Elsevier

Journal of Hydrology

Volume 511, 16 April 2014, Pages 335-349
Journal of Hydrology

Intercomparison of five lumped and distributed models for catchment runoff and extreme flow simulation

https://doi.org/10.1016/j.jhydrol.2014.01.050Get rights and content

Highlights

  • Application of lumped and distributed hydrological model ensemble.

  • Multi-objective calibration of the models.

  • Calibration includes focus on flow extremes and runoff subflows.

  • Assessment of the models' predictive capacities under high rainfall intensities.

Summary

Five hydrological models with different spatial resolutions and process descriptions were applied to a medium sized catchment in Belgium in order to assess the accuracy and differences of simulated hydrological variables, including peak and low flow extremes and quick and slow runoff subflows. The models varied from the lumped conceptual NAM, PDM and VHM models over the intermediate detailed and distributed WetSpa model to the highly detailed and fully distributed MIKE SHE model. The latter model accounts for the 3D groundwater processes and interacts bi-directionally with a full hydrodynamic MIKE 11 river model. A consistent protocol to model calibration was applied to all models. This protocol uses information on the response behavior of the catchment extracted from the river flow and input time series and explicitly focuses on reproducing the quick and slow runoff subflows, and the extreme high and low flows next to testing the conventional model performance statistics. Also the model predictive capacity under high rainfall intensities, which might become more extreme under future climate change was explicitly verified for the models. The tail behavior of the extreme flow distributions was graphically evaluated as well as the changes in runoff coefficients in relation to the changing rainfall intensities.

After such calibration, all tested models succeed to produce high performance for the total runoff and quick and slow runoff subflow dynamics and volumes, peak and low flow extremes and their frequency distributions. Calibration of the lumped parameter models is much less time consuming and produced higher overall model performance in comparison to the more complex distributed models.

Introduction

Hydrological models are widely applied in water engineering for design and scenario impact investigations. Depending on the type of application, the catchment characteristics and the data availability, different spatial and temporal scales, different model conceptualizations and parameterizations are considered. In some cases, the most appropriate model is selected based on these criteria. However, in many applications the model selection seems subject to the common practice of the modeller. Rarely an objective model selection seems conducted (Najafi et al., 2011). Moreover, hydrological studies are often based on one particular hydrological model. The selected model structure might, however, strongly affect the study results, as was shown before by Breuer et al., 2009, Viney et al., 2009, Huisman et al., 2009, Ludwig et al., 2009, Maurer et al., 2010, Bae et al., 2011, Gosling et al., 2011, Smith et al., 2012, Van Steenbergen and Willems, 2012, Velázquez et al., 2012, among others. However, these studies did not draw much attention to the performance of the models under extreme conditions. The calibration was based on statistics evaluating the overall runoff performance, whereas it is known that this does not necessarily lead to good model performance for high and low flow extremes (Westerberg et al., 2011). It is more appropriate to consider multiple objectives that focus on the different aspects of the fit between simulated and observed discharges. Freer et al. (1996) used several performance measures in their Generalised Likelihood Uncertainty Estimation (GLUE) framework. Boyle et al., 2000, Madsen, 2000, Yu and Yang, 2000, Wagener et al., 2001, Wagener et al., 2003, Ferket et al., 2010, Zhang et al., 2011 applied performance measures on the subflow components of the runoff discharges or on periods covering different catchment response modes, e.g. wet periods, draining periods, dry periods; or high and low flows above or below a threshold. Westerberg et al. (2011) developed a calibration method including flow-duration curves. Model calibration based on multiple objectives is qualitatively more balanced but does not necessarily statistically perform the best (Westerberg et al., 2011, Willems, 2014). This might raise concern that uncertainty in the impact predictions is additionally induced by the calibration of the models.

Within this paper the influence of the model structure on the model performance for catchment runoff, including high and low flow conditions, is investigated by an ensemble of five hydrological models with different spatial resolutions and process descriptions. In order to obtain consistent and reliable models for use in water engineering (design) applications or scenario-based impact assessment, all models are consistently calibrated by a given systematic but time demanding calibration protocol. The protocol relies on information of runoff subflows and various types of runoff responses derived from the observed river flow, rainfall and potential evapotranspiration (ETo) time series. Explicit focus is given to the high and low flow extremes. It is analyzed whether the models produce reliable estimates of the flow regimes under the current climate and how well they simulate the changes in quick runoff coefficient under changing rainfall intensities. To cover a wide set of model complexities, the selected models in this study vary from the lumped conceptual models NAM, PDM and VHM, over the intermediate detailed and distributed model WetSpa, to the highly detailed and fully distributed model MIKE-SHE. The latter model simulates next to the catchment runoff also internal discharges and groundwater heads.

The Grote Nete catchment in Belgium is taken as case study. It is recognized that next to testing different model structures also different catchments with different meteorological and hydrological characteristics should be studied. Practical barriers, however, prevented us from repeating the approach on other catchments. High quality data and good knowledge on the case study processes and particularities are indeed required to make exhaustive studies on model structures behavior.

