Low-Frequency Signals in Long Tree-Ring Chronologies for Reconstructing past

Temperature Variability
Author(s): Jan Esper, Edward R. Cook and Fritz H. Schweingruber
Source: Science, New Series, Vol. 295, No. 5563 (Mar. 22, 2002), pp. 2250-2253
Published by: American Association for the Advancement of Science
Stable URL: http://www.jstor.org/stable/3076339
Accessed: 17-09-2016 07:45 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/3076339?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

American Association for the Advancement of Science is collaborating with JSTOR to digitize,
preserve and extend access to Science

This content downloaded from 128.119.168.112 on Sat, 17 Sep 2016 07:45:37 UTC
All use subject to http://about.jstor.org/terms
 REPORTS

These growth trends occur almost universally

Low-Frequency Signals quently

indescribeLong a downward trend of ring
in "raw" tree-ring measurement series and fre-

width with increasing age. Dendrochronologists

Tree-Ring Chronologies for
usually eliminate these growth trends by de-
trending each tree-ring width series with a fitted

Reconstructing Past smooth mathematical growth function. In such

cases, the maximum wavelength of recoverable

Temperature Variability
by the segment lengths of the individual de-
climatic information is fundamentally limited

trended series (7). Thus, a 100-year-long tree-

Jan Esper,' Edward R. Cook,2* Fritz H. Schweingruber1
ring series will not contain any climatic vari-
Preserving multicentennial climate variability in long tree-ring records
ance at periods is crit-
longer then 100 years if it is
explicitly detrended
ically important for reconstructing the full range of temperature by a fitted
variability over growth curve.
the past 1000 years. This allows the putative "Medieval Warm Period"
Consequently, (MWP)
the problem of missing long-
to be described and to be compared with 20th-century warming
term trends in in modeling
millennia-length tree-ring chro-
and attribution studies. We demonstrate that carefully selected tree-ring
nologies is due chro-series that
to using detrended
are short
nologies from 14 sites in the Northern Hemisphere (NH) relative to the multicentennial
extratropics can fluctu-
ations duetrends
preserve such coherent large-scale, multicentennial temperature to climateif
(8).prop-
Exceptions are chro-
er methods of analysis are used. In addition, we show that thebuilt
nologies average
with 1000of
yearthese
or longer individ-
occurrence of the MWP over the NH
chronologies supports the large-scale ual tree-ring series (9, 10) and chronologies
extratropics. developed by Regional Curve Standardization
(RCS) (11) or Age Banding (12) methods.
Much of the current debate on the Earth's cli- Several of the tree-ring collections analyzed
distributed over a large part of the NH extrat-
mate variability is driven by the observation of ropics (Fig. 1). Tree species represented inhere
thishave been described and used previously
a modem, century-long temperature increase, collection are from the genera Picea, Pinus, in individual and large-scale temperature recon-
culminating with the last decade of the 20th Larix, and Abies. Using these data, we demon-structions and related studies (11, 13-19). Ring
widths of trees growing in cold environments
century as the warmest since 1856 (1). These strate that multicentennial temperature informa-
