Geoscience Reference
In-Depth Information
produced a highly significant calibration R 2 of 0.436 and significant ( p < 0.05)
verification RSQ, but the RE and CE were weakly negative.
The individual species results are an interesting contrast to the joint reconstruc-
tion model (Table 4.5 ) . None of the single-species models calibrated as much
variance as the joint model. Single-species R 2 ranged from 0.128 for QUVE to
0.286 for BELE. Thus, a significant amount of additional calibrated variance has
been captured by using the mutual climate information contained in the seven tree
species chronologies. In terms of verification RSQ, the joint model outperforms all
single-species models except that for QUPR. The RE and CE statistics show more
variability between the models, but most are
0.05, a range that is indeterminate
with respect to assessing differences in the significance of the models (Gordon and
LeDuc 1981 ) . The QUPR model is the most anomalous single-species model. It
has substantially higher verification statistics even though its calibration R 2 is the
second lowest one. Since the 1950s, the forests around Mohonk Lake have been
subjected to episodic gypsy moth ( Lymantria dispar ) infestations that have resulted
in the occasional defoliation of certain tree species, and chestnut oak is among the
most favored host species. Years of notable gypsy moth population buildup and
defoliation in the vicinity of Mohonk Lake were 1957, 1965-1966, 1971, 1981, and
1987-1988 (Smiley and Huth 1982 ; Huth 2005 ) . So it is possible that the weak
QUPR calibration is due to gypsy moth defoliation effects on the radial growth of
chestnut oak.
Overall, the results presented in Table 4.5 support the use of multiple trees
species for reconstructing past climate. This finding is not terribly surprising,
because, as we have seen from our response function modeling results (Table 4.4 )
and the results from other studies (e.g., Graumlich 1993 ; Cook et al. 2001 ) , tree
species at the genus and subgenus levels can have different phylogenetic responses
in their ring widths to nearly the same macroclimatic influences on growth. These
statistically expressed differences in response to a common climate forcing are
likely to enter into any climate reconstruction if it is based on only one tree species.
The introduction of certain non-climatic biases into such reconstructions is there-
fore likely. This ought to be avoided whenever practical through the use of multiple
tree species in climate reconstructions.
±
4.9 Concluding Remarks
Tree-ring analysis is one of the most powerful tools available for the study of envi-
ronmental change and the identification of fundamental relationships between tree
growth and climate. At every stage of analysis there is both statistical and biolog-
ical uncertainty. In statistical analysis, we wish to reduce the uncertainty of our
inferences as best as our data and analysis methods can reasonably allow, always
taking into account the fact that we could be horribly wrong. To reduce the chances
of making false statistical inferences, some form of model validation should there-
fore be conducted whenever possible (Snee 1977 ) . Validation based on the analysis
of withheld data as described here and on new tree-ring data from the same or
Search WWH ::




Custom Search