Contact
Deltares | R&D Highlights 2015 Flood Risk
Real-time hydrological forecasts can reduce but not eliminate uncertainty about future hydrological conditions such as future streamflow rates or water levels. The uncertainty can be caused by many factors such as imperfect hydrological models for streamflow generation and streamflow propa-gation and weather forecasts. These models and weather forecasts themselves are inherently uncertain and so hydro-logical forecasts may be ‘wrong’ in that events (such as floods) may not be noticed in time or in that a forecast event may not materialise.
Uncertainty can be managed in multiple ways. In the long term, it may be reduced through research. Indeed, improvements to hydrological and meteorological forecasts have achieved just this: forecasting skills have gradually improved over the years. A more immediate approach involves the assessment of the remaining uncertainties in order to produce probabilistic forecasts. Although probabilistic forecasts cannot prevent missed events or false alarms, the probability will be known in advance, allowing for risk-based decision-making. The research has provided evidence that allows for larger reductions of flood risk than would be the case with single-valued, deterministic forecasts.
The hydro-meteorological sciences have multiple approaches for estimating predictive uncertainty. Many are based on either ensemble techniques (the implementation of the Monte Carlo principle) or on statistical post-processing techniques. The aim of the first approach is to produce multiple forecasts by using multiple inputs that vary from one another while remaining plausible. Post-processing techniques modify forecasts on the basis of prior performance. The study described here explored variations on existing techniques. These included the application of post-processing techniques to both weather forecasts and hydrological forecasts, the combination of ensemble techniques with statistical post-processing techniques, and improvements to statistical post-processing techniques. Many of these experiments were conducted with the Delft-FEWS forecasting system and its inherent facilities for ‘hind-casting’ or reforecasting.
Estimating real-time
predictive hydrological
uncertainty
Further reading:
Verkade (2015), Estimating Real Time Predictive Hydrological Uncertainty, PhD thesis Delft UT, via http://repository.tudelft.nl jan.verkade@deltares.nl
T +31(0)88 335 8348
‘Ensemble dressing’ forecast for Meuse at St Pieter
Pr ob ab ili ty o f t hr es ho ld ex cee de n ce [-] D isch ar ge [m 3/s] 31/1 1/1 2/1 3/1 4/1 2500 2000 1500 1000 500 0 1 0,8 0,6 0,4 0,2 0 no w Prediction Thresholds