Geoscience Reference
In-Depth Information
also did this for each species separately to see how much might be gained by using
a multi-species assemblage.
Using the same principal components regression program that calculated the
response functions, we reversed the order of dependence and used the seven
tree-ring chronologies as predictors of May-July total precipitation. The tree-ring
chronologies were lagged on themselves by one year (
t
and
t
+1 for each series) to
allow for the carryover of the climate signal from one year to the next, thus creat-
ing a matrix of 14 candidate predictors. Lagging the tree-ring data resulted in the
loss of one year for calibration (now 1931-1995), and the use of the now unlagged
climate data resulted in a gain of one year for verification (now 1896-1930). The
14 candidate predictors were next screened for simple correlation with May-July
precipitation over the calibration period. Those that did not correlate significantly
(
p
< 0.05; one-tailed test) were removed from this candidate predictor pool. Of the
14
t
,
t
+1 candidates, only 3 were rejected: PIRI
t
+1, QUPR
t
+1, and QUVE
t
+1.
Those that passed the screening had correlations with May-July precipitation rang-
ing from 0.272 to 0.468. The retained 11 candidate predictors were subjected to
principal components analysis and the EV1 cutoff was used to retain the candidate
eigenvectors. The EV1 cutoff retained 3 of 11 eigenvectors, which accounted for
65.5% of the total variance. These were used as candidate predictors in regression
analysis using the minimum AIC for selecting the best-fit model. See Table
4.5
for details. The result was the entry of the first two tree-ring eigenvectors, which
is consistent with what we found earlier in our evaluation of the most important
climate eigenvectors associated with tree growth. The reconstruction model
Table 4.5
Comparisons of May-July total precipitation reconstructions at Mohonk Lake
Calibration period statistics
Verification period statistics
R
2
PREDICTORS
NEIG
AIC
RSQ
RE
CE
0.126
∗
ALL (11)
3/2
0.436
−
30.83
−
0.047
−
0.112
0.098
∗
0.040
∗
0.020
∗
TSCA (2,
t
,
t
+1)
2/1
0.266
−
15.91
0.072
∗
PIRI (1,
t
)
1/1
0.213
−
11.36
−
0.033
−
0.097
0.257
∗
0.256
∗
0.244
∗
QUPR (1,
t
)
1/1
0.195
−
9.88
0.014
∗
QUVE (1,
t
)
1/1
0.128
−
4.72
0.051
−
0.002
CAGL (2,
t
,
t
+1)
2/1
0.284
−
17.48
0.019
−
0.137
−
0.155
0.069
∗
0.008
∗
LITU (2,
t
,
t
+1)
2/1
0.237
−
13.42
−
0.008
−
17.69
−
0.262
−
0.282
BELE (2,
t
,
t
+1)
2/1
0.286
0.011
The calibration period is 1931-1995 and the verification period is 1896-1930.
PREDICTORS
the tree-ring variables used in each principal components regression model
that passed the correlation screening with climate (all but the first row are single-species predic-
tors); NEIG
=
the number of candidate tree-ring eigenvectors/the number of tree-ring eigenvectors
entered into the model;
R
2
=
=
cumulative fractional variance of the model; AIC
=
Akaike infor-
mation criterion; RSQ
=
square Pearson correlation; RE
=
verification reduction of error; CE
=
verification coefficient of efficiency. Higher RSQ, RE, and CE mean better verification of the
fitted model.
∗
Means significant
p
< 0.05, RE > 0, or CE > 0.
