On Dec 10, 10:21 am, Elkunn <Wasit.Weat...@gmail.com> wrote:
> On Dec 10, 7:54 am, Jeremy Bailin <astroco...@gmail.com> wrote:
>
>
>
>> On Dec 10, 12:12 am, Elkunn <Wasit.Weat...@gmail.com> wrote:
>
>>> Hello,
>>> I have a data array like this. The 1st value a[0] is somewhat
>>> contaminated and need to be removed.
>
>>> a=[0.0382000, 0.3919000, 0.3843000, 0.3880000, 0.3720000, 0.4221000,
>>> 0.5966000, 0.8063000,0.7955000
>>> 0.8022000,0.7941000,0.8149000,0.8170000,0.7212000,0.7299000, 0.7644000,0.7738000,0.7574000
>>> 0.6756000, 0.6122000,0.5646000,0.5595000,0.5151000]
>
>>> I want to remove 1st pixel and replace it by a predicted value from
>>> curve fitting and smooth the overll data, return them back into
>>> spatial domain. I do not have error values. Is there any simpler
>>> method to do that?
>
>>> Thanks a lot!
>
>>> Elkuun
>
>> And do you expect whatever the curve you fit to have compact support
>> (probably a good idea for most applications, but I have no clue about
>> in your case) or to depend on all of the other data?
>
>> -Jeremy.- Hide quoted text -
>
>> - Show quoted text -
>
> Thanks for your reply!
> This is NDVI data of one pixel over one year. I think Gussian curve
> fit works for that. Some pixels has no clouds, then I just need to
> smooth the curve, but one for like this, I want to remove the cloud
> pixel, then predict its value from least-square fitting, then smooth
> the whole curve.
>
> Thank you!
Hmmm. Well, you can certainly fit a Gaussian to that, if you have good
reason to think that it should be a good parameterization... but it
doesn't *look* that good to me (of course, I have no idea what the
errors on those points are). It would look something like this:
na = n_elements(a)
gfit = gaussfit(lindgen(na-1)+1, a[1:*], gparms, nterms=3)
; replace a[0] with value from Gaussian. x=0 at this point, which
makes it simpler.
a[0] = gparms[0] * exp(-0.5*(gparms[1]/gparms[2])^2)
If you want to add more terms, just increase nterms and add gparms[3]
at the end. For example, to add a linear term (possibly a good idea,
but I don't know what you'd theoretically expect):
gfit = gaussfit(lindgen(na-1)+1, a[1:*], gparms, nterms=5)
a[0] = gparms[0] * exp(-0.5*(gparms[1]/gparms[2])^2) + gparms[3]
-Jeremy.
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
