Estimating extreme wind speed based on regional frequency analysis

• Apply regional frequency analysis to extreme wind speed for Canadian wind records.
• Use k-means, hierarchical and self-organizing map clustering to identify clusters.
• k-Means method is the preferred exploratory tool for the considered data.
• The GEV distribution fits the normalized data within a cluster better.
• The GEV distributions are associated with unrealistic low values of the upper bound.

The quantiles of wind speed at spatially distributed locations within a region that are needed for codifying wind load can be estimated based on the at-site analysis of the annual maximum wind speed using records at a number of meteorological stations. The historical wind records and available meteorological stations, however, are often short and insufficient or unavailable; a decreased sample size increases the uncertainty in the estimated quantiles. To overcome the problem with data insufficiency, the use of the regional frequency analysis applied to annual maximum wind speed is investigated in this study by using wind records from 235 Canadian meteorological stations. The analysis uses the k-means, hierarchical and self-organizing map clustering to explore potential clusters or regions; statistical tests are then applied to identify homogeneous regions for subsequent regional frequency analysis. Results indicate that the k-means is the preferred exploratory tool for the considered data, and that the discordancy measure is valuable to identify stations with wind records that may require further scrutiny. Results also indicate that the generalized extreme value distribution provides a better fit to the normalized data within a cluster than the Gumbel distribution. However, the former is associated with low values of the upper bound that influence significantly the return period values with return period greater than 500 years.