62 3.5.2.2. Exercise 4b: A range of expert estimations Results for Exercise 4b (with models using different water circulation and own set of parameters) are presented in Figures 42 and 43. The decrease in surface water 137Cs concentrations produced by models in Exercise 4a around day 50 at some points (using JCOPE2 water circulation) is now not produced by the JAEA model (e.g. see Figure 42, T4), which reproduces the measured 137Cs concentrations very well. This model uses water circulation from the University of Kyoto hydrodynamic model, which has a higher spatial resolution in the area of Fukushima than JCOPE2. This higher resolution may be leading to more accurate modelling of water circulation and a less noisy concentration time series. Except in the case of JAEA, models tend to underestimate water 137Cs concentrations. In the case of Exercise 4b, results for sediment are presented in Figure 44. The use of di?erent water circulation by the models results in di?erent distributions of 137Cs activity concentration in sediment. It is particularly interesting to observe that the University of Kyoto circulation (in the JAEA model) leads to a very narrow contaminated band along the coast and in the Bay of Senday. There is not any extension of 137Cs north of 38.5° latitude, which does not agree with measurements. However, water activity concentrations calculated with this model were in the best agreement with measurements. Thus, the JAEA model performs better than the others when calculating surface water concentrations, but worse than the others for sediment. This situation cannot be attributed to the water–sediment interaction description, since in the case of Exercise 4a (see Figure 41) output of this model was similar to the others. Instead, it seems that the University of Kyoto circulation model does not accurately reproduce deep circulation. In this sense, as mentioned previously [37], models may perform di?erently depending on the target variable. For instance, one model may predict radionuclide concentrations in sediment in good agreement with measurements, but it may provide not so close agreement for water. For another model, the situation may be the opposite. The di?erences between the Eulerian (I/K-E and USEV) and Lagrangian (JAEA and KAERI) models may be clearly appreciated from the maps of sediment concentrations (see Figures 41 and 44). Eulerian models introduce artificial (numerical) di?usion which leads to smoother concentrations maps, with 137Cs present over the whole domain. It is noted that, while Exercises 1 to 3 are completely blind model tests, this is not entirely true for Exercise 4b due to the possible influence of model results by existing knowledge of measured data. In the case of the I/K model, the desorption rate from the sediment was fitted in order to reproduce measured concentrations in sediment. However, model results can be considered as blind for surface water. Exactly the same occurs in the case of the USEV model. The KAERI model was slightly modified with respect to Exercises 1–3 in order to obtain a better agreement with observations. This modification consisted of making the release in a single point instead of into an Eulerian grid cell. Moreover, it was also found that the best agreement with observations was obtained with parameters defined in Exercise 3. Consequently, results for KAERI model Exercises 4a and 4b are the same and the KAERI model results cannot be considered blind. Finally, the JAEA model application has been a blind exercise for both water and sediment. In spite of some contamination of model results by knowledge of data, model results are in general consistent with observations. The range of computed values for a given variable may be regarded as an estimation of model uncertainty. Overall, 137Cs concentrations in surface water tended to be underestimated, while a good representation of sediment was generally obtained.