Hydrological consistency in modeling - Improving model structure and parameter estimation with temporal diagnostic analyses
Final Report Abstract
Hydrological consistency in modeling is an emerging and challenging goal to improve the understanding of how hydrological processes are represented in models and how to handle model parameters so that the corresponding processes are accurately reproduced. This project contributes to an achievement of hydrological consistency by representing several diagnostic approaches for a better spatio-temporal analysis of parameter and process understanding. For this, several diagnostic analyses were carried out with the eco-hydrological model SWAT in four contrasting German catchments. These catchments follow an increasing elevation gradient from lowlands (Treene) via Uplands (Saale, Kinzig) to alpine (Ammer) catchments. The handling of parameters in models is investigated using a temporally resolved parameter sensitivity analysis (TEDPAS). As major contribution of this project part to the hydrological community, the understanding of reasons for temporal variations in the dominance of model parameters is improved. This is realised by using different hydrological components as target variables in the sensitivitiy analysis so that each model parameter can be precisely related to its corresponding processes. Moreover, parameter sensitivities are separately analysed for different hydrological conditions under seasonal variations. This results in a clear determination of hydrological conditions leading to high parameter sensitivities. By repeating these analyses for the four catchments, the heterogeneity in process relevance is derived. Subsequently, it is shown that TEDPAS is an useful diagnostic tool to investigate how parameter sensitivities are impacted by changing climate conditions. Following on a better understanding of model parameters and their role in a hydrological model, the identification of parameter values is analysed. For this, the suitability of multiple performance measures in identifying the parameter values is investigated. Using regression trees, appropriate performance measures are determined for model parameters to show how the relationship between model parameters and performance measures vary depending on catchment conditions. Furthermore, parameter identifiability plots are used to detect the best parameter values. Hereby, appropriate parameter values can vary depending on the selection of the performance measure. To achieve a consistent handling of performance measures, a principal component analysis is used to determine an accurate number of performance measures for each catchment. The better knowledge of which performance measures are appropriate for the parameter identification is used to detect the best (set of) model runs. A multi-variable calibration approach shows that only a suitable combination of classical performance measures and signature measures leads to an overall good presentation of discharge and nitrate. The use of expert knowledge on process behaviour in the catchment can even more improve the selection of accurate parameter sets. In the last step, both achievements are combined by inserting the sensitivity time series for each model parameter into the calculation of a performance measure. In this way, days of higher sensitivities were higher weighted. This leads to a more precise parameter identification also for model parameters of lower relevance. The joint interpretation of all these diagnostic model approaches leads to a better understanding on how hydrological processes are represented in models. Based on this, different methods are provided how to handle and to identify model parameters such that the selection of optimal parameter sets is improved in the sense of an accurate process representation. All these approaches are of general applicability and are or will be published to achieve a benefit for the entire hydrological community and to provide ideas for future research.
Publications
- (2015): Detection of dominant nitrate processes in ecohydrological modelling with temporal parameter sensitivity analysis, Ecol. Model., 314, 62-72
Haas, M.; Guse, B.; Pfannerstill, M.; Fohrer, N.
(See online at https://doi.org/10.1016/j.ecolmodel.2015.07.009) - (2015): Process verification of a hydrological model using a temporal parameter sensitivity analysis, Hydrol. Earth Syst. Sci., 19, 4365-4376
Pfannerstill, M.; Guse, B.; Reusser, D.; Fohrer, N.
(See online at https://doi.org/10.5194/hess-19-4365-2015) - (2016): A joined multi-metric calibration of river discharge and nitrate loads with different performance measures, J. Hydrol., 536, 534-545
Haas, M.; Guse, B.; Pfannerstill, M.; Fohrer, N.
(See online at https://doi.org/10.1016/j.jhydrol.2016.03.001) - (2016): Demasking the integrated information of discharge: Advancing sensitivity analysis to consider different hydrological components and their rates of change, Water Resour. Res., 52, 8724-8743
Guse, B.; Pfannerstill, M.; Gafurov, A.; Fohrer, N.; Gupta, H.
(See online at https://doi.org/10.1002/2016WR018894) - (2016): On characterizing the temporal dominance patterns of model parameters and processes, Hydrol. Process., 30(13), 2255-2270
Guse, B.; Pfannerstill, M.; Strauch, M.; Reusser, D.; Lüdtke, S.; Volk, M.; Gupta, H.; Fohrer, N.
(See online at https://doi.org/10.1002/hyp.10764) - (2017): How to constrain multiobjective calibrations of the SWAT model using water balance components, J. Am. Water Resour. A., 1-15
Pfannerstill, M.; Bieger, K.; Guse, B.; Bosch, D.; Fohrer, N.; Arnold, J.G.
(See online at https://doi.org/10.1111/1752-1688.12524)
