest 0.01 mm. The samples were measured under a binocular
microscope on a purpose-built moving stage with a linear transducer, attached to a desktop
computer. Measurements and subsequent analysis were carried out using DENDRO for
WINDOWS, written by Ian Tyers (Tyers 2004).
ANALYSIS & RESULTS

Following the sampling by DM of the heads in 2005 with the micro-borer (samples dev1 – dev8)
it became evident that some of them had already been sampled by MB in 1989. Although the
earlier samples (hd2 – hd9) could not be immediately located in the Portsmouth University
archives, the ring width data sheets were available. As five of the 1989 samples had initially dated,
it seemed logical to re-analyse all the 1989 data with the new material. Details of the samples taken
are given in Table 1.

First, all the data were compared with each other, and all samples originating from the same parent
timber were combined together to make same-timber means. Two cores were taken from the same
head (dev3a and dev3b) and these were firstly matched independently to other samples to
establish their short overlap, both matching very well against hd5. The two series were then
combined to form a single mean for the head they represented (dev3). Given that sample hd5
matched with a t-value of 22 with the mean dev3, they were considered to have originated from the
same parent timber, and therefore both sequences were combined together to form the mean
hd5dev3 (Table 2).

Next, all samples that appeared to have originated from the same parent tree were combined.
Usually, this is based on t-values of 10 or more, although shorter sequences may have a lower
threshold if the visual comparisons between the plots show exceptionally good correlation.
Therefore, two same-tree groups were found. The first is the same-tree mean hd58dv36 which is
comprised of hd5dev3, hd8, and dev6 (Table 3). The second group is hd7dev5, made up of
hd7 and dev5 (Table 4). These two same-tree means were then used in the subsequent analysis.

Together with hd58dv36 and hd7dev5, eight other individual samples were found to match
(Table 5). Four were from the 1989 samples (hd3, hd4, hd6, and hd9), and four from the 2005
cores (dev1, dev2, dev7, and dev8). All ten sequences were therefore combined to form the new
195-ring site master DEVHEADX, (Table 6).

This site master was then positively dated by comparison with a wide range of dated reference
material, which established its date range as 1213-1407, the evidence being presented in Table 6.
The master chronology gave substantially higher matches then with the 1989 site master
DEVHEADS, which is now superseded. However, all the previously assigned dates determined in
1989 have been confirmed.

INTERPRETATION AND DISCUSSION

In 1989, only five samples dated, and based on the recorded sapwood present on the samples, a
felling date range of 1370-1405 was given. The present analysis has allowed a further two samples
from 1989 to be successfully cross-matched, together with seven samples from the 2005 work.
These newly-dated samples had the advantage of producing samples with later heartwood/sapwood
boundaries which has refined the estimated felling date range to 1408-1430. This is based on the
assumption that the timbers to have come from a single batch of timbers felled within a short period
of time, and that all the carved heads are coeval. There are no obvious differences in the carpentry
details to suggest otherwise. Sample hd7 has the latest ring present at 1407, but unfortunately the
original sample has not as yet been retrieved from the archive of Portsmouth Polytechnic (now
Portsmouth University), and therefore the sapwood compliment for this sample is not known.
However, as sample dev5, which has come from the same parent tree as hd7, had a
heartwood/sapwood boundary of 1395, this suggests that some sapwood might have been present
on sample hd7. Therefore, the early part of the felling date range has truncated at 1408 on the
assumption that at least 9 sapwood rings were present on hd7. If further research shows that the
number of sapwood rings present on hd7 was less than this figure, then the earlier part of the
felling date range would have adjusted correspondingly later. Therefore it seems likely either that
hd58dv36 had more than the normally expected number of sapwood rings, or that it may have been
felled a few years before the other trees used.

The matches with independent chronologies, as laid out in Table 6, show that equally high matches
are found with chronologies from both Hampshire and Somerset. Thus dendro-provenancing would
suggest that the roof from which the carved heads come from originally would be somewhere in
this region. Although the carved heads were in Devizes Castle as early as the 1970s, there are not
many chronologies for Wiltshire to allow any further conclusions to be drawn, except to say that the
roof would have been located either in Wiltshire or in a neighbouring county. There was no
evidence to suggest that the carved heads had originated from the Continent; the timber certainly
originated from the south of England. The heads also were carved at the same time as the roof was
originally constructed, as subsequent distortions and shakes clearly demonstrate that the wood was
unseasoned when carved.

Given the disparity between the dendrochronological dating and the typological dating by Stalley
(1970), it was decided to obtain a second opinion. As a matter of quality control amongst the
established dendrochronological laboratories in the country, data are sometimes exchanged blind
(that is to say just the ring-width data is sent, with no information on its source or the expected
dates) and checked by a second laboratory. In this case Cathy Groves of the Sheffield University
Tree-Ring Laboratory kindly undertook this work and confirmed the findings.

