Site or regional design wind speeds?

Abstract

Estimates of high return period design wind speeds are required to achieve an acceptable probability of failure. These can be found by fitting an appropriate distribution to observed data but is complicated by spatial variation of the wind climate, therefore it is unclear which data are relevant at a particular location. We aim to frame this problem in terms of model selection and the bias variance trade-off. Existing models for incorporating site versus regionally averaged statistics are expressed in terms of key parameters, which allow the errors associated with each model to be derived. It is found that using a characteristic wind speed in codification combines the site statistics (low bias) with regionally averaged statistics (low variance). The return period of the characteristic wind speed is shown to act as a parameter controlling the relative weighting of these models. This insight is used to develop an optimal (minimum mean square error) estimator of the design wind speed, which is shown to vary based on both the available quantity of data at a particular site and how well this data corresponds to the regional average. Some practical advantages of this optimal model are then demonstrated at several South African stations.