Projected wave climate temporal variability due to climate change

This work proposes a new general procedure to stochastically analyze multi-model multivariate wave climate time series projections at different temporal scales. For every projection, it characterizes significant wave height, peak period and mean direction by means of univariate non-stationary distributions capable of capturing cyclic climate behavior over a reference time interval duration. The temporal dependence between the values at a given sea state and previous short-term wave climate is described with a vector autoregressive model (VAR). The multi-model ensemble wave climate characterization is based on a compound distribution of the individual non-stationary distributions and a weighted averaged VAR model. The methodology is applied to bias-adjusted wave climate projections derived using WaveWatch III forced by wind field data from EURO-CORDEX models at a location close to the Mediterranean Spanish coast. Results are compared to hindcast data which shows a clear bi-seasonal behavior. Different temporal references were considered, starting with a 1-year reference period to analyze overall changes in wave climate at scales ranging from days, months and seasons with respect to historic conditions. The results show that the projected wave climate has a very different temporal behavior than hindcast data, delaying and widening/shortening the start and duration of the two main seasons and including shorter term variations. Regarding the energetic content of the sea states, the compound variable highest percentiles of the significant wave height present lower values than the hindcast (approximate to 3-10%) during the traditionally more severe period (November-March) but higher values (approximate to 10-35%) during the calmer months. The projected peak period presents a similar temporal pattern to the hindcast data, while the mean wave direction shows a significant change from the historical bi-modal behavior towards more likely easterly waves throughout the year. Additionally, a 10-year analysis is done to find larger temporal variabilities such as decadal variations associated with the North Atlantic Oscillation. The observed temporal variability in the yearly seasonal pattern throughout the century is addressed by analysing 20-year rolling windows in all the model projections and in the compound variable. The compound distribution shows significant temporal variabilities throughout the century with the most severe periods and more likely severe waves during summer at the end of the century.