Dendroclimatology in High-Resolution Paleoclimatology - Dendroclimatology - page 23

networks contain robust synoptic-scale patterns (Fig. 2.3 upper and middle), which
may often be interpreted in terms of climatic conditions (Fig. 2.3 lower).Itisthe
existence of such emergent patterns that permits crossdating of tree rings, not only
between trees derived from the same site but also over distances of hundreds of
kilometers (Hughes and Brown 1992 ; Hughes et al. 2001 ; Kelly et al. 1989 , 2002 ) .
Clearly, these tree-ring networks provide a very different kind of proxy record from
that extracted from a single ice core or sediment core—they are massively repli-
cated, precisely and accurately dated to the calendar year throughout, and eminently
suitable for testing against instrumental climate data.
2.3 Climate Signal in Tree-Ring Properties
2.3.1 Identifying Signal—An Empirical-Statistical Approach
The primary methods of identifying a potential climate signal in tree-ring records
have been inference based on external knowledge of the conditions controlling the
formation of the rings—for example, LaMarche ( 1974 ) or Esper et al. ( 2002b ) —
and statistical exploration of the relationships of the tree-ring data with candidate
climate factors.
Fritts et al. ( 1971 ) developed a general scheme in which regression models were
used to relate climate variables and tree-ring series. In the first instance, models
using climate variables as predictors and tree-ring records as predictands were used
to identify the aspects of climate most likely to be controlling tree-ring formation.
Such models were termed 'response functions.' The response functions, consid-
ered in the light of available ecological and physiological knowledge, were then
used in the design of another kind of regression model known as a 'transfer func-
tion.' In this case, tree-ring records were used as candidate predictors for target
climate predictands. The approach was designed to be applied on a broad spatial
scale, using the climate-relevant data covering the longest possible period and most
broadly available at that time. These were monthly mean temperatures and total
precipitation from meteorological stations.
This approach was considered appropriate, even though Fritts was well aware
of the influence of local site and ecological conditions (see Fritts 1974 ) , because
the target was the reconstruction of subcontinental-scale spatiotemporal patterns in
climate, and because he was also aware of the existence of emergent patterns of tree-
ring growth on scales similar to synoptic patterns in climate data (LaMarche and
Fritts 1971 ) . Moreover, more directly ecologically relevant climate data covering a
sufficiently long period were extremely rare. The need to use monthly data from
meteorological stations was further supported because the length of records was
extremely important. Vaganov et al. ( 2006 , p 191) point out that 'for any derived
statistical model to be valid, it must have the largest possible ratio between the
number of cases (years) and the number of variables (months or seasons per year