Assessing seasonal hydrological model parameters for a better real-time forecasting performance
Nsubuga, R. (2021). Assessing seasonal hydrological model parameters for a better real-time forecasting performance. MSc Thesis. VUB/KU Leuven: Brussels. vi, 59 pp.
|
| Beschikbaar in | Auteurs |
|
Documenttype: Doctoraat/Thesis/Eindwerk
|
| Author keywords |
CRR-model; Seasonally-calibrated model parameters; Seasonal variation. |
| Auteurs | | Top |
- Nsubuga, R.
- Nossent, J., revisor, meer
|
|
|
| Abstract |
CRR models are commonly calibrated with a single global parameter set, that determines the behavior of the model in a wide range of different situations, such as very long dry periods, high intensity storms after a dry period, long wet periods, etc. And although the use of a single parameter set has the advantage of being an easier and more straightforward approach, it can lead to a tradeoff in the performance of the model for these specific hydrologic circumstances and to a varying seasonal-response performance. Moreover, some of the model parameters, conceptually relate to catchment properties, such as surface storages, which can vary with changing seasonal conditions. In this research, we investigate the possibility of improving the simulations from a CRR model, by introducing seasonally-calibrated model parameters. Two seasons are defined, the wet and dry, and the CRR model used is the NAM. The study is carried out on 3 catchments located in Belgium.
The model parameters are assessed for seasonal-sensitivity, and the most seasonally-sensitive parameters are: Umax, CQOF, TIF and TG. Two methods for splitting seasons are studied:- based on (1) dates; and (2) L/Lmax; and a model is developed for each method. A model calibrated with a uniform single parameter set is also developed and used as the base-model to assess the seasonal models’ performance. A multi criteria model performance evaluation is used, which includes using goodness-of-fit statistics, hydrograph comparison and extreme value analysis.
From the results, the models with seasonally-calibrated parameters show improved simulations over the single parameter set model for at least 2 of the 3 studied catchments. All the goodness-of-fit statistics showed modest improvements. The seasonal models simulate the winter peaks with relatively reduced bias, respond significantly better to the summer peaks, and show a reduction in the water balance error. For the extreme flow analysis, the seasonal models show mostly reduced bias in simulating the extreme flows. Both the date- and L/Lmax-split seasonal models alternately perform superior to each other, and they mostly show very similar behavior on average. Overall, the seasonally-calibrated models showed an interesting possibility for increasing CRR model performance. It is especially important for catchments where a single parameter set CRR model shows performance tradeoffs during the varying seasons. |
|
