1
$\begingroup$

I have 10 years worth of mean wind speed meteorological data (timely 10-minutes data). I want to compare the observed wind speed data against a Weibull distribution on a graph of the probability density distribution. I used function in R:fitdistr(). But the Weibull paramets, which gave me this function is probably wrong? Because when I made qqPlot with this paramets, I had something like this: enter image description here

enter image description here

Help me please and tell me, what I have been wrong? What should I do?

$\endgroup$
  • 1
    $\begingroup$ You have made the comparison and found your data do not look like they have a Weibull distribution. What that means to you depends on why you were doing this. What was your purpose? $\endgroup$ – whuber Feb 17 '15 at 22:44
  • 1
    $\begingroup$ 1. Some advice: Show much narrower bins in your histogram so you can see more detail. Since you have plenty of data, I'd suggest bins equal to the discrete intervals your data are already binned to. 2. Your data show a heavier right tail than the Weibull that would describe the middle part of the data, and a heavier left tail as well. 3. If your data were close to a Weibull ... what you would do? $\endgroup$ – Glen_b -Reinstate Monica Feb 18 '15 at 2:05
  • $\begingroup$ I'm writing my Master Thesis at the University about Extreme Wind Speed in Poland. I have a kind of data like I wrote above and I would like to analyse my data by means of Weibul distribution (You know, I want to find the percentiles [0,9; 0,95; 0,999] by menas of qweibull() and I need to do this the paramets of Weibull distribuation). I'm not to good with statistic and I need some help. $\endgroup$ – Arek Feb 18 '15 at 16:25
  • $\begingroup$ A heavier-tailed distribution seems to be called for; it may be for example that a scale-mixture of Weibulls might be sufficient. $\endgroup$ – Glen_b -Reinstate Monica Oct 9 '16 at 2:27
1
$\begingroup$

I want to find the percentiles [0,9; 0,95; 0,999]

You don't need to fit this using a Weibull-distribution. Given $N$ observation, sort the vector of observations and take the 1st, the $N0.9$th,$N0.95$th, etc

Also as @glen_b noted, you need to add more bins to the histogram before even considering if your visual inspection has any merit

$\endgroup$
  • 1
    $\begingroup$ Certainly for the histogram it's almost impossible to tell, but the QQ plot is fairly unequivocal -- that's far too heavy a right tail compared to the middle of the distribution. $\endgroup$ – Glen_b -Reinstate Monica Oct 9 '16 at 2:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.