R help - May 2008 - Alternative options: nonlinear model &autocorrelation?

Dear R community,



Using nlme library I have developed a nonlinear mixed model. Incorporating
an autoregressive model gives me an error that I can't allocate vector of
size X. The problem is that my computer does not have enough physical memory
most probably due to a large number of observations (17,000).



I was wondering what alternative options I might use:



1)      To use ARIMA and calculate residuals, and then to apply my nonlinear
model to the residuals. I have a feeling it does not sound reasonable.

2)      If I could reduce the number of observations, my original plan would
work well and I would not have a problem with physical memory. How to reduce
the number of observations? My data consist of six sites with two plots per
site. A) To average data from two plots? It seems to me we can loose some
information; B) To model the effect of plots and remove it from the
observations, then to average the residuals for two plots within each of the
sites and apply the nonlinear model?



Your suggestions are appreciated.



Thanks,

Julia



PS: My data are daily data of soil temperature for six years, six sites and
two plots per site.

	[[alternative HTML version deleted]]

Dear R community,
 
Using nlme library I have developed a nonlinear mixed model. Incorporating an
autoregressive model gives me an error that I can’t allocate vector of size X.
The problem is that my computer does not have enough physical memory most
probably due to a large number of observations (17,000).
 
I was wondering what alternative options I might use:
 
1)      To use ARIMA and calculate residuals, and then to apply my nonlinear
model to the residuals. I have a feeling it does not sound reasonable.
2)      If I could reduce the number of observations, my original plan would
work well and I would not have a problem with physical memory. How to reduce the
number of observations? My data consist of six sites with two plots per site. A)
To average data from two plots? It seems to me we can loose some information; B)
To model the effect of plots and remove it from the observations, then to
average the residuals for two plots within each of the sites and apply the
nonlinear model?
 
Your suggestions are appreciated. 
 
Thanks,
Julia
 
PS: My data are daily data of soil temperature for six years, six sites, two
plots per site.
_________________________________________________________________


Refresh_family_safety_052008
	[[alternative HTML version deleted]]