Before commencing the actual ERA production, it was necessary to define the assimilation system. The aim was to have a proven, but modern data assimilation, not necessarily identical to the then operational suite. The project work thus began in 1993 with a comprehensive set of experiments in the form of parallel data assimilations and forecasts, usually over three week periods and with extensive diagnostics.

The only predetermined limitation was that the reanalyses should be carried out with a horizontal resolution of T106, this was for reasons of resources. For the vertical resolution a test assimilation made us decide upon 31 levels, rather than 19, since the higher resolution produced clearly superior analyses particularly around the tropopause.

A comprehensive test of the "First Guess at Appropriate Time" (FGAT) algorithm was also carried out since FGAT was considered quite expensive in computing time and caused a rather cumbersome file handling. FGAT was indeed shown to give a marginally positive impact in the test, but the benefit was not considered large enough to justify the increased costs.

At the time the experimentation took place, envelope orography was being used in daily operations to parametrize the effects of sub-gridscale mountains. A new parametrization of the effects of sub- gridscale orography based on mean orography, and including a revised formulation of the gravity wave drag, developed by Lott and Miller, was also available. Test assimilations using this scheme showed no negative effects, while up to 10-15% more observations were accepted at 1000hPa and 925hPa. In consequence, this scheme was chosen.

Using a prescribed soil climatology, as in the pre-1995 operational system, has the obvious risk of forcing a reanalysis towards its climate, which is based on very sparse information and may suffer from inconsistencies. Hence the new four level self-contained soil parametrization scheme developed by Beljaars & Viterbo for operational implementation was also selected for ERA.

Ongoing work in the research department on a new variational assimilation scheme (3D-Var), and a new cloud parametrization with cloud water and cloud fraction as predictive parameters, were not sufficiently mature at the time of decision and were not selected for the reanalysis. The ERA production then began in earnest in December 1994.

However ...

The spring time warm surface temperature biases found in the April 1994 ECMWF operations, were also noticed in the reanalyses of the first year, 1979. These biases were attributed to deficiencies in the cloud cover, particularly over land in spring time, leading to unrealistic drying and heating of the soil. As a quick remedy a very simple soil moisture nudging scheme was introduced, in which the near-ground atmospheric humidity analyses are used to adjust the soil moisture. Production was stopped and new assimilation experiments were run in cooperation with the Research Department in order to test and tune the scheme. Although the nudging reduced the bias problem, the more fundamental problem of too few clouds was still apparent. Since in the meantime the new prognostic cloud scheme of Tiedtke and Jakob had approached maturity and had been shown to yield more realistic clouds, it was decided to run parallel assimilations for July 1985 and January 1986 with the new cloud scheme. The results were so encouraging that it was decided to go back and restart the entire reanalysis from the beginning of 1979 with the prognostic clouds included.

The two complete 1979 assimilations, with the old and the new cloud schemes are quite different, not only in the actual cloud amounts, but also in important derived quantities such as surface energy fluxes and precipitation. This one-year sensitivity 'experiment' demonstrates the very important fact that many aspects of a reanalysis are to a very high degree defined by the assimilating forecast model, and particularly by its physical parametrization.

16.11.2001