Abstract

Many astronomical time series are observed within intervals determined by the acquisition schedule, so that the recorded time series is fragmented into blocks of observations separated by long periods without exposure. We introduce a modelling framework for fragmented time series of counts, in which the dynamics of the latent intensity are allowed to vary across blocks. Block-specific local parameters are estimated jointly and shrunk towards a common source-level behaviour, and the degree of similarity between two blocks is allowed to decay with the length of the unobserved interval separating them. We apply the proposed modelling framework to photon-counting observations of astronomical X-ray sources collected by the NICER telescope, where valid exposure is restricted to Good Time Intervals.