Re: Question about the OLR space-time Spectra

Hi,

This is a case of -999 being a bad choice of a missing value to use
for the input. The first input argument to dtrend_msg is an integer,
and since the missing value isn't explicitly set, -999 is used
internally.

With the test below, the value -999 ended up as a calculation
internally, and this caused a crucial "if" statement to be skipped,
because it thought this was a missing value.

I believe the fix is to simply not use -999 as a missing value. This
means you will need to change your code a little:

   t = ispan(0,ntime,1)*1.0
   t@_FillValue = 1e20 ; If you don't do this, -999 will be used
   b = dtrend_msg(t,temp,False,False)

We may change dtrend_msg to use something other than -999 for a
default missing value for floats and integers.

Also, one of our big plans for the next release of NCL is to change
some of the default missing values, as "-999" for floats and integers
is a bad choice. We may make this something like 1e20.

--Mary

On Apr 21, 2010, at 9:34 PM, HUANG, Ping wrote:

> Hi, Carl
>
> Thanks for your reply. It help me a lot.
>
> I used the input data starting from 1980. There are no missing
> values. The abnormal large zonal mean appears only if the input
> data is longer than 1998 days. For example, there is no problem
> when the input data is from 1jan1980 to 31dec1985 or from 1jan1986
> to 31dec1990. However, if the input data is from 1jan1980 to
> 31dec1990, the abnormal large zonal mean appears.
>
> After checking the function wkSpaceTime detailedly, I found that
> the error is from the function "dtrend_msg". When the dimension
> sizes is larger than 1998, the dtrend_msg output a missing value at
> the first location. And wkSpaceTime delete the missing value
> attribution to adapt the FFT function. Thus, abnormal large power
> at wavenumber zero are created.
>
> The missing value output may be a bug of the function dtrend_msg.
>
> For example:
>
> begin
> ntime = 1999
> temp = random_normal(10, 5, ntime+1)
> print(num(ismissing(temp)))
> b = dtrend_msg(ispan(0,ntime,1)*1.0,temp , False, False)
> print(num(ismissing(b)))
> print(ind(ismissing(b)))
> end
>
> This simple script will get the result as:
>
> NCAR Command Language Version 5.1.1
> The use of this software is governed by a License Agreement.
> See http://www.ncl.ucar.edu/ for more details.
> (0) 0 ; there is no missing value before detrending
> (0) 1 ; there is a missing value after detrending!
> (0) 0 ; the missing value place in the first location.
>
>
>
> Regards,
>
> Huang Ping
>
>
>
> 2010-04-22
> HUANG, Ping
> 发件人： Carl J. Schreck, III
> 发送时间： 2010-04-21 21:45:40
> 收件人： HUANG, Ping; ncl forum
> 抄送：
> 主题： Re: [ncl-talk] Question about the OLR space-time Spectra
> [SEC=UNCLASSIFIED]
> Dear Huang Ping,
>
> In your longer (> 2000 days) plots, is it possible that you are
> including times from before 1 January 1979? The OLR data contain a
> lot of missing data before this date, but none after it. For the
> missing periods, the entire globe would have missing values of
> 32766, which is a substantial zonal mean! I do not know if/how
> wkSpaceTime handles these missing points, but treating them as
> regular data could cause that issue.
>
> Hope this helps.
>
> Cheers,
> Carl
>
> 2010/4/21 HUANG, Ping <ph0007@ustc.edu>
> Dear Dr. Wheeler,
>
> Thanks for your reply.
>
> I found that the problem is from the contributed function
> wkSpaceTime in NCL. My input data don't have a large zonal mean
> component. And only if the time range of input data are larger than
> 2000 days, there are the abnormal powers in wavenumber zero. In
> wkSpaceTime, all the steps operate the 96 and 65 overlapping days
> segment except removing the long term linear trend and annual
> cycle. Actually, the annual cycle of input data has been removed
> and the linear trend of each segment has been removed too. Thus, I
> deleted this two processes and the problem was figured out. I don't
> know why these two steps create so strong zonal mean when the input
> data have long time range.
>
> Regards,
>
> Huang Ping
>
>
> 发件人： Matthew Wheeler
> 发送时间： 2010-04-21 09:15:35
> 收件人： 'HUANG, Ping'
> 抄送：
> 主题： RE: Question about the OLR space-time Spectra
> [SEC=UNCLASSIFIED]
> Dear Huang Ping,
>
> Wavenumber zero is the zonal mean. Hence, your spectrum appears to
> be indicating that you have a large zonal mean component, and this
> component has strong variability at a range of frequencies. Can you
> think of a reason why your input data may have an abnormally large
> zonal mean component? You could try removing the zonal mean at each
> time step, but this would then give zero power at wavenumber 0. It
> would be a good test to try.
>
> Regards,
>
> -----------------------------------------------------------
> Matthew C. Wheeler
> Centre for Australian Weather and Climate Research (CAWCR)
> A partnership between the Bureau of Meteorology and CSIRO
> Post: GPO Box 1289, Melbourne, VIC 3001
> Street: 700 Collins Street, Melbourne, VIC 3008
> Phone: +61 (0)3 9669 4068 Fax: +61 (0)3 9669 4660
> m.wheeler@bom.gov.au
> http://cawcr.gov.au/staff/mwheeler (new)
> http://cawcr.gov.au/bmrc/clfor/cfstaff/matw.htm (old)
> -----------------------------------------------------------
>
>
>
> From: HUANG, Ping [mailto:huangping@mail.iap.ac.cn]
> Sent: Wednesday, 21 April 2010 00:34
> To: Matthew Wheeler
> Subject: Question about the OLR space-time Spectra
>
> Dear Dr. wheeler,
>
> I am a research from China. Recently, I analyzed the Wheeler-
> Kiladis space-time spectra using the NCL function wkSpacetime and
> OLR daily data. However, the symmetric power has an abnormal value
> at wavenumber zero. Please see the following figure.
>
> The annual cycle of input daily data has been removed.
>
> Could you give some suggestions? Thanks.
> <Catch3(04-22-09-49-35).jpg>
>
>
> Regards,
>
>
>
> _______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
>
>
>
>
> --
> Carl J. Schreck, III
> PhD Student
> Department of Atmospheric and Environmental Sciences
> University at Albany, ES 330
> carl@atmos.albany.edu
> http://www.atmos.albany.edu/student/carl/
> <Catch3
> (04-22-09-49-35).jpg>_______________________________________________
> ncl-talk mailing list
> List instructions, subscriber options, unsubscribe:
> http://mailman.ucar.edu/mailman/listinfo/ncl-talk

_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Fri Apr 23 14:39:03 2010