www.elsevier.com/locate/gloplacha

Global and Planetary Change 40 (2004) 11–26

Large-scale temperature inferences from tree rings: a review

K.R. Briffaa,*, T.J. Osborna, F.H. Schweingruberb

aClimatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UKbSwiss Federal Institute for Forest, Snow and Landscape Research, Zurcherstrasse 111, CH-8903 Birmensdorf, Switzerland

Received 21 October 2002; accepted 7 May 2003

Abstract

This paper is concerned with dendroclimatic research aimed at representing the history of very large-scale temperature

changes. It describes recent analyses of the data from a widespread network of tree-ring chronologies, made up of ring width

and densitometric measurement data spanning three to six centuries. The network was built over many years from trees selected

to maximise their sensitivity to changing temperature. This strategy was adopted so that temperature reconstructions might be

achieved at both regional and very large spatial scales. The focus here is on the use of one growth parameter: maximum

latewood density (MXD). The detailed nature of the temperature sensitivity of MXD across the whole network has been

explored and the dominant common influence of mean April–September temperature on MXD variability is demonstrated.

Different approaches to reconstructing past temperature for this season include the production of detailed year-by-year gridded

maps and wider regional integrations in the form of subcontinental and quasi-hemispheric-scale histories of temperature

variability spanning some six centuries.

These ‘hemispheric’ summer series can be compared with other reconstructions of temperature changes for the Northern

Hemisphere over the last millennium. The tree-ring-based temperature reconstructions show the clear cooling effect of large

explosive volcanic eruptions. They also exhibit greater century-timescale variability than is apparent in the other hemispheric

series and suggest that the late 15th and the 16th centuries were cooler than indicated by some other data.

However, in many tree-ring chronologies, we do not observe the expected rate of ring density increases that would be

compatible with observed late 20th century warming. This changing climate sensitivity may be the result of other environmental

factors that have, since the 1950s, increasingly acted to reduce tree-ring density below the level expected on the basis of summer

temperature changes. This prevents us from claiming unprecedented hemispheric warming during recent decades on the basis of

these tree-ring density data alone. Here we show very preliminary results of an investigation of the links between recent changes

in MXD and ozone (the latter assumed to be associated with the incidence of UV radiation at the ground).

D 2003 Elsevier B.V. All rights reserved.

Keywords: Dendroclimatology; Tree-ring density; Northern Hemisphere; Temperature reconstructions; Global warming

0921-8181/$ - see front matter D 2003 Elsevier B.V. All rights reserved.

doi:10.1016/S0921-8181(03)00095-X

* Corresponding author. Tel.: +44-1603-593909; fax: +44-

1603-507784.

E-mail address: k.briffa@uea.ac.uk (K.R. Briffa).

1. Introduction

A large amount of tree-ring research is concerned

with very localised site studies (Dean et al., 1996),

necessarily reflecting the complex ecological process-

es that operate on small scales in forest ecosystems.

Depending on the specific situation, dendrochronolo-

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2612

gy can focus on the study of many different factors

that influence tree growth. Examples include the

following: the frequency of insect defoliation, the

occurrence of severe frosts or fire, or the general

competitional interactions in forest dynamics. The

challenge for the tree-ring researcher is to establish

an optimal representation or reconstruction of the past

variability of the particular factor under study. This

should involve providing realistic estimates of uncer-

tainty, given that in practice many factors can act

together to produce the changing patterns of tree

growth that are measured.

From its beginnings, dendroclimatology recog-

nised the need to undertake careful site selection in

order to maximise the potential climate sensitivity

within sampled tree-ring data. Dendroclimatologists

also had to take account of the bias that might be

introduced into tree-ring chronologies by the temporal

distribution of data drawn from different age trees:

younger trees lay down wider and generally more

dense rings than older trees (Fritts, 1976; Schweing-

ruber, 1988).

This paper sets out to share some insights into how

one particular extensive set of tree-ring data has been

used as evidence of regional and hemispheric-scale

temperature variability, going back hundreds of years

from today. The intention is to illustrate the important

contribution that such data can make in providing well

dated and relatively accurate indications of the vari-

ability of climate and its causes, here using the

evidence of interannual density variations recorded

in wood from locations spread across the Northern

Hemisphere. Issues relating to the statistical process-

ing and climatic calibration of these data are dis-

cussed. These must be borne in mind when reviewing

the accuracy of density-based reconstructions and in

particular when considering the apparent magnitude

of 20th century and other, longer timescale, deduced

temperature changes. As requested, this selected

review draws mainly on our own recent research—

some published, some unpublished—to demonstrate

how the interpretation of tree-ring records like other

proxy climate data is, to some extent, a compromise

and may be subject to recent confounding influences.

For wider dendroclimatic reviews, the reader is

referred to Cook and Kairiukstis (1990), Dean et

al. (1996), and Kaennel Dobbertin and Braker

(2001).

