HR: 1340h
AN: H13H-1681    [Abstracts]
TI: Dynamic Multicriteria Evaluation of Conceptual Hydrological Models
AU: * de Vos, N J
EM: n.j.devos@tudelft.nl
AF: Water Resources Section, Delft University of Technology, Stevinweg 1, Delft, 2600GA, Netherlands
AU: Rientjes, T H
EM: rientjes@itc.nl
AF: Department of Water Resources, International Institute for Geo-Information Science and Earth Observation, Hengelosestraat 99, Enschede, 7500AA, Netherlands
AU: Fenicia, F
EM: fenicia@lippmann.lu
AF: Water Resources Section, Delft University of Technology, Stevinweg 1, Delft, 2600GA, Netherlands
AU: Fenicia, F
EM: fenicia@lippmann.lu
AF: Public Research Center – Gabriel Lippmann, 41 Rue du Brill, Belvaux, L-4422, Luxembourg
AU: Gupta, H V
EM: hoshin.gupta@hwr.arizona.edu
AF: Department of Hydrology and Water Resources, University of Arizona, 1133 E James E. Rogers Way, Tucson, AZ AZ85721, United States
AB: Accurate and precise forecasts of river streamflows are crucial for successful management of water resources and under the threat of hydrological extremes such as floods and droughts. Conceptual rainfall–runoff models are the most popular approach in flood forecasting. However, the calibration and evaluation of such models is often oversimplified by the use of performance statistics that largely ignore the dynamic character of a watershed system. This research aims to find novel ways of model evaluation by identifying periods of hydrologic similarity and customizing evaluation within each period using multiple criteria. A dynamic approach to hydrologic model identification, calibration and testing can be realized by applying clustering algorithms (e.g., Self-Organizing Map, Fuzzy C-means algorithm) to hydrological data. These algorithms are able to identify clusters in the data that represent periods of hydrological similarity. In this way, dynamic catchment system behavior can be simplified within the clusters that are identified. Although clustering requires a number of subjective choices, new insights into the hydrological functioning of a catchment can be obtained. Finally, separate model multi-criteria calibration and evaluation is performed for each of the clusters. Such a model evaluation procedure shows to be reliable and gives much-needed feedback on exactly where certain model structures fail. Several clustering algorithms were tested on two data sets of meso-scale and large-scale catchments. The results show that the clustering algorithms define categories that reflect hydrological process understanding: dry/wet seasons, rising/falling hydrograph limbs, precipitation-driven/ non-driven periods, etc. The results of various clustering algorithms are compared and validated using expert knowledge. Calibration results on a conceptual hydrological model show that the common practice of single-criteria calibration over the complete time series fails to perform adequately in all periods or on all criteria. Subsequently, improved model structures are constructed and the evaluation repeated. We conclude that a dynamic, multi-criteria approach to model identifying and testing is effective in constructing models that are more accurate and precise in forecasting streamflow.
DE: 0520 Data analysis: algorithms and implementation
DE: 0555 Neural networks, fuzzy logic, machine learning
DE: 1805 Computational hydrology
DE: 1846 Model calibration (3333)
SC: Hydrology [H]
MN: 2007 Fall Meeting