hi Sam,
in case you have verified the properties of the signal in the above comment,we can proceed as the following :
We have a signal of length L=3950 taken every 12 hours for 2563.5 days. So formally the sampling rate is Fs= 3950/2563 =~1.54 Hz .
For these types of times series , the range is dynamically changing therefore we plot the frequency in dB to expand the low values and compress the high ones :
1)Solution 1:
Fs=1.55;
L=length(Thermocline);
N=ceil(log2(L));
FFTherm=fft(Thermocline,2^N);
f=(Fs/2^N)*(0:2^(N-1)-1);
Power=FFTherm.*conj(FFTherm);
figure,
semilogy(f,Power(1:2^(N-1)))
xlabel(' Frequency (Hz)'), ylabel(' Magnitude (w)'),
title(' Power Spectral Density'), grid on;
2) Solution : 2
So the frequency is in the range [0,...,0.1] Hz. To see if this result, physically, makes sens, try to check :
"Stochastic climate models, Part I1, Application to sea-surface temperature anomalies and thermocline variability" by CLAUDE FRA NKIG NOUL :
And another paper: "MEAN AND EDDY DYNAMICS OF THE MAIN THERMOCLINE" by GEOFFREY K. VALLIS
These two papers give a general idea on how much an estimate of the frequency must be .
I hope this helps .
KHMOU Youssef.
Best Answer