Interactive demo · Statistical Methods for Modern Biology · Chapter 11

Spatial point patterns: Ripley's K, L, the pair correlation and G

Is a point pattern clustered, regular, or as random as a homogeneous Poisson process (complete spatial randomness, CSR)? Ripley's K(r) is the expected number of further points within distance r of a typical point, divided by the intensity; under CSR it is πr², so L(r) − r = √(K/π) − r is 0, the pair correlation g is 1 and the nearest-neighbour distribution is G(r) = 1 − exp(−λπr²). Points near the window edge have neighbours we cannot see; the edge corrections repair that. The grey band is the range of the same estimate over simulated CSR patterns with the same number of points (Holmes and Huber 2019, Sections 11.16 and 11.17).

Pattern

expected number of points

cluster spread σ and number of parents

hard-core distance δ

Summary function

Edge correction

CSR envelope

Intensity under the points

σ

Scoreboard

Try these

The window is the square [0, 1200]² (units of the scanner's coordinates). Estimators, with λ̂ = (n − 1)/|W|: none, |W| Σi≠j 1{dij ≤ r} / (n(n − 1)); border, the same count over the points at least r from the edge divided by their number and λ̂; isotropic (Ripley) and translation, each pair weighted by the inverse fraction of the circle, or of the shifted window, that lies inside W. The pair correlation is a translation-corrected kernel estimate (Epanechnikov, bandwidth 0.15/√λ̂). The envelope uses CSR patterns with exactly n points; simulations are seeded.

In the lecture notes