Many communities on the east coast of Australia are subject to flash flooding. Due to the convective rainfall typical of Eastern Australia, combined with local catchment characteristics, flash flooding can be very intense and occur with little warning.
Flash flood forecasting is used to predict the impact of these high rainfall events before they occur. A typical Flash Flood Forecasting System (FFFS) uses meteorological forecasts, soil moisture estimates, sea level conditions and numerical modelling to understand how much rain will turn to runoff and how it will flow through a catchment.
Traditional systems in Australia have used the meteorological forecasts within the Australian Digital Forecast Database (ADFD), which provide a seven-day rainfall forecast with a temporal resolution of 3-hours. However, ongoing performance testing of existing systems shows the need to move towards higher resolution datasets. The following graph shows an example of ADFD 3-hour forecast rainfall against observed rainfall in South East Queensland, showing hyetographs and cumulative totals. Over the initial 48-hour period to the 17th December, the forecast and recorded rainfall are almost identical in terms of cumulative totals; each being around 90mm. However, due to the 3-hour resolution of the ADFD rainfall forecasts, the small storm bursts are not predicted to occur, and instead the rain is predicted to occur at a lower intensity over a longer period.
The impact of using the 3-hour temporal forecasts can be estimated numerically using the URBS semi-distributed hydrological model. The graph below shows the resulting flood conditions from the forecast vs the observed datasets. Even with approximately the same total rainfall, the 3-hourly forecast rainfall predicts a river level far below the actual river levels. In this instance the short bursts of observed rainfall were the greatest contributor to the local catchment runoff.
New data is emerging to better capture short bursts of rainfall. Recently the Bureau of Meteorology released the gridded rainfall product RAINFIELDS, which has a timestep of 5 minutes over the upcoming 12 hours. However, unlike the ADFD which has years of archived historical forecasts, it will take time to build a long-term database of the RAINFIELDS forecasts. Understanding the performance of the RAINFIELDS product in forecasting future rainfall is currently a focus of our work and we believe an important step in improving flash flood forecasting in Australia.