[R-sig-Geo] countingweights?

[Adrian.Baddeley at csiro.au]

salma anwar <salmaaries at hotmail.com> writes:

> Hi, I am trying  to fit a "Area-Interaction" point process model 
> to my data consisting of log-landing locations in a forest. 
> I get the following warning message along with the result of 
> the fitting: 
> "1: In countingweights(id, areas) :   Some tiles with zero area contain points
> 2: In countingweights(id, areas) : Some weights are zero" 
> Could you please guide me as to what error this message indicates? Thanks!   

This is about the package 'spatstat' and specifically about the model-fitting function for point processes, 'ppm'.

Basically this warning means that there might be some inaccuracy in the fitted model and you should consider whether to use a finer resolution.

'ppm' uses a discrete approximation to the likelihood or composite likelihood that involves a set of 'dummy' or sample points in the study region (called the "quadrature scheme"). By default, this approximation uses 'counting weights' which are computed by dividing the study region into pieces called 'tiles' and counting the number of data or dummy points that fall in each tile. 

If this process is performed using the default settings without any control, then it can happen that one of the tiles is very small - for example if your study region is an irregular shape, or if you use a spatial covariate on a very irregular region. In the latter case, it can happen that one of the tiles has zero digital area (digital area is the number of pixels in the tile) but nevertheless is not empty. Then you can get this error message.

The default settings for the discrete approximation are set to a relatively coarse approximation so that the package can be checked in a reasonable time (otherwise CRAN complains). These settings are not necessarily appropriate for every dataset. 

To increase the resolution of the discrete approximation you can increase the argument 'nd' in the call to ppm, or you can set the value of spatstat.options(ndummy.min). Alternatively you can use quadscheme() to control the approximation completely.

Adrian Baddeley

Prof Adrian Baddeley FAA
CSIRO Mathematics, Informatics & Statistics
Leeuwin Centre, 65 Brockway Road, Floreat WA 6014, Australia
Tel: 08 9333 6177 | Fax: 08 9333 6121 | Skype: adrian.baddeley