Application of Streamflow Reconstruction to Water Resources Management - Dendroclimatology - page 253

archaeological samples. Methods for quality-controlling and adjusting chronolo-
gies for time-varying sample size are available (Wigley et al. 1984 ; Osborn et al.
1997 ) , and should be more routinely adopted in streamflow reconstruction. Another
possible strategy is to tailor reconstruction models so that the equation yielding
a reconstructed flow in any given year is based on a calibration model using the
identical (or similar) tree-ring cores (e.g., Meko 1997 ) .
As tree-ring chronologies are updated and additional chronologies are developed,
reconstructions for the same flow record can be expected to change: the revised
flow estimates will be a different linear combination of tree-ring indices from the
estimates in previous reconstructions. Illustrations of such changes for reconstruc-
tions of streamflow for the Colorado River can be found in Woodhouse et al. ( 2006 ) .
Moreover, choices of tree-ring processing (e.g., residual or standard indices, lags or
no lags, indices or principal components) can lead to different reconstructions from
the same basic tree-ring measurements. Because the 'true' model relating flow to
tree-ring indices is an abstraction and is unknown, it is important that more research
address the sensitivity of reconstructed streamflow features to modeling choices.
Uncertainty in the low-frequency component of streamflow variability is another
aspect of streamflow reconstruction that cannot be satisfactorily addressed with cal-
ibration and validation statistics of reconstruction models. First, the flow record for
the period used to calibrate and validate the model simply may not be represen-
tative of the low-frequency behavior of the long-term record. In that case, we do
not know how well the tree rings might track the low-frequency flow variations,
and we must assume that low-frequency features—such as broad swings above
and below the mean in tree growth—reflect similar variations in flow. Second,
the detrending operation in conventional standardization places a lower limit on
the frequency of climatic variation resolvable with the tree-ring index. That limit
depends on the length of the tree-ring series and choices of detrending curve by
the researcher developing the chronology (Cook et al. 1990 , 1995 ) . Regional curve
standardization (RCS), which depends on identification of a generally applicable
function of expected ring width with tree age, has been applied in dendrohydrology
in an attempt to circumvent the frequency-response limitation (e.g., St. George and
Nielsen 2002 ) . Unfortunately, RCS requires intensive sampling, with trees of vari-
ous ages represented throughout the period of record (Briffa et al. 1996 ) . Few river
basins may afford such a luxury of moisture-sensitive trees.
Uncertainty can never be completely eliminated from streamflow reconstruc-
tions. A streamflow reconstruction relies on a statistical relationship between
streamflow (observed or adjusted to natural flows) and tree-ring chronologies dis-
tributed over the basin. Increased tree-ring site coverage and improved statistical
methodology may increase the strength of the relationship. Uncertainty may eventu-
ally be reduced by incorporating information from tree-ring variables other than ring
width index in the reconstruction model. Variables might include wood density (e.g.,
Briffa et al. 1988 ) , stable isotope ratios in tree rings (e.g., Leavitt and Wright 2002 )
and the anatomical features of cambial cells (e.g., Vaganov 1990 ; Vaganov et al.
2006 ) . Such efforts can never arrive at a perfect reconstruction, but improvements
in accuracy may enhance the usefulness of the reconstruction for water resource
management.