Nasibeh Rashidi - Institute of Geophysics, University of Tehran, Tehran, Iran
Sarmad Ghader - Institute of Geophysics, University of Tehran, Tehran, Iran
S. Abbas Haghshenas - Institute of Geophysics, University of Tehran, Tehran, Iran
Edris Delkhosh - Institute of Geophysics, University of Tehran, Tehran, Iran

The wave climate either for a period of years in the past or the daily variations of wave parameters in the next couple of days are of great importance for structural designas well as for marine operations. Wave measurements data are essential for studying wave climate in a certain area; however, measurements are not usually availableeverywhere and on a long-term basis; thus reliability of wave simulation is a crucial issue in coastal, harbor and ocean engineering. Hence, numerical simulations are to beadopted to obtain such data in locations of interest. Due to the lack of wind/wave measured data, predicted wind waves are used to determine wave parameters in manyregions. Numerical wind wave modeling has been practiced since the mid 1950’s for a variety of purposes, such as operational forecasting, coastal engineering design andcoastal evolution studies, and general wind wave research. Moreover, this type of data is essential for environmental studies (Huang et al., 2016). Traditionally, wind wavemodeling has been most actively pursued by coastal engineers. Nowadays spectral wind waves model are the most practical tool for wave predictions. Since the outputsof these models generally contain some errors, their results should modify verifying against available measured data. In this study, a new approach based on the errorprediction of simulated wind/wave data at the observing stations and distributing the errors in the computational domain, was implemented for updating the model outputs.This aims to improve the accuracy of reproduced measured wind and wave data by adopting machine learning methods. To do so, wind wave simulation ofmeteorological model was carried out over the Persian Gulf and the results were compared with the measured data. A machine learning approach is used to find out the relationship between the output of numerical models and the measured data to be applied for improving model results.