2. The database and inherent temperature signals

Fig. 1 shows an extensive network of nearly 400

locations. From each, multiple tree core samples have

been collected from coniferous tree species and pro-

cessed to provide separate ring width and maximum

(latewood) ring density chronologies (Schweingruber,

1988; Schweingruber and Briffa, 1996). These chro-

nologies are time series of growth indices that repre-

sent the average of the interannual measurements for

one or the other parameter from selected trees at that

site. The choice of which sites to sample was guided

by a belief that the year-to-year growth of trees in

cool, relatively moist areas is likely to reflect a

predominant changing limitation of growing season

temperature, as evidenced by the width of the annual

rings and the density of the wood formed late in the

growing season (Schweingruber et al., 1993; Jacoby

and D’Arrigo, 1995). Hence a range of different tree

species was sampled in northern or relatively high

elevation areas. In this discussion we focus on the

maximum latewood density chronologies (MXD).

Each location in Fig. 1 is represented by a symbol

that shows the magnitude of the local correlation

between MXD and co-located instrumental tempera-

ture records, averaged over an extended (April–Sep-

tember) warm season. Detailed analyses of the

associations between each site chronology and ‘local’

average temperature (Jones et al., 1999) and precipi-

tation (New et al., 2000) measurements for individual

months, preceding and during the tree growing season,

have established that the temperature during an April–

September season is an optimum choice to represent

the climate forcing of tree-ring density variation, when

considering the network as a whole (Briffa et al.,

2002a). [We use ‘local’ to mean the closest grid-box

average climate record.] There are regions where

correlations between MXD series and a shorter sea-

sonal (summer) average temperature are systematically

higher (e.g. June–August or July–August in locations

in central and eastern Siberia) than those for this longer

warm season. There is also some variability in the

strength of correlations with the longer season within

all regions, in part reflecting variability in the temper-

ature responses of different tree species. However, a

ubiquitous statistically significant positive response, in

the form of high density associated with above average

spring and summer warmth, is clearly present over the

Fig. 1. Locations of tree-ring maximum latewood density chronologies, with symbols indicating the correlation between each chronology and its

local grid-box temperature record for the April –September season. The key defines the correlation range associated with each symbol and lists

the number of sites falling in each range. Nine arbitrary regions are defined by the black lines and identified by the acronyms: NEUR= northern

Europe; SEUR= southern Europe; NSIB = northern Siberia; ESIB = eastern Siberia; CAS= central Asia; TIBP=Tibetan Plateau; WNA=wes-

tern North America; NWNA=northwestern North America; ECCA= eastern and central Canada.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 13

whole network. When the MXD and instrumental

observations are each separately averaged within dif-

ferent regions (such as those arbitrarily defined and

shown in Fig. 1), to represent large-scale integrated

series, there is invariably a very high correlation

between them. This is evidence of a likely common

response across the region that is frequently of equiv-

alent, or even greater, magnitude than the strongest

response shown at any of the local sites contained

within the region (Briffa et al., 2002a).

This averaging of different site data to enhance the

expression of an underlying common regional growth-

influence on trees (even of different species) is exem-

plified in Fig. 2. This summarises the seasonal extent

and strength of association between temperature and

tree-ring density across the entire network shown in

Fig. 1. Correlations between all of the density chronol-

ogies and ‘local’ temperature records are summarised

for each of the months, leading up to and during the

period when ring formation occurs. Correlations are

also shown for various mean temperature ‘seasons’.

The shaded arrows (5th and 95th percentile ranges)

indicate the range of individual site temperature corre-

lations. The histograms show the average values of all

these individual site associations. The open circles

show the results achieved when all the local density

Fig. 2. Summary of correlations between individual chronologies

and their monthly or seasonal local grid-box temperature record for

months from the previous June to September of the year of growth,

for the October to September annual mean, and for four seasonal

means. Bars indicate the mean of the local correlations, with the 5th

and 95th percentiles marked by triangles. The open circles are

correlations between the average of all density chronologies and the

average of all the grid-box temperatures at the chronology sites.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2614

chronologies are averaged to form a single series and

this is correlated with the equivalent single record made

up by averaging all of the local temperature series. In

effect, these are the correlations between ‘whole net-

work’ chronologies and averaged temperatures for

nearly the entire Northern Hemisphere land series for

that month or season. We can see, in Fig. 2, little if any

significant influence on tree-ring density during the

months of late summer or winter preceding the forma-

tion of the ring. There is, though, a near universal

positive response across the network to warmth in each

of the spring and summer months.

In some regions, such as northeastern Siberia, a

shorter seasonal (June–August or even July–August)

temperature average displays a stronger association

with density variability (see Briffa et al., 2002a) but

the mean temperature for April–September can be

clearly identified as a very significant influence on

tree-ring latewood density across the whole network,

and as such it represents a justifiable common ‘target’

to use as a predictand in attempts to reconstruct past

temperature variability with these data.

Such a choice is, of course, a statistical compromise

and at a very local or even larger regional scale, other

targets could be used. Defining a rigid calendar-based

season obviously takes no account of any direct

physiological mechanism by which the seasonal

growth of trees is triggered or terminated. The use of

mean monthly data averages is, perhaps, crude but it is

dictated by the need for widely available data. The start

and end of ring growth and seasonal density changes

do not correspond to any particular fixed season. Nor

are they controlled directly or solely by temperature.

