67 general features. As a consequence, it could be concluded that a significant part of the variability between models is caused by the description of hydrodynamics. In the case of 137Cs sediment concentrations, model variability is increased due to the di?erent descriptions of water–sediment interactions used by each dispersion model. Exercise 3 consisted of using the same description for water–sediment interactions in the case of 137Cs. Thus, the same kd, or equivalent kinetic ratios (to be used in cases of equilibrium or dynamic models respectively), were used. An additional harmonization of models in this exercise consisted of using exactly the same topographic data for the Pacific Ocean and the same values for the horizontal and vertical di?usion coe?cients. This experiment was again carried out for a tracer and for 137Cs. In the latter case, the e?ects of considering the presence of suspended matter in the water column were also investigated. In the case of the tracer, agreement between models improved with respect to Exercise 2. The main factor in producing model discrepancies is water circulation since model agreement improvement is higher from Exercise 1 to 2 than from Exercise 2 to 3. This confirms a previous finding where di?erences between models are mainly due to hydrodynamics, although no systematic study was performed in order to assess this [60]. In the case of 137Cs, the use of the same water–sediment parameterization also led to a better agreement between model outputs in sediment. Calculated 137Cs concentration maps for water and sediment were also similar, with models producing the same behaviour. In this respect, it was also clear that a good description of contamination in the deepest water is essential for a good description of radionuclide adsorption by bed sediments. The presence of suspended matter in the water column does not a?ect the calculated dissolved concentrations. This is not surprising given the relatively low a?nity of 137Cs to be fixed to the solid phase and the low suspended matter concentration in open ocean waters. E?ectively, for the considered kd of 137Cs and suspended matter concentration (5 mg/L), the partition coe?cient indicates that around 99% of this radionuclide remains in solution. At this stage, given the model harmonization which was carried out, it does not seem possible to achieve a better agreement between models. Di?erences in model outputs were now due to intrinsic di?erences between models, i.e.: ? Lagrangian versus Eulerian models; ? Di?erent numerical schemes which may be used for each model category mentioned above. In this sense, the method used to reconstruct concentrations from the density of particles in Lagrangian models may be relevant. Nevertheless, the overall agreement between models which, generally, has been achieved, may be considered as satisfactory. The first three exercises assumed a constant hypothetical source of radionuclides. The final exercise consisted of the use of a realistic source term, allowing the comparison of model outputs with 137Cs measurements in water and sediment. The source terms for both direct releases to the Pacific Ocean and atmospheric deposition on the sea surface had to be reconstructed. The direct release source term was reconstructed by the JAEA model based on 137Cs measurements carried out by TEPCO at the southern and northern