Minimum number of samples for Generalised Extreme Wind Climate?

[julmou]

Good morning,

When creating our Generalised EWC, is there a minimum number of points/samples to consider it reliable and usable?

I mean, in some cases, only four points/samples are extracted from the time series, which seem a bit light to be able to have a good Generalised EWC...

And in some cases, I can have 22 samples, which seem more reliable to estimate our 50 year wind speed.

Thank you for your help,

Julien

[Morten]

You are right. Sometimes the software enables you to calculate EWC and U50 estimates even with sparse data and you wonder about the accuracy. Both annual-maximum and peak-over-threshold methods have associated uncertainty estimates. In WAsP Engineering, these uncertainties are indicated by the curves around the extreme wind versus return period plot. This statistical uncertainty depends on sample size and Gumbel alpha parameter. However, there is even uncertainty on the uncertainty estimate itself, so take care with very small samples.

The IEC 61400-1 standard includes an annex on measure-correlate-predict methods based on correlation with a reference station. It also discusses extreme-wind estimates based on the MCP long-term-corrected series, and it recommends at least seven years for the reference station. For extreme winds in tropical storms, it recommends methods based on satellite-tracking. I think you need something similar if you only have local data, but it also depends on the uncertainty you are willing to accept. 

In WAsP Engineering, I recommend to use the spectral correction method, whenever possible. This combines local information with a database using 21 years of modelled data with a short time series of local observations, e.g. only one year long. Unfortunately, the database does not yet cover Australia.