The April–September mean temperature is, neverthe-

less, highly correlated with other thermal measures,

even those not calendrically fixed (e.g. cumulative

degree-day sums: Jones and Briffa, 1995), and offers

a very defensible focus for attempts at large-scale

reconstruction.

3. Temperature patterns over six centuries

Figs. 3 and 4 are presented here to illustrate some

initial results that exploit the spatial dimension of this

network in the form of estimates of past April–

September mean temperature patterns. Density chro-

nologies, averaged over individual 5j latitude by 5jlongitude boxes, have been regressed against modern

instrumental temperatures (Jones et al., 1999), simi-

larly averaged in the same boxes, to provide estimates

of past temperature that, in some areas, extend back to

AD 1400 (Briffa et al., 2002b; Osborn et al., in

preparation). Fig. 3 shows maps for the 12 coolest

and 12 warmest years, ranked by the value obtained

when available data are averaged across the whole

network. The spatial coverage of the maps is reduced

in earlier years when fewer locations have tree-ring

data, particularly before the 17th century and particu-

larly in northwest and northeast North America. As the

individual grid-box estimates are independent of each

other, consistency in the anomalies across wide areas

gives additional confidence in the realism of the

inferred changes, above that provided by the high

regression significance. Even in the coldest/warmest

summers there are regions of relative warmth/cold.

Indeed, the maps frequently divide into subcontinen-

tal-scale regions of alternating sign, with western and

Fig. 3. Estimates of warm-season temperature anomalies (jC with respect to the 1961–1990 mean) for the 12 coolest (coldest first) and the 12

warmest (hottest last) summers in the tree-ring density reconstructions.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 15

Fig. 4. Longitude versus time (1400–1960) diagram of reconstructed high-latitude warm-season temperature anomalies (jC with respect to the

1961–1990 mean). Each value represents the average of any reconstructed temperatures poleward of 50jN at each longitude, with decadal

smoothing applied to the time series. The locations of the three main continents are given at the top.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2616

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 17

eastern North America often showing contrasting

anomalies and, similarly, northern Eurasia dividing

into three or four regions of alternating warm and

cold. In some of the coolest overall years, northwest

Europe, including Britain and southern Scandinavia, is

sometimes clearly warm (e.g., in 1698, 1699 and 1884)

while this area is often cool when the overall network

mean is warm (e.g., in 1478, 1686, 1722 and 1931).

In Fig. 4, the reconstructions have been smoothed

using a decadal low-pass filter and meridionally aver-

aged to highlight decadal timescale changes, effective-

ly between 50j and 75jN, at different longitudes overthe time period from AD 1400 to 1960. [The data in

Fig. 4 have additional low-frequency variability to that

shown both in Fig. 3 and by Briffa et al. (2002a)

because the low-frequency regional temperature vari-

ability reconstructed by Briffa et al. (2001) has been

superimposed on the data (Osborn et al., in prepara-

tion)]. The complexity and longitudinal variation of

these temperature changes is apparent. In Europe and

western Siberia, beside the warmth of much of the first

half of the 20th century, the relative warmth of the 15th

and much of the early 16th centuries is clear. There are

also very distinct periods of widespread contempora-

neous cool conditions: the most prominent and persis-

tent began during the last few decades of the 16th

Fig. 5. Estimates of warm-season temperature (jC anomalies from the 196

the 25-year low-pass filtered reconstruction produced using the Age-Ba

Explosivity Index (VEI) is indicated by the arrows at the bottom; ‘?’ mar

century and continued through to the mid-17th century

in Eurasia, perhaps persisting longer, to the end of the

17th century, in North America. It is possible that this

cold occurred earlier in eastern Siberia (near the mid-

16th century) and ‘propagated’ westward. There are

several other, much shorter but very distinct, periods

when cool conditions appeared abruptly and simulta-

neously across all or much of the network: during the

1810s, 1830s, late 1860s and late 1880s (except in

Alaska).

When we average the chronology data across the

whole network and calibrate the resulting series

against average land temperatures north of 20jN(Fig. 5), these periods with extremely cool summers

now stand out. Comparison with years of known large,

explosive volcanic eruptions (identified by Volcanic

Explosivity Index, VEI, values greater than 4 in Fig.

5), provides extremely strong circumstantial evidence

that many widespread cold excursions were a response

to known volcanoes, while other cold summers (such

as in 1695) are strongly suggestive of major eruptions

that have not, as yet, been identified on the basis of

existing historical or geological evidence (Briffa et al.,

1998a).