ACKNOWLEDGEMENTS

We would like to thank Mrs Annie Kemp for commissioning this second study and Richard
Haddlesey for providing the photos. We also thank Cathy Groves, University of Sheffield, for
carrying out a blind test on the data to confirm the dating. Finally, we would also like to thank our
fellow dendrochronologists for permission to use their data.
Table 1: Details of the cores taken from the carved heads in 1989 and July 2005
Sample Head Dates AD H/S Sapwood No of Mean Std Mean Felling seasons and
number spanning bdry complement rings width devn sens dates/date ranges (AD)
mm mm mm
1989 samples
hd2 s Head 2 unknown - - 48 1.40 0.53 0.184 unknown
* hd3 s Head 3 1254-1307 - - 54 2.40 0.83 0.145 after 1316
* hd4 s Head 4 1230-1299 - - 70 0.99 0.35 0.239 after 1308
hd5 s Head 5 (= Head 13 2005) 1230-1363 1359 4 134 1.13 0.42 0.263 (1383-1405)
* hd6 s Head 6 1338-1389 - - 52 1.87 0.84 0.222 after 1398
hd7 s Head 7 1317-1407 - - 91 1.34 0.41 0.227 (1408-1436)
hd8 s Head 8 1258-1368 1357 11 111 1.08 0.42 0.241 (1383-1405)
* hd9 s Head 9 1213-1349 - - 137 1.86 0.65 0.140 after 1358

2005 samples
* dev1 mc Head 10 1344-1390 1390 H/S 47 1.39 0.58 0.204 1399-1431
* dev2 mc Head 5 1290-1388 1387 1 99 1.37 0.57 0.199 1396-1428
dev3a mc Head 13 1215-1304 - - 90 1.23 0.52 0.300
dev3b mc ditto 1292-1382 1382 H/S 91 1.06 0.41 0.257
dev3 Mean of dev3a + dev3b 1215-1382 1382 H/S 168 1.13 0.45 0.274 (1383-1405)
dev4 mc Head 11 undated - - <40NM - - - unknown
dev5 mc Head 6 1315-1395 1395 H/S 81 1.48 0.67 0.212 (1408-1436)
dev6 mc Head 1 1246-1353 - - 108+11NM 1.21 0.63 0.231 (1383-1405)
* dev7 mc Head 15 1243-1305 - 63+21NM 1.58 0.43 0.186 after 1335
* dev8 mc Head 12 1302-1392 1389 3 91 1.15 0.43 0.275 1398-1430

Same timber mean

hd5dev3 Mean of hd5 + dev3 1215-1382 1371 168 1.12 0.42 0.262

Same tree means

* hd58dv36 Mean of hd5dev3 + hd8 + dev6 1215-1382 1364 168 1.14 0.41 0.236 1383-1405
* hd7dev5 Mean of hd7 + dev5 1315-1407 1395 93 1.38 0.51 0.215 1408-1436

* = DEVHEADX Site Master 1213-1407 195 1.46 0.47 0.159 1408-1430

Key: *, †, = sample included in site-masters; c = core; mc = micro-core; ¼C, ½C, C = bark edge present, partial or complete ring: ¼C = spring (ring not measured),
½C = summer/autumn, or C = winter felling (ring measured); H/S bdry = heartwood/sapwood boundary – last heartwood ring date; std devn = standard deviation;
mean sens = mean sensitivity. Sapwood estimate (95% confidence) of 9 – 41 used for English timbers (Miles 1997a)
Table 2: Matrix of t –values and overlaps for components of hd5dev3
Sample: dev3a dev3b
Last ring 1304 1382
date AD:

hd5 17.63 13.28

75 72

dev3a 3.18
13

Table 3: Matrix of t -values and overlaps for components of hd58dv36

Sample: hd8 dev6
Last ring 1368 1364
date AD:

dev3hd5 10.95 13.95

111 119

hd8 14.23
107

Table 4: Matrix of t -values and overlaps for components of hd7dev5

Sample: dev5
Last ring 1395
date AD:

hd7 9.22
79
Table 5: Matrix of t -values and overlaps for components of DEVHEADX
Sample: hd4 hd6 hd9 dev1 dev2 dev7 dev8 hd7dev5 hd58dv36
Last ring 1299 1389 1349 1390 1388 1305 1392 1407 1382
date AD:

hd3 2.96 0.00 4.60 0.00 0.91 3.45 0.00 0.00 4.27
46 0 54 0 18 52 6 0 54

hd4 0.00 5.87 0.00 0.00 3.56 0.00 0.00 4.62

0 70 0 10 57 0 0 70

hd6 0.00 8.51 1.17 0.00 6.74 3.93 2.66

12 46 51 0 52 52 45

hd9 0.00 3.55 4.84 2.39 2.62 5.23

6 60 63 48 35 135

dev1 2.50 0.00 7.60 5.07 5.22

45 0 47 47 39

dev2 0.97 1.95 3.66 4.05

16 87 74 93

dev7 0.00 0.00 5.02

93 0 63

dev8 3.88 3.27

78 81

hd7dev5 2.41
68
Figure 1: Bar chart showing the relative positions of overlap of the dated samples in the site
chronology DEVHEADX, showing their sapwood complements (hatched areas), additional
unmeasured rings (narrow bars) and their likely felling dates