usually reflect the influence of warm-season
dramatic recent temperature changes have been tion can be preserved in long tree-ring records
related to those of the last millennium by the provided that the data are properly processed
temperatures
to on growth most strongly. Howev-
Mann-Bradley-Hughes (MBH) multiproxy re- preserve such low-frequency information due er, to
in some cases, they also reflect temperatures
from the cool-season months before the radial
construction of NH annual temperatures (2). By climate. We also show that the MWP was likely
combining instrumental temperature data with to have been a large-scale phenomenon ingrowththe season as well (20). Here, we will not
long, temperature-sensitive proxy records, the NH extratropics that appears to haveexplicitly
ap- model the temperature signals of the
MBH reconstruction indicates that the 20th- individual tree-ring chronologies, because this
proached, during certain intervals, the magni-
century warming is abrupt and truly exception- tude of 20th-century warming, at least up hastomostly been done already (11, 13-19).
al. It shows an almost linear temperature de-1990. Rather, we will demonstrate the preservation of
crease from the year 1000 to the late 19th Most millennia-long tree-ring chronologies coherent multicentennial variability among the
century, followed by a dramatic and unprece- are averages of many tree-ring series from14 tree-ring sites, which is inferred to reflect
liv-
ing and dead trees. The segment lengths
dented temperature increase to the present time. of
large-scale, multicentennial temperature chang-
The magnitude of warmth indicated in the es over the past 1000 years in the NH extrat-
these series are typically 200 to 400 years long,
MBH reconstruction for the MWP, -1000- ropics. These inferred changes are almost cer-
and the overlapping individual series are exact-
ly aligned by calendar year and connected
1300 (3, 4), is uniformly less than that for most tainly
in weighted toward the warm-season
of the 20th century. time using a method known as "cross-dating"
months, as some previous studies have shown
The MBH reconstruction has been criticized (6). The difficulty of preserving multicenten-
(11, 13, 18, 19). Even so, low-frequency warm-
(5) for its lack of a clear MWP. Critics argue season
nial variation in such tree-ring series, when the and annual temperature trends recorded
segment lengths are substantially shorter in
that tree-ring records, the substantial basis of the NH instrumental data are statistically indis-
than
MBH reconstruction before the 17th century, tinguishable
the length of the overall chronology being de- (8).
cannot preserve long-term, multicentennial tem- To
veloped, results from the removal of age-related preserve possible multicentennial
perature trends. This contention is of fundamen- growth
biological growth trends that represent noise for trends due to climate, we analyzed the
tal importance because if tree-ring reconstruc-the purpose of climatic reconstruction (6).
individual raw ring-width measurements using
tions are limited in this way, then including such
records in hemispheric estimates of past temper-
atures would bias the results as argued (5). To
illustrate that this need not be the case, we
0)
present the analysis of centuries-long ring-width
trends in 1205 radial tree-ring series from 14
high-elevation and middle-to-high latitude sites -j

