Cohen-Solal, Emmanuelle S., Peter J. Gleckler, Michael
F. Wehner, Benjamin D. Santer, Karl E. Taylor and Charles Doutriaux
September 1998, 27 pp.
Climate simulation experiments involving atmospheric general circulation
models (AGCMs) are routinely forced with prescribed ocean surface boundary
conditions. For instance, in the standard experiments of the Atmospheric
Model Intercomparison Project (AMIP), sea surface temperature (SST) and
sea ice fraction are defined in terms of the monthly averages of an observed
data set. Although this is the traditional approach for AGCM simulations,
higher frequency averages of this surface forcing are now available.
In this study, we investigate the statistically significant differences
in a simulated climate by performing two ensembles of AGCM integrations
with the surface forcing defined by different averaging periods.
We construct two surface forcing data sets of daily SST and sea ice fraction
from the same set of observations. The first data set is obtained
by linearly interpolating between the mid-points of observed weekly means;
the second is similarly obtained, but from observed monthly mean values.
The largest differences in the two reconstructions of daily SST are found
in places and seasons where the temporal variability is largest. Over broad
areas of the mid-latitude ocean, the daily forcing data derived from monthly
means fail to capture the sharp summertime maxima found in the forcing
series based on weekly mean data. In addition, some evolving structures
with time scales less than one month in the tropical oceans cannot be well
reproduced by the monthly mean data. For sea ice, the differences are found
mostly at the ice margins during months of rapid change in ice coverage.
The impact of the different temporal specifications of surface boundary
conditions on the simulated climate is assessed in relation to observational
uncertainty and model errors. Two ensembles were generated with a version
of the LLNL AGCM, each comprising 18 simulations, one forced by the boundary
conditions based on weekly data, and the other by the monthly data. Grid-point
t-tests were then used to examine the statistical significance of differences
in the ensemble means of various quantities. For variables such as atmospheric
boundary layer temperature, surface pressure and latent heat flux, the
overall differences in the "weekly versus monthly" ensemble means were
judged to be statistically significant (under the assumption that at least
5% of the total number of grid-points were statistically independent).
However, differences in ensemble means were still smaller than the observational
errors in the most skillful of current models. Moreover, for many variables,
the differences between the two ensemble means were smaller than estimates
of uncertainties in the observations themselves. Thus, given the
current observational uncertainty and model errors, we conclude that for
AMIP-like simulations it is sufficient to prescribe SST and sea ice fraction
with monthly resolution. (pdf file)