Validation of Historical Operational Data of the Marine Forecasting Centre 2018-2021

This report presents the validation of the hydrodynamic and wave operational models run by the Marine Forecasting Centre of Belgium, RBINS (MFC) for the period 2018-2021. These models are run twice daily. The hydrodynamic models suite used in this study are set-up with the COHERENS software, a three-dimensional hydrodynamic multi-purpose model for coastal and shelf seas, which resolves mesoscale to seasonal scale processes. The model suite consists of three one-way nested models: the Continental Shelf model (CSM, 2D) covering the European Continental Shelf to its break, the North Sea Shelf (NOS,3D) model, covering the North Sea and part of the English channel and the Belgian Coastal Zone model (BCZ, covering Belgian waters and their immediate surroundings). Two wave models are set-up with the SHYPRO operational system and are called Hypas and Refra, where Refra refines the results of Hypas. There is also a third operational wave model set-up with WAM software (referred to as the WAM-model in this report). The models are validated by measurement data made available on the CMEMS server. Globally, the wave models perform consistently at the different stations, all Taylor diagrams are similar. The correlation is relatively poor, and ranges from 0.3 to 0.8 with an average of ~0.55. The ratio of the standard deviations is usually below 1: the models introduce more variability than is present in the data. The analysis of time series of actual data indicates that the models tend to generate events of high average frequency that do not reflect the reality. On the other hand, the observations show the zero up-crossing frequency varies rapidly in the short term; variability that is not reproduced by the model. In terms of bias, the stations seem to differ, creating a group of off-shore stations (Europlatform, K13) with larger bias or coastal stations (Akkaert, Wandelaar and Westhinder) with more moderate bias. This might reflect the impact of the fetch, given that the wind is usually south-westerly in that part of the North Sea. The RMSE is usually larger than the Bias. The RMSE seems to be sensitive to the overall overestimation of frequencies, and especially to the high frequencies events. There is no clear clustering of stations like was the case for the bias, the tests on the differences in average RMSE are usually not conclusive. Salinity is very sensitive to river plume influences. When studying the background sea surface salinity (filtering out the tidal component), the models show a stable offset in salinity. This offset is best measured by the median of the error distribution. The correlation between data and forecasts is poor because the river discharge uses monthly climatologies. Hence, the models do not have the ability to forecast a drop of salinity due to increased river discharge. On the other hand, a variation in salinity due to hydro-dynamical conditions is reproduced. Statistical tests indicate that the BCZ model forecasts the background salinity with a better accuracy than NOS. The bias analysis of temperature shows that the forecasts are close to the observations and relatively balanced around 0 degrees. The monthly distributions of the error are narrow, with an average span (Q75 – Q25) of 0.268 degrees. Globally, the error seems to remain in the range [-1 ; +1.5] degrees. The sea level forecasts are globally very efficient, with ratio close to 1, and high correlation. The RMSE remains below 30 cm and the bias below 8 cm. The usual statistics are gathered in table 11.4, for all the stations available for each model. According to the metrics, the models perform equally well during low and high tide. The phase shift (tidal timing) however is consistently stronger for the high tides (max 34 minutes) than the low tides (max 17 minutes).

Models, Modelling, Marine environment, Belgian Continental Shelf, Belgian coastal zone

This is the bi-yearly validation report of MFC, www.marineforecasts.be