1Swiss Federal Research Institute WSL, Zuercher-

strasse 111, 8903 Birmensdorf, Switzerland. 2Lamont-
Doherty Earth Observatory, Palisades, NY 10964, Longitude
USA.
Fig. 1. Map of the 14 tree ring sampling sites. Ath, Athabasca; Bor, Boreal; Mac, Mackenzie
Quebec; Upp, Upperwright; Got, Gotland; Jae, Jaemtland; Tir, Tirol; Tor, Tornetraesk; Man
*To whom correspondence should be addressed. E-
mail: drdendro@ldeo.columbia.edu gazeja; Mon, Mongolia; Pol, Polar Urals; Tai, Taimir; and Zha, Zhaschiviersk.

2250 22 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org

This content downloaded from 128.119.168.112 on Sat, 17 Sep 2016 07:45:37 UTC
All use subject to http://about.jstor.org/terms
 REPORTS

RCS, a method that has been used previously "linear" form (443 series) and one with age
1300), although
for reconstructing long-term temperature vari- of
trends that are more "nonlinear" (762 peak
series) (8). condit
ability from tree rings (11). Successful use of The data sets were divided this weakness
way because of sa
the RCS method generally requires a large (Fig.
differences in growth levels and slopes can bias 2B), and
number of ring-width series because the meth- resentation
resulting RCS chronologies (21). For example, as
od of detrending is not based on any explicit back
according to their mean age trends (8), the in time (
curve-fitting to the individual series as de- young nonlinear trees grow 2 to 3higher times faster spatial
scribed above. Rather, a single mean biological than the linear trees up until 200 years in the
of age. NH ex
growth curve, estimated from all the data, is Two smoothed RCs were estimated from collections tha
used (8). Consequently, systematic departures the averaged biological age-aligned data in the are long and w
of actual growth of individual series from the linear and nonlinear groups. Tree-ring depar- hypothesis is c
regional curve (RC) are common. Indeed, this tures from each RC were calculated as ratios for After the ye
is why the RCS method can preserve trends in the linear and nonlinear data sets following linear RCS ch
excess of the lengths of the individual records standard procedures (6). The resulting tree-ring markably wel
being detrended (7, 11). When systematic de- indices were then averaged into linear and non- time scales. T
partures from the RC occur commonly among linear mean value functions to produce two ferred below
series in a given time period, this may reflect a nearly independent tree-ring chronologies cov- much of the
low-frequency change in climate forcing on ering the years 800-1990 (Fig. 2A). For each be regarded a
growth. chronology, the changes in sample size each of the Little
To build RCS chronologies from the whole year are indicated in Fig. 2B, with a more year 1850, la
data set that contains different sites and species, detailed description of changing individual site extratropics i
we analyzed the growth levels and trends of the contributions provided in (8). The two chronol- instrumental
individual ring-width series after aligning them ogies are very similar over the past -1200 all, the broad
by cambial age and classifying them into two tennial
years. Each shows evidence for inferred above- varia
groups: one with age trends that have a weakly average temperatures during the MWP (900- nonlinear ch
properly sele
records can p
Fig. 2. RCS chronolo- A Linear and non climate
linear variab
RCS
gies of linear and non- 2
linear classified trees
x
- LINEAR
contention (
(A), the yearly sample c 1.5
NON LINEAR| records can no
tuations due to

y i l|^^
size for each chronol-
C

ogy (B), the 20-year We also provi

smoothed NH extrat- *I
linear and non
0.5
ropics reconstruction with two-tailed
of radial stem produc- 0.5
vals (8). In thi
tivity in high elevation
and high latitude forest
t B Sample size per year - 300 - and its confid
environments since /-^ \^^^^^-^~-~- 200 aa rescaled to re
-800 (black) and two- 1 00 E increment (m
tailed 95% bootstrap , .I II ,. II I,I II II ,. I
I I I
. I II II II II I II II I, ,I -
I I 0 c
-
some idea abou
confidence intervals
sequestration in
(blue) (C), and the
number of chronolo- sensitive trees
gies available for the growth that
reconstruction each from the long
year (D). The NH re- clude episodes
construction was de-
century. The c
rived from the 14 site
dence limits a
RCS chronologies after
each was smoothed anomalous red
with a 20-year low- found at some
pass filter to emphasize - D Number of chronologies per year -- 15 sites (24). Reg
multidecadal to multi- 10 a
provide compe
centennial time scales. -5 quency growth
The two-tailed 95% a)
14 tree-ring si
confidence limits were 800 1000 1200 1400 1600 1800 2000
estimated with the use Years
to large-scale
of a bootstrap proce- NH extratropi
dure (8). Even though
signal is undo
warm-season
Table 1. Correlations between instrumental mean annual temperatures (0? to 900N) and the MBH
brate
reconstruction and mean RCS tree-ring chronology before and after digital filtering. The filtered series are
it as an
40-year low-pass (MBH series ends in 1977), 20-year low-pass, and 20-year high-pass. The comparison
time period to
is 1856-1980. Orig., unfiltered data; LP, low-pass; HP, high-pass. correlation (
and warm-sea
Series Orig. 40-year LP 20-year LP 20-year HP tions between
NH (0? to 90?
MBH 0.82 0.96 0.95 0.59
RCS 0.59 0.88 0.84 0.15
temperatures
period indicat

www.sciencemag.org 2251
SCI

This content downloaded from 128.119.168.112 on Sat, 17 Sep 2016 07:45:37 UTC
All use subject to http://about.jstor.org/terms
 REPORTS

