Analysing the temporal dynamics of model performance for hydrological models

Reusser, Dominik E.; Blume, Theresa; Schaefli, Bettina; Zehe, Erwin

doi:10.5194/hess-13-999-2009

Cited by 96 publications

(80 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides the likelihood used for parameter calibration, there are many metrics for evaluating model performance of hydrological and water quality models (Jachner et al, 2007;Smith and Rose, 1995;Reusser et al, 2009;Moriasi et al, 2007). Out of those, we have selected some frequently used statistics ( Table 3) that allow for a comparison with other studies.…”

Section: Model Evaluationmentioning

confidence: 99%

Modelling biocide and herbicide concentrations in catchments of the Rhine basin

Möser

Wemyss

Scheidegger

et al. 2018

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. Impairment of water quality by organic micropollutants such as pesticides, pharmaceuticals or household chemicals is a problem in many catchments worldwide. These chemicals originate from different urban and agricultural usages and are transferred to surface waters from point or diffuse sources by a number of transport pathways. The quantification of this form of pollution in streams is challenging and especially demanding for diffuse pollution due to the high spatio-temporal concentration dynamics, which require large sampling and analytical efforts to obtain representative data on the actual water quality.Models can also be used to predict to what degree streams are affected by these pollutants. However, spatially distributed modelling of water quality is challenging for a number of reasons. Key issues are the lack of such models that incorporate both urban and agricultural sources of organic micropollutants, the large number of parameters to be estimated for many available water quality models, and the difficulty to transfer parameter estimates from calibration sites to areas where predictions are needed.To overcome these difficulties, we used the parsimonious iWaQa model that simulates herbicide transport from agricultural fields and diffuse biocide losses from urban areas (mainly façades and roof materials) and tested its predictive capabilities in the Rhine River basin. The model only requires between one and eight global model parameters per compound that need to be calibrated. Most of the data requirements relate to spatially distributed land use and comprehensive time series of precipitation, air temperature and spatial data on discharge. For larger catchments, routing was explicitly considered by coupling the iWaQa to the AQUASIM model.The model was calibrated with datasets from three different small catchments (0.5-24.6 km 2 ) for three agricultural herbicides (isoproturon, S-metolachlor, terbuthylazine) and two urban biocides (carbendazim, diuron). Subsequently, it was validated for herbicides and biocides in Switzerland for different years on 12 catchments of much larger size (31-35 899 km 2 ) and for herbicides for the entire Rhine basin upstream of the Dutch-German border (160 000 km 2 ) without any modification. For most compound-catchment combinations, the model predictions revealed a satisfactory correlation (median r 2 : 0.5) with the observations. The peak concentrations were mostly predicted within a factor of 2 to 4 (median: 2.1 fold difference for herbicides and 3.2 for biocides respectively). The seasonality of the peak concentration was also well simulated; the predictions of the actual timing of peak concentrations, however, was generally poor.Limited spatio-temporal data, first on the use of the selected pesticides and second on their concentrations in the river network, restrict the possibilities to scrutinize model performance. Nevertheless, the results strongly suggest that input data and model structure are major sources of predictive uncertainty. The latter is for exampl...

show abstract

Section: Model Evaluationmentioning

confidence: 99%

Modelling biocide and herbicide concentrations in catchments of the Rhine basin

Möser

Wemyss

Scheidegger

et al. 2018

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

show abstract

“…Visual hydrograph inspection continues to be the most widely used technique for hydrologic model evaluation (Seibert et al, 20 2016;Reusser et al, 2009;Ehret and Zehe, 2011). Dividing the hydrograph into distinct time periods and adjusting parameters by trial and error is still commonly done during model calibration (Singh and Frevert, 2002).…”

Section: Meaningful Measures For Evaluating Model Performance 10mentioning

confidence: 99%

“…Automated and intervention type calibration in hydrology generally rely on a lumped model performance metric such as NSE (a normalized measure ranging between -inf to 1.0) or R 2 (Krause et al, 2005;Reusser et al, 2009). Such metrics are useful insofar as they are objective measures that can be used to quantify model behavior and can be readily and quantitatively compared among models (Krause et al, 2005).…”

