NEW FRAMEWORK TO WORK TOWARDS MORE ACCURATE FLOOD FORECASTS
NEWS
No matter if you are a civilian or a flood manager, you need accurate information at the earliest moment in order to adequately prepare for a flood. Preferably days before the flood actually occurs. But as anybody knows who has ever been caught off guard by a rainstorm, weather forecasts can be inaccurate. The same is true for the models that predict where the rain will end up and how it will affect flood levels. Researchers in the Flood Control program are developing different methods to increase the accuracy of these forecasts. They are developed alongside a common framework for uncertainty reduction in operational flood forecasting systems.
Close to reality
Albrecht Weerts
Albrecht Weerts, researcher at Deltares, leads the team working on this ‘Uncertainty Framework’. “In operational flood forecasting, we want to predict water levels as close as possible to how they occur in reality together with an estimate of the (un)certainty around this estimate. This is called ‘predictive uncertainty,” Weerts explains. “By measuring the water levels that actually occurred, we can establish if the forecasting methods were accurate, unbiased and precise. Or in other words, we can find out if the methods on average predict water levels close to reality, with clearly defined and small deviations from this average.”
"These deviations from the average, or errors, are essential in establishing the uncertainty in the forecast. In flood forecasting, a series of models are run such as weather models, rainfall-runoff models and hydrodynamic models. The output from one model is the input for the next. So when every model is only slightly inaccurate, the resulting end prediction can be highly inaccurate."
Calibration
“Getting feedback from reality, in our case the real flood levels, is essential in establishing the predictive uncertainty of an operational forecasting system. Without some form of forecast calibration, forecasts are little more than sensitivity analyses of the models being used. Informative, but not very practical in real-life conditions.” For this reason, different techniques are investigated, such as data assimilation, error/bias correction, and forecast calibration methods like quantile regression. “The techniques are now available through Delft-FEWS. A study of the operational forecasting system and the relevant catchment areas will reveal which techniques will help most with the reduction of uncertainties”.
Example of Quantile Regression for the Severn River, UK
Uncertainty Framework
An uncertainty framework for flood and storm surge forecasting has been defined which can help in getting a quick and complete overview of the main sources of uncertainty that play a role in the flood forecasts. The framework is build around procedural and operational constraints that may vary per organisation. The framework can help to get insight into which future investments makes most sensein for some existing flood forecasting system. Or the framework can help to decide which method, in which part of the model chain, is most suitable for increasing the accuracy or quantifying the (predictive) uncertainty of the flood forecast.
“In every model used for flood forecasting, there are basically three sources of uncertainty. There is an uncertainty about the input boundary conditions for the prediction, about the initial conditions of the area or model, and about the behaviour of the model during the prediction phase. Furthermore, the resulting uncertainty in the model results, e.g. in the weather conditions, is then used as input for another model, e.g. to calculate the resulting storm surge in a certain coastal area.”
International Workshop
To share results of this research and to discuss it with the international research community, Deltares is the host of an “International workshop on data assimilation for operational hydrologic forecasting and water resources management”, to be held on November 1-3 this year in Delft, the Netherlands. Co-organised with the US NOAA National Weather Service and co-sponsored by the HEPEX initiative and Flood Control 2015, this workshop will bring together the state of the art in uncertainty reduction.
Registration and abstract submission for this event is now open, via the website www.floodcontrol2015.com/daworkshop2010
The uncertainty framework offers a structured approach to reduce the predictive uncertainty