Note the century-timescale variance in Fig. 5,

emphasised by the smooth curve, which shows a

1–1990 mean) for land areas north of 20jN. The smoothed curve is

nd Decomposition approach of Briffa et al. (2001). The Volcanic

ks those eruptions whose date is uncertain.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2618

gradual cooling trend throughout the 16th century,

persistent cool conditions throughout much of the

17th century, and a slow warming through the 18th

and 19th centuries. This degree of long-timescale

variance is preserved in these data by the use of a

new statistical processing technique (Age-Band De-

composition, ABD, standardisation), required to re-

move age-related bias inherent in the direct

measurements of the tree-ring densities (Briffa et al.,

2001). The bias, which is commonly expressed as a

negative trend in radially measured tree growth param-

eters, is associated with tree life-cycle changes: trees

often laying down wider and denser rings in their

youth and progressively narrower and less dense rings

as they age (Fritts, 1976; Schweingruber, 1988). Some

earlier methods used to correct for this potential bias

removed long-timescale variability in the resulting

chronologies and with it any potential for preserving

slowly evolving climate trends in the reconstructions

produced from them (Cook et al., 1990, 1995; Briffa et

al., 1996). The MXD chronologies used to produce the

Fig. 6. (a) Instrumental temperatures (red) and tree-ring density reconstructi

50jN, smoothed with a 5-year low-pass filter. (b) Pattern of regression coef

curves in (a) and the difference between the grid-box temperature reconstru

decline in tree-ring density over recent decades, and regression slopes >1

local decline is weaker, and < 0 that there is no local decline. (c) Corre

temperatures and area-average of all tree-ring densities, both high-pass filte

year sliding window. The shaded area is bounded by the 5th and 95th perce

horizontal line represents the overall mean of the correlations.

maps in Fig. 3 did not use these new ‘standardization’

techniques. So while they do represent up to multi-

decadal timescale variability, they do not show evi-

dence of multi-century changes. This will not

significantly affect the spatial patterns in relative

temperatures reconstructed, but the regional average

time series produced from them (see Briffa et al.,

2002a) do not show the same extent of long-timescale

variability exhibited in Figs. 4 and 5. For further

discussion of the important ‘standardization’ issue as

it relates to these data see also Briffa et al. (1992, 1996)

and Cook et al. (1995).

4. An inconsistency in the relationship between

tree-ring density and temperature

The near hemispheric scale record of temperature

estimates in Fig. 5 displays a clear underlying cooling

trend during the second half of the 20th century. No

such trend is seen in the summer (or any other

ons of temperature (black) averaged over all land grid boxes north of

ficients between the difference between the smoothed, area-averaged

ctions and observations. The difference is dominated by the relative

indicate the local decline is stronger than average, < 1 indicates the

lation between area-average of all instrumental (April –September)

red to retain only subdecadal variability, and then correlated in a 20-

ntile values of multiple sample correlations of sample size 20 and the

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 19

seasonal) instrumental record. Fig. 6a shows how the

trend in latewood density averaged across all northern

site trees (i.e. selecting the chronologies from north of

50jN) begins to diverge from the April–September

mean temperature record for the same northern land

areas, after about 1960. This phenomenon can be

recognised on larger geographical scales, virtually

across the whole northern sampling network, and has

been illustrated and discussed previously for subcon-

tinental scale regions (Briffa et al., 1998b). Fig. 6b

shows a detailed geographical breakdown of where the

trend of late 20th century tree-ring density falls in

relation to coincident regional temperatures, expressed

relative to the average decline shown in Fig. 6a.

In only a very few grid boxes (in southern Europe,

eastern Canada, northwest and southwest United

States and central Asia) is there no relative density

decline. Over virtually the whole of the remaining

area, it is easily discernible and at three areas in

particular, all at higher latitudes, the decline in density

relative to local temperatures is even stronger than that

represented by the average high-latitude decline: in

north central Canada, northwest Siberia south of

Novaya Zemlya, and northwest central Siberia just

east of the Taimyr Peninsula. Though these differ-

ences in trend are real, they are largely masked at local

scales by the high interannual variability of the

temperature and MXD data that is largely coincident

in both, even in the second half of the 20th century.

This is demonstrated (at the overall hemispheric scale)

in Fig. 6c, which shows the temporal stability in the

strength of the interannual-timescale correlation be-

tween the mean network MXD chronology and

corresponding regionally averaged April–September

mean temperatures through the late 19th and 20th

centuries. The link between these is clearly very firm

and apparently time stable during all of the second

half of the 20th century. However, similar analyses

repeated for each of the major regions that make up

the network show that the strength of the correlations

at high frequency (i.e., based on 10-year high-pass

filtered time series) also displays a degradation from

about 1970 onwards, in a few regions, particularly in

northern Europe and northwest North America (see

also Fig. 14 of Briffa et al., 2002a).

The above facts seem to support an inference that

some slowly varying factor began to exert a very

widespread negative influence on the trend of these

MXD data from around the middle of the 20th

century, with effects at higher frequency also becom-

ing noticeable in some high-latitude regions. For the

time being, we circumvent this problem by restricting

the calibration of the density data to the period before

1960. This reduces the potential overlap between

temperature observations and density measurements

and means that less data can be reserved for indepen-

dent tests of the validity of predictive equations. This

situation is far from ideal, but the alternative, using

data after 1960 and thus incorporating non-tempera-

ture-related bias when fitting regression equations as a

function of density variability, would invariability

produce earlier estimates of past temperature that, to

some extent, too warm.

5. A possible link between tree-ring density

changes and ozone?

