[R-sig-Geo] Lattice with autocorrelation and predetermined values

[Virgilio Gomez Rubio]

Dear Simon,


Perhaps you could model your data using a multivariate Normal with
spatial autocorrelation (using a SAR or CAR specification). In this way,
you can fix some of the values and then simulate the rest. You can tune
the autocorrelation coefficient to make sure that the spatial
correlation is large enough to give high values of Moran's I.

Hope this helps,

Virgilio

El dom, 19-04-2009 a las 21:16 +0000, Simon Chamaillé escribió:
> Dear all,
> 
> I want to distribute a vector of known values onto a lattice, while still making sure that the Moran autocorrelation is higher than a specific threshold.
> 
> I know of several ways of getting a lattice with some pre-determined levels of approximate spatial autocorrelation, but all involve some random number generation of one sort or another, whereas I want cell values to have some pre-determined values.
> 
> I assume that I could:
> 1) generate a lattice
> 2) randomly distribute my pre-determined values
> 3) compute Moran's I, then randomly permutate two cell values
> 5) recompute Moran's I and, and if higher that the one calculated in 3) keep the new lattice
> 6) back to 3) for permutation, or stop when Moran's I higher than my threshold, or kill the simulation when i've reached a specific number of iterations.
> 
> The above is obviously so inefficient that it is likely to be of no use for a lattice of even a moderate size or for a high autocorrelation threshold to be reached. I intend to work on approx. 100*100 lattice, with values equally distributed between 0-100, and autocorrelation threshold of approx. 0.3.
>  
> Any suggestions (hum..., even 'don't even think of doing something like this'), would be appreciated.
> Thanks,
> simon
> 
> 
>       
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo