A simulation analysis of climatic and basin factors affecting the flood frequency curve
A single site Neyman-Scott point process model of rainfall, with convective and stratiform cells (Cowpertwait, 1994; 1995), has been employed to generate synthetic rainfall inputs to a rainfall runoff model. The time series of the potential evapotranspiration (ETp) demand has been represented through an AR(n) model with seasonal component, while a simplified version of the ARNO rainfall-runoff model (Todini, 1996) has been employed to simulate the continuous discharge time series. The model incorporates a saturation excess runoff production component, and a routing component based on a network width function and a linear parabolic transfer function. All these models have been parameterised in a realistic manner using observed data and results from previous application, to obtain ‘reference’ parameter sets.
Continuous hourly time series of rainfall and potential evapotranspiration, of length 10000 years, have been generated using the reference parameter sets for both models. They have been then used as inputs to the rainfall runoff model, and an hourly time series of discharges has been generated from which the annual maximum flood peaks have been extracted and plotted against the Gumbel variate. Subsequently, perturbations to the model parameters have been made through two approaches: a) the analysis initially was performed by perturbing the parameters one-at-a-time and the sensitivities of the generated annual maximum rainfall and flood frequency curves (unstandardised, and standardised by the mean) have been assessed graphically and with the assistance of several statistics (a) of the annual maximum rainfall and peak flood values and (b) of the soil moisture content at the storm arrival time.
Overall, the sensitivity analysis described in this research suggests that the soil moisture regime, and, in particular, the probability distribution of soil moisture content at the storm arrival time, can be considered as a unifying link between the perturbation to the several parameters and their effects on the standardised and unstandardised ffcs, thus revealing the physical mechanism through which their influence is exercised. However, perturbations to the parameters of the linear routing component affect only the unstandardised ffc; b) the second approach which can be considered as a full sensitivity analysis, the effect of model parameters on the ffc has been assessed through an analysis of variance (ANOVA) of the results obtained from a formal experimental design, where all the parameters are allowed to vary simultaneously, thus providing deeper insight into the interactions between the different factors. This approach allows a wider range of climatic and basin conditions to be analysed and suggests that further investigations are needed to understand better the factors affecting the links between climate and basin characteristics and the ffc properties.
Finally, as a complementary study, the simulation modelling approach is evaluated using several properties of the streamflow and annual maximum flood.