A stochastic space-time rainfall forecasting system for real time flow forecasting I: Development of MTB conditional rainfall scenario generator

摘要

The need for the development of a method for generating an ensemble of rainfall scenarios, which are conditioned on the observed rainfall, and its place in the HYREX programme is discussed. A review of stochastic models for rainfall, and rainfall forecasting techniques, is followed by a justification for the choice of the Modified Turning Bands (MTB) model in this context. This is a stochastic model of rainfall which is continuous over space and time, and which reproduces features of real rainfall fields at four distinct scales: raincells, cluster potential regions, rainbands and the overall outline of a storm at the synoptic scale. The model can be used to produce synthetic data sets, in the same format as data from a radar. An inversion procedure for inferring a construction of the MTB model which generates a given sequence of radar images is described. This procedure is used to generate an ensemble of future rainfall scenarios which are consistent with a currently observed storm. The combination of deterministic modelling at the large scales and stochastic modelling at smaller scales, within the MTB model, makes the system particularly suitable for short-term forecasts. As the lead time increases, so too does the variability across the set of generated scenarios.