R help - Jan 2007 - correlation value and map

Dear R-help community,

I have 2 different arrays of precipitation data each of the same dimensions of 
[longitude, latitude, time] dim=[30,32,43], called array1 and array2. I need to 
correlate them. This is the code I used to get one overall correlation value for
the whole of the area of interest:
> result <- cor(array1,array2,use="complete.obs")
> result
This give me a single value but I'm not convinced it is actually a
correlation
value for the total area for the total time period of 43 years....can anybody 
tell me if I am indeed wrong in my coding and/or indeed my low knowledge of the 
statistics of correlation.

Also, I wanted to produce a correlation map over the 43 years. Could you also 
advise me if this is correct, I am more confident that this is than the above 
code:
> result <- array(NA, c(30,32))
> 
> for(i in 1:30){
> for(j in 1:32){
>	array1.ts <- array1[i,j,]
>	array2.ts <- array2[i,j,]
>	result[i,j] <- cor(array1.ts,array2.ts,use= "complete.obs")
> }
> }
I appreciate your time very much. If I don't iron out this problem now the 
ground-work for my entire PhD will not be stable at all,

Many thanks for reading my problem, happy 2007 :-)

Jenny Barnes





~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jennifer Barnes
PhD student - long range drought prediction
Climate Extremes
Department of Space and Climate Physics
University College London
Holmbury St Mary, Dorking
Surrey
RH5 6NT
01483 204149
Web: http://climate.mssl.ucl.ac.uk

Hi Jenny!

So if i understand your datafile corect you have 960 case for a year. Any
you have 43 years.. Yes?

I'm not sure you should use correlation in this situation because of the
autocorrelation of the data. There are big autocorrelation on spatial data's
like what you use, and there are also a very big autocorrelation in time
series data. I think you have to decompose your time series, and you have to
cut down, the trend (and maybe some kind of sesonality), and than for the
residuals you should do a correlation. You have to filter out the
autocorrelation on the spatial data too, some way..

And because of the above problems, don't calculate correlation for the
entierly databases!

bye,
Zoltan

	[[alternative HTML version deleted]]

Hi Zoltan,

Right, I have 30x32=960 data points per year (It is actually the mean febuary 
precipitation total in case you were wondering) at each grid point over the 
world, so I have 960 data points each of the 43 years. Therefore can I do 
anything with a trend and residuals? I don't think I can if it's just
mean feb
precipitation, one data point per grid square per year... I apreicate your help 
though very much.....although I do still need to perform a spatial correlation 
if anyone else can help?

Many thanks,

Jenny



Hi Jenny!

So if i understand your datafile corect you have 960 case for a year. Any
you have 43 years.. Yes?

I'm not sure you should use correlation in this situation because of the
autocorrelation of the data. There are big autocorrelation on spatial data's
like what you use, and there are also a very big autocorrelation in time
series data. I think you have to decompose your time series, and you have to
cut down, the trend (and maybe some kind of sesonality), and than for the
residuals you should do a correlation. You have to filter out the
autocorrelation on the spatial data too, some way..

And because of the above problems, don't calculate correlation for the
entierly databases!

bye,
Zoltan

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jennifer Barnes
PhD student - long range drought prediction
Climate Extremes
Department of Space and Climate Physics
University College London
Holmbury St Mary, Dorking
Surrey
RH5 6NT
01483 204149
07916 139187
Web: http://climate.mssl.ucl.ac.uk