Design and testing of a real-time flood forecasting system for urban catchments

Abstract

In the late afternoon on Friday 9 March 2001, a severe storm occurred on the south side of Brisbane. Peak intensity reached more than 210 mm/h, and total rainfall up to 200 mm occurred over a period of 3 hours. Extensive flooding was recorded at a number of locations in the Norman Creek catchment. This and other localised storm events have highlighted the need for real-time flood forecasting to increase the awareness of and preparedness for flash floods in urbanised areas, and to minimise the damage and disruption as a result of flash floods in the tropical and sub-tropical environments. A real-time forecasting system is developed using ensemble probability-based rainfall forecasts coupled with an efficient raster-based hydrologic model to forecast water levels at critical locations for dissemination on the internet. Forecasting of flash flood involves interrogating ALERT/telemetry data on gauge rainfall and water levels in the catchment, processing and interpreting radar scans to determine the storm’s spatial structure and movement, numerical weather prediction up to 30–90 min into the future, generating multiple realisations of likely rain fields, predicting multiple hydrographs from forecast rain fields, and converting hydrographs into probability-based water level forecasts at critical locations in the catchment. The design is critically constrained by the requirement to update the forecast on the internet every 15 minutes. The paper reports a retrospective test using the gauge and radar data for the March 2001 event in the 28 km2 Norman Creek catchment.