Geoscience Reference
In-Depth Information
is indicated, some simulations may have drought events longer than 5 years, and
some may have more than one event. The null hypothesis is phrased to treat all
such series equally, in that each series with at least one event contributes equally
to the probability. It may seem surprising that the noise-added reconstructions give
such a high probability when the reconstruction itself has no events. The reason
is that in years of reconstructed flow above but near the threshold, the chance is
not negligible that the true flow was actually below the threshold. The noise-added
series incorporate this uncertainty.
Probability estimates of hydrologic events from tree-ring reconstructions are
unlikely to be used by water resources professionals without some context of com-
parable information from the better-understood observed flow record. The Colorado
River example is extended here to illustrate how this may be done. The observed
flows are limited to the period 1906-2004, but the statistics of the observed flows
can be used to simulate flows as long as the tree-ring reconstruction. Differences
in probabilities based on the simulated flows and the noise-added reconstruction
may highlight the 'new' information provided by the tree-ring record. Conversely,
similarities may attest to longer-term relevance of the short observed record.
Because the most recent years of the Colorado River observed flows have been
among the lowest on record, flow statistics for designing simulation are sensitive to
the truncation year for analysis. Histograms of observed flows for subperiods 1906-
1999 and 1906-2004 show barely noticeable differences solely due to the low flows
Fig. 8.3
Histograms and gamma-fit probability density functions (PDFs) of Colorado River at
Lees Ferry annual flows for different analysis periods. Dropping the years of the most recent
drought (2000-2004) results in a barely noticeable rightward shift in the fitted density function
