Validation of the Extremal Behaviour of Rainfall Generators
This project investigates the validation of the extremal behaviour of rainfall generators using extreme-value theory techniques. The motivation behind this study is the growing frequency and severity of extreme rainfall events globally, largely due to climate change, which presents significant challenges for hydro-meteorological modeling.
The research focuses on the Advanced WEather GENerator (AWE-GEN), calibrated on observed hourly rainfall data spanning 30 years from San Francisco. The primary objective is to validate the generator’s ability to replicate extreme events. This is achieved by applying two key methodologies:
Block maxima models: Utilizing the Generalized Extreme Value (GEV) distribution to fit observed and simulated rainfall data. A modified approach incorporating the r largest values per block is employed to improve parameter estimation.
Threshold models: Implementing the Generalized Pareto Distribution (GPD) on declustered time series. An alternative method is used to account for clusters of extremes, by incorporating the extremal index. Additionally, two hierarchical models for the GPD are evaluated for better temporal dependence management.
The findings indicate that while the GEV distribution is the current standard for validating extremal behaviour in rainfall generators, the GPD approach shows greater promise in accounting for temporal dependencies. Using the extremal index or hierarchical processes enhances the comparison between observed and simulated data.
This study makes a methodological contribution to hydrology by introducing advanced extreme-value theory tools for validating rainfall generators, thus addressing a crucial requirement for hydro-meteorologists.