Rainfall nowcasting models: state of the art and possible future perspectives

This paper represents a short review of the most widely adopted types of models for rainfall nowcasting, which is an important component for early warning systems (EWSs). Specifically, the authors focus on: (1) extrapolation techniques from remote sensing observations; (2) numerical weather prediction (NWP) models; (3) stochastic models; (4) deep learning models. Moreover, the possibility of realizing blended systems with two or more different kinds of models is also described, in order to overcome the limitations of using only a single model and to take advantage of pros from the considered model ensemble. However, both single and blended models are affected by uncertainty, a topic characterized by ongoing debates in the scientific community. In this context, evaluation of predictive uncertainty (PU) is also discussed, as it provides an important perspective for an EWS, enabling informed decision making based on the forecasts of one or multiple models.