A Multivariate Statistical Approach based on a Dynamic Moving Storms (DMS) Generator for Estimating the Frequency of Extreme Storm Events

Abstract:

Challenges of fully considering the complexity among spatially and temporally varied rainfall always exist in flood frequency analysis. Conventional approaches that simplify the complexity of spatiotemporal interactions generally undermine their impacts on flood risks. A previously developed stochastic storm generator called Dynamic Moving Storms (DMS) aims to address the highly-dependent nature of precipitation field: spatial variability, temporal variability, and movement of the storm. The authors utilize a multivariate statistical approach based on DMS to estimate the occurrence probability or frequency of extreme storm events. Fifteen years of radar rainfall data is used to generate a large number of synthetic storms as basis for statistical assessment. Two parametric retrieval algorithms are developed to recognize rain cells and track storm motions respectively. The resulted parameters are then used to establish probability density functions (PDFs), which are fitted to parametric distribution functions for further Monte Carlo simulations. Consequently, over 1,000,000 synthetic storms are generated based on twelve retrieved parameters for integrated risk assessment and ensemble forecasts. Furthermore, PDFs for parameters are used to calculate joint probabilities based on 2-dimensional Archimedean-Copula functions to determine the occurrence probabilities of extreme events. The approach is validated on the Upper Trinity River watershed and the generated results are compared with those from traditional rainfall frequency studies (i.e. Intensity-Duration-Frequency curves, and Areal Reduction Factors).