This study aims to develop a stochastic method (SM_GSTR) for generating short-time (i.e., hourly) rainstorm events at all grids (named gridded rainstorm events) in a region. The proposed SM_GSTR model is developed by the non-normal correlated multivariate Monte Carlo simulation (MMCS) method (Wu et al. 2006) with the statistical properties and spatiotemporal correlation structures of the four event-based gridded rainfall characteristics. The radar-based rainfall data on 20 typhoon events at 336 grids in a basin located in north Taiwan, Nankan River watershed, are used in the model development and demonstration. The results from the model demonstration indicate that the proposed SM_GSTR model can reproduce a great number of gridded rainfall characteristics, of which, the statistical properties in time and space have a good fit to those from the observations in association with the acceptable deviation; thus, it can reasonably emulate the behavior of the rain field in both time and space. It is expected that the resulting massive rainfall-induced disasters (e.g., inundation and landslide) from the physical-based numerical model with the simulated gridded rainstorms by the proposed SM_GSTR model can be applied to establish an alternative artificial intelligence (AI) model for effectively forecasting the hydrologic variables (e.g., runoff and water-level).
A stochastic model for generating a rain field is develop under consideration of correlation of rainfalls in time and space.
The rainstorm event consists of five gridded rainfall characteristics.
The resulting rain field is composed of simulated rainstorm at all grids.
A correlated multivariate Monte Carlo method is applied.
The statistical properties of simulated rainstorms at all grids can be persisted.