Statistical Models for Rainfall:characterisation and forecasting

Abstract

Weather forecasting is a prediction of what the weather to come will be like. This can be done in a deterministic or probabilistic way. We are interested in nowcasting, which is the production of short-range (0 to 6 hours) forecasts. The goal of this thesis is to create and develop a statistical model for precipitation forecast with radar data only. This forecast will be probabilistic, which means that we will provide a forecast mean and covariance for the precipitation. This study starts with the research of a space reduction for the data, as time required to run models is crucial in weather forecasting. Then, we build in a Bayesian framework a general model, which is developed in the latter part of the thesis. We need dynamics to represent the evolution of the rainfall field. The dynamics of this model come from the advection equation, which links the rainfall field to the advection field. The model has been tested with simulated and real data.