A high resolution large-scale Gaussian random field rainfall model for Australian monthly rainfall

摘要

Australian rainfall conditions are highly variable in time and in space. This variability of rainfall is responsible for floods and droughts that impact rural and urban catchments alike. These extreme rainfall conditions are dependent on rainfall patterns at a large spatial scale not considered by current stochastic models. This paper proposes a stochastic random field rainfall model for the simulation of monthly rainfall events at a high spatial resolution across the Australian continent. Monthly rainfall data, gridded at a 25km resolution across the continent, were used to determine monthly means and covariances in time and space. To correctly reproduce the occurrences of zero rainfall, the data were transformed according to a latent variable approach. The model then simulates Gaussian random fields using a Fourier technique which is then back-transformed to recover the statistical rainfall properties. Simulated data were validated against recorded data using comparison tests between sample means and standard deviations and comparing the simulated cross- and autocorrelation.with observed correlations.