Estimation of extremes of non-Gaussian wind pressure on building roof: Sampling error in moment-based translation process model with no monotonic limit

Abstract

Estimation of extremes of non-Gaussian wind pressure on building roof is necessary for cladding design. When limited length of non-Gaussian wind pressure is used for calculation, the estimated extreme involves sampling error. The moment-based Hermite polynomial model is extensively applied for estimation of extreme wind pressure due to the straightforwardness and accuracy, however, Hermite polynomial model has a monotonic limit resulting in a restricted application region of skewness and kurtosis combination. However, another two moment-based translation process models with no monotonic limit including Johnson transformation model and piecewise Hermite polynomial model have attracted some attention as these two models can be applied to a broader region of skewness and kurtosis combination. The sampling error in estimation of extremes of non-Gaussian wind pressure on building roof by Hermite polynomial model is proposed in the literature recently. Nevertheless, the sampling errors in Johnson transformation model and piecewise Hermite polynomial model have not been addressed. In this study, sampling errors in estimation of extremes of non-Gaussian wind pressures by Johnson transformation model are investigated. Formulations for estimating sampling errors of newly defined statistical moments and subsequent extremes in piecewise Hermite polynomial model are presented. The performance of sampling errors in Hermite polynomial model, Johnson transformation model, and piecewise Hermite polynomial model are finally compared with each other. Based on very long wind pressures from wind tunnel tests, it is shown that the sampling error of minimum wind pressure (suction) in Hermite polynomial model is generally the smallest compared to Johnson transformation model and piecewise Hermite polynomial model, while that of maximum wind pressure in piecewise Hermite polynomial model seems to be the smallest.