Section 11.5: Applications to Alternative Climate Simulation
211
the index stations to be relatively elose to the eoast, but when the model
was applied to the middle Atlantie region, CART weather states eould be
obtained only when the index stations were quite elosely spaeed, and then
only in winter. The diffieulty appeared to be the shorter spatial seale of
summer precipitation, and weaker eoupling of loeal preeipitation with the
regional cireulation patterns.
One of the motivations for development of stoehastie models that eouple
large area atmospherie variables with loeal variables, such as preeipitation,
is to provide a means of downsealing simulations of alternative elimates for
effects assessments. However, as no ted above, most of the applieations to
date have been to historie data, for instanee, loeal precipitation has been
simulated using either an historie sequenee ofweather states (e.g., Hughes et
al. , 1993) or via a stoehastie model of the historie weather states (e.g., Hay
et al., 1991, Bardossy and Plate, 1992; Wilson et al., 1992).
For most of the models reviewed, it should be straightforward to produee
a sequenee of weather states eorresponding to an alternative elimate scenario
(e.g., from a lengthy GCM simulation). There are, nonetheless, eertain eomplieations. Selection of the variables to use in the weather state elassifieation
is problematie. Wilson et al. (1991) elassified weather states using sea level
press ure and 850 hPa temperature. However, if this seheme is used with an
alternative, warmer elimate, the temperature change dominates the elassifieation, resulting in a major change in the stoehastie strueture of the weather
elass sequenee that may not be physieally realistie. Although this problem
is resolved by use of variables, such as sea level pressure, that more directly
refleet large area eireulation patterns, elimination of temperature from eonsi der at ion as a elassifying variable is somewhat arbitrary. A related problem
is the effect of the strength of the linkage between the weather states and the
loeal variables. In asense, the problem is analogous to multiple regression. If
the regression is weak, i.e., it does not explain roueh of the varianee in the dependent variable (e.g., loeal preeipitation), and ehanges in the independent
variables (e.g., weather states) will not be evideneed in predietions of the
loeal variable. Therefore, one might erroneously eonelude that ehanges in,
for instanee, preeipitation would be smalI, merely beeause of the absence of
strong linkages between the large scale and loeal eonditions (see, for example,
Zorita et al. , 1995).
Applieation of all of the models for alternative elimate simulation requires
that eertain assumptions be made about what aspeet of the model strueture will be preserved under an alternative elimate. All of the models have
parameters that link the large area weather states with the prob ability of
oeeurrenee, or amount of, loeal preeipitation. For instanee, in the model
of Wilson et al. (1992) there are parameters that eontrol the prob ability
of precipitation for each eombination of weather state and the preeipitation
state at the higher order stations. In the model of Bardossy and Plate (1991)
there is a Markov parameter that deseribes the persistenee of precipitation
211
the index stations to be relatively elose to the eoast, but when the model
was applied to the middle Atlantie region, CART weather states eould be
obtained only when the index stations were quite elosely spaeed, and then
only in winter. The diffieulty appeared to be the shorter spatial seale of
summer precipitation, and weaker eoupling of loeal preeipitation with the
regional cireulation patterns.
One of the motivations for development of stoehastie models that eouple
large area atmospherie variables with loeal variables, such as preeipitation,
is to provide a means of downsealing simulations of alternative elimates for
effects assessments. However, as no ted above, most of the applieations to
date have been to historie data, for instanee, loeal precipitation has been
simulated using either an historie sequenee ofweather states (e.g., Hughes et
al. , 1993) or via a stoehastie model of the historie weather states (e.g., Hay
et al., 1991, Bardossy and Plate, 1992; Wilson et al., 1992).
For most of the models reviewed, it should be straightforward to produee
a sequenee of weather states eorresponding to an alternative elimate scenario
(e.g., from a lengthy GCM simulation). There are, nonetheless, eertain eomplieations. Selection of the variables to use in the weather state elassifieation
is problematie. Wilson et al. (1991) elassified weather states using sea level
press ure and 850 hPa temperature. However, if this seheme is used with an
alternative, warmer elimate, the temperature change dominates the elassifieation, resulting in a major change in the stoehastie strueture of the weather
elass sequenee that may not be physieally realistie. Although this problem
is resolved by use of variables, such as sea level pressure, that more directly
refleet large area eireulation patterns, elimination of temperature from eonsi der at ion as a elassifying variable is somewhat arbitrary. A related problem
is the effect of the strength of the linkage between the weather states and the
loeal variables. In asense, the problem is analogous to multiple regression. If
the regression is weak, i.e., it does not explain roueh of the varianee in the dependent variable (e.g., loeal preeipitation), and ehanges in the independent
variables (e.g., weather states) will not be evideneed in predietions of the
loeal variable. Therefore, one might erroneously eonelude that ehanges in,
for instanee, preeipitation would be smalI, merely beeause of the absence of
strong linkages between the large scale and loeal eonditions (see, for example,
Zorita et al. , 1995).
Applieation of all of the models for alternative elimate simulation requires
that eertain assumptions be made about what aspeet of the model strueture will be preserved under an alternative elimate. All of the models have
parameters that link the large area weather states with the prob ability of
oeeurrenee, or amount of, loeal preeipitation. For instanee, in the model
of Wilson et al. (1992) there are parameters that eontrol the prob ability
of precipitation for each eombination of weather state and the preeipitation
state at the higher order stations. In the model of Bardossy and Plate (1991)
there is a Markov parameter that deseribes the persistenee of precipitation