This work introduces a stochastic-dynamical rainfall prediction model suitable for use on the hydrologic prediction scales (1-6 hours lead time and 10-30km spatial resolution). The developed model creates a link between the scales of operational numerical weather prediction models and operational hydrologic models. The model is designed to provide short-term distributed rainfall prediction in real-time to equip surface hydrologic prediction models with additional prediction lead time. The microphysical parameterizations of the rainfall prediction model are based on the station rainfall model by Georgakakos and Bras (1984). The synoptic scale operational numerical weather prediction model (Nested Grid Model) wind prediction output is used for rainwater advection simulation. The numerical scheme for solving the conservation of condensed water mass governing equation is designed in a way that minimizes numerical diffusion errors and admits implementation of a Kalman filter. Inclusion of the Kalman filter state estimator enables the model for real-time simulation error correction as well as for estimation of prediction uncertainty. Required model inputs are the surface temperature, dew point temperature and pressure observed on a operational basis at meteorological stations. Sensitivity analyses indicated that the "optimal" model parameters of the stochastic-dynamical rainfall prediction model do not vary during different rainfall events. Initial tests of the model with summer rainfall events in Oklahoma shows that, with one-hour lead time, the rainfall predictions, verified by radar data, explained 88% (30km^2^atial resolution) to 93% (10km^2 spatial resolution) of the observed rainfall variances with a cross-correlation coefficient between predictions and observations that ranged from 0.55 to 0.62.
- Pub Date:
- February 1991