Addressing the change of support problem with non Gaussian distributions: an application to hourly rainfall prediction

We address the change of support problem in a non Gaussian model; in particular, we predict occurrence and accumulation of hourly rainfall in the Emilia-Romagna region by modeling the relationship between rain gauge and radar data. A basic formulation exploits radar information only at the grid cells containing rain gauges locations; an enrichment consists in exploiting neighbourhood information via a weighted mean of a latent spatial process for addressing spatial misalignment. A Bayesian hierarchical model is specified for each formulation, both sharing a zero-inflated likelihood, and with gamma distribution on the positive semiaxis. Results are assessed and compared in terms of point and probabilistic forecasts.