Sensitivity of the regionalised inputs in the Monte Carlo simulation technique to design flood estimation for New South Wales

In Australia, there is a growing interest to move away from the Design Event Approach (DEA) and to apply more holistic approaches of design flood estimation such as Monte Carlo Simulation Technique (MCST). Recent research on the MCST has demonstrated that it can overcome the major limitations associated with the DEA, which ignores the stochastic nature of the model inputs except rainfall depth. The MCST explicitly accounts for the stochastic nature of the input variables in rainfall runoff modelling. In practical application of MCST, the regionalisation of stochastic input variables is desirable. This paper presents a sensitivity analysis involving various regional stochastic input variables in the application of the MCST for the state New South Wales (NSW) in Australia. In this application of MCST, complete storm duration (DCS), inter-event duration (IED), initial loss (IL) and runoff routing storage delay parameter (k) data are approximated by gamma distribution. Continuing loss (CL) data is approximated by an exponential distribution. An arbitrary increase and decrease of 5%, 10%, 20% and 50% are applied to the mean and standard deviation values of the marginal distributions of the input variables to generate various possible sets of model inputs. The intensityfrequency- duration (IFD) and temporal patterns (TP) data are regionalised based on IFD curves and TP data of 86 pluviograph stations. Hence, for these two model inputs, the numbers of nearby pluviograph stations are varied from one to thirteen to achieve a possible variation in the stochastic IFD and TP data. It is found that the DFFCs are sensitive to the individual regional stochastic model inputs/parameters in the following order (most sensitive to the least sensitive ones): k (-30% to 95%), IED (-29% to 60%), DCS (-30% to 50%), IL (-40% to 40%), IFD (10% to 24%), TP (9% to 15%) and CL (-10% to 14%). It is found that up to about 10% variations in the stochastic model inputs/parameters do not make any notable differences in the DFFCs.