In the present paper, ten-year long three-hourly time series of wind and wave data, based on hindcasts of WAVEWATCH III model, are analyzed and modelled as nonstationary stochastic processes. The study area covers the region [100 E, 70 W] × [60 S,66 N]. The initial time series is decomposed by means of the nonstationary modelling, and the residual stationary part is used as input to a Fuzzy Inference System (FIS) in combination with an Adaptive Neuro-Fuzzy Inference System (ANFIS) for the prediction of future values of wind and wave parameters. For comparison purposes, the FIS/ANFIS models are also applied to the initial nonstationary series without making any decomposition. The performance of both forecasting procedures is assessed by means of well-known error measures. It should be noted that FIS/ANFIS models are coupled for the first time with a nonstationary time series modelling for forecasting purposes.
Wind and wave data are very important for a number of applications connected with the open ocean and coastal activities. Thus, shortand/ or long-term forecasting of them is of great practical utility in areas ranging from navigation safety and oil response contingency planning to coastal erosion and ocean climate change. Third generation spectral wave models (the WAMDI group 1988; Tolman 1991; Booij et al 1999) run on a regular basis providing us with long-term data of good quality with high spatial and time resolution without gaps, and, thus, can be used for forecasting purposes (either off-line or in near realtime); see, e.g., Roulston et al (2005), Reikard and Rogers (2011). However, and because their numerical implementation is quite complicated, they require great computational power and high CPU time.
On the other hand, various researchers treat the forecasting problem by means of various soft computing techniques. Some of them utilize Artificial Neural Networks (ANN); see, e.g., Deo et al (2001), Rao and Mandal (2005), Jain and Deo (2007). Some others use Fuzzy Inference Systems (FIS) in combination with Adaptive Neuro-Fuzzy Inference Systems (ANFIS); see, e.g., Kazeminezhad et al (2005), Özger and Sen (2007), Mahjoobi et al (2008), Zamani et al (2008), Sylaios et al (2009), Akpinar et al (2014). These techniques require less computational effort and they are easy to be applied.
