Browsing by Author "Bárdossy, András (Prof. Dr. rer.nat. Dr.-Ing. habil.)"

Now showing 1 - 2 of 2

Open Access
Multi-objective automatic calibration of hydrodynamic models - development of the concept and an application in the Mekong Delta
(2011) Nguyen, Viet-Dung; Bárdossy, András (Prof. Dr. rer.nat. Dr.-Ing. habil.)
Automatic and multi-objective calibration of hydrodynamic models is still underdeveloped, in particular, in comparison with other fields such as hydrological modeling. This is for several reasons: lack of appropriate data, the high degree of computational time demanded, and a suitable framework. These aspects are aggravated in large-scale applications. There are recent developments, however, that improve both the data and the computing constraints. Remote sensing, especially radar-based techniques, provide highly valuable information on flood extents, and in case high precision Digital Elevation Models (DEMs) are present, also on spatially distributed inundation depths. With regards to computation, the use of parallelization techniques brings significant performance gains. In the presented study, we build on these developments by calibrating a large-scale one-dimensional hydrodynamic model of the whole Mekong Delta downstream of Kratie in Cambodia: We combine in-situ data from a network of river gauging stations, i.e. data with high-temporal but low-spatial resolution, with a series of inundation maps derived from ENVISAT Advanced Synthetic Aperture Radar (ASAR) satellite images, i.e. data with low-temporal but high-spatial resolution, in a multi-objective automatic calibration process. It is shown that this kind of calibration of hydrodynamic models is possible, even in an area as large-scale and complex as the Mekong Delta. Furthermore, the calibration process reveals deficiencies in the model structure, i.e. the representation of the dike system in Vietnam, which would be difficult to detect by a standard manual calibration procedure. In the last part of the dissertation the established hydrodynamic model is combined with flood frequency analysis in order to assess the flood hazard in the Mekong Delta. It is now common to state that climate change can lead to a change in flood hazard. Starting from this assumption, this study develops a novel approach for flood hazard mapping in the Mekong Delta. Typically, flood frequency analysis assumes stationarity and is limited to extreme value statistics of flood peaks. Both, the stationarity assumption and the limitation to univariate frequency analysis remain doubtful in the case of the Mekong Delta, because of changes in hydrologic variability and because of the large relevance of the flood volume for the impact of flooding. Thus, besides the use of the traditional approach for flood frequency analysis, this study takes non-stationarity and bivariate behavior into account. Copula-based bivariate analysis is used to model the dependence and to generate pairs of maximum discharge and volume, by coupling their marginal distributions to gain a bivariate distribution. In addition, based on cluster analysis, groups of characteristic hydrographs are identified and synthetic flood hydrographs are generated. These hydrographs are the input for the calibrated large-scale hydrodynamic model of the Mekong Delta, resulting in flood hazard maps for the whole Mekong Delta. To account for uncertainty within the hazard assessment, a Monte Carlo framework is applied yielding probabilistic hazard maps.
Open Access
Regionalization of an event based Nash cascade model for flood predictions in ungauged basins
(2008) Patil, Sachin Ramesh; Bárdossy, András (Prof. Dr. rer.nat. Dr.-Ing. habil.)
This study was aimed at developing a practical, robust and physically reasonable methodology for estimation of design flood under data scarce conditions. Due to lack of sufficient discharge data and inconstant hydrological conditions, application of discharge-frequency analysis or calibration of a hydrological model is not always viable. In such cases model parameters are obtained through regionalization procedure. The main objective of this study was to derive a regionalization methodology for flood predictions in ungauged catchments, which is strictly based on physically reasonable transfer functions and adequately addresses the problem of parameter equifinality. A possibility of assessment of impact of land use changes on flood characteristics was also investigated during the study. An event based hourly Nash cascade model was developed to derive the direct runoff hydrograph from rainfall time series. The model was implemented on a semi-distributed scale, i.e. the direct runoff hydrograph is estimated at sub-catchment scale, and then it is routed to the outlet of the catchment using the Muskingum routing procedure. The model uses three parameters, the runoff coefficient (RC), the number of reservoirs (N) and a reservoir constant (K). For ungauged catchments, the model parameters RC, N and K must be estimated through a regionalization procedure. The study was conducted using 209 rainfall-runoff events from 41 mesoscale catchments in the south-west region of Germany. Among the 41 catchments, 22 were used for optimization of the regionalization methodology and 19 were used for its validation. Areal rainfall time series for the events were estimated through external drift kriging. Various event and catchment specific hydrological characteristics, describing event specific conditions were estimated for each of the events. Four different approaches were employed to derive four different transfer functions for RC: a multiple linear transfer function (MLTF), an artificial neural network transfer function (ANNTF), a fuzzy logic transfer function (FLTF) and a logistic transfer function (LoTF). ANNTF, FLTF and LoTF exhibited a high goodness fit performance for the events in the regionalization as well validation set. In order to investigate whether the transfer functions are physically reasonable, validation for physical relationships was carried out by comparing the signs of derivatives of the functional relationships between the predictors and RC, as featured in the transfer functions, with the signs of derivatives of the physical relationships. The validation revealed that the response of ANNTF and FLTF to hydrological changes often conflicts with the response of the physical relationships. On the other hand, the response of LoTF was consistent with that of the physical relationships, which indicates that it is physically reasonable. The Nash cascade parameters N and K exhibit strong inter-parameter relationship which can be represented by a power function with an exponent (α = -1.0) and a coefficient β. Therefore, regionalization of K and the coefficient β was carried out, where N can be estimated by using K and the inter-parameter function. The transfer functions were optimized by using mean Nash-Sutcliffe coefficient as an aggregated goodness fit measure for a set of gauged catchments. During the optimization and the validation of the transfer functions, highly acceptable aggregated goodness fit performance was achieved, which indicates that the transfer functions are both reliable and efficient at transferring the model parameters to ungauged catchments. The validation of the transfer functions for physical relationships was carried out by comparing the change in the shape of modeled unit hydrograph, due to change in the hydrological characterisics, with the change anticipated from the a priori knowledge of runoff propagation processes. The comparisons with the existing common practices, such as SCS curve number method and the Lutz procedure, revealed that for the study area under consideration, the regionalization methodology performs better than the existing practices. The regionalization methodology is built on physically reasonable relationships with event as well as catchment specific hydrological characteristics. Therefore, it is robust and suitable for both the temporal as well as the spatial transfer of the model parameters. The assessment of impact of land use changes on flood characteristics was carried out for three different land use scenarios in the catchment Tübingen. The attempt led to the conclusion that there is a reasonable chance of using such methodology for assessment of impact of land use changes.