I've got some wind speed measurements and I would like to find out which test is the most reasonable one to clarify if the data is stationary or not. Your help is highly appreciated.
Dear Thomas
I am not aware of the most recent progresses in this context, but when I was working on climate change in time series I was using tests based on the cumulative sum (Pettitt's test able to detect one change, and the Lombard's test able to detect multiple changes, see references in Vannitsem and Nicolis, 1991, in Contributions to Atmospheric Physics). I was also using the Mann-Kendall test also able to detect one change (whatever the nature). More recent development of these techniques were made by Olivier Mestre and colleagues at Météo-France. You can also find nice applications of the Mann-Kendall test in a paper of Reinhard Bohm (https://www.researchgate.net/profile/Wolfgang_Schoener2/publication/229879931_Regional_temperature_variability_in_the_European_Alps_17601998_from_homogenized_instrumental_time_series/links/5440ed370cf251bced6149f5.pdf)
Stéphane
Article Regional temperature variability in the European Alps: 1760-...
Note that there are several different definitions for "stationarity" and in particular it is not usually a question of whether the "data" is stationary but whether the random process generating the data is stationary. Also are you asking about stationarity in time or in space? Most statistical tests are based on an assumption of random sampling but random selection of data locations is not the same kind of "random sampling"
Hi Thomas,
I am not sure if you already have a satisfactory answer, but I would like to provide an answer anyway. First of all, I am glad to see that researchers/engineers across the world are still on a look out for a robust method to identify stationarity in wind speed time series. It gives me some comfort that I am not alone in this.
After parsing through the literature in the past, one of the better methods that I could think of is from Andreas et.al. (2008). It somehow takes care of correlations on a very short time scale, which I believe they do by estimating the integral time scale. Please find the details in this publication.
In case you have found a new/better method then I am very interested to know that too.
Kind Regards,
Ameya
Article Identifying Nonstationarity in Turbulence Series
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics