212
Chapter 11: Stochastic Modeling of Precipitation
occurrences given the weather state. These parameters, once estimated using
historical data, must then be presumed to hold under a different sequence of
weather states corresponding, for instance, to a GCM simulation. Likewise,
many of the models (e.g., Bardossy and Plate, 1992; Hughes, 1993) have
spatial covariances that are conditioned on the weather state. The historical values of these parameters likewise must be assumed to hold under an
alternative climate.
Another complication in application of these models to alternative climate
simulation is comparability of the GCM predictions with the historie observations. For instance, McCabe (1990) used a weather classification scheme
based on surface wind direction and cloud cover. The resulting weather
classes were shown to be weH related to precipitation at a set of stations
in the Delaware River basin. Unfortunately, however, GCM predictions of
cloud cover and wind direction for current climate are often quite biased as
compared to historie observations, and these biases will be reflected in the
stochastic structure of the weather class sequence.
11.6 Conclusions
The coupling of weather state classification procedures, either explicitly or
implicitly, with stochastic precipitation generation schemes is a promising
approach for transferring large-area climate model simulations to the local
scale. Most of the work reported to date has focused on the simulation of daily
precipitation, conditioned in various ways on weather classes extracted from
large-area atmospheric features. The approach has been shown to perform
adequately in most of the studies, although there remain quest ions as to
how best to determine the weather states. Further , no useful means has
yet been proposed to determine the strength of the relationship between
large-area weather classes and local precipitation, and to insure that weak
relationships do not result in spurious downward biases in inferred changes
in local precipitation at the local level. This is an important concern, since
at least one of the studies reviewed (Zorita et al., 1995) found conditions
under which weather classes well-related to local precipitation could not be
identified.
There have been relatively few demonstration applications of these procedures for climate effects interpretations. One of the major difficulties is
accounting for biases in the GCM present climate, or "base" runs. In addition, few of the models reviewed presently simulate variables other than
precipitation needed for hydrological studies. Temperature simulation is especially important for many hydrological modeling applications, but methods
of preserving stochastic consistency between local and large-scale simulations
are presently lacking.
Chapter 11: Stochastic Modeling of Precipitation
occurrences given the weather state. These parameters, once estimated using
historical data, must then be presumed to hold under a different sequence of
weather states corresponding, for instance, to a GCM simulation. Likewise,
many of the models (e.g., Bardossy and Plate, 1992; Hughes, 1993) have
spatial covariances that are conditioned on the weather state. The historical values of these parameters likewise must be assumed to hold under an
alternative climate.
Another complication in application of these models to alternative climate
simulation is comparability of the GCM predictions with the historie observations. For instance, McCabe (1990) used a weather classification scheme
based on surface wind direction and cloud cover. The resulting weather
classes were shown to be weH related to precipitation at a set of stations
in the Delaware River basin. Unfortunately, however, GCM predictions of
cloud cover and wind direction for current climate are often quite biased as
compared to historie observations, and these biases will be reflected in the
stochastic structure of the weather class sequence.
11.6 Conclusions
The coupling of weather state classification procedures, either explicitly or
implicitly, with stochastic precipitation generation schemes is a promising
approach for transferring large-area climate model simulations to the local
scale. Most of the work reported to date has focused on the simulation of daily
precipitation, conditioned in various ways on weather classes extracted from
large-area atmospheric features. The approach has been shown to perform
adequately in most of the studies, although there remain quest ions as to
how best to determine the weather states. Further , no useful means has
yet been proposed to determine the strength of the relationship between
large-area weather classes and local precipitation, and to insure that weak
relationships do not result in spurious downward biases in inferred changes
in local precipitation at the local level. This is an important concern, since
at least one of the studies reviewed (Zorita et al., 1995) found conditions
under which weather classes well-related to local precipitation could not be
identified.
There have been relatively few demonstration applications of these procedures for climate effects interpretations. One of the major difficulties is
accounting for biases in the GCM present climate, or "base" runs. In addition, few of the models reviewed presently simulate variables other than
precipitation needed for hydrological studies. Temperature simulation is especially important for many hydrological modeling applications, but methods
of preserving stochastic consistency between local and large-scale simulations
are presently lacking.