[ncl-talk] Question about Spectral Analysis(computation of variance)

Mon Jan 16 08:04:37 MST 2017

My previous email addressed the major peak in your original email.

Data with high auto correlation create a "red spectrum". This has more
power at lower frequencies. Please read:

http://www.nws.noaa.gov/om/csd/pds/PCU2/statistics/Stats/part2/Noise_red.htm

Without knowing the details, I would say the spectrum is likely correct.

I think you should talk with your advisor for details.

Good luck

On Sun, Jan 15, 2017 at 10:47 PM, Lyndon Mark Olaguera <
olagueralyndonmark429 at gmail.com> wrote:

> Dear Sir Dennis,
>
> Many thanks for the reply. I really thought that the iopt in the
> specx_anal() function would remove the mean annual cycle (should be in
> terms of anomalies!).
> I misunderstood this. :-(
>
> I followed your suggestions above and rerun the script (detrend and use
> anomalies). The new output is attached in this email.
>
> The variance/freq is still very large (~600).
>
>
>
> Many thanks for the help.
>
> *Lyndon Mark P. Olaguera*
>
>
> On Mon, Jan 16, 2017 at 7:55 AM, Dennis Shea <shea at ucar.edu> wrote:
>
>> The "power" at the lowest frequencies is because:
>>
>> [a] you have not removed the mean annual cycle
>>      and/or
>> [b] you not removed linear trend.
>>
>> Both contribute to "power" at low frequencies.
>>
>> Removing the influence of known signals that may obscure the signals is
>> called "prewhitening". Please look this up.
>>
>> I speculate [a] is the primary reason.
>>
>> http://www.ncl.ucar.edu/Applications/climo.shtml
>> http://www.ncl.ucar.edu/Applications/Scripts/climo_5.ncl
>>
>> http://www.ncl.ucar.edu/Applications/mjoclivar.shtml
>> http://www.ncl.ucar.edu/Applications/mjoclivar.shtml#ex2
>>
>> Please read:
>> http://www.ncl.ucar.edu/Document/Functions/Contributed/
>> calcDayAnomTLL.shtml
>>
>> Good Luck
>>
>> On Thu, Jan 12, 2017 at 7:59 PM, Lyndon Mark Olaguera <
>> olagueralyndonmark429 at gmail.com> wrote:
>>
>>> Dear Sir Dennis,
>>>
>>> I tried to perform the spectral analysis in different regions using a
>>> different data set from my previous post. For example, Western Africa and
>>> South China Sea but I still get the same spike in variance near the 0 mark.
>>> I am using gpcp daily data. Attached are the images and my script.
>>>
>>> Any suggestion on how can I fix this?
>>>
>>>
>>> Here is the link to the data that I am using.
>>>
>>> https://drive.google.com/drive/folders/0B9faET7Bc2o8R0ZsZXM1
>>> bkxUWlE?usp=sharing
>>>
>>>
>>> Many thanks,
>>>
>>> Lyndon
>>>
>>>
>>> On Mon, Jan 9, 2017 at 11:14 AM, Lyndon Mark Olaguera <
>>> olagueralyndonmark429 at gmail.com> wrote:
>>>
>>>> Hi Sir Dennis,
>>>>
>>>> Thanks for the help. I was looking for peaks with periods of 30 to 60
>>>> days.
>>>> The data is not good for this analysis. I also tried in the raw daily
>>>> values (attached in this email) but there is still a very large
>>>> variance/frequency.
>>>>
>>>>
>>>>
>>>> On Sat, Jan 7, 2017 at 1:42 AM, Dennis Shea <shea at ucar.edu> wrote:
>>>>
>>>>> [1]
>>>>> As noted in the documentation:
>>>>>    "The units are variance/(unit frequency interval)."
>>>>>     **not**
>>>>>    "variance"
>>>>>
>>>>> [2]
>>>>> Look at the data ... literally
>>>>>    plot = gsn_csm_y(wks,rain,False)                      ; see
>>>>> attachment
>>>>>
>>>>> Basically, the annual cycle of rain.
>>>>>
>>>>> What is it you expect from a time series like this? The lag-1 day auto
>>>>> correlation is 0.96. A ***very red** process.
>>>>>
>>>>> Assuming this is a climatology, I did a [four)info] on the data, The
>>>>> 1st harmonic has amplitude 7.3 with the phase max at day 220 and it
>>>>> explains 82% of the variance. The 2nd harmonic has amplitude 2.3; phase day
>>>>> 32; 8% of the variance.
>>>>>
>>>>>
>>>>>
>>>>> On Fri, Jan 6, 2017 at 1:57 AM, Lyndon Mark Olaguera <
>>>>> olagueralyndonmark429 at gmail.com> wrote:
>>>>>
>>>>>> Hi All,
>>>>>>
>>>>>> I'm trying to perform spectral analysis using the specx_anal function
>>>>>> in ncl.
>>>>>> I'm getting a variance around 1000 near the zero mark.I dont know if
>>>>>> I'm missing a step prior to performing the spectral analysis. I would like
>>>>>> to ask for any suggestions on how can I improve this.
>>>>>>
>>>>>>
>>>>>> Attached are the data set and script.
>>>>>>
>>>>>> load "$NCARG_ROOT/lib/ncarg/nclscripts/csm/gsn_code.ncl"
>>>>>> load "$NCARG_ROOT/lib/ncarg/nclscripts/csm/gsn_csm.ncl"
>>>>>> load "$NCARG_ROOT/lib/ncarg/nclscripts/csm/contributed.ncl"
>>>>>> load "$NCARG_ROOT/lib/ncarg/nclscripts/csm/shea_util.ncl"
>>>>>>
>>>>>> begin
>>>>>> filename = "type1_clim.csv"
>>>>>> lines    = asciiread(filename,-1,"string")
>>>>>> data     = lines(1:)
>>>>>> rain    = tofloat(str_get_field(data,10,","))
>>>>>>
>>>>>> d = 1
>>>>>> sm = 12
>>>>>> pct = 0.10
>>>>>>
>>>>>> spec= specx_anal(rain,d,sm,pct)
>>>>>> printVarSummary(spec)
>>>>>>
>>>>>> wks  = gsn_open_wks("png","test_fldmean")             ; Opens a ps
>>>>>> file
>>>>>> res = True
>>>>>> res at tiMainString = "Type1_Spectrum"                     ; title
>>>>>> f = spec at frq
>>>>>> res at tiXAxisString = "Frequency (cycles/day)"  ; xaxis
>>>>>> res at tiYAxisString = "Variance"                       ; yaxis
>>>>>>
>>>>>> splt = specx_ci(spec, 0.05, 0.95)
>>>>>>
>>>>>> res at xyLineColors        = (/"foreground","green","blue","red"/)
>>>>>> plot = gsn_csm_xy(wks,f,splt,res)     ; create plot
>>>>>> end
>>>>>>
>>>>>> I'll appreciate any help.
>>>>>>
>>>>>> Many thanks,
>>>>>>
>>>>>> Lyndz
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> ncl-talk mailing list
>>>>>> ncl-talk at ucar.edu
>>>>>> List instructions, subscriber options, unsubscribe:
>>>>>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20170116/fb572339/attachment.html 