Articles | Volume 21, issue 11
Hydrol. Earth Syst. Sci., 21, 5663–5679, 2017
https://doi.org/10.5194/hess-21-5663-2017
Hydrol. Earth Syst. Sci., 21, 5663–5679, 2017
https://doi.org/10.5194/hess-21-5663-2017

Research article 15 Nov 2017

Research article | 15 Nov 2017

Identifying the connective strength between model parameters and performance criteria

Björn Guse1,2, Matthias Pfannerstill1, Abror Gafurov2, Jens Kiesel3,1, Christian Lehr4,5, and Nicola Fohrer1 Björn Guse et al.
  • 1Christian Albrechts University of Kiel, Institute of Natural Resource Conservation, Department of Hydrology and Water Resources Management, Kiel, Germany
  • 2GFZ German Research Centre for Geosciences, Section 5.4 Hydrology, Potsdam, Germany
  • 3Leibniz Institute of Freshwater Ecology and Inland Fisheries (IGB), Berlin, Germany
  • 4Leibniz Centre for Agricultural Landscape Research (ZALF), Institute of Landscape Hydrology, Müncheberg, Germany
  • 5University of Potsdam, Institute for Earth and Environmental Sciences, Potsdam, Germany

Abstract. In hydrological models, parameters are used to represent the time-invariant characteristics of catchments and to capture different aspects of hydrological response. Hence, model parameters need to be identified based on their role in controlling the hydrological behaviour. For the identification of meaningful parameter values, multiple and complementary performance criteria are used that compare modelled and measured discharge time series. The reliability of the identification of hydrologically meaningful model parameter values depends on how distinctly a model parameter can be assigned to one of the performance criteria.

To investigate this, we introduce the new concept of connective strength between model parameters and performance criteria. The connective strength assesses the intensity in the interrelationship between model parameters and performance criteria in a bijective way. In our analysis of connective strength, model simulations are carried out based on a latin hypercube sampling. Ten performance criteria including Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE) and its three components (alpha, beta and r) as well as RSR (the ratio of the root mean square error to the standard deviation) for different segments of the flow duration curve (FDC) are calculated.

With a joint analysis of two regression tree (RT) approaches, we derive how a model parameter is connected to different performance criteria. At first, RTs are constructed using each performance criterion as the target variable to detect the most relevant model parameters for each performance criterion. Secondly, RTs are constructed using each parameter as the target variable to detect which performance criteria are impacted by changes in the values of one distinct model parameter. Based on this, appropriate performance criteria are identified for each model parameter.

In this study, a high bijective connective strength between model parameters and performance criteria is found for low- and mid-flow conditions. Moreover, the RT analyses emphasise the benefit of an individual analysis of the three components of KGE and of the FDC segments. Furthermore, the RT analyses highlight under which conditions these performance criteria provide insights into precise parameter identification. Our results show that separate performance criteria are required to identify dominant parameters on low- and mid-flow conditions, whilst the number of required performance criteria for high flows increases with increasing process complexity in the catchment. Overall, the analysis of the connective strength between model parameters and performance criteria using RTs contribute to a more realistic handling of parameters and performance criteria in hydrological modelling.

Download
Short summary
Performance measures are used to evaluate the representation of hydrological processes in parameters of hydrological models. In this study, we investigated how strongly model parameters and performance measures are connected. It was found that relationships are different for varying flow conditions, indicating that precise parameter identification requires multiple performance measures. The suggested approach contributes to a better handling of parameters in hydrological modelling.