Final Activity Report Summary - MODELLING COMPETENCE (Improving competence in integrated hydrological modelling by gaining innovative uncertainty analysis skills and software engineering capabilities)
Model identification, parameter estimation and related uncertainties have become topics of interest in hydrological modelling in the last years. In the MODELLING COMPETENCE project new statistical uncertainty analysis methodologies were developed to investigate model uncertainties for different hydrological and water quality models in river catchments with different site characteristics.
Both generalised likelihood uncertainty estimation (GLUE) and Monte Carlo Markov chain (MCMC) techniques were applied to assess uncertainty of the so called 'Topmodel'. A numerical Bayesian multi response calibration approach was developed to include information on discharge, silica and calcium stream water concentrations simultaneously. Furthermore, the Bayesian approach was used to compute the uncertainty in input forcing data and model parameters simultaneously. Spatial discretization, particularly regarding the dependence of the performance of Topmodel on the grid size, was thoroughly investigated. The results showed that the proposed MCMC methodology could provide additional insights into the model behaviour.
The main difference between GLUE and MCMC consisted in the choice of the likelihood function. In case the same likelihood was used by both methods the results should be very similar. The multi response approach allowed reducing uncertainty of the estimated parameters and contributed towards an improved understanding of the role of the internal variables. In addition, the results showed that the proposed methodology was a valuable tool to assess different sources of uncertainty in hydrological modelling and also demonstrated the effects of uncertainty in the input forcing when a fully distributed physically based hydrological model was used.
Moreover, the suggested methodology proved to be very useful in selecting an appropriate spatial discretization, e.g. the grid size, of the selected hydrological model. In the scope of the research nitrogen turnover was simulated with the WASP5 river water quality model. Uncertainty analysis was carried out using a Monte Carlo analysis including all 39 parameters of the submodel EUTRO. Climate scenarios were used to characterise changing flow and climatic conditions. Under low flow conditions denitrification rate was about 50 % higher in the 2050 to 2054 period compared to the reference year 2000.
Overall, the results of the study revealed significance of climate change in regulating the magnitude, seasonal pattern and variability of the nitrogen retention. The results of the study provided improved understanding of seasonal and spatial changes of nitrogen retention, which was an important sink of nitrogen in riverine systems. This was a rewarding subject because, by the time of the project completion, it was still unclear how denitrification was influenced by climate induced changes.
Both generalised likelihood uncertainty estimation (GLUE) and Monte Carlo Markov chain (MCMC) techniques were applied to assess uncertainty of the so called 'Topmodel'. A numerical Bayesian multi response calibration approach was developed to include information on discharge, silica and calcium stream water concentrations simultaneously. Furthermore, the Bayesian approach was used to compute the uncertainty in input forcing data and model parameters simultaneously. Spatial discretization, particularly regarding the dependence of the performance of Topmodel on the grid size, was thoroughly investigated. The results showed that the proposed MCMC methodology could provide additional insights into the model behaviour.
The main difference between GLUE and MCMC consisted in the choice of the likelihood function. In case the same likelihood was used by both methods the results should be very similar. The multi response approach allowed reducing uncertainty of the estimated parameters and contributed towards an improved understanding of the role of the internal variables. In addition, the results showed that the proposed methodology was a valuable tool to assess different sources of uncertainty in hydrological modelling and also demonstrated the effects of uncertainty in the input forcing when a fully distributed physically based hydrological model was used.
Moreover, the suggested methodology proved to be very useful in selecting an appropriate spatial discretization, e.g. the grid size, of the selected hydrological model. In the scope of the research nitrogen turnover was simulated with the WASP5 river water quality model. Uncertainty analysis was carried out using a Monte Carlo analysis including all 39 parameters of the submodel EUTRO. Climate scenarios were used to characterise changing flow and climatic conditions. Under low flow conditions denitrification rate was about 50 % higher in the 2050 to 2054 period compared to the reference year 2000.
Overall, the results of the study revealed significance of climate change in regulating the magnitude, seasonal pattern and variability of the nitrogen retention. The results of the study provided improved understanding of seasonal and spatial changes of nitrogen retention, which was an important sink of nitrogen in riverine systems. This was a rewarding subject because, by the time of the project completion, it was still unclear how denitrification was influenced by climate induced changes.