Dendroclimatology in High-Resolution Paleoclimatology - Dendroclimatology - page 28

the common climatic signal, and a dilution of the individual effects of competition
and disturbances on the resulting mean chronology, as was discussed above.
As networks were built in more mesic regions characterized by denser forests and
much greater influences of competition and disturbances—in Western Europe and
eastern North America, for example—it became clear that additional approaches
were needed. In order to fit and hence remove growth surges or suppressions unique
to single trees, more 'flexible' curves were used in detrending; for example, poly-
nomials (Fritts 1976 ) , flexible cubic smoothing splines (Cook and Peters 1981 ) , and
digital filters (Briffa 1984 ) . An accessible and thorough account of these develop-
ments is given by Cook ( 1990 ) and Cook et al. ( 1990 ) . It was acknowledged at an
early stage that this approach could well result in the loss of climatic information
on the same and longer timescales as those of the age-size-related trend (LaMarche
1974 , pp 1045-1046; Fritts 1976 , p 264), or on even shorter timescales when the
more 'flexible' approaches were applied. Cook et al. ( 1995 ) also noted that, in
building a millennia-long chronology from overlapping short segments, 'the maxi-
mum length of recoverable climate information is ordinarily related to the lengths
of the individual tree-ring series used to reconstruct the millennia-long chronology.'
They named this problem the 'segment length curse.' In some cases this problem
was avoided by performing no detrending, specifically where there was reason to
believe an age-size trend was largely absent (see, for example, LaMarche 1974 ) . Of
course, there was a concomitant need to exclude samples showing evidence of a sud-
den surge or suppression from inclusion prior to pretreatment, and in the example
mentioned, the rapid growth of the early centuries of the tree's life was discarded.
This option is not available when individual segments are only a few centuries
long, as a large fraction of each series will likely be in the fast-growing early part of
the age-size curve. In such situations, those seeking to reconstruct lower-frequency
climate fluctuations—for example, on multidecadal to multicentennial timescales—
have long chosen to use 'stiff' detrending curves such as straight lines, the modified
negative exponential, the Hugershoff curve, and very stiff splines, in detrending the
original tree-ring measurement series, so as to conserve these slower variations to
as great an extent as possible. Where individual segments are exceptionally long
(many centuries), it has been possible to capture multicentennial features by using
such techniques (for example, Hughes and Funkhouser 1998 ) . More commonly,
chronologies with segment lengths of 300 or 400 years cannot be expected to cap-
ture variability on timescales much greater than multidecadal, no matter how long
the chronology.
It was in response to this problem that the approach known as regional curve
standardization (RCS) was developed, or at least modified for use in dendroclima-
tology, initially by Briffa et al. ( 1992 ) . An empirically defined age/ring-width or
age/maximum latewood density curve is developed and used in place of, for exam-
ple, the modified negative exponential curve to detrend each measurement series.
Unlike the fitted curves, whether deterministic (e.g., negative exponential, polyno-
mial) or stochastic (e.g., digital filters or splines), the empirical age/growth curve
is positioned relative to the cambial age, not the calendar date, of each ring in the
sample series being treated.