Figure 2: Plots of the site chronology DEVHEADX and the relevant section of the dated reference
chronology HANTS02, showing the very good agreement between the two curves with the outside ring
of the site chronology at 1407.
Table 6: Dating evidence for chronology DEVHEADX, AD 1213-1407 – Regional multi-site chronologies are in BOLD

County or region: Chronology name: Short publication reference: File name: Spanning: Overlap t-value:
(yrs):
Hampshire Hampshire Master Chronology (Miles 2003) HANTS02 443-1972 195 12.7
Somerset Somerset Master Chronology (Oxford Dendrochronology Lab) SOMRST04 770-1979 195 11.8
† Somerset George Inn, Norton St Philip (Miles and Worthington 1998) GEORGIN1 1258-1457 150 10.8
London London Master Chronology (Tyers pers comm) LONDON 413-1728 195 9.9
Southern England Southern England Master (Bridge 1998) SENG98 944-1790 195 9.4
Southern England South Master Chronology (Hillam and Groves 1994) SOUTH 406-1594 195 8.7
Oxfordshire Oxfordshire Master Chronology (Haddon-Reece et al 1993) OXON93 632-1987 195 8.4
Gloucestershire The Cottage, Stanley Pontlarge (Miles and Worthington 2001) STNLYPLG 1231-1387 157 8.2
Gloucestershire New Inn House (Arnold, Howard and Litton 2004) KGWBSQ01 1191-1519 195 8.2
* Oxfordshire Bayllols Manor, Harwell (Miles and Haddon-Reece 1991) BAYLLOLS 1170-1370 158 8.1
‡ Hampshire Winchester Cathedral (Miles and Worthington 1998) WINCATH2 1098-1317 105 7.6
Hampshire The Close, Winchester (Miles et al 2003) WCCLOSE2 1284-1443 124 7.6
‡ Hampshire Summers Farm, Long Sutton (Miles and Worthington 2002) SMMRSFRM 1270-1440 138 7.5
Oxfordshire New College Oxford (Worthington & Miles in prep) NWCOLLG1 1271-1396 126 7.0
Wiltshire Saxon House, Malmsbury (Miles et al 2003) MALMSBRY 1304-1486 104 7.0
Wiltshire Burghope Manor, Winsley (Miles in prep) BURGHOPE 1191-1316 104 6.9

‡ = component of HANTS02
† = component of SOMRST04
* = component of OXON93
REFERENCES

Arnold, A. J., Howard, R. E. and Litton, C. D. (2004) Tree-ring analysis of timbers from New Inn House, 7
Wotton Road, Kingswood, Gloucestershire, Centre for Archaeology Report, 62/2004.

Baillie, M.G.L. and Pilcher, J.R. (1973) A simple cross-dating program for tree-ring research. Tree Ring
Bulletin, 33, 7-14.

Bridge, M. C. (1989) Dendrochronological dating of carved oak corbels from Devizes Castle, Wilts.
Unpublished report.

Bridge, M. C. (1998) Compilation of master chronologies from the South, unpubl computer file SENG98,
University of London Dendrochronology Laboratory.

English Heritage (1998) Guidelines on producing and interpreting dendrochronological dates, English
Heritage, London.

Haddon-Reece, D., Miles, D. H., Munby, J. T. and the late Fletcher, J. M. (1993) Oxfordshire Mean Curve -
a compilation of master chronologies from Oxfordshire, unpubl computer file OXON93, Oxford
Dendrochronology Laboratory.

Hillam, J. and Groves, C. (1994) Compilation of master chronologies from the South, unpubl computer file
SOUTH, Sheffield Dendrochronology Laboratory

Miles, D. (1997) The interpretation, presentation, and use of tree-ring dates, Vernacular Architecture, 28,
40-56.

Miles, D. (2003) Dating Buildings and Dendrochronology in Hampshire, in Hampshire Houses 1250 -
1700: Their Dating and Development (ed E Roberts), 220-6, Southampton (Hampshire County Council).

Miles, D. H., with Haddon-Reece, D. (1991) The tree-ring dating of Bayllols Manor, Harwell, Oxfordshire
unpublished report.

Miles, D. H. and Worthington, M. J. (1998) Tree-ring dates, Vernacular Architecture, 29, 111-129.

Miles, D. H. and Worthington, M. J. (2001) Tree-ring dates, Vernacular Architecture, 32, 74-86.

Miles, D. H. and Worthington, M. J. (2002) Tree-ring dates, Vernacular Architecture, 33, 81-102

Miles, D. H., Worthington, M. J. and Bridge, M. C. (2003) Tree-ring dates, Vernacular Architecture, 34,
109-113.

Tyers, I. (2004) Dendro for Windows Program Guide 3rd edn, ARCUS Report, 500b.