Short-term rainfall prediction using a multifractal model

Abstract

This study develops a method to predict multifractal measure of temporal rainfall intensity by using Kalman filter, and gives some examples of prediction for generated rainfall. The model for the rainfall generation proposed here is established using a continuous-time, discrete-scale lognormal cascade (CLC) with AR(1) process for each component. This model allows us to simulate rainfall field with the property of the multifractality, which indicates the invariance for scaling of rainfall measure. Through the observation from the synthetic rainfall simulated by this model, Kalman filter is used as the tool for short-term rainfall prediction. We compare different results of predictions made under different simulations and discuss the extensions of this study, prediction for the wet/dry process while looking at real rainfall and issues about space-time rainfall modeling. Keywords: Multifractality, Bayesian estimation, Kalman filter.