There is some, though limited, experimental evi-

dence of the negative effect of enhanced ultra violet

radiation (UV-B) on the photosynthetic process of

some higher plants (e.g., see references in Sullivan,

1994; Tevini, 1994). It has been suggested that an

‘unusual’ increase in the incidence of UV at near-

ground level, a possible consequence of falling con-

centrations of ozone in the stratosphere (DeLuisi et

al., 1994), might be associated with a reduction in tree

productivity (Briffa et al., 1998b; Briffa, 1999). Un-

fortunately, there are virtually no systematic direct

measurements of UV for any length of time across the

hemisphere. Therefore, as a basic first exploration of

the empirical evidence for such an effect, we have

investigated whether there is any correlation between

observations of ozone concentrations and our MXD

data. As no widespread data on stratospheric ozone

exist, we use satellite-based estimates of total column

ozone, provided by the Total Ozone Mapping Spec-

trometer (TOMS: McPeters et al., 1996). These data

do not distinguish between tropospheric and strato-

spheric ozone concentrations, but they are available

for the whole of the Northern Hemisphere, though

unfortunately only from 1979 onwards. Fig. 7a illus-

trates the mean April field for 1979–1993 along with

the spatial distribution of ozone trends over the same

period. Virtually the whole land area north of 40jNdisplays a reduction in total ozone, with the greatest

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2620

falls evident over the northwest and the extreme east

of Siberia, and over the western Arctic Ocean, includ-

ing the northern part of Greenland. This pattern of

negative trends across much of the high Northern

Hemisphere is, in some part, forced by the pattern

of relatively severe ozone reduction that occurred in

1993 (Fig. 7a). However, there is a high interannual

variation in the pattern of ozone concentrations and

relatively high values (above the 1979–93 mean)

occurred over some regions in some years; such as

over Novaya Zemlya in 1988, and north of the Taimyr

peninsula in 1993 (see also Fig. 7a).

Note that all of the single-year patterns and longer-

period mean maps shown in Fig. 7a refer to the

average of April measurements only. We have chosen

April, as it is early in the daylight season in the area of

the northern tree line and so it represents the time

when maximum ozone reduction would be expected

to occur. The recent time-dependent change in the

nature of the seasonal association between tempera-

ture and MXD has also been shown to occur during

the spring (Briffa et al., 2002a).

Obviously, the overlap period between the ozone

measurements and our currently available tree-ring

density chronologies is very short and precludes any

possibility of testing coincidence in common regional

trends, but given the large year-to-year variability in

the ozone and MXD patterns, we considered it worth-

while exploring whether there was any possible sta-

tistically significant link between them.

Rather than compare ozone variability with MXD

data directly, we chose instead to use residuals from a

regression of April–September mean temperature on

MXD data. Hence we seek to test whether years with

low ozone (assumed to be related to higher UV

radiation) are associated with MXD values below

those expected on the basis of April–September mean

seasonal temperatures.

We chose three regions for this initial comparison

(NEUR, NSIB and CAS: see Fig. 1) because they

offered the largest periods of data overlap, and be-

cause they represent a range of regional associations

Fig. 7. (a) Total Ozone Mapping Spectrometer (TOMS) measurements of

trend; middle row, 1988 and 1989 anomalies; and bottom row, 1990 and 19

MXD chronologies (black) superimposed on estimates of these based o

predictor (red), for three regions shown in Fig. 1: northern Europe, northe

are also shown in black with local regional ozone concentration anomalie

between temperature and MXD (strong association in

NEUR and NSIB and weaker association in CAS).

Also note that the interannual correlations in the

NEUR and NSIB regions degrade progressively after

about 1960, as mentioned earlier. The results are

shown in Fig. 7b.

Good correspondence between the temperature and

MXD series is apparent during the 19th and early 20th

centuries, for both the NEUR and NSIB regions, but a

common systematic divergence can be clearly seen in

NEUR beginning in the early 1960s and, similarly, in

NSIB in the mid-1960s. Though the general MXD/

temperature link in CAS (a large and less homoge-

neous temperature region with many fewer chronolo-

gies) is clearly weaker, it is also possible to detect a

systematic difference beginning around 1970. The

short ozone series (the average of April data extracted

over roughly adjacent regions to the tree-ring areas)

exhibit qualitatively similar declining trends within

each region. The correlations with the MXD residual

data, which are strongly influenced by the high

interannual variability as well as the trend, are all

positive, consistent with the notion that reduced ozone

might be associated with reduced MXD. To be sig-

nificant, these correlations would need to be of the

order of 0.44 or above (at a p = 0.05 level, using a one-

tailed test and 13 degrees of freedom). This is not the

case for NEUR (r = 0.25) or NSIB (0.34), though the

correlation for CAS (0.45) is just at this level. Hence

the results are equivocal, though not negative.

Other possible causes of, or contributors to, the

apparent recent dissociation between this, admittedly

crude, seasonal temperature parameter and the MXD

data are possible (Briffa et al., 1998b). An increasing

influence of later snow cover delaying the onset of

seasonal tree growth is one suggestion (Vaganov et al.,

1999). This theory, like that of a possible UV influence,

requires further investigation. This could involve

smaller spatial scales and more refined specification

of the regional ozone data and the use of data for

months other than April. Detailed exploration of the

intercorrelations between temperature and ozone, and

total column ozone in April: top row, 1979–1993 mean and linear

93 anomalies. All anomalies are from the 1979–1993 mean. (b) The

n regression using local April –September mean temperature as a

rn Siberia and central Asia. For each region, the regression residuals

s (arbitrarily scaled here) shown in green.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 21

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2622

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 23

ozone and MXD are also needed, and, where possible,

local studies should be undertaken where longer direct

measures of UVradiation are available (Weatherhead et

al., 1997; Staehelin et al., 1998). Other factors need to

be considered also, such as the effect of cloud cover on

UV penetration (Schafer et al., 1996).

