[R] STL- TimeSeries Decomposition

[Prof Brian Ripley]

On Wed, 30 Jul 2003, Jan Verbesselt wrote:

> Dear R Helpers,
> 
> Currently I'm working with the ts package of R and created a TimeSerie
> from pixels extracted from satellite imagery(S10 NDVI data, 10 daily
> composites). I'm trying to decompose this signal in different signals
> (seasonal and trend).
> 
> When testing out the STL method is says => Only univariate timeseries
> are allowed, but the current Timeserie I'm using is univariate! => The
> problem is probably that this time series has to much noise so that it
> consequently gives the following error.
> > plot(stl(Timeserie))
> Error in stl(Timeserie) : only univariate series are allowed.  I also
> import the data as an ts object.

No, the problem *is* that the time series is a matrix, and so not 
univariate.  Try dim(Timeserie) to see.  If it has one column (as I 
suspect), you need to remove that (dim(Timeserie) <- NULL).

> A solution would be to eliminate the noise (sensor and atmospheric) with
> a filter (kalman/ holt-Winters/TsSmooth? Or FFT.) or the BISE method in
> R? 
> 
> Is the BISE (Best index slope extraction) function already programmed in
> R I couldn't find it?

I've never even heard of it.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595