Dynamical Seasonal Prediction (DSP) is an informally coordinated multi-inst
itution research project to investigate the predictability of seasonal mean
atmospheric circulation and rainfall. The basic idea is to test the feasib
ility of extending the technology of routine numerical weather prediction b
eyond the inherent limit of deterministic predictability of weather to prod
uce numerical climate predictions using state-of-the-art global atmospheric
models. Atmospheric general circulation models (AGCMs) either forced by pr
edicted sea surface temperature (SST) or as part of a coupled forecast syst
em have shown in the past that certain regions of the extratropics, in part
icular, the Pacific-North America (PNA) region during Northern Hemisphere w
inter, can be predicted with significant skill especially during years of l
arge tropical SST anomalies. However, there is still a great deal of uncert
ainty about how much the details of various AGCMs impact conclusions about
extratropical seasonal prediction and predictability.
DSP is designed to compare seasonal simulation and prediction results from
five state-of-the-art U.S. modeling groups (NCAR, COLA, GSFC, GFDL, NCEP) i
n order to assess which aspects of the results are robust and which are mod
el dependent. The initial emphasis is on the predictability of seasonal ano
malies over the PNA region. This paper also includes results from the ECMWF
model, and historical forecast skill over both the PNA region and the Euro
pean region is presented for all six models.
It is found that with specified SST boundary conditions, all models show th
at the winter season mean circulation anomalies over the Pacific-North Amer
ican region are highly predictable during years of large tropical sea surfa
ce temperature anomalies. The influence of large anomalous boundary conditi
ons is so strong and so reproducible that the seasonal mean forecasts can b
e given with a high degree of confidence. However, the degree of reproducib
ility is highly variable from one model to the other, and quantities such a
s the PNA region signal to noise ratio are found to vary significantly betw
een the different AGCMs. It would not be possible to make reliable estimate
s of predictability of the seasonal mean atmosphere circulation unless caus
es for such large differences among models are understood.