Section: Lumped Metricsmentioning

confidence: 99%

“…Studies have highlighted the need for other measures in a multi-objective 5 criteria approach for a robust evaluation of model performance Singh and Frevert, 2002). The use of a single measure serves well in an automatic calibration context, but does not necessarily capture the most relevant aspects of model performance (Reusser et al, 2009). For example, the quadratic formulation of NSE or R 2 emphasizes higher magnitude flows, relative to lower flows, the latter of which may be critical for many ecological and water quality applications (Criss and Winston, 2008).…”

Section: Lumped Metricsmentioning

confidence: 99%

See 1 more Smart Citation

Multiple domain evaluation of watershed hydrology models

Kumarasamy

Belmont

2017

Preprint

View full text Add to dashboard Cite

Abstract. Watershed scale models simulating hydrology and water quality have advanced rapidly in sophistication, process 5 representation, flexibility in model structure, and input data. Given the importance of these models to support decision-making for a wide range of environmental issues, the hydrology community is compelled to improve the metrics used to evaluate model performance. More targeted and comprehensive metrics will facilitate better and more efficient calibration and will help demonstrate that the model is useful for the intended purpose. Here we introduce a suite of new tools for model evaluation,packaged as an open-source Hydrologic Model Evaluation (HydroME) Toolbox. Specifically, we demonstrate the use of box 10 plots to illustrate the full distribution of common model performance metrics, such as R 2 , use of Euclidian distance, empirical Quantile-Quantile (Q-Q) plots and flow duration curves as simple metrics to identify and localize errors in model simulations.Further, we demonstrate the use of magnitude squared coherence to compare the frequency content between observed and modeled streamflow and wavelet coherence to localize frequency mismatches in time. We provide a rationale for a hierarchical selection of parameters to adjust during calibration and recommend that modelers progress from parameters with the most 15 uncertainty to the least uncertainty, namely starting with pure calibration parameters, followed by derived parameters, and finally measured parameters. We apply these techniques in the calibration and evaluation of models of two watersheds, the Le Sueur River Basin (2880 km 2 ) and Root River Basin (4300 km 2 ) in southern Minnesota, USA.

show abstract

“…Because of its internal structure,similarly to other complete hydrological models, it is possible to map different processes (Reusser et al, 2009) and therefore different parts of the hydrograph, onto the parameters, so that different parts of the hydrograph time series can be used separately to better identify model parameter values. Further analysis is needed to show sensitivity to spatial and temporal resolution.…”

Section: Influence Of the Calibration Parameters On Simulationsmentioning

confidence: 99%

Exploiting remote sensing land surface temperature in distributed hydrological modelling: the example of the Continuum model

Silvestro

Gabellani

Delogu

et al. 2013

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. Full process description and distributed hydrological models are very useful tools in hydrology as they can be applied in different contexts and for a wide range of aims such as flood and drought forecasting, water management, and prediction of impact on the hydrologic cycle due to natural and human-induced changes. Since they must mimic a variety of physical processes, they can be very complex and with a high degree of parameterization. This complexity can be increased by necessity of augmenting the number of observable state variables in order to improve model validation or to allow data assimilation.In this work a model, aiming at balancing the need to reproduce the physical processes with the practical goal of avoiding over-parameterization, is presented. The model is designed to be implemented in different contexts with a special focus on data-scarce environments, e.g. with no streamflow data.All the main hydrological phenomena are modelled in a distributed way. Mass and energy balance are solved explicitly. Land surface temperature (LST), which is particularly suited to being extensively observed and assimilated, is an explicit state variable.A performance evaluation, based on both traditional and satellite derived data, is presented with a specific reference to the application in an Italian catchment. The model has been firstly calibrated and validated following a standard approach based on streamflow data. The capability of the model in reproducing both the streamflow measurements and the land surface temperature from satellites has been investigated.The model has been then calibrated using satellite data and geomorphologic characteristics of the basin in order to test its application on a basin where standard hydrologic observations (e.g. streamflow data) are not available. The results have been compared with those obtained by the standard calibration strategy based on streamflow data.

show abstract

Analysing the temporal dynamics of model performance for hydrological models

Cited by 96 publications

References 41 publications

Modelling biocide and herbicide concentrations in catchments of the Rhine basin

Modelling biocide and herbicide concentrations in catchments of the Rhine basin

Multiple domain evaluation of watershed hydrology models

Exploiting remote sensing land surface temperature in distributed hydrological modelling: the example of the Continuum model

Contact Info

Product

Resources

About