[R] Coordinate systems for geostatistics in R

[Rubén Roa-Ureta]

imicola wrote:
> Hi,
>
> I read somewhere that when carrying out geostatistical analysis in R you
> should not use latitude and longitude...can anyone expand on this a little
> for me, and what would be the best coordinate system to use?
>   
Not only in R. In most systems, the inter-point distances are assumed to 
be planar (distances over an Euclidean space), whereas latitude and 
longitude are spherical. I guess there could be a geostatistical 
analysis based on spherical distances, but why to make things more 
complicated when projecting the spherical coordinates into planar 
coordinates b4 the analysis produces a good approximation and simplifies 
the analysis significantly? I use UTM coordinates, and transform from 
geodetic to metric with Eino Uikkanen's GeoConv program (it's free).

> I have my data in a geographic coordinate system, WGS84, decimal
> degrees....is this the wrong format for such analyses?
>   
If the distances are short, it is not so wrong, and the wrongness 
increases with increasing distance.
> I have also converted my data in the UTM projection and so have it in
> metres....(ranging from 480,000 to 550,000 E and 170,000 to 230,000 N).  
>
> If I was to use the UTM coordinates, should I be using the actual
> coordinates in metres, or should I convert this into an arbitrary coordinate
> system (i.e. from 0 - 1) somehow?
It would be convenient to use km rather than m, so the range parameter 
would be closer in magnitude to the other parameters of the model. A 
very large range parameter in metres may cause numerical instability 
during minimization of the negative support function in likelihood-based 
models such as that implemented in geoR.
>   
>
> I have noticed that running an analysis on the data gives different results
> depending on which type of system you use, so I want to make sure I have the
> correct system.  I should also probably note that I am a geostatistical
> novice!
>
> Thanks,
>   
Bottomline, your geostatistical software is probably based on distance 
calculations on an Euclidean space so it is wrong to input locations in 
spherical coordinates.
HTH
Ruben