Spectral estimation for spatial point patterns
share
This article determines how to implement spatial spectral analysis of point processes (in two dimensions or more), by establishing the moments of raw spectral summaries of point processes. We establish the first moments of raw direct spectral estimates such as the discrete Fourier transform of a point pattern. These have a number of surprising features that departs from the properties of raw spectral estimates of random fields and time series. As for random fields, the special case of isotropic processes warrants special attention, which we discuss. For time series and random fields white noise plays a special role, mirrored by the Poisson processes in the case of the point process. For random fields bilinear estimators are prevalent in spectral analysis. We discuss how to smooth any bilinear spectral estimator for a point process. We also determine how to taper this bilinear spectral estimator, how to calculate the periodogram, sample the wavenumbers and discuss the correlation of the periodogram. In parts this corresponds to recommending suitable separable as well as isotropic tapers in d dimensions. This, in aggregation, establishes the foundations for spectral analysis of point processes.
READ FULL TEXT
