Application of Streamflow Reconstruction to Water Resources Management - Dendroclimatology - page 240

record as a long-term estimator of autocorrelation of streamflow (Meko and Graybill
1995 ) . In reconstruction methods, much attention has been paid to minimizing the
distortion of autocorrelation in streamflow reconstructions. For example, in a recon-
struction of annual flows of the Sacramento River, California, Meko et al. ( 2001 )
filtered standard tree-ring indices in a preliminary step with a first-order autore-
gressive model, iteratively fit such that the flow series and filtered tree-ring series
had approximately the same first-order autocorrelation coefficient for their overlap
period.
Another approach to dealing with autocorrelation of tree-ring data was taken by
Cleaveland and Stahle ( 1989 ) in their reconstruction of the White River, Arkansas.
Autoregressive (AR) modeling was applied separately to tree-ring series and flow
to produce AR residual time series. The AR residual flows were then regressed on
the AR residual tree-ring series, the residual flows were reconstructed, and the AR
flow model was applied to reintroduce persistence into the reconstruction. A third
method of dealing with autocorrelation in flow reconstruction, applied to the Gila
River, Arizona, was to use residual tree-ring chronologies as predictors of flow in a
distributed-lag regression model (Meko and Graybill 1995 ) . It is important to note
that streamflow reconstruction models employing residual tree-ring chronologies as
predictors without lags in the model are prone to underestimation of the persistence
in reconstructed flows when gauged values do contain significant persistence. This
point is emphasized in a sensitivity analysis in reconstruction of the Colorado River
at Lees Ferry (Woodhouse et al. 2006 ) .
Tree-ring information on streamflow has been extended in some studies to
include the complete empirical distribution function of annual flow. For example,
Jain et al. ( 2002 ) identified differences in the probability density function (PDF)
signatures of two prominent 31-year reconstructed low-streamflow periods for mid-
dle Boulder Creek, Colorado. One period showed a general shift of the PDF to the
left, reflecting generally lower mean flows. Another period showed a focused shift
toward an increase in more severe low flows, but with only slightly lower average
flows than the full record. The differences were presented as an example of infor-
mation that could be important to water resources system management on typical
planning horizons (30-50 years).
The technique of 'runs analysis' has gained popularity in recent decades as a
tool for summarizing drought properties of reconstructions of streamflow and other
hydroclimatic variables. In the terminology of runs analysis, a run is a series of
consecutive values below some threshold; the run length is the number of consec-
utive years in the run; the severity, or run sum, is the sum of departures from the
threshold; and the average intensity is the quotient of the run sum and run length
(Dracup et al. 1980 ; Salas et al. 1980 ) . Runs analysis was first applied descriptively
in dendrohydrology to summarize the co-occurrence of drought in different parts of
the western United States (Meko et al. 1995 ) . A shortcoming of the method is the
somewhat artificial delimitation of the temporal extent of a drought, which can be
'ended' with a single year of normal moisture conditions. The drought tally is also
sensitive to the subjective choice of a drought threshold. The theory of runs analysis
has recently been extended in a tree-ring context to develop a method to place any