A stochastic space-time rainfall model for engineering risk assessment.

Abstract

The temporal and spatial variability of Australia’s climate affects the quantity and quality of its water resources, the productivity of its agricultural systems, and the health of its ecosystems. This variability should be taken into account when assessing the risks associated with flooding. Continuous simulation rainfall models are one means for doing this, whereby sequences of storms are generated for an arbitrarily long time period and over some region of interest. The simulated rainfall should reproduce observed statistics in time and space so that it can be used as a suitable input for hydrologic models at the catchment scale, with particular emphasis on extreme events. There are a variety of approaches to modelling rainfall, including a broad range of singlesite and multi-site rainfall models. By way of contrast there are few models that aim to simulate rainfall across all points within a region at daily or sub-daily increments. This thesis focuses on models calibrated solely to rain gauges, and a specific type known as Neyman-Scott Rectangular Pulse (NSRP) models. Existing NSRP models have a mature history of modelling developments including calibration methodology and an ability to reproduce key statistics across a range of timescales. Nonetheless, these models also have several limitations (and other space-time models not withstanding) that are addressed in this thesis. These developments include improvements to the conceptual representation of rainfall and improvements to calibration and simulation techniques. Specifically these improvements include (i) the development of an efficient simulation technique, (ii) assessing the impact of monthly parameter changes on on rainfall statistics, (iii) the use of simulated statistics within calibration to overcome reliance on derived model properties (iv) incorporating a storm extent parameter to better match spatial correlations, (v) incorporating long term climatic variability and developing a methodology to assess climatic and seasonal variability in simulated extremes (vi) incorporating inhomogeneity of rainfall occurrence across a region. Numerous case studies are used at various locations about Australia to illustrate these improvements and highlight the applicability of the model under varied climatic conditions.