10.6.2.3 Weather typing
This synoptic downscaling approach relates “weather classes” to local
and regional climate variations. The weather classes may be defined synoptically
or fitted specifically for downscaling purposes by constructing indices of airflow
(Conway et al., 1996). The frequency distributions of local or regional climate
are then derived by weighting the local climate states with the relative frequencies
of the weather classes. Climate change is then estimated by determining the
change of the frequency of weather classes. However, typing procedures contain
a potentially critical weakness in assuming that the characteristics of the
weather classes do not change.
In many cases, the local and regional climate states are derived by sampling
the observational record. For example, Wanner et al. (1997) and Widmann and
Schär (1997) used changing global to continental scale synoptic structures
to understand and reconstruct Alpine climate variations. The technique was applied
similarly for New Zealand (Kidson and Watterson, 1995) and to a study of changing
air pollution mechanisms (Jones and Davies, 2000).
An extreme form of weather typing is the analogue method (Zorita et al., 1995).
A similar concept, although mathematically more demanding, is Classification
And Tree Analysis (CART) which uses a randomised design for picking regional
distributions (Hughes et al., 1993; Lettenmaier, 1995). Both analogue and CART
approaches returnapproximately the right level of variance and correct spatial
correlation structures.
Weather typing is also used in statistical-dynamical downscaling (SDD), a hybrid
approach with dynamical elements (Frey-Buness et al., 1995 and see references
in Appendix 10.4). GCM results of a multi-year climate
period are disaggregated into non-overlapping multi-day episodes of quasi-stationary
large-scale flow patterns. Similar episodes are then grouped in classes of different
weather types, and, members of these classes are simulated with an RCM. The
RCM results are statistically evaluated, and the frequency of occurrence of
the respective classes determines their statistical weight. An advantage of
the SDD technique over other empirical downscaling techniques is that it specifies
a complete three-dimensional climate state. The advantage over continuous RCM
simulations is the reduction in computing time, as demonstrated in Figure
10.17.
10.6.3 Issues in Statistical Downscaling
10.6.3.1 Temporal variance
Transfer function approaches and some weather typing methods suffer from an
under prediction of temporal variability, as this is related only in part to
the large-scale climate variations (see Katz and Parlange, 1996). Two approaches
have been used to restore the level of variability: inflation and randomisation.
In the inflation approach the variation is increased by the multiplication of
a suitable factor (Karl et al., 1990). A more sophisticated version is “expanded
downscaling”, a variant of Canonical Correlation Analysis that ensures
the right level of variability (Bürger, 1996; Huth, 1999; Dehn et al.,
2000). In the randomisation approachs, the unrepresented variability is added
as noise, possibly conditioned on synoptic state (Buma and Dehn, 1998; Dehn
and Buma 1999; Hewitson, 1999; von Storch, 1999b).
Often weather generators have difficulty in representing low frequency variance,
and conditioning the generator parameters on the large-scale state may alleviate
this problem (see Katz and Parlange, 1996; Wilby, 1998; Charles et al., 1999a).
For example, Katz and Parlange (1993, 1996) modelled daily time-series of precipitation
as a chain-dependent process, conditioned on a discrete circulation index. The
results demonstrated that the mean and standard deviation of intensity and the
probability of precipitation varied significantly with the circulation, and
reproduced the precipitation variance statistics of the observations better
than the unconditioned model. The method describes the mean precipitation as
a linear function of the circulation state, and the standard deviation as a
non-linear function (Figure 10.18).
Figure 10.17: Similarity of time mean precipitation distributions
obtained in a continuous RCM simulation and through statistical-dynamical
downscaling (SDD) for different levels of disaggregation. Black line:
mean absolute error (mm/day), grey line: spatial correlation coefficient.
Horizontal axis: computational load of SDD. N is the number of days
simulated in SDD, Ñ the number of days simulated with the continuous
RCM simulation. |
|
Figure 10.18: Hypothetical changes in mean and standard deviation
of January total precipitation at Chico, California, as a function
of changing probability that January mean sea level pressure is above
normal. |
|
