Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Civil and Environmental Engineering

Supervisor

Dr. Hanping Hong

2nd Supervisor

Dr. Jinfei Wang

Joint Supervisor

Abstract

Consideration of wind load is important for design of engineered structures. Codification of wind load for structural design requires the estimation of the quantiles or return period values of the annual maximum wind speed. The extreme wind speeds are estimated based on the measured wind records at different meteorological stations and affected by the length of the wind record (i.e., sample size) and other factors such as the surrounding terrain and so on. This study is focused on 1) the spatial interpolation of wind speed statistics, 2) the potential of using regional frequency analysis in estimating the extreme wind speed, and 3) the reliability of designed structure at sites with and without sample size effects.

For the spatial interpolation, both code recommended values of the wind speed as well as those based on at-site analysis are used, and commonly used spatial interpolation methods including 8 deterministic methods and 6 geostatistical methods have been applied. The preferred methods for each data set are determined based on the (leave-one-out) cross validation analyses. It is shown that the preferred method depends on the considered data set; the use of the ordinary kriging is preferred if a single method is to be selected for all considered data sets.

The historical wind records and available meteorological stations are often short and insufficient or unavailable, and the limited sample size will cause the uncertainty in the estimated quantiles. To deal with the data insufficiency in the wind speed records at the meteorological stations, the regional frequency analysis was applied to the data from the same 235 Canadian meteorological stations as mentioned above to calculate the quantiles of the annual maximum wind speed for Canada. The obtained estimates of the quantiles of the extreme wind speed based on the regional frequency analysis are compared with those obtained based directly on the at-site analysis. The analysis uses the k-means, hierarchical and self-organizing map clusteringtoexplore potential clusters or regions; statistical tests are then appliedtoidentify homogeneous regions for subsequent regional frequency analysis. Results indicate that the k-means is the preferred exploratory tool for the considered data and the generalized extreme value distribution provides a better fit to the data than the Gumbel distribution for regional frequency analysis. However, the former is associated with low values of the upper bound that do not influence the estimation of 10- to 50-year return period values of annual maximum wind speed but do influence the return period values with return period greater than 500 years. Based on these observations, regional frequency analysis may not be needed as an alternative to the at-site analysis.

Furthermore, since the estimated quantiles of the extreme wind speed at a site are uncertain due to the limited sample size, the effect of this statistical uncertainty on the estimated return period value of the wind speed and structural reliability is investigated and two strategies (i.e. (i) a low return period for the nominal wind speed combined with a wind load factor greater than one and (ii) a high return period for the nominal wind speed combined with unity wind load factor) for specifying the factored design wind load are also evaluated to determine the optimal one. Results indicate that at least 20 years of useable data are needed for a station to be included in the extreme value analysis, and the first strategy is preferred to cope with sample size effect for the design at a particular site or in a region with statistically homogeneous wind climate, while the second strategy is recommended for the code making for a country with spatially varying coefficient of variation of annual maximum wind speed since it leads to better reliability consistency.


Share

COinS