signal in the tree rings at interdecadal and RCS. 'Correlations between RCS and MBH uncalibrated tree-ring data. MBH has two data
longer time scales (40-year low-pass, 20-year (Table 2) indicate that this disagreement issets in common with RCS (Tometraesk and
low-pass), but not for interannual to decadal largely confined to periods longer than 200Polar Urals) that are largely restricted to the
variability (20-year high-pass). The low-pass years. At shorter periods, the RCS and MBHnonlinear group (8). Yet the linear and nonlinear
RCS correlations also compare favorably series are almost synchronous. These results areRCS chronologies are extremely similar (Fig.
with those for the MBH reconstruction (Ta- consistent with comparisons of MBH to other2A). Thus, the noted similarities between RCS
ble 1). In contrast, the RCS 20-year high-pass long proxy temperature reconstructions (25). and MBH are unlikely to be due to the small
correlation is noticeably weaker than that of We have demonstrated how multicentennialamount of data overlap between the series.
MBH. However, MBH contains some long- temperature variability in long tree-ring records Where the two series disagree is on
term instrumental measurements (2), which can be preserved if the appropriate tree-ringmulticentennial time scales, which relates
undoubtedly improves its relation with NH data and proper methods of analysis are used.to the criticism noted earlier (5). The MBH
instrumental temperatures at all frequencies. This result is not new (11, 13, 18), but the wayreconstruction includes temperature esti-
Figure 3 shows the mean RCS chronology, it has been demonstrated is somewhat novel mates from the tropical and subtropical NH
rescaled now to estimates of NH annual tem- and instructive. In so doing, evidence for a (2), which is not represented in the RCS
perature (to be compatible with MBH) and the large-scale MWP (sensu lato) has been recon- record. This may explain some of the ob-
MBH temperature reconstruction itself (8), structed, and it approaches the magnitude of served differences. Much of the multicen-
each with 95% confidence limits. Each series 20th-century warming in the NH up to 1990. tennial variability in MBH has also been
has been smoothed with a 40-year low-pass Consistent with other analyses of the MWP (4, replicated by an energy balance model that
filter to emphasize the multidecadal to multi- 22) and with our comparisons of the linear and includes solar, volcanic aerosol, anthropo-
centennial changes in temperature. The 40-year nonlinear RCS chronologies (Fig. 2A), the genic aerosol, and greenhouse gas forcing
low-pass filtered mean RCS chronology was MWP appears to be more temporally variable (26). Therefore, the large multicentennial
scaled to NH annual temperatures using 1900- than the warming trend of the last century. Our differences between RCS and MBH are real
1977 as the calibration period. This time period analysis also indicates that the MWP in NH and would seem to require a NH extratropi-
was chosen because it represents the interval extratropics may have begun in the early 900s cal forcing to explain them, one that atten-
most similar between RCS and MBH (r = (Fig. 2C). The warmest period covers the inter- uates toward the equator. One candidate is
0.94). For comparison, the correlations of RCS val 950-1045, with the peak occurring around the 1000- to 2000-year climate rhythm
and MBH with 40-year low-pass NH (0? to 990, a result consistent with another analysis (1470 ? 500 years) in the North Atlantic,
90?) annual temperatures are 0.98 and 0.97, using some of the same tree-ring data (18). This which may be related to solar-forced
respectively. These very high correlations are finding suggests that past comparisons of the changes in thermohaline circulation (27,
