Simulation of rainfall events

Efe-Eyefia, Eferhonore 2023. Simulation of rainfall events. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (3MB) | Preview
	PDF - Supplemental Material Restricted to Repository staff only Download (123kB)

Abstract

Rainfall events play an essential role in ecosystems worldwide. While water in good quality and quantity is vital for all life on earth, rainfall can also be the cause of adverse events such as floods, debris movements and droughts. The design of hydrological structures (dams/drainage), urban planning or climate change adaptation needs to take into account such risks. However, as rainfall data are usually only available and measured for a few places of interest, one often relies on simulations of rainfall events that can accurately reproduce realistic rainfall patterns. In order to also capture risks imposed by flash floods (short duration-high intensity events), it is crucial to have simulators with a high-frequency output (e.g. sub-hourly). It is the purpose of this thesis to build a novel stochastic parsimonious high-frequency rainfall simulator from high-frequency data that can accurately represent key characteristics of rainfall events in the data: duration (D), intensity (I), maximum intensity (M), and volatility (V), collectively referred to as DIMV, as well as temporal patterns of inter-event times. Therefore, this thesis works with a unique dataset of high-resolution (6-minute) rainfall gauge data from Sunbury, Australia, spanning 36 years from the Australian Bureau of Meteorology. We use a 1-hour minimum inter-event time to extract rainfall events from these data. First, we analyse the univariate marginal distributions of the above characteristics. Our studies addressed the skewed nature of the DIMV data using log transformations, leading to effective modelling. The skew t was identified as the best fit for duration and volatility, while the generalised extreme value distribution was the best fit for intensity and maximum intensity. We also developed a novel univariate hybrid model, F-Exp-GPD, designed to model rainfall events. By generalising existing hybrid distributions, the F-Exp-GPD showcased versatility, offering a harmonious representation of both bulk and tail behaviour. The model was used to fit duration and intensity to affirm the efficacy of this model, with the GEV-Exp-GPD variant standing out. This knowledge facilitated sophisticated compound distribution and copula modelling. To capture the interdependence among the variables, we utilised the vine copula methodology. Among the various vine copula structures, the D-vine copula proved to be the most formidable in representing the dependencies of the DIMV characteristics. This was validated by successful simulations that maintained intricate sample dependencies, drawing a striking resemblance between the copula simulated and observed data. Lastly, from the fitted models, we developed a flexible rainfall event simulator that also incorporates accurate rainfall temporal intensity patterns using IET data information. It effectively simulates rainfall across long time intervals, reproducing statistical properties of DIMV patterns from the data. This model iterates through specified times, integrates real-world statistical properties by utilising the rain event simulator, and generates detailed rain events, considering consistent and irregular intervals between them. The developed model for simulating an array of rainfall events promises a high degree of authenticity, making it a cornerstone for future hydrological studies, urban planning, and climate change modelling.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Mathematics
Date of First Compliant Deposit:	15 May 2024
Last Modified:	15 May 2024 11:51
URI:	https://orca.cardiff.ac.uk/id/eprint/168922

Actions (repository staff only)

Edit Item