29.01.2015 Views

Analysing spatial point patterns in R - CSIRO

Analysing spatial point patterns in R - CSIRO

Analysing spatial point patterns in R - CSIRO

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

210 Gibbs models for multitype <strong>po<strong>in</strong>t</strong> <strong>patterns</strong><br />

34 Gibbs models for multitype <strong>po<strong>in</strong>t</strong> <strong>patterns</strong><br />

Gibbs <strong>po<strong>in</strong>t</strong> process models (section 26) are also available for marked <strong>po<strong>in</strong>t</strong> processes, and can<br />

be fitted to data us<strong>in</strong>g ppm. Currently the methods are only implemented for multitype <strong>po<strong>in</strong>t</strong><br />

processes (categorical marks), so we restrict attention to this case.<br />

34.1 Gibbs models<br />

Much of the theory of Gibbs models described <strong>in</strong> Section 26 carries over immediately to multitype<br />

<strong>po<strong>in</strong>t</strong> processes.<br />

34.1.1 Conditional <strong>in</strong>tensity<br />

The conditional <strong>in</strong>tensity λ(u,X) of an (unmarked) <strong>po<strong>in</strong>t</strong> process X at a location u was def<strong>in</strong>ed<br />

<strong>in</strong> section 26.5. Roughly speak<strong>in</strong>g λ(u,x)du is the conditional probability of f<strong>in</strong>d<strong>in</strong>g a <strong>po<strong>in</strong>t</strong><br />

near u, given that the rest of the <strong>po<strong>in</strong>t</strong> process X co<strong>in</strong>cides with x.<br />

For a marked <strong>po<strong>in</strong>t</strong> process Y the conditional <strong>in</strong>tensity is a function λ((u,m),Y) giv<strong>in</strong>g a<br />

value at a location u for each possible mark m. For a f<strong>in</strong>ite set of marks M, we can <strong>in</strong>terpret<br />

λ((u,m),y)du as the conditional probability f<strong>in</strong>d<strong>in</strong>g a <strong>po<strong>in</strong>t</strong> with mark m near u, given the rest<br />

of the marked <strong>po<strong>in</strong>t</strong> process.<br />

The conditional <strong>in</strong>tensity is related to the probability density f(y) by<br />

λ((u,m),y) =<br />

f(y ∪ {u})<br />

f(y)<br />

for (u,m) ∉ y.<br />

For Poisson processes, the conditional <strong>in</strong>tensity λ((u,m),y) co<strong>in</strong>cides with the <strong>in</strong>tensity<br />

function λ(u,m) and does not depend on the configuration y. For example, the homogeneous<br />

Poisson multitype <strong>po<strong>in</strong>t</strong> process or “CSRI” (Section 33.1.1) has conditional <strong>in</strong>tensity<br />

λ((u,m),y) = β m (50)<br />

where β m ≥ 0 are constants which can be <strong>in</strong>terpreted <strong>in</strong> several equivalent ways (section 26.5).<br />

The sub-process consist<strong>in</strong>g of <strong>po<strong>in</strong>t</strong>s of type m only is Poisson with <strong>in</strong>tensity β m . The process<br />

obta<strong>in</strong>ed by ignor<strong>in</strong>g the types, and comb<strong>in</strong><strong>in</strong>g all the <strong>po<strong>in</strong>t</strong>s, is Poisson with <strong>in</strong>tensity β =<br />

∑<br />

m β m. The marks attached to the <strong>po<strong>in</strong>t</strong>s are i.i.d. with distribution p m = β m /β.<br />

34.1.2 Pairwise <strong>in</strong>teractions<br />

A multitype pairwise <strong>in</strong>teraction process is a Gibbs process with probability density of the form<br />

⎡ ⎤ ⎡<br />

⎤<br />

n(y)<br />

∏<br />

f(y) = α⎣<br />

b mi (x i ) ⎦ c mi ,m j<br />

(x i ,x j ) ⎦ (51)<br />

i=1<br />

where b m (u),m ∈ M are functions determ<strong>in</strong><strong>in</strong>g the ‘first order trend’ for <strong>po<strong>in</strong>t</strong>s of each type,<br />

and c m,m ′(u,v),m,m ′ ∈ M are functions determ<strong>in</strong><strong>in</strong>g the <strong>in</strong>teraction between a pair of <strong>po<strong>in</strong>t</strong>s of<br />

given types m and m ′ . The <strong>in</strong>teraction functions must be symmetric, c m,m ′(u,v) = c m,m ′(v,u)<br />

and c m,m ′ ≡ c m ′ ,m. The conditional <strong>in</strong>tensity is<br />

⎣ ∏ i

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!