[R] Savitzky-Golay Smoother

[Erick Okuto]

Dear Gabor, Paul and Ryan,
Many thanks for all your guidance concerning  the application of the
filter/smoother. I need to temporally smooth 32 years 5 km resolution
bi-monthly global NDVI dataset. On average each pixel have got up to 768
cases/images with data missing at random. I consider the data to be large
and with the nature of missing data, applying the filter manually might be
tricky. I have run a function which kind of combines Gabor and  Paul's
suggestion " sgolayfilt(zoo::na.spline(ydata)) " as suggested by Ryan (cc'd
here) on a few sampled time series data and observed/projections are
generally  matching well. If i understand the function, it first fill in
NAs with spline curve fits from zoo package and run Savitzky-Golay filter
on a continuous data. The problem is that data/observations on the end
points (like first and last 5 or 10) are poorly projected with some
projections being off the possible values that NDVI can take. Can we adjust
the algorithm  such that the problem is controlled? Thanks for your time.

Kind regards,
Erick.


On Sat, Sep 27, 2014 at 2:23 AM, Gabor Grothendieck <ggrothendieck at gmail.com
> wrote:

> On Fri, Sep 26, 2014 at 3:32 AM, Erick Okuto <erickokuto at gmail.com> wrote:
> > Dear Paul and Henrik,
> > I have a time series with some missing data points that i need smoothed
> > using Savitzky-Golay filter. Related question was asked  here
> >
> http://thr3ads.net/r-help/2012/11/2121748-Savitzky-Golay-filtering-with-missing-data
> > but no straight forward answer was posted. However, Henrik (cc'd here)
> did
> > ask related question on smoothing for reflectance here
> >
> http://thr3ads.net/r-help/2004/02/835137-Savitzky-Golay-smoothing-for-reflectance-data
> > which i have as well been unable to follow up. I will be glad if you
> could
> > assist me with some insights on the way forward or point to a relevant
> > source of help.
>
>
> Not Savitzky-Golay but if z is a time series then
>
> library(zoo)
> na.spline(z)
>
> will fill in NAs with spline curve fits.  See ?na.spline
>
> There are other NA filling routines in zoo too:
>
> > ls(pattern = "^na[.]", "package:zoo")
>  [1] "na.aggregate"         "na.aggregate.default" "na.approx"
>  [4] "na.approx.default"    "na.fill"              "na.fill.default"
>  [7] "na.locf"              "na.locf.default"      "na.spline"
> [10] "na.spline.default"    "na.StructTS"          "na.trim"
> [13] "na.trim.default"      "na.trim.ts"
>
>
>
> --
> Statistics & Software Consulting
> GKX Group, GKX Associates Inc.
> tel: 1-877-GKX-GROUP
> email: ggrothendieck at gmail.com
>



-- 
*Erick Okuto, Ph.D.*
* Candidate*
*School of Mathematics & Actuarial Science*
*Jaramogi Oginga Odinga University of Science & Technology (JOOUST)/ World
Agroforestry Centre (ICRAF), **Climate Change Unit, Nairobi-Kenya. *
Voice:  +254207224154​    Mobile: +254725005276    Skype id:  erickokuto
Email: erickokuto at gmail.com

	[[alternative HTML version deleted]]