This remains idle until a new set of satellite data is available on the SatBaltyk server, at which time the system switches to assimilation mode. It performs data assimilation, sets the assimilated data as the new initial state of the model and performs new calculations from the time of the satellite data’s appearance until the current forecast ending time. Afterwards the system uploads new

results in the same way as in the regular mode. Then it switches back to regular mode. The Fig. 1 outlines the scheme of how the system operates. The test run of the model was performed on the historical data covering the years 2011 and 2012. Independent calculations were performed for the model with and without satellite SST assimilation, respectively referred to in Panobinostat this paper as 3D CEMBS_A and 3D CEMBS. The results of both runs were compared selleck compound with each other as well as with satellite data and different in situ measurements. Validation of the satellite data assimilation with the 3D CEMBS model consisted of two parts. Firstly, the results of both models were compared with the satellite data to check whether the assimilation algorithm was working properly and to examine the impact of the assimilation on the model results. Then, the results from both

model test runs were compared with different in situ data to check whether Amobarbital the assimilation actually improved the overall model accuracy. For a preliminary

assessment of the correctness of the assimilation algorithm, sample images from the satellite were compared with the results of both models from different days. Fig. 2 shows the sample scene from January 1st, 2011. The figure consists of the model data before assimilation, the satellite data used for assimilation and the model data after satellite data assimilation. The picture at bottom right shows the difference between the two models. In this example the satellite measured temperature is mostly lower than the one calculated by the model before assimilation. Assimilation lowers the temperature in the model surface layer, as expected. The same results were obtained for other scenes, which indicates that the assimilation algorithm is working properly. Of course, visual comparison is not sufficient, so additional tests were performed. In order to assess the accuracy of the assimilation algorithm and model accuracy, statistical parameters such as the correlation coefficient r, the mean systematic error 〈ɛ〉 and the standard deviation 〈σ〉 between both models and satellite data were calculated for all data from the years 2011 and 2012, as were the mean values and differences between the models. After validation of the assimilation algorithm, the same methods were used to assess the model error with respect to in situ data.