Project

General

Profile

Forums » Support »

Time series filtering

Added by Dirk Olonscheck almost 4 years ago

Hello,

I am unsure how to use the lowpass filter correctly.
I have a yearly timeseries and like to transform it do decadal, i.e. omitting all variability on timscales smaller than 10 years.
I tried with

cdo lowpass,0.1 data.nc data_decadal.nc

Is this correct? The description in the CDO Manual on how to determine fmin and fmax confused me.

Thanks,
Dirk


Replies (7)

RE: Time series filtering - Added by Ralf Mueller almost 4 years ago

hi Dirk!

could you upload the data?

cheers
ralf

RE: Time series filtering - Added by Dirk Olonscheck almost 4 years ago

Hi Ralf,
here is the data, the first file without the filtering, the second one with the lowpass filter applied. Is this correct?
Afterwards I do (with both files)
cdo timstd tsc_control_yr_detrend.nc tsc_control_yr_detrend_timstd.nc
to estimate the variability, comparing interannual and decadal.
I am simply not sure if "cdo lowpass,0.1" really removes all variability on timescales smaller than 10 years.
Thanks for your support!
Dirk

Download all files
tsc_control_yr_detrend.nc (13.7 MB) tsc_control_yr_detrend.nc annual, unfiltered
tsc_control_yr_detrend_decadal.nc (13.7 MB) tsc_control_yr_detrend_decadal.nc filtered, correctly?

RE: Time series filtering - Added by Ralf Mueller almost 4 years ago

Hi Dirk!

here is a plot of the original and the filtered data:

I didn't use the second upload, but called CDO instead. From my point of view it looks good. But I need to check the frequencies itself in more depth.

tsc_VS_lowpass.png (597 KB) tsc_VS_lowpass.png

RE: Time series filtering - Added by Ralf Mueller almost 4 years ago

Forgot to mention, that I used 

cdo -sellonlatbox,-6,-2,-2,0
to select a single location from your input file. This location I used to timeseries analysis.

RE: Time series filtering - Added by Dirk Olonscheck almost 4 years ago

Thanks Ralf!

This looks correct indeed. I can reconstruct your plot. I would be happy to hear whether the frequencies are as intended, but the result is promising. So "cdo lowpass,0.1" seems to be correct for omitting frequencies higher than 10 years for annual data.

Best,
Dirk

RE: Time series filtering - Added by Ralf Mueller almost 4 years ago

IMO 0.1 is the right value for 10-year frequency according to the documentation. '1' represents 'yearly'. Hence a 10-times smaller frequency must represent decadel.

The examples in the documentation use a value of 0.5 for 2-years variability. So, your value of 0.1 makes perfectly sense to me.

cheers
ralf

RE: Time series filtering - Added by Dirk Olonscheck almost 4 years ago

Thanks a lot! That's what I wanted to ensure. Makes totally sense for me now :)
Have a nice weekend,
Dirk

    (1-7/7)