A. Galambosi et al., EVALUATION AND ANALYSIS OF DAILY ATMOSPHERIC CIRCULATION PATTERNS OF THE 500 HPA PRESSURE FIELD OVER THE SOUTHWESTERN USA, Atmospheric research, 40(1), 1996, pp. 49-76
Daily atmospheric circulation patterns (CP's) are defined and analyzed
on the basis of the 500 hPa pressure field for the purpose of describ
ing and then later generating local hydroclimatological variables such
as precipitation under possible climate change over the southwestern
USA. To obtain the CP's we first use a so-called objective or automate
d clustering method, namely, principal com ponent analysis (PCA) coupl
ed with K-means clustering algorithm. To obtain a set of CP types that
are more distinguishable and so more useful for our purpose we follow
two possible ways: (1) reduce subjectively the number of CP's from PC
A coupled with K-means clustering analysis by aggregating the types on
the basis of the precipitation producing characteristics, (2) perform
K-means clustering analysis with fewer types. Thus we have three diff
erent cluster systems: original types from the result of PCA coupled w
ith K-means clustering (8 or 9 types depending on the season), types f
rom the K-means clustering analysis with fewer types (5 or 6 types in
each season) and the subjectively aggregated types (3 types in any sea
sons). We compare them from the point of view of information content f
or precipitation modeling. An analysis is made for these types for com
parison: statistical properties of these patterns are evaluated and an
alyzed using first observed data, and then General Circulation Model (
GCM) outputs for 1 X CO2 and 2 X CO2 scenarios to estimate climate cha
nge effects, On the basis of the historical circulation pattern catalo
gue and observed precipitation data in Arizona, simple calculations ar
e provided to find the ''precipitation-producing'' types of each syste
m in each season. Three indices are evaluated using the same observed
precipitation data from ten Arizona stations in order to measure objec
tively the information content of each type in the three cluster syste
ms. It turns out that the larger number of types in a given season giv
es higher information content as we expected.