because the filtered data have similar trends in MWP with the 20th-century warming back to 28). The degree to which this mode of
the 20th century. the year 1000 (19, 22) have not included all of climate forcing is responsible for the mul-
When the mean RCS chronology is ex- the MWP and, perhaps, not even its warmest ticentennial variations in RCS requires fur-
pressed this way, striking differences in low- interval. ther investigation.
frequency behavior with MBH are revealed. The good agreement between the RCS and
This comparison suggests that MBH is not MBH records at all time scales except multicen-
References and Notes
necessarily missing a MWP. Rather, it has a tennial is remarkable, given that the mean RCS 1. P. D. Jones, M. New, D. E. Parker, S. Martin, I. G. Rigor,
reduced expression of the LIA compared with chronology is simply a large-scale average of Rev. Geophys. 37, 173 (1999).
2. M. E. Mann, R. S. Bradley, M. K. Hughes, Geophys. Res.
Let. 26, 759 (1999).
Fig. 3. Comparison of 2 . . . . . . . .. 3. H. H. Lamb, Palaeogeogr. Palaeoclim. Palaeoecol. 1,
the MBH NH temper- - - MBH (40YR LOW-PASS) 13 (1965)
ature reconstruction -RCS (40YR LOW-PASS) 4. M. K. Hughes, H. F. Diaz, Clim. Change 26, 109
with the mean RCS ,, , (1994).
chronology scaled to ,, ^. ,- - 5. W. S. Broecker, Science 291, 1497 (2001).
the MBH record using ? ' 6. E. R. Cook, L. A. Kairiukstis, Eds., Methods of Dendro-
chronology (Kluwer Academic Publishers, Dordrecht,
the 1900-1977 period o -/ , -\^ ;' ,
Netherlands, 1990).
for calibration. The -1-" -
7. E. R. Cook, K. R. Briffa, D. M. Meko, D. A. Graybill, G.
RCS confidence inter- ;--'
Funkhouser, The Holocene 5, 229 (1995).
vals are estimated us-
8. Supplementary details are available on Science On-
ing the bootstrap 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 line at www.sciencemag.org/cgi/content/full/295/
method (8). Each se- Year 5563/2250/DC1.
ries (solid lines) and 9. V. C. LaMarche, Science 183, 1043 (1974).
its confidence intervals (dashed lines) have been smoothed
10. J. Esper, with
F. H. Schweingruber, M. 40-year
Winiger, The Holo-
emphasize the multidecadal to multicentennial timecene 12, 267 (2002).
scales. The filtered M
11. K. R. Briffa et al., Clim.
supplied by M. E. Mann. Correlations between unfiltered and Dyn.filtered
7, 111 (1992). versio
12. K. R. Briffa et al.,J. Geophys. Res. 106, 2929 (2001).
(Table 2) show that the mean RCS chronology is well related to the MBH
13. K. R. Briffa, P. D. Jones, F. H. Schweingruber, S. G.
multidecadal to centennial time scales. However, the two disagree at multicen
Shiyatov, E. R. Cook, Nature 376, 156 (1995).
14. B. H. Luckman, K. R. Briffa, P. D. Jones, F. H. Schwein-
Table 2. Correlations between the mean RCS chronologygruber, and the
The Holocene MBH
7, 375 (1997). temperatur
two time periods and for different bandwidths of 15.
variability. The
A. H. Lloyd, L. J. Graumlich, first
Ecology interval
78, 1199 (1997).
associated with the MWP and 20th-century warming 16.
where
R. D'Arrigo the greatest
et al., Geophys. Res. Lett. differences
28, 543 (2001).
are evident. The second time period covers the 17.
fullJ. M. Szeicz,
overlap G. M. MacDonald,
of the Quat. Res.
two 44, 257reco
(1995). do we see a lack of agreem
multicentennial fluctuations (200 YR LP) are correlated
18. K. R. Briffa, Quat. Sci. Rev. 19, 87 (2000).
19. P. D. Jones, K. R. Briffa, T. P. Barnett, S. F. B. Tett, The
Interval Orig. 200-year LP 200-year HP 20- to 200-year BP 20-year HP
Holocene 8, 477 (1998).
1100-1850 0.30 -0.37 0.47 0.68 0.24 20. G. C. Jacoby, R. D. D'Arrigo, T. Davaajamts, Science
273, 771 (1996).
1000-1980 0.40 0.43 (0.09*) 0.41 0.60 0.24
21. J. Esper, E. R. Cook, K. Peters, Tree-Ring Res., in press.
*Correlation minus the 1851-1980 data. 22. T. J. Crowley, T. S. Lowery, Ambio 29, 51 (2000).

