Simulating rainfall and drainage response using CON-SST-RAIN - a stochastic areal rainfall generator

Stochastic rainfall generators have been commonly used in the field of hydrological and hydrodynamic modeling for a long time. These generators allow for an extensive ensemble of rainfall scenarios and continuous time series that is applicable for risk assessment and response variability studies under current and future climate conditions. Most rainfall generators simulate rainfall at daily scale and at point values. Recently some generators have been developed to produce gridded rainfall products. With advancement in weather radar technology a much more detailed representation of rainfall fields is now possible. This is especially needed in the field of urban hydrology.

We developed the stochastic rainfall generator CON-SST-RAIN that is based on traditional dry/wet sequencing using Markov Chains and rainfall field generation by Stochastic Storm Transposition (SST), a time-in-space resampling method. CON-SST-RAIN was developed utilizing a 17-year long C-band radar dataset, with a spatio-temporal resolution of 500m x 500m and 10 minutes, discontinuous in time (discard of data) and Markov Chains are derived from rain gauges.

CON-SST-RAIN can recreate continuous areal time series that captures the mean annual precipitation while also retaining seasonal and inter-annual variances. Extreme rain rates are likewise preserved and comparable to rain gauge data with +40 years of record.

We test the CON-SST-RAIN on stochastically generated artificial hydrological networks to examine the importance of spatio-temporal dynamic rainfall fields. The networks are generated by a Gibbs sampling approach where the modeler can choose the extent and complexity of the generated network. Runoff from these networks is coupled with a simple detention pond model to estimate return periods for rainfall storage.