6. Late Holocene Northern Hemisphere

Temperature

All of the tree-ring data taken from the wider

network shown in Fig. 1, processed to retain evidence

of low-frequency temperature forcing (as shown in

Fig. 5), provide information on the variability of warm

season temperatures over a large area of the Northern

Fig. 8. Average temperature over land areas north of 20jN, as observed (b

published series by Jones et al. (1998) in red; Mann et al. (1999) in purple

Esper et al. (2002) in pink. The series used from Mann et al. (1999) wa

resolved reconstructions. Each series was recalibrated over 1881–1960 ag

temperature. Note the effect on the temperature magnitudes in the two sets

predictands.

Hemisphere land, though with increasing uncertainty,

back to AD 1400 (Briffa et al., 2001). Longer series of

similar tree-ring MXD data, but for much more

restricted regions (e.g. individual sites in Sweden,

Russia or western Canada), also contribute to our

knowledge of the history of hemispheric temperature

change through their incorporation within different

multi-proxy amalgamations representing the last 1000

years (e.g., Jones et al., 1998; Mann et al., 1999). A

selection of temperature reconstructions is shown in

Fig. 8. Tree-ring-based reconstructions contained in

these compilations represent a significant proportion

of the palaeoclimate data, especially for the period

before AD 1600. Fig. 8 also includes two reconstruc-

tions made up exclusively from tree-ring width meas-

urements: one (Briffa and Osborn, 1999) is an average

lack) and reconstructed by a simple linear regression recalibration of

; Briffa and Osborn (1999) in green; Briffa et al. (2001) in blue; and

s an average of land grid boxes north of 20jN from their spatially

ainst (a) annual-mean temperature and (b) April–September mean

of series caused by calibrating the same data against these alternative

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2624

of only three regional chronologies from northwest

Sweden (Grudd et al., 2002), and the Yamal (Hante-

mirov and Shiyatov, 2002) and Taimyr (Naurzbaev et

al., 2002) regions of northern Siberia. The other

(Esper et al., 2002) uses data from 14 widespread

sites, many (though not all) independent of those used

in the other compilations.

Many (but in some cases not all) tree-ring width

and MXD data contained in these reconstructions

have been processed to retain long-timescale variance

and together they demonstrate how temperatures

have changed over the centuries, prior to the clear

20th century warming shown in the instrumental

curve. The relative magnitude of these changes varies

between the different compilations, despite the ele-

ment of common data input (see Briffa and Osborn,

1999, 2002), because of different regional concen-

trations of data and because of different approaches

used to assemble and express them quantitatively in

terms of different temperature ‘targets’. Those based

solely on tree-ring data (Briffa and Osborn, 1999;

Briffa et al., 2001; Esper et al., 2002) tend to show

greater relative cold overall, though most clearly in

the 13th and 17th centuries, and possibly slightly

more warmth around AD 1000, but the large uncer-

tainty (not shown here, but see Mann et al., 1999;

Briffa et al., 2001) associated with all of the records

should caution against over-interpretation of these

differences. It is also likely that many of the tree-

ring data respond more to summer rather than winter

conditions and represent more northerly, rather than

full, Northern Hemisphere temperatures. The Mann

et al. (1999) data were calibrated directly against

annual temperature records and contain tree-ring, and

non-tree-ring series that are influenced to a greater

extent by winter conditions than are our MXD data.

Hence, this is likely to be a better indicator of mean

annual temperature change than any of the MXD data

alone (see also Briffa and Osborn, 2002). The ques-

tion of different seasonal responses in palaeoclimate

data is an important one, not least because regres-

sion-based reconstruction of different seasonal data

can affect the relative magnitude (and uncertainty) of

past changes, even using similar predictor data, as

shown in Fig. 8.

However, even allowing for differences in these

various records, they provide strong evidence that the

hemisphere has warmed over the last century, and

particularly the last two decades, to a level that

appears unusual. Even considering the uncertainties

associated with all of these curves, it is not possible to

discount the possibility that this warming is unprece-

dented in the context of the last millennium.

7. Conclusions

This paper was intended as a brief summary of the

authors’ recent work and an illustration of the prog-

ress and potential in using tree-ring density data from

around the Northern Hemisphere. It is, hopefully,

apparent that such work is contributing useful infor-

mation on the past natural variability of late Holocene

climate, but we have clearly pointed to some inter-

pretational problems associated with the use of these

data. While real progress is being made in overcoming

them, several require further study. Tree-ring density

and tree-ring width data will continue to enhance our

detailed knowledge of past temperature and other

climate changes, but a major priority, besides the need

to further develop and expand the networks, is an

urgent requirement to systematically update the exist-

ing sample collections with more recent data to allow

better exploration of the implications of different

chronology production and climate calibration techni-

ques, and to enable more holistic exploration of the

nature and expression of the climate sensitivity of the

data to be undertaken.

