Statistical downscaling of precipitation through nonhomogeneous stochastic weather typing

We present a novel statistical downscaling method that provides accurate and relatively transparent simulations of local-scale precipitation characteristics. The method combines large-scale upper-air circulation with surface precipitation fields, and is based on a nonhomogeneous stochastic weather typing approach. Here we applay the method to downscale precipitation at 37 rain gauges in the state of Illinois, USA. Regional climate conditions are categorized in terms of 2 different types of weather states: (1) ’precipitation patterns‘ developed by a hierarchical ascending clustering (HAC) method with an original metric applied directly to the observed rainfall characteristics in Illinois, and (2) ’circulation patterns‘ developed by a mixture model applied to large-scale NCEP reanalysis fields. We modeled the transition probabilities from one pattern to another by a nonhomogeneous Markov model that is influenced by large-scale atmospheric variables such as geopotential height, humidity and dew point temperature depression. Our results indicate that including the precipitation states in the statistical model allows us to simulate important precipitation features such as conditional distributions of local simulated rainfall intensities and wet/dry spell behavior more accurately than with a traditional approach based on upper-air circulation patterns alone.