33 3.3.3.1. Methodology The computed fields were interpolated on the same regular grid with 200 × 200 cells for a domain extending 140.5–144°E in longitude and 35.5–38°N in latitude. The same coastal line mask was applied to all data and the daily satellite SST fields were averaged, after which remaining gaps in observations were filled by linear interpolation from surrounding areas. The field was smoothed and interpolated into the same grid as computation data. The resulting observed SST field is shown in Figure 18, where the computed fields of SST are also shown in the same scale as the observed field and the ensemble averaged SST field is also given. A simple method of classification (e.g. etalon-field [63]) was applied and, at a first step, the data matrix for SST anomalies was obtained by subtracting the model ensemble SST field from each model field. At a second step the geometric similarity between field pairs was identified using the criterion based on a sign of anomaly of two fields: ? = ????? ? , ?1 < ? < 1 (1) where: M+ is the number of grid points in which sign of anomaly in the field pair coincides; M– is the number of grid points in which sign of anomaly in the field pair is opposite; M++M– = M is total number of grid points. The square symmetric matrix R represents similarity between fields of anomalies: (2) where: N is number of fields in the ensemble. At a third step the mean square distance between two fields is used as second criterion of similarity: ??,?? = ? ? ? ????,? ? ???,?? ?? ??? (3) where: ?Ti,k and ?Tj,k are values of SST in point k of the grid from two fields i and j.