How much additional model complexity do the use of catchment hydrological signatures, additional data and expert knowledge warrant?
Abstract
In the frequent absence of sufficient suitable data to constrain hydrological models, it is not uncommon to represent catchments at a range of scales by lumped model set-ups. Although process heterogeneity can average out on the catchment scale to generate simple catchment integrated responses whose general flow features can frequently be reproduced by lumped models, these models often fail to get details of the flow pattern as well as catchment internal dynamics, such as groundwater level changes, right to a sufficient degree, resulting in considerable predictive uncertainty. Traditionally, models are constrained by only one or two objectives functions, which does not warrant more than a handful of parameters to avoid elevated predictive uncertainty, thereby preventing more complex model set-ups accounting for increased process heterogeneity. In this study it was tested how much additional process heterogeneity is warranted in models when optimizing the model calibration strategy, using additional data and expert knowledge. Long-term timeseries of flow and groundwater levels for small nested experimental catchments in French Brittany with considerable differences in geology, topography and flow regime were used in this study to test which degree of model process heterogeneity is warranted with increased availability of information. In a first step, as a benchmark, the system was treated as one lumped entity and the model was trained based only on its ability to reproduce the hydrograph. Although it was found that the overall modelled flow generally reflects the observed flow response quite well, the internal system dynamics could not be reproduced. In further steps the complexity of this model was gradually increased, first by adding a separate riparian reservoir to the lumped set-up and then by a semi-distributed set-up, allowing for independent, parallel model structures, representing the contrasting nested catchments. Although calibration performance increased, additional parameters also increased model equifinality, leading to substantially increased predictive uncertainty, thus indicating that these model complexities are not warranted by the chosen calibration strategy. In a next step the calibration strategy was reviewed and the three model set-ups of increasing complexity gradually were constrained with more information, starting from catchment signatures extracted from available flow data (e.g. Flow Duration Curve), over additional data (observed groundwater levels) to expert knowledge. The results suggest that in the study catchments, already the efficient extraction and use of additional information from available flow data together with anecdotal expert knowledge warrants relatively high degrees of model complexity: such well constrained semi-distributed models reflect real process heterogeneity to a higher degree, resulting in higher model performances, more realistic reproductions of system internal dynamics as well as significantly reduced predictive uncertainties illustrating the advantages of models with increased complexity even in the absence of additional data
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H33J..04H
- Keywords:
-
- 1847 HYDROLOGY Modeling;
- 1804 HYDROLOGY Catchment;
- 1846 HYDROLOGY Model calibration;
- 1873 HYDROLOGY Uncertainty assessment