From the 1999 Report of the Working Group on Numerical Experimentation (WGNE) on "Transpose AMIP"(1):
"WGNE is continuing to develop the concept of what is termed a "Transpose AMIP", in which climate models would be run in NWP mode, and the evolution of the forecast and of various variables examined, as well as the behaviour of parameterizations before the forecast state diverges too far from the truth. More specifically, predicted variables will be compared with values from reanalyses over regions where these variables are known to be correct from comparison with observations (i.e. data rich areas over the US and/or Europe) in forecasts of only a few days during which the state may be considered 'correct'. The intention is to try and learn why there are model errors, rather than just what the errors are. WGNE recognized that the initialization and spin up of the forecasts were likely to be critical aspects of whether useful results could be obtained, especially in trying to assess model treatments of cloud and radiation. Nevertheless, a pilot project is being undertaken at NCAR with the CCM model using initial data provided by ECMWF (which then have to be interpolated to the CCM grid)."
and from the 1999-2009 Plan of the European Centre for Medium-Range Weather Forecasts (ECMWF) (2):
"One can have confidence in simulated climate scenarios only if one has confidence in the physical formulations and feed-back loops of the GCMs. A strong case could be made that every GCM should be equipped with a data assimilation system, so that one can diagnose its performance with field experiment data and in medium- and extended-range forecasts."
- Tony Hollingsworth