Section snippets

Study area

The Grote Nete catchment is located in the northeast of Belgium, with an area of 385 km2 at the outlet limnigraphic station of Geel-Zammel (Fig. 1). The long term mean annual precipitation in the catchment ranges from about 600 to 1100 mm with an areal average of 828 mm based on the years 2002–2008. The precipitation is almost equally distributed during the winter and summer periods. The long term average annual ETo is about 670 mm. The topography is flat, ranging from 12 m in the west to 69 m in the

Presentation of the calibration protocol

Given that this study explicitly focuses on the high and low flow extremes, the model calibration explicitly focused on these extremes rather than only on the simulation of the general runoff characteristics and response modes of the hydrographs. This requires multiple objectives to be considered. Multi-objective calibration strategies have been investigated and automated for conceptual rainfall–runoff models (Gupta et al., 1998, Madsen, 2000, Madsen, 2003, Wagener et al., 2001, Madsen et al.,

Results

The five hydrological models in this study were applied under the same meteorological and catchment conditions. They were run at an hourly time step, using the same precipitation and ETo inputs, and were calibrated against the same measured river flow data at the catchment outlet. The calibration data cover the period from September 2002 to the end of 2005 which covers a wide range of meteorological and hydrological conditions, including a very wet winter with several high peak flows and

Conclusions

Although the structure complexity (spatial resolution and process details) play a role on the model efficiency, all five models appear to be fairly adequate for the catchment. The overall runoff processes and streamflow dynamics are well captured by the models. Winter events are generally better estimated than the summer ones, and the lumped models perform better than the distributed ones. Also in terms of overall water balance and the subflows the lumped models are slightly superior with very

Acknowledgement

This research was supported by Flanders Hydraulics Research, Ministry of Mobility and Public Works, Flanders, Belgium.

References (60)

  • H. Madsen et al.

    Comparison of different automated strategies for calibration of rainfall–runoff models

    J. Hydrol.

    (2002)
  • M.K. Muleta et al.

    Sensitivity and uncertainty analysis coupled with automatic calibration for a distributed watershed model

    J. Hydrol.

    (2005)
  • J.E. Nash et al.

    River flow forecasting through conceptual models, Part I – A discussion of principles

    J. Hydrol.

    (1970)
  • S. Reed et al.

    Overall distributed model intercomparison project results

    J. Hydrol.

    (2004)
  • M.B. Smith et al.

    Results of the DMIP 2 Oklahoma experiments

    J. Hydrol.

    (2012)
  • M.T. Taye et al.

    Identifying sources of temporal variability in hydrological extremes of the upper Blue Nile basin

    J. Hydrol.

    (2013)
  • N.R. Viney et al.

    Assessing the impact of land use change on hydrology by ensemble modelling (LUCHEM) II: Ensemble combinations and predictions

    Adv. Water Resour.

    (2009)
  • Z.-M. Wang et al.

    A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa)

    Phys. Chem. Earth

    (1996)
  • P. Willems

    A time series tool to support the multi-criteria performance evaluation of rainfall–runoff models

    Environ. Model. Softw.

    (2009)
  • P. Willems

    Parsimonious rainfall–runoff model construction supported by time series processing and validation of hydrological extremes – Part 1: Step-wise model-structure identification and calibration approach.

    J. Hydrol.

    (2014)
  • P. Willems et al.

    Bias correction in hydrologic GPD based extreme value analysis by means of a slowly varying function

    J. Hydrol.

    (2007)
  • P. Willems et al.

    Parsimonious rainfall–runoff model construction supported by time series processing and validation of hydrological extremes – Part 2: Intercomparison of models and calibration approaches

    J. Hydrol.

    (2014)
  • P.S. Yu et al.

    Fuzzy multi-objective function for rainfall–runoff model calibration

    J. Hydrol.

    (2000)
  • H. Zhang et al.

    Multi-period calibration of a semi-distributed hydrological model based on hydroclimatic clustering

    Adv. Water Resour.

    (2011)
  • N.K. Ajami et al.

    Calibration of a semi-distributed hydrologic model for streamflow estimation along a river system

    J. Hydrol.

    (2004)
  • Apip et al.

    Spatial lumping of a distributed rainfall sediment-runoff model and its effective lumping scale

    Hydrol. Process.

    (2012)
  • P. Baguis et al.

    Climate change scenarios for precipitation and potential evapotranspiration over central Belgium

    Theor. Appl. Climatol.

    (2009)
  • A. Bauwens et al.

    Hydrological response to climate change in the Lesse and the Vesdre catchments: contribution of a physically based model (Wallonia, Belgium)

    Hydrol. Earth Syst. Sci.

    (2011)
  • G.E.P. Box et al.

    An analysis of transformations

    J. Roy. Stat. Soc.

    (1964)
  • D.P. Boyle et al.

    Toward improved calibration of hydrologic models: combining the strengths of manual and automatic methods

    Water Resour. Res.

    (2000)
  • Cited by (80)

    • Simulating hydrologic pathway contributions in fluvial and karst settings: An evaluation of conceptual, physically-based, and deep learning modeling approaches

      2022, Journal of Hydrology X
      Citation Excerpt :

      Examples of intermediate pathways include interflow and unsaturated flow in the vadose zone. Numerous methods exist for simulating the varying contribution of quick, intermediate, and slow flow to overall discharge, ranging from recursive digital filters (RDF) and recession curves for data-driven hydrograph separation (Singh and Stenger, 2018; Stoelzle et al., 2020) to conceptual or explicit numerical modeling (Al Aamery et al., 2021; Vansteenkiste et al., 2014). Few studies, however, have assessed the uncertainty associated with the choice of method for pathway modeling, which has implications for understanding the timing, extent, and connectivity of catchment fluxes.

    • Climate change impact on salinization of drinking water inlets along the Campine Canals, Belgium

      2022, Journal of Hydrology: Regional Studies
      Citation Excerpt :

      This hydrological model is an implementation of the conceptual rainfall-runoff model NAM and was calibrated to daily Meuse river discharge data at Monsin. This was done in the framework of other impact studies in the Meuse basin (Vansteenkiste et al., 2009), following the model calibration methodology described in Vansteenkiste et al. (2014). This involves model parameter optimization based on simulated versus observed peak flows, low flows, and cumulative volumes.

    View all citing articles on Scopus
    View full text