A point rainfall model for risk-based design

Abstract

The point rainfall model presented extends previous work on event-based rainfall models and overcomes some of their shortcomings. The model uses event-based data and can be calibrated using rainfall data substantially affected by missing or corrupted values. Particular attention was given to adequately simulating extreme storm rainfall events for use in hydrological risk assessment. The model is capable of simulating the inter-event time, storm duration, average event intensity and intra-event temporal characteristics. Conditioning of the average event rainfall intensity on rainfall duration and time of year is a feature of the model. Rainfall events are disaggregated using a conditional random walk on a dimensionless mass curve. Pluviograph data in 6 min increments from three Australian capital cities (Sydney, Brisbane and Melbourne) was used to calibrate the model parameters. It was found that the constrained random walk parameters were almost identical for the three cities. The model was tested using statistics not used in its calibration and the simulated intensity–frequency–duration extreme rainfall statistics compared very favorably with observed values. In addition, simulated aggregated statistics compared favorably with observed statistics from 30 min to monthly durations. The simulated annual rainfalls significantly underestimated the observed variability for Brisbane and Sydney, whereas satisfactorily reproduced the Melbourne variability. An explanation is offered for these differences.