2252 22 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org

This content downloaded from 128.119.168.112 on Sat, 17 Sep 2016 07:45:37 UTC
All use subject to http://about.jstor.org/terms
 REPORTS

23.
23.J. M.
J.Grove,
M. The
Grove,
Uttle Ice Age
The (Methuen,
Uttle London,
Ice Age (Methuen,
Graumlich London,
and A. Lloyd (Boreal and Upperwright), J. with K. Briffa, P. Jones, and T. Osbom on error estima-
1987). Szeicz (Mackenzie), S. Payette and L Filion (Quebec), T. tion and thank other colleagues for vigorous discussions
24. K. R. Briffa et al., Nature 391, 678 (1998). Bartholin and W. Karlen (Gotland, Jaemtland, Tome- on various aspects of this paper. This work was support-
25. P. D. Jones, T. J. Osborn, K. R. Briffa, Science 292, 662 traesk), V. Siebenlist-Kemer (Tirol), S. Shiyatov (Polar ed by the Max Kade Foundation, Inc, during a research
(2001). Urals, Mangazeja), G. Jacoby (Mongolia), M. Naurzbaev visit at the Lamont-Doherty Earth Observatory of Co-
26. T. J. Crowley, Science 289, 270 (2000). (Taimir), and F. Schweingruber (Zhaschiviersk), who lumbia University in New York (J.E.). Lamont-Doherty
27. G. Bond et al., Science 278, 1257 (1997). sampled the tree-ring data sets in 14 different regions Earth Observatory Contribution No. 6301.
28. G. Bond et al., Science 294, 2130 (2001). of the Northem Hemisphere and/or made them avail-
29. We gratefully acknowledge B. Luckman (Athabasca), L able to this study. We also appreciate the discussions 12 September 2001; accepted 11 February 2002

Coda Wave Interferometry for

The perturbation
perturbation in
in the
themedium
mediumcan
canbe
be
retrieved from the cross correlation of the
coda waves recorded before and after the

Estimating Nonlinear Behavior perturbation. The unperturbed wave field

u,p(t) can be written as a Feynman path
in Seismic Velocity summation (10) over all possible paths P

Roel Snieder, Alexandre Gr6t, Huub Douma, John Scales np(t) = ApS(t - tp), (1)

In coda wave interferometry, one records multiply scattered waves at a limited where
wherea apath
path
is defined
is defined
as a sequence
as a sequence
of of
number of receivers to infer changes in the medium over time. With this scatterers
scatterers that
that
is encountered,
is encountered,
tp is the
tptrav-
is the trav-
technique, we have determined the nonlinear dependence of the seismic ve- el
el time
timealong
alongpath
path
P, Ap
P,isAptheiscorrespond-
the correspond-
locity in granite on temperature and the associated acoustic emissions. This ing
ingamplitude,
amplitude,andand
S(t) is
S(t)
theissource
the source
wave- wave-
technique can be used in warning mode, to detect the presence of temporal let.
let.When
When thethe
perturbation
perturbationof theofscatterer
the scatterer
changes in the medium, or in diagnostic mode, where the temporal change in locations (or source location) is much
the
themedium
medium is quantified.
is quantified. smaller than the mean free path, the effect
of this perturbation on the geometrical
In
In many
manyapplications,
applications,
suchsuch
as nondestruc-
as nondestruc- tion
tionisis1/30
1/30of of
thethe
dominant
dominant
wavelength and is spreading
wavelength and is and the scattering strength can be
tive
tivetesting
testingor or
monitoring
monitoringof volcanoes
of volcanoes
or or uncorrelated
uncorrelated between
between
scatterers
scatterers
(9). (9). ignored, and the dominant effect on the
radioactive
radioactive waste
wastedisposal
disposal
sites,sites,
one is one
pri- is pri- In
In this
thisexample,
example,the the
scatterers'
scatterers'
locations waveform arises from the change in the
locations
marily
marilyinterested
interested in detecting
in detecting
temporal
temporal are
areperturbed.
perturbed.In general,
In general,
a perturbation
a perturbation
can travel
cantime Tp of the wave that travels along
changes
changesinin thethe
structure
structure
of theofmedium.
the medium. involve
involveotherotherchanges
changes
in the
inmedium
the medium
or a eachor a
path
Temporal
Temporalchanges
changesin Earth's
in Earth's
structure
structure
that that change
changeinin source
source
location.
location.
We refer
We torefer
the to the
accompany earthquakes have been ob- waveform
waveform before
before
the the
perturbation
perturbation
as the as the
Uper(t) = ApS(t - tp - Tp). (2)
P
served on the basis of the attenuation of
unperturbed
unperturbed signal
signal
and and
to thetowaveform
the waveform
coda waves (1), on the arrival times of thethe
after
after theperturbation
perturbationas theasperturbed sig- Thesig-
the perturbed time-windowed correlation coefficient is
directly arriving waves (2), on velocity
nal.
nal.For
Forearly
early
times
times
(t < 0.04
(t < s),
0.04
thes),
waves
the computed
waves from
changes inferred from later arriving waves
in
in Fig.
Fig.1 1
have
havenotnot
scattered
scattered
often,often,
rendering
rendering
R(t ,.(ts) =
(3) [see also (4)], and on changes in seis-
the
thepath
path lengths
lengthsof these
of these
waveswaves
insensitive
insensitive
t t+T
+ T
mic anisotropy (5). Here, we introduce to
to the
thesmall
small perturbations
perturbationsof theofscatterers
the scatterers
coda wave interferometry whereby multi- (small
(smallcompared
compared withwith
the dominant
the dominant
wave- wave- unp(t')uper(t' + ts)dt'
ply scattered waves are used to detect length
length
tem- X X = 2.5
= 2.5
m),m),
whichwhich
causescauses
the unper-
the unper-t-T
poral changes in a medium by usingturbed the and
turbed andperturbed
perturbedsignals
signals
to be similar.
to be similar.

