Information and complexity-based evaluation of spatial point process models
Quantifying randomness and heterogeneity is an essential objective in spatial pattern analysis. Information and complexity measures provide a valuable insight into structural properties. We propose a framework for assessing goodness of fit, with a focus on log-Gaussian Cox processes. In a simulation study, we generate spatial patterns under different parameter settings and use kernel density estimation (KDE) to approximate intensity functions with varying accuracy levels. Information and complexity measures are then employed to evaluate the model relative performance with respect to the reference pattern. We apply this approach to wildfire real data, comparing a KDE-based model with a log-Gaussian Cox process model that explicitly incorporates stochastic second-order spatial characteristics. By analyzing discrepancies between observed and predicted patterns in terms of information and complexity measures, we illustrate its practical applicability for model fitting and selection.
Keywords: complexity entropy goodness of fit kernel density estimation spatial log-Gaussian Cox process