Improving operational flood forecasting using data assimilation
toon extra info.
|Wageningen : Wageningen University|
|138 pagina’s figuren, grafieken|
|Proefschrift Wageningen University ter verkrijging van de graad van doctor in het jaar 2014 toon alle annotatie(s)
Met literatuuropgave. - Met samenvatting in het Engels en Nederlands
|Uijlenhoet, Prof. dr. R. ; Weerts, Dr. A.H.|
|Samenvatting door auteur||
Reliable and accurate flood early warnings are very important because they can mitigate the number of casualties and reduce economic damage. Understanding the behaviour of extreme hydrological events and the ability of hydrological modellers to improve their forecast skills are distinct challenges of applied hydrology. Since models simplify the real world complexity, hydrological forecasts are prone to many sources of uncertainties, such as in initial conditions, boundary conditions, and model input, structure and parameters.
Currently, most hydrological forecast systems employ lumped hydrological models (with deterministic or manual state updating), but there is a noticeable trend to move towards distributed hydrologic models with ensemble forecasting. Contrary to deterministic forecasts, which provide only one predicted value, ensemble forecasts enable quantitative estimates of the associated uncertainty. Besides improving on the boundary conditions of the hydrologic forecast, the hydrologic forecasts can be made more reliable and less uncertain by improving initial conditions and/or the model being used (either by improved model structure or better parameters). A common way of improving the initial conditions is to make use of data assimilation (DA), a feedback algorithm or update methodology which merges model estimates with available real world observations.
This thesis contributes to improved understanding and quantification of hydrological model uncertainty especially related to the initial conditions of the model and to a lesser extent to the model structure and parameters.
A plausible rainfall ensemble generator based on time-dependent multivariate spatial conditional simulations is presented. It is defined in such a way that the rainfall fields provide proper information on the spatial error structure. Additionally, it captures the temporal coherence for each realization of a sequence of spatial rainfall fields at an hourly time step. The advantage of taking the temporal coherence in hydrological model states into account is that it avoids the necessity to smooth possible extreme state values, which can occur when neglecting temporal coherence.
To efficiently and effectively evaluate parameter sensitivity, a new computationally frugal method is proposed: Distributed Evaluation of Local Sensitivity Analysis (DELSA). DELSA distinguishes important and unimportant parameters and evaluates how model parameter importance changes as parameter values change. Two test cases are compared with the popular global, variance-based Sobol' method. The results indicate that Sobol' and DELSA identify similar important and unimportant parameters, with DELSA enabling more detailed insight at much lower computational cost. For example, in the real-world problem the time delay in runoff is considered to be the most important parameter in all models, but DELSA shows that for about 20% of the parameter sets it is not important at all and alternative mechanisms and parameters dominate. Moreover, the time delay was identified as important in regions producing poor model fits, whereas other parameters were identified as more important in regions producing better model fits.
In the second part of this thesis, benefits of data assimilation for improved flood forecasting are presented. Several computational experiments are carried out to update model states of grid-based hydrological models by sequentially assimilating streamflow observations. For example, the configuration of the discharge observation vector and the updating frequency with which the observed data becomes available are scrutinized using the Ensemble Kalman filter (EnKF). The results show that the hydrological forecast at the catchment outlet is improved by assimilating interior gauges. This augmentation of the observation vector improves the forecast more than increasing the updating frequency.
Subsequent analysis investigates how the capabilities of the DA procedure can be improved by applying alternative Kalman-type methods, e.g., the Asynchronous Ensemble Kalman Filter (AEnKF) (comparable to 4D VAR), which calculates the analysis from several previous time steps up to the time of forecast, instead of mapping the instantaneous covariance between states and discharge. The results show that including past predictions and observations in the AEnKF improves the model forecasts as compared to the traditional EnKF. Additionally we show that elimination of the strongly non-linear relation between the soil moisture storage and assimilated observations can become beneficial for improved operational forecasting.
A related study shows the effect of noise specification on the quality of hydrological forecasts via an advanced DA procedure based on a rainfall ensemble generator and lagged particle filtering. The strength of the proposed procedure is that it requires less subjectivity in implementation of DA compared to conventional methods and therefore subjective use of additional random noise in DA is mitigated.
|Trefwoorden (cab)||overstromingen / hoogwaterbeheersing / voorspellen / data-assimilatie / gegevensanalyse / hydrologie / modellen|
|Toelichting||Hoogwatervoorspellingssystemen die betrouwbaar en nauwkeurig overstromingen kunnen voorspellen zijn erg belangrijk, omdat dit het aantal slachtoffers en de economische schade van overstromingen kan beperken. Het begrijpen van het gedrag van extreme hydrologische gebeurtenissen en het vermogen van de modelleur om betere en nauwkeurigere prognoses te krijgen, zijn uitdagingen binnen de toegepaste hydrologie. Omdat modellen slechts een versimpelde weergave van de complexe werkelijkheid geven, kleven er aan hydrologische voorspellingen veel onzekerheden. Dit proefschrift draagt bij aan een verbeterd begrip en kwantificatie van hydrologische modelonzekerheid, vooral gekoppeld aan de initi¨ele condities van het model, en in mindere mate aan de modelstructuur en de parameters.|
WUR, Sectie Waterhuishouding