Here and throughout this article, X^ denote an estimate of X  . Wang and Swail, 2006 and Wang et al., 2010 applied this model to simulate seasonal mean or 12-hourly HsHs in the global oceans and in the North Atlantic, respectively, with spatial resolution of 2°°. Recently, Wang et al. (2012) extended the set of predictors in model (1), adding the principle components (PCs) of P(t,m)P(t,m)

and of G(t,m)G(t,m) over a domain that is larger than the wave field in question to represent the swell component of waves, as well as p  -lagged dependent variables, Hs(t-p,m)Hs(t-p,m), to account for serial correlation in the predictand (dependent variable) HsHs. They also proposed a data adaptive Box–Cox Crizotinib transformation procedure to diminish the departure of HsHs and SLP gradients from a normal distribution. They have shown that their new model is more skillful, resulting in less biased simulations of 6-hourly HsHs, than model (1). The methodological developments we propose below include physical and statistical aspects. On the physical aspects, we modify the way to account for swell waves by using the term ΔswΔsw as defined later in Section 4.2, and the way to account

for serial correlation in HsHs using the term ΔtΔt defined later in Section 4.3. Thus, our new model is of the form: equation(2) H^s(t,m)=aˆ(m)+aˆP(m)P(t,m)+aˆG(m)G(t,m)+Δsw(t,m)+Δt(t,m). The last term makes the statistical model more coherent with ocean wave physics, because it can be interpreted Selumetinib ic50 as a discrete approximation of the first order derivative that appears in the spectral energy balance governing equation (e.g. Holthuijsen, 2007). Such temporal dependence is especially important for high temporal resolution data as in the present study. In fact, it is closely related to the large autocorrelation found in the 3-hourly HsHs time series. More

details about the inclusion of this term are given in Section 4.3. On the statistical aspects, we take into account the data scale and explore the effects of deviation from the Gaussian distribution Interleukin-2 receptor assumption in the multiple linear regression analysis by transforming the data in different ways, as detailed below in Section 4.4. Since different regimes dominate in different seasons (see Section 2.1), waves in different seasons should be modeled, separately. In this study, we focus on the winter (most energetic) season, which is defined here as December–January–February. Swell waves are waves propagating across the ocean, after being generated remotely during a storm. As explained in Section 2.2, the Catalan coast is often affected by an important swell component coming from E. Ignoring swell waves would lead to a significant underestimation of HsHs.