Acknowledgements

KRB and TJO acknowledge current support from

the European Commission (under EVK2-CT-2002-

00160{SOAP}) and the UK NERC (under NER/T/S/

2002/00440). We thank Ed Cook and Mike Baillie for

comments on the initial draft of this paper.

References

Briffa, K.R., 1999. Annual climate variability in the Holocene:

interpreting the message of ancient trees. Quaternary Science

Reviews 19, 87–105.

Briffa, K.R., Osborn, T.J., 1999. Seeing the wood from the trees.

Science 284, 926–927.

Briffa, K.R., Osborn, T.J., 2002. Blowing hot and cold. Science

295, 2227–2228.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–26 25

Briffa, K.R., Jones, P.D., Bartholin, T.S., Eckstein, D., Schweing-

ruber, F.H., Karlen, W., Zetterberg, P., Eronen, M., 1992. Fen-

noscandian summers from A.D. 500: temperature changes on

short and long timescales. Climate Dynamics 7, 111–119.

Briffa, K.R., Jones, P.D., Schweingruber, F.H., Karlen, W., Shiyatov,

S.G., 1996. Tree-ring variables as proxy-climate indicators:

problems with low-frequency signals. In: Jones, P.D., Bradley,

R.S., Jouzel, J. (Eds.), Climate Variations and Forcing Mecha-

nisms of the Last 2000 Years. Springer-Verlag, Berlin, pp. 9–41.

Briffa, K.R., Jones, P.D., Schweingruber, F.H., Osborn, T.J.,

1998a. Influence of volcanic eruptions on Northern Hemisphere

summer temperatures over the past 600 years. Nature 393,

450–455.

Briffa, K.R., Schweingruber, F.H., Jones, P.D., Osborn, T.J., Shiya-

tov, S.G., Vaganov, E.A., 1998b. Reduced sensitivity of recent

tree-growth to temperatures at high northern latitudes. Nature

391, 678–682.

Briffa, K.R., Osborn, T.J., Schweingruber, F.H., Harris, I.C., Jones,

P.D., Shiyatov, S.G., Vaganov, E.A., 2001. Low-frequency tem-

perature variations from a northern tree ring density network.

Journal of Geophysical Research 106, 2929–2941.

Briffa, K.R., Osborn, T.J., Schweingruber, F.H., Jones, P.D., Shiya-

tov, S.G., Vaganov, E.A., 2002a. Tree-ring width and density

around the Northern Hemisphere: Part 1. Local and regional

climate signals. Holocene 12, 737–757.

Briffa, K.R., Osborn, T.J., Schweingruber, F.H., Jones, P.D., Shiya-

tov, S.G., Vaganov, E.A., 2002b. Tree-ring width and density

around the Northern Hemisphere: Part 2. Spatio-temporal varia-

bility and associated climate patterns. Holocene 12, 759–789.

Cook, E.R., Kairiukstis, L.A. (Eds.), 1990. Methods of Dendrochro-

nology: Applications in the Environmental Sciences. IIASA/

Kluwer, Dordrecht. 394 pp.

Cook, E.R., Briffa, K.R., Shiyatov, S.G., Mazepa, V.S., 1990. Tree-

ring standardisation and growth-trend estimation. In: Cook,

E.R., Kairiukstis, L.A. (Eds.), Methods of Dendrochronology:

Applications in the Environmental Sciences. IIASA/Kluwer,

Dordrecht, pp. 104–123.

Cook, E.R., Briffa, K.R., Meko, D.M., Graybill, D.A., Funkhouser,

G., 1995. The ‘Segment Length Curse’ in long tree-ring chro-

nology development for palaeoclimatic studies. Holocene 5,

229–237.

Dean, J.S., Meko, D.M., Swetnam, T.W. (Eds.), 1996. Tree Rings,

Environment and Humanity. Radiocarbon. Dept. of Geoscien-

ces, University of Arizona, Tucson.

DeLuisi, J.J., Mateer, C.L., Theisen, D., Bhartia, P.K., Longenecker,

D., Chu, B., 1994. Northern middle-latitude ozone profile fea-

tures and trends observed by SBUYand UMKEHR, 1979–1990.

Journal of Geophysical Research 23, 2625–2628.

Esper, J., Cook, E.R., Schweingruber, F.H., 2002. Low-frequency

signals in long tree-ring chronologies for reconstructing past

temperature variability. Science 295, 2250–2253.

Fritts, H.C., 1976. Tree Rings and Climate. Academic Press, London.

567 pp.

Grudd, H., Briffa, K.R., Karlen, W., Bartholin, T.S., Jones, P.D.,

Kromer, B., 2002. A 7400-year tree-ring chronology in northern

Swedish Lapland: natural climatic variability expressed on an-

nual to millennial timescales. Holocene 12, 657–666.

Hantemirov, R.M., Shiyatov, S.G., 2002. A continuous multimillen-

nial ring-width chronology in Yamal, northwestern Siberia. Hol-

ocene 12, 717–726.

