35 3.3.3.2. Results and Conclusions Initially seven model fields were compared. The matrices R and ? are given in Tables 11 and 12, where the symmetric matrix R describes the correlation in range (-1; +1) between pairs of fields. Positive values mean that anomalies on more than 50% of the area have the same sign and the bottom row is the model averaged value, ?j, describing similarity of a given field to the rest of the fields. The overall averaged value for the matrix R is ??? = 0.21. The matrix ? is the mean square distance of a given field to the rest of fields and the overall averaged value is ??? = 2.95°C. The first class of fields can be separated for ?j > ???. Five fields belong to the first class, i.e. HYCOM, JAEA, JCOPE2, NCOM and SELFE and all of these fields (except JAEA) also satisfy another condition, i.e. ??? > ?j. The remaining fields belong to other classes. From the first class two fields can be chosen as etalon-fields, i.e. NCOM and SELFE. Regarding similarity with the observed SST field, the modelled field and observed field were subtracted from the average of 8 fields and the matrices R and ? are summarized in Tables 13 and 14. The overall averaged values are ??? = 0.222 and ??? = 2.86°C. Table 13 shows that observed SST has the highest correlation with HYCOM, NCOM and SELFE fields and that the closest distance is again for the HYCOM, NCOM and SELFE fields. The first class of fields can be separated by criterion ?j > ???, with five fields belonging to the first class, i.e. HYCOM, JAEA, NCOM, SELFE and observed SST, with JCOPE2 in the field ?j = ???. With the exception of JAEA, these fields also satisfy another condition, i.e. ??? > ?j. The remaining fields belong to another class. From the first class, one field can be chosen as etalon- fields, i.e. NCOM. Some general observations concluded from this study are as follows: ? Visual inspection and objective classification agree in the conclusions derived from the analysis of SST field similarity in the chosen domain; ? The presented results show the ability of the method of classification to determine field analogues; ? The set of fields can be used to obtain a statistically valid classification of ensemble results; ? The method of classification can also be applied to the analysis of fields of calculated radionuclide concentrations. During previous modelling work concerning the Fukushima Daiichi NPP accidental releases into the Pacific Ocean [56, 57], it was found that a better agreement between model calculations and measured 137Cs concentrations in water and sediment was obtained if JCOPE2 circulation was used than with NCOM or HYCOM. SELFE results for the first dispersion exercise are significantly noisy. Moreover, the JCOPE2 field is within the limit ?j = ??? and satisfies the additional condition ??? > ?j. Consequently, it is concluded that JCOPE2 is a reasonable selection to carry out the dispersion exercises described in the following sections, in which all models will use the same hydrodynamics.