Sampling Bias in the Estimation of Significant Wave Height Extreme Values

Abstract

It has been shown before, and it is intuitively evident, that in a Significant Wave Height (SWH) time series, the longer the sampling interval, the lower is the number of events which are above a given threshold value. As a consequence, the use of data with a low time resolution (such as a 3 h sampling, for instance) causes a considerable undervaluation of the extreme SWH values for a given return time RT. In this paper an example of such a bias is provided, and a method is suggested to estimate it on a regional basis. Results may help to improve the use of historical wave meters data which were often collected with a low time resolution, and may also provide a tool to improve the application of Numerical Meteo-Wave models to the evaluation of extremes.

An Experimental Assessment of Extreme Wave Evaluation by Integrating Model and Wave Buoy Data

“…An inherent negative bias in the model has thus been completely removed. It is worth mentioning that the presence of such a negative bias confirms what had been highlighted in previous work [1,3,25,38], i.e., that the weather/wave models underestimate the extremes despite the constant assimilation of satellite measurement, which due to their coarse temporal and spatial resolution are likely to miss the strongest peaks of the storms.…”

“…Most of the assimilation procedures are carried out with satellite altimeter data, which are scattered in time (at many hours' intervals) and wide apart in space (tens or hundreds of kilometers), so extreme SWH values may often be missed. It is also worth noting that the sampling time of the models, i.e., the time interval at which data are stored and released, is often higher than the standard sampling time of buoys, thus causing a negative bias on the estimated extreme values [1][2][3].In order to overcome these problems, an integrated procedure [4] was been proposed by some of the authors of the present paper whereby the curves of extreme SWH as a function of the return time T R (in the following: SWH(T R )) deriving from synthetic data are compared and calibrated with the similar curves computed from buoy data in different locations. This provides a way of deriving SWH(T R ) curves for sites where no experimental data are available.The present paper presents an extension of the same technique and provides an experimental authentication of the methodology based on a new large set of reliable data along the coasts of the USA.The determination of the probability of extreme SWH is one of the main problems of coastal, offshore, and marine engineering, so that the relevant literature is not only extensive, but also increasing with time as the technology improves and the requirements become more stringent.…”

“…Most of the assimilation procedures are carried out with satellite altimeter data, which are scattered in time (at many hours' intervals) and wide apart in space (tens or hundreds of kilometers), so extreme SWH values may often be missed. It is also worth noting that the sampling time of the models, i.e., the time interval at which data are stored and released, is often higher than the standard sampling time of buoys, thus causing a negative bias on the estimated extreme values [1][2][3].…”

Extreme Wave Analysis by Integrating Model and Wave Buoy Data

“…Models generally underestimate extreme values compared to experimental data, partly due to the inevitable smoothing of the results due to the numerical interpolation, and partly due to inherent limitations of the model chain, as it was shown quantitatively in [13][14][15][16]. As a consequence, it is to be expected for a given return time T R that the corresponding significant wave height return value computed with the indirect data HM(T R ) is lower than the value obtained by making use of direct data HD(T R ).…”

“…The results shown for instance in [10,16], prove that use of data with a low time resolution (such as a 3 or 6 h) causes a considerable undervaluation of the extreme SWH values for a given return time T R . The following Figure 1 [16] illustrates this point-by degrading the original buoy data from a sampling rate of 30 (full data set) to a sampling rate of 6 h, there is an important reduction of the estimated H S (T R ). This raises the problem of deciding what would be the "right" interval to choose in order to compute the H S (T R ) curves.…”