Ensembles of hindcasts from seven models are analyzed to evaluate dynamical
seasonal predictability of 850-hPa wind and rainfall for the Asian summer
monsoon (ASM) during 1987, 1988, and 1993. These integrations were performe
d using observed sea surface temperatures and from observed initial conditi
ons. The experiments were designed by the Climate Variability and Predictab
ility, Working Group on Seasonal to Interannual Prediction as part of the S
easonal Prediction Model Intercomparison Project. Integrations from the Eur
opean Union Prediction of Climate Variations on Seasonal to Interannual Tim
escales experiment are also evaluated.
The National Centers for Environmental Prediction-National Center for Atmos
pheric Research and European Centre for Medium-Range Weather Forecasts rean
alyses and observed pentad rainfall form the baseline against which the hin
dcasts are judged. Pattern correlations and root-mean-square differences in
dicate errors in the simulation of the time mean low-level flow and the rai
nfall exceed observational uncertainty. Most models simulate the subseasona
l EOFs that are associated with the dominant variations of the 850-hPa flow
during the ASM, but not with the fidelity exhibited by the reanalyses as d
etermined using pattern correlations. Pattern correlations indicate that th
e first EOF, associated with the tropical convergence zone being located ov
er the continental landmass, is best simulated. The higher-order EOFs are l
ess well simulated, and errors in the magnitude and location of their assoc
iated precipitation anomalies compromise dynamical seasonal predictability
and are related to errors of the mean state. In most instances the models f
ail to properly project the subseasonal EOFs/principal components onto the
interannual variability with the result that hindcasts of the 850-hPa flow
and rainfall are poor. In cases where the observed EOFs are known to be rel
ated to the boundary forcing, the failure of the models to properly project
the EOFs onto the interannual variability indicates that the models are no
t setting up observed teleconnection patterns.