loading page

Stochastic Storm Simulation Using Optimal Estimation and Non-Parametric Generation of Ensemble Parameters
  • Shang Gao,
  • Zheng Nick Fang
Shang Gao
University of Oklahoma

Corresponding Author:[email protected]

Author Profile
Zheng Nick Fang
The University of Texas at Arlington
Author Profile


The historical record of rainfall observation rarely provides sufficient record length and resolution required in many applications. The work described here presents a stochastic framework for long-term simulations of non-tropical storms at high spatial and temporal resolution. The framework adopts optimal estimation for spatio-temporal modeling of rain fields. A non-parametric approach featuring K-Nearest Neighbor Resampling (KNNR) plus the Genetic Algorithm (GA) mixing process is utilized for generating parameters in the long-term simulation. A case study is conducted in Dallas-Fort-Worth metroplex as the simulation domain. Ensemble parameters are generated using the KNNR+GA method from adjacent homogeneous areas and 10 years of radar rainfall observation. One hundred most rainy days in the 10 years are simulated at the resolutions of 4 × 4 km and 1 hour for 50 ensemble members. The simulated rainfall is thoroughly evaluated against the observed radar rainfall with respects to statistical moments, spatio-temporal structure, and frequency distribution of rainfall at both near-point scale and domain scale. The results indicate that ensemble simulations successfully reproduce key statistical properties of the observed rainfall. In addition, the approach is also effective and flexible in capturing heavy rainfall values, which is important for many hydrologic/hydraulic practices. As essentially a downscaling tool, this stochastic rainfall generator can have many applications where rainfall needs to be represented at finer spatiotemporal resolution.