An adaptive kernel estimator for the intensity function of spatio-temporal point processes

Abstract: In spatio-temporal point pattern analysis, one of the main statistical objectives is to estimate the first-order intensity function, i.e., the expected number of points per unit area and unit time. This estimation is usually carried out non-parametrically through kernel functions, where one of the most frequent handicaps is the selection of kernel bandwidths prior to estimation. This work presents an intensity estimation mechanism in which the spatial and temporal bandwidths change at each data point in a spatio-temporal point pattern. This class of estimators is called adaptive estimators, and although there have been studied in spatial settings, little has been said about them in the spatio-temporal context. We define the adaptive intensity estimator in the spatio-temporal context and extend a partitioning technique based on the bandwidths quantiles to perform a fast estimation. We demonstrate through simulation that this technique works well in practice with the partition estimator approximating the direct estimator and much faster computation time. Finally, we apply our method to estimate the spatio-temporal intensity of fires in the Amazonia basin.