Jacoby, G.C., D’Arrigo, R.D., 1995. Tree ring width and density

evidence of climate and potential forest change in Alaska. Global

Biogeochemical Cycles 9, 227–234.

Jones, P.D., Briffa, K.R., 1995. Growing season temperature over

the former Soviet Union. International Journal of Climatology

15, 943–959.

Jones, P.D., Briffa, K.R., Barnett, T.P., Tett, S.F.B., 1998. High-

resolution palaeoclimate records for the last millennium: inter-

pretation, integration and comparison with general circulation

model control-run temperatures. Holocene 8, 455–471.

Jones, P.D., New, M., Parker, D.E., Martin, S., Rigor, L.G., 1999.

Surface air temperature and its variations over the last 150 years.

Reviews of Geophysics 37, 173–199.

Kaennel Dobbertin, M., Braker, O.U. (Eds.), 2001. Tree Rings and

People. International Conference on the Future of Dendrochro-

nology, Davos, 22–26 September 2001, Abstracts. Swiss Fed-

eral Research Institute WSL, Birmensdorf. 278 pp.

Mann, M.E., Bradley, R.S., Hughes, M.K., 1999. Northern Hemi-

sphere temperatures during the past millennium: inferences, un-

certainties and limitations. Geophysical Research Letters 26,

759–762.

McPeters, R.D., Bhartia, P.K., Krueger, A.J., Herman, J.R., Schle-

singer, B.,Wellemeyer, C.G., Seftor, C.J., Jaross, G., Taylor, S.L.,

Swissler, T., Torres, O., Labow, G., Byerly, W., Cebula, R.P.,

1996. Nimbus 7 total ozone mapping spectrometer (TOMS) data

products user’s guide. NASA Reference Publication, vol. 1384.

April.

Naurzbaev, M.M., Vaganov, E.A., Sidorova, O.V., Schweingruber,

F.H., 2002. Summer temperatures in eastern Taimyr inferred

from a 2427-year late-Holocene tree-ring chronology and earlier

floating series. Holocene 12, 727–736.

New, M., Hulme, M., Jones, P.D., 2000. Representing twentieth-

century space– time climate variability: Part II. Development of

1901–1996 monthly grids of terrestrial surface climate. Journal

of Climate 13, 2217–2238.

Osborn, T.J., Briffa, K.R., Schweingruber, F.H., Jones, P.D., 2003.

Annually resolved patterns of summer temperatures over the

Northern Hemisphere since A.D 1400 from a tree-ring-density

network. In preparation.

Schafer, J.S., Saxena, V.K., Wenny, B.N., Barnard, W., DeLuisi,

J.J., 1996. Observed influence of clouds on ultraviolet-B radia-

tion. Geophysical Research Letters 23, 2625–2628.

Schweingruber, F.H., 1988. Tree Rings Basics and Applications of

Dendrochronology. Kluwer Academic Publishing, Dordrecht.

276 pp.

Schweingruber, F.H., Briffa, K.R., 1996. Tree-ring density networks

for climate reconstruction. In: Jones, P.D., Bradley, R.S., Jouzel,

J. (Eds.), Climate Variations and Forcing Mechanisms of the Last

2000 Years. Springer-Verlag, Berlin, pp. 43–66.

Schweingruber, F.H., Briffa, K.R., Nogler, P., 1993. A tree-ring

densitometric transect from Alaska to Labrador: comparison of

ring-width and maximum-latewood-density chronologies in the

conifer belt of northern North America. International Journal of

Biometeorology 37, 151–169.

K.R. Briffa et al. / Global and Planetary Change 40 (2004) 11–2626

Staehelin, J., Renaud, A., Bader, J., McPeters, P., Viatte, P.,

Hoegger, B., Burgnion, M., Giroud, M., Schill, H., 1998.

Total ozone series at Arosa (Switzerland) homogenisation and

data comparison. Journal of Geophysical Research (D5) 103,

5827–5841.

Sullivan, J.H., 1994. Temporal and fluence responses of tree foliage

to UV-B radiation. In: Hilton Biggs, R., Joyner, M.E.B. (Eds.),

Stratospheric Ozone Depletion/UV-B Radiation in the Bio-

sphere. NATO ASI Series, vol. I 18. Springer-Verlag, Berlin,

pp. 67–76.

Tevini, M., 1994. Physiological changes in plants related to UV-B

radiation: an overview. In: Hilton Biggs, R., Joyner, M.E.B.

(Eds.), Stratospheric Ozone Depletion/UV-B Radiation in the

Biosphere. NATO ASI Series, vol. I 18. Springer-Verlag, Berlin,

pp. 38–56.

Vaganov, E.A., Hughes, M.K., Kirdyanov, A.V., Schweingruber,

F.H., Silkin, P.P., 1999. Influence of snowfall and melt timing

on tree growth in subarctic Eurasia. Nature 400, 149–151.

Weatherhead, E.C., Tiao, G.C., Reinsel, G.C., Frederick, J.E., De-

Luisi, J.J., Choi, D.S., Tam, W.K., 1997. Analysis of long-term

behaviour of ultraviolet radiation measured by Robertson–

Berger meters at 14 sites in the United States. Journal of Geo-

physical Research (D7) 102, 8737–8754.