(
t+ T t+ T 11 1/2

scattering medium as an interferometer.However,

However,
For thethemultiply
multiplyscattered
scattered
waves are
waves are UMnp (t
quasi-random perturbations of the positions
increasingly
increasingly sensitive
sensitive
withwith
time to
time
the to
per-the per-
fi t-T I t-T

of point scatterers, or for a change inturbations

the of the scatterer locations because
(3)
source location or the wave velocity,the esti-
waves bounce more often among scat-
terers as time increases. The correlation
mates of this perturbation can be derived where the time window is centered at time t with
from multiply scattered waves by a between
cross the unperturbed and perturbed duration
sig- 2T and ts is the time shift used in the
correlation in the time domain. nals, therefore, decreases with increasingcross correlation. When Eqs. 1 and 2 are insert-
time.
In the numerical example (Fig. 1), the wave ed, double sums Spp, over all paths appear. In
field for a medium consisting of isotropic point
scatterers is computed with the use of a deter-
Fig.
Fig.
1. Location 1.
of 100 Location of 100
scatterers
ministic variant (6, 7) of Foldy's method (8).scatterers before and o o , before and o o ,
after
Given the mean free path (1 = 20.1 m) and the after
the perturbation
the_% 5o o perturbation _% 5o o
wave velocity (v = 1500 m/s), one can infer
that after t = 5.4 X 10-2 s the waves are on
average scattered more than three times. The
later part of the signal is called the coda. Sup-
pose that one repeats this multiple scattering the sake of laritythe 4 , ? ? Strong multiple scattering
experiment after the scatterer locations are per- scatterer displace- d o IIIp
turbed. The perturbation in the scatterer loca-
ment is exaggerated
by a factor 40. The 0.02 0.04 0.06 0.08 0.10
scatterers are placed t(s)
Department of Geophysics and Center for Wave Phe-in an area of 40 m by
nomena, Colorado School of Mines, Golden, CO
80401, USA. right with a solid and dashed line, respec

www.sciencemag.org SCIENCE 2253 295

VOL

This content downloaded from 128.119.168.112 on Sat, 17 Sep 2016 07:45:37 UTC
All use subject to http://about.jstor.org/terms