Analysing spatial point patterns in R - CSIRO
Analysing spatial point patterns in R - CSIRO
Analysing spatial point patterns in R - CSIRO
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
27.2 Fitt<strong>in</strong>g Gibbs models <strong>in</strong> spatstat 163<br />
all other parameters are ‘irregular’ parameters. For example, the Strauss process conditional<br />
<strong>in</strong>tensity (42) can be recast as<br />
log λ(u,x) = log β + (log γ)t(u,x)<br />
so that θ = (log β,log γ) are regular parameters, but the <strong>in</strong>teraction distance r is an irregular<br />
parameter (technically called a ‘bloody nuisance parameter’).<br />
In spatstat we split the conditional <strong>in</strong>tensity <strong>in</strong>to first-order and higher-order terms:<br />
log λ θ (u,x) = η · S(u) + ϕ · V (u,x). (46)<br />
The ‘first order term’ S(u) describes <strong>spatial</strong> <strong>in</strong>homogeneity and/or covariate effects. The ‘higher<br />
order term’ V (u,x) describes <strong>in</strong>ter<strong>po<strong>in</strong>t</strong> <strong>in</strong>teraction.<br />
The model with conditional <strong>in</strong>tensity (46) is fitted by call<strong>in</strong>g ppm <strong>in</strong> the form<br />
ppm(X, ~ terms, V)<br />
The first argument X is the <strong>po<strong>in</strong>t</strong> pattern dataset. The second argument ~terms is a model<br />
formula, specify<strong>in</strong>g the first order term S(u) <strong>in</strong> (46), <strong>in</strong> the manner described <strong>in</strong> Section 15.<br />
Thus the first order term S(u) <strong>in</strong> (46) may take very general forms.<br />
The third argument V is an object of the special class "<strong>in</strong>teract" which describes the<br />
<strong>in</strong>ter<strong>po<strong>in</strong>t</strong> <strong>in</strong>teraction term V (u,x) <strong>in</strong> (46). It may be compared to the ‘family’ argument<br />
which determ<strong>in</strong>es the distribution of the responses <strong>in</strong> a l<strong>in</strong>ear model or generalised l<strong>in</strong>ear model.<br />
Only a limited number of canned <strong>in</strong>teractions are available <strong>in</strong> spatstat, because they must be<br />
constructed carefully to ensure that the <strong>po<strong>in</strong>t</strong> process exists.<br />
To fit the Strauss process to the cells data us<strong>in</strong>g ppm,<br />
> data(cells)<br />
> ppm(cells, ~1, Strauss(r = 0.1))<br />
Stationary Strauss process<br />
First order term:<br />
beta<br />
762.6005<br />
Interaction: Strauss process<br />
<strong>in</strong>teraction distance: 0.1<br />
Fitted <strong>in</strong>teraction parameter gamma: 0.008<br />
Relevant coefficients:<br />
Interaction<br />
-4.825006<br />
Here Strauss is a special function that creates an ‘<strong>in</strong>teraction’ object (class "<strong>in</strong>teract")<br />
describ<strong>in</strong>g the <strong>in</strong>teraction structure of the Strauss process. Notice that we had to specify the<br />
value of the irregular parameter r (more about that later).<br />
To fit the <strong>in</strong>homogeneous Strauss process with conditional <strong>in</strong>tensity<br />
λ(u,x) = b(u)γ t(u,x)<br />
where, say, b(u) is logl<strong>in</strong>ear <strong>in</strong> the Cartesian coord<strong>in</strong>ates,<br />
we simply type<br />
log b((x,y)) = β 0 + β 1 x + β 2 y<br />
Copyright<strong>CSIRO</strong> 2010