A Comparison of Various Numerical Wave Prediction Techniques

In recent years, the number of wave prediction models has increased greatly. These models range from relatively simple parameterizations of significant wave height as a function of wind, duration, and fetch to rather sophisticated solutions for the generation, propagation, and dissipation of two-dimensional (2D) wave spectra. It sometimes is suggested that any wave model will provide reasonable answers when properly applied, and that provide reasonable answers when properly applied, and that most of the deviations between measured waves and predicted waves can be explained by discrepancies predicted waves can be explained by discrepancies between actual and estimated wind fields. Although much of the error in wave prediction almost certainly is related to problems in determining a wind field, this paper examines the specific question of whether there are differences among these models such that even if the wind field were specified perfectly, there would remain significant deviations among predicted waves. First, wave generation under uniform wind fields is compared by use of nondimensional parameters. Then the models are compared again under conditions of time-varying, space-varying wind fields and with irregular fetch boundaries. We concluded that, in the open ocean with a long-duration, slowly varying weather system, most models produce similar results; however, near a coast or in produce similar results; however, near a coast or in regions with rapidly varying weather systems, marked differences can be expected from the use of different models. Introduction The need for wave data has led increasingly to the use of wave hindcast techniques to produce wave climates, and a number of major hindcast efforts are under way in the U.S. alone. Numerous techniques are available, ranging from significant wave techniques in which wave parameters can be estimated from nomograms, to parameters can be estimated from nomograms, to directional spectral models, which usually are run on large-core, high-speed digital computers. Table 1 lists some of these techniques. A common underlying assumption of practicing engineers is that each of the techniques will practicing engineers is that each of the techniques will produce similar results when properly applied with produce similar results when properly applied with correct wind input. This paper demonstrates that this is not always the case. Instead, various models can be shown to have theoretical differences that in climatological as well as specific applications might lead to significant discrepancies in estimates of sea state.Since all wave hindcasts begin with reconstruction of past wind fields from historical records, a baseline error past wind fields from historical records, a baseline error present in all wave estimates comes from inaccuracies in present in all wave estimates comes from inaccuracies in available meteorological data. Often it seems as though investigators tacitly assume that the wind error dominates the total error term in hindcast studies and, hence, that the absolute accuracy of the wave model is not that important. A consequence of this might be that, where available meteorological data are high-quality, a wave model of high quality should be used; but where available meteorological data are low-quality (or sparse in time and space), a simple wave model will suffice. This logic assumes that any errors introduced by the wave model should be of comparable magnitude to those implicit in the meteorological input. It is not clear, however, that this is a reasonable argument with respect to errors, since they tend to be additive. Thus, the root mean square error will increase by the square root of 2 when a wave model with independent error characteristics of equal magnitude to the meteorological data is applied. If the error is already large, adding 40% to it could be detrimental to the final results. SPEJ p. 764