Adapt return period for extreme value estimation to time duration of data

[Arthur Pecher]

Hey,

it might sound like an easy or stupid question, but I have a 30 year timeseries on which I have performed  a POT analysis.

Based on these peaks, I want to perform an extreme value analysis and thereby obtain the return period of extreme events.

When performing a Gumbel or GEV analysis, I would expect to have the largest peak of my 30 year timeseries, as value for the 30 year return period. Is this correct?

At the moment, the return period for the maximum of my timerseries depends on the amount of peaks that are included to do the analysis, not on the time duration of my timeseries...

To give an idea of the script (mainly based on Wafo - which is really a super help!!) and the resulting image is attached:

the peaks of the timeseries are in Hs

gum = plotgumb(Hs)

T = 1:100;          % return period [years]

sT = gum(2) - gum(1)*log(-log(1-1./T));

fig_num=fig_num+1;figure(fig_num);

semilogx(T,sT), hold on

N = 1:length(Hs); Nmax = max(N);

plot(Nmax./N,sort(Hs,'descend'),'.')

% plot(L_data./N,sort(Hs,'descend'),'.')

title('Return values in the Gumbel model')

xlabel('Return period')

ylabel('Return value'), 

grid on; hold off

Do I really need the right amount of data samples to represent correctly the timeseries, or what is the logic behind it?

Thank you a lot for your idea/help.

Best regards,

Arthur