Statistical framework to simulate daily rainfall series conditional on upper-air predictor variables

Abstract

We propose a statistical framework to generate synthetic rainfall time series at daily resolution, conditional on predictor variables indicative of the atmospheric circulation at the mesoscale. We do so by first introducing a dimensionless measure to assess the relative influence of upper-air variables at different pressure levels on ground-level rainfall statistics, and then simulating rainfall occurrence and amount by proper conditioning on the selected atmospheric predictors. The proposed scheme for conditional rainfall simulation operates at a daily time step (avoiding discrete approaches for identification of weather states), can incorporate any possible number and combination of predictor variables, while it is capable of reproducing rainfall seasonality directly from the variation of upper-air variables, without any type of seasonal analysis or modeling. The suggested downscaling approach is tested using atmospheric data from the ERA-Interim archive and daily rainfall measurements from western Greece. The model is found to accurately reproduce several statistics of actual rainfall time series, at both annual and seasonal levels, including wet day fractions, the alternation of wet and dry intervals, the distributions of dry and wet spell lengths, the distribution of rainfall intensities in wet days, short-range dependencies present in historical rainfall records, the distribution of yearly rainfall maxima, dependencies of rainfall statistics on the observation scale, and long-term climatic features present in historical rainfall records. The suggested approach is expected to serve as a useful tool for stochastic rainfall simulation conditional on climate model outputs at a regional level, where climate change impacts and risks are assessed.