Statistical post processes for the improvement of the results of numerical wave prediction models

Abstract

A new mathematical technique for adaptation of the results of numerical wave prediction models to local conditions is proposed.The main aim is to reduce the systematic part of the prediction error in the direct model outputs by taking advantage of the availability of local measurements in the area of interest. The methodology is based on a combination of two different statistical tools: Kolmogorov-Zurbenko (KZ) and Kalman filters.The first smoothes the observation time series as well as that of model direct outputs in order to be comparable via a Kalman filter.This is not the case in general, since forecasted values are smoothed spatially and temporarily by the model itself while observations are point records where no smoothing procedure is applied.The direct application of a Kalman filter to such qualitatively different series may lead to serious instabilities of the method and discontinuities in the results. The proper utilisation of KZ-filters turn the two series into a compatible mode and, therefore, makes possible the exploitation of Kalman filters for the identification and subtraction of systematic errors.The proposed method was tested in an open sea area for significant wave height forecasts using the wave model WAM and six buoys as observational stations.