Re: ftcurvs

Well the issue is more data driven. I have two datasets ... One with near
hourly and in some cases 20 minute data and special metar observations, and
the other is the 1 minute dataset. I need something that works across that
range.

My purpose is to compare 3 hour Temperature tendencies from observations and
a set of wrf model runs (nearest grid point). I do the tendencies in 3 hour
sets, so I can get a feel for how the model tendencies compare over the
diurnal cycle.

I suppose a Lanczos filter would do the trick, but the data spacing with the
hourly data may have some missing data and thus not be equally spaced. I
figured the cubic spline is a good way to curve fit out some of the 1 minute
noise, and yet still offer the goodness of fit with the hourly observations.

Mary-
The data set name is ia_awos.dat and the ncl file is ia_ts1.ncl. I commented
out the sections of wrf model data. The relevant code is located after line
1086. My apologies for messy code...
jimmyc

On 8/3/09 10:04 AM, "Dennis Shea" <shea@ucar.edu> wrote:

> My suggestion ....
>
> Define a filter that has the characteristics you want.
>
> http://www.ncl.ucar.edu/Applications/filter.shtml
>
> See Example 6
>
> Here you can use an fft *affter* getting the appropriate weights.
>
> ====
>
> FYI: If u are doing the fit for plotting only, ftcurvs
> is fine. However, if u r doing this for some physical reasons,
> somebody [eg, a reviewer] might say "What is the half power
> point of the filter?" ftcurvs will not help you there!
>
> D
>
>
>
> Mary Haley wrote:
>> James,
>>
>> Can you provide the full script and data offline?
>>
>> If it's large, you can ftp it:
>>
>> ftp ftp.cgd.ucar.edu
>> <log in as "anonymous">
>> <Use email address as password>
>> cd incoming
>> put [your files]
>> quit
>>
>> I will need to know the exact filenames in order to download them.
>>
>> Thanks,
>>
>> --Mary
>>
>> On Aug 2, 2009, at 9:50 PM, Correia, James wrote:
>>
>>> Hi all-
>>> I am using ftcurvs to produce a smooth line through an observed
>>> temperature time series, sort of like a filter. This works great when
>>> the number of points is small (hourly time series), but not so great
>>> when I use 1 minute observations.
>>>
>>> In all I have 31 time series that I run the ftcurvs function on. 29 of
>>> them turn out just fine. 2 of them have similar yet strange output.
>>> Attached is the example of a bad fit. The dashed line is the output
>>> from ftcurvs and the solid line is the observations. I have checked
>>> and checked again for missing data or strange values. There are none.
>>> Time increments by 1 minute with 1440 points, as do the other time
>>> series without problem.
>>>
>>> I have d=0.4 to get a close fit. I am requesting the line fit with 96
>>> points.
>>>
>>> Does anyone know what might be going wrong to produce such a terrible
>>> fit?
>>>
>>> James Correia Jr
>>> Post Doc
>>> PH: 372-6463
>>> james.correia@pnl.gov
>>>
>>> "Wisdom. Strength. Courage. Generosity. Each of us are born with one
>>> of these. We must find the other three inside of us."
>>> -from "Into the West"
>>>
>>>
>>> <bad.pdf>_______________________________________________
>>> ncl-talk mailing list
>>> List instructions, subscriber options, unsubscribe:
>>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Mon Aug 3 12:34:08 2009