Implementation and validation of a wave hindcast/forecast model for the Western Mediterranean

In this work we present a study of the performances of the WAWEWATCH III (WWIII) model in the Western Mediterranean basin. The model is implemented using the last generation source terms proposed by Ardhuin et al. (2008). Model validation is carried out referring to eight case studies corresponding to heavy wave storms in the northern Tyrrhenian basin (Ligurian sea) in the last twenty five years. Input wind data for wave generation are elaborated by means of WRF meteorological model assimilating CFSR reanalysis. We realize an analysis of the performances of different sets of source terms and parameterizations through a comparison between numerical simulation results and buoy data provided by Rete Ondametrica Nazionale (RON). In particular simulations are realized using the widespread Tolman-Chalikov (1996) source terms, and Bidlot et al. (2005) parameterization, the latter being very similar to the reference parameterization (Janssen et al., 1994) of the famous WAM cycle-4 model. A sensitivity analysis in parameters space is performed on ACC350 parameterization (Ardhuin et al. 2009). This task is achieved varying each parameter of ACC350 around its reference value, leaving the other ones at their reference value, obtaining 41 different parameterizations. Buoy and simulation data are compared through characteristic statistical error measures (bias, scatter index, correlation coefficient and square root mean error). The statistical error analysis is performed considering all the buoys together, and grouping those of Tyrrhenian, Ionian and Adriatic basins in order to test the reliability of the model chain in the different basins. For the examined cases, simulations realized using ACC350 perform generally better than those performed using other parameterizations, especially in the Tyrrhenian basin where storms are usually more severe, while the reference parametrization of Tolman-Chalikov (1996) source terms, and to a less extent Bidlot et al. (2005), tend to underestimate wave height. Moreover, parametric variations of ACC350 don't improve significantly the results of the simulations. Finally, in order to analyze the role in wave generation of higher frequencies, the investigation in parameters space is performed employing two different spectral ranges. The first range considers 25 frequencies separated by a factor 1.1, the second considers 30 frequencies, and for both of them the minimum frequency corresponds to a period of about 15 s. The usage of an enlarged spectral range towards higher frequencies does not produce an improvement in the simulations results.