Application of Methods of Systematic Initial Model Tendency Error Detection to the Deduction of Underlying Sources of Systematic Errors

The causes of systematic errors in the time mean fields of a GCM are difficult to detect from an analysis of the errors themselves. Direct estimates of systematic initial tendency errors (SITEs) determined by assimilation of reanalysis data will on the other hand point more directly to the model components causing the errors. The ECMWF Reanalysis (ERA-15) data for the period 1982 until 1994 were assimilated in different ECHAM versions using different techniques, the SNMI with window, nudging and for a part of the period using the Total Insertion (TI) with window methods, introduced and validated in an accompanying presentation. The techniques uncover SITEs of the model in question, visible as rather similar structures. In the TI experiments, the SITEs are generally less weakened than in the SNMI experiment by balancing during the assimilation, in particular at upper levels, but contains more unbalanced spurious SITEs, caused by truncation and interpolation errors. Using these methods assimilation runs were made and SITEs were determined. These were analysed with the purpose to deduce the causes of the systematic mean errors found in AMIP runs with the model. To try to distribute the total SITEs on different model processes we use methods based on comparisons of the three dimensional structure of the SITEs and the three dimensional structure of the time mean tendency fields of each of the model processes in question. This reviles deficiencies in the representation of key physical processes in the climate model, which together form a basis for improvement of the model.