Decision support experiments and evaluations using seasonal to ...

The U.S. Climate Change Science Program Chapter 2
44
A question not addressed in this Product relates
to the probabilistic skill of the forecasts:
How reliable are the confidence limits around
the median forecasts that are provided by the
published forecast quantiles (10th and 90th
Figure 2.13 Percentages of stations with various correlation skill scores in
the various panels (forecast dates) of Figure 2.12.
Figure 2.14 Potential contributions of antecedent snowpack conditions,
runoff, and Niño 3.4 sea-surface temperatures to seasonal forecast skills
in hydrologic simulations under historical, 1950 to 1999, meteorological
conditions (left panels) and under those same conditions but with a 2ºC
uniform warming imposed (Dettinger, 2007).
percentiles, for example)? In a reliable forecast,
the frequencies with which the observations
fall between various sets of confidence
bounds matches the probability interval set by
those bounds. That is, 80 percent of the time,
the observed values fall between the 10th and
90th percentiles of the forecast. Among the few
analyses that have been published focusing on
the probabilistic performance of United States
operational streamflow forecasts, Franz et al.
(2003) evaluated Colorado River basin ESP
forecasts using a number of probabilistic measures
and found reliability deficiencies for many
of the streamflow locations considered.
2.2.3.2 the i m plic ation S oF decadal
variability and long term change
in climate For SeaSonal hydrologic
prediction Skill
In the earlier discussion of sources of watersupply
forecast skill, we highlighted the
amounts and sources of skill provided by snow,
soil moisture, and antecedent runoff influences.
IPCC projections of global and regional warming,
with its expected strong effects on western
United States snowpack (Stewart et al., 2004;
Barnett et al., 2008), raises the concern that
prediction methods, such as regression, that
depend on a consistent relationship between
these predictors, and future runoff may not perform
as expected if the current climate system
is being altered in ways that then alters these
hydro-climatic relationships. Decadal climate
variability, particularly in precipitation (e.g.,
Mantua et al., 1997; McCabe and Dettinger,
1999), may also represent a challenge to such
methods, although some researchers suggest
that knowledge of decadal variability can be
beneficial for streamflow forecasting (e.g.,
Hamlet and Lettenmaier, 1999). One view (e.g.,
Wood and Lettenmaier, 2006) is that hydrologic
model-based forecasting may be more robust to
the effects of climate change and variability due
to the physical constraints of the land surface
models, but this thesis has not been comprehensively
explored.
The maps shown in Figure 2.14 are based on
hydrologic simulations of a physically-based
hydrologic model, called the Variable Infiltration
Capacity (VIC) model (Liang et al., 1994),
in which historical temperatures are uniformly
increased by 2ºC. These figures show that the