空间统计-ch6.point.process.pptVIP

* Chapter 6 – Point process models In this chapter we will introduce some important point process models and the methods of simulation. In principle, any observed point data (e.g., a tree point pattern) can be viewed as a realization of some point process. This will allow us to infer (from the observed point data) properties of the underlying structure of the points and help us understand the statistical mechanisms behind. Suppose we observed the locations of points in a bounded region D. For any subset A ? D, define N(A) as the number of points in A. If for any n, A1, A2, …, An are disjoint sets in D, then the collection of probabilities p(N(A1) = k1, …, N(An) = kn) are referred to as the finite multi-dimensional distributions of the point process N. 0 20 40 60 80 100 0 20 40 60 80 100 A D Homogeneous Poisson process Generating mechanism – A Poisson point process entails notions of both independence and the Poisson distribution. Assume N(A) counts the number of events occurring in a region A. N(A) is a homogeneous Poisson process if: 1. The number of events happening in disjoint regions A1, A2, …, An are independent, i.e., the random variables N(A1), N(A2), …, N(An) are independent. 2. N(A) has the Poisson distribution where l is the intensity of the Poisson point process. The above mechanisms suggest both uniformity and independence in point distribution. Uniformity refers to the fact that events in question exhibit no tendency to occupy particular regions of the space, while independence suggests that the location of a given event is determined without reference to that of any other events. Simulation of the Poisson process Conditional – Fixed A and given N(A) = m, generate N(A1), N(A2), …, N(An) iid uniform(A) random variates. For a regularly shaped area, say A = [0, a]?[0, b], it is easy to implement the simulation. Because of the uniformity independence, the x-coordinates of the points have a uniform distribution at 0 ? x ? a and the y-coordin