Towards a new generation of high-resolution meteorological input data for small-scale hydrologic modeling

https://doi.org/10.1016/j.jhydrol.2011.03.026Get rights and content

Summary

Current and future challenges of hydrologic sciences are to accurately predict and assess climate-driven impacts on water resources for the relevant scales of planning. However, process-based small-scale hydrologic modeling is data demanding and large uncertainties exist in data-sparse areas.

The aim of our study was to test the applicability of the COSMO-DE analysis data (COSMO-DE-A) for hydrologic modeling. COSMO-DE-A data are a new meteorological data set with high temporal and spatial resolution that originates from the German Weather Service data assimilation system using the COSMO-DE weather prediction model. We collected field parameters in a small (10 km2) mountainous catchment in the Upper Middle Rhine Valley (west Germany) to parameterize the static boundary conditions of the hydrologic model CATFLOW. We evaluated error components of hourly COSMO-DE-A fields in comparison to interpolated hourly meteorological station data (i.e. reference data), applied two bias-correction methods for precipitation (i.e. linear correction method and quantile–quantile mapping technique), calibrated (in a 36 ha large subcatchment) and tested (in another 48 ha large subcatchment) the CATFLOW model using the reference data for input, and compared stream flow predictions using uncorrected and bias-corrected COSMO-DE-A data for input. Moreover, we compared soil moisture and latent heat flux from COSMO-DE-A with values simulated by CATFLOW.

Relative error between COSMO-DE-A precipitation and interpolated precipitation is ca. 50%. Other climatic variables from COSMO-DE-A are almost unbiased, though errors for global radiation and temperature are autocorrelated. Nash and Sutcliffe efficiencies accounted for 0.83 and 0.33 for the simulated vs. observed stream flow of the calibration and test catchment, respectively. The use of uncorrected COSMO-DE-A precipitation leads to poor performances of the hydrologic model. In contrast, if either of both bias-correction methods is applied to COSMO-DE-A precipitation, predicted hydrographs and soil moisture are almost the same as if interpolated reference data is used. Soil moisture and latent heat flux are simulated consistently by the independent models COSMO-DE-A and CATFLOW.

We conclude that COSMO-DE-A data are suitable for hydrologic modeling of longer periods (e.g. seasons), provided that bias correction of precipitation is done prior to model application. Further research is required to test the regional and temporal stability of bias-correction terms.

Highlights

► Meteorological analysis data (COSMO-DE-A) was used in a catchment model (CATFLOW). ► Bias-corrected COSMO-DE-A precipitation can be used for hydrologic modeling. ► Correction of other COSMO-DE-A parameters (G, T, u, and rH) is not necessary. ► COSMO-DE-A and CATFLOW simulate soil moisture and latent heat flux consistently. ► Land-use × hillslope classes allow regionalization of soil physical parameters.

Introduction

Process-based small-scale hydrologic modeling is a data-demanding task. The ongoing IAHS decade on predictions in ungauged basins calls for innovations to reduce predictive uncertainty (Sivapalan et al., 2003) in data-sparse areas. Climatic input data from either meteorological station measurements or numerical models were identified as one of three main components that drive a hydrologic prediction system and thus cause uncertainty because consistent meteorological station measurements are not always available in the immediate vicinity of a catchment of interest (Sivapalan et al., 2003). Usually, the spatial resolution of climate models is too coarse for hydrologic modeling (Prudhomme et al., 2002).

The call of the Intergovernmental Panel on Climate Change (IPCC) for climate impact studies at the regional scale has stimulated researchers to develop both downscaling methods and bias-correction methods to account for spatial scale effects and systematic errors that are included in climate model outputs (e.g. Krahe et al., 2009, Görgen et al., 2010). However, bias-correction methods were developed and applied for the correction of long-term averages (often climatic normals) within climate change studies. The application and consequent analysis of bias-correction methods to hourly records for hydrologic applications are not reported, hitherto. Recently, a new meteorological data set for Germany with high spatial and temporal resolution has become available, which gives the opportunity to test bias-correction methods. The data set is provided by the German Weather Service data assimilation system using the numerical COSMO-DE weather forecast model (Doms and Schättler, 2002, Steppeler et al., 2003) and is referred to as COSMO-DE analysis (COSMO-DE-A).

Another component causing predictive uncertainty that was specified by Sivapalan et al. (2003) are parameters representing landscape properties used for parameterization of hydrologic models. The fact that spatially distributed parameters, such as soil physical properties, can only be measured at specific points in space results in the necessity to regionalize these parameters. Therefore, factors that influence these parameters have to be identified and an appropriate regionalization method has to be chosen (Kleeberg, 1999). In former studies a broad range of methods, including geostatistical and classification approaches, have been used for soil mapping as described by McBratney et al., 2000, McBratney et al., 2003, Scull et al., 2003. In many of these studies soil properties were spatially predicted using topographical data processed in geographical information systems (GIS) (e.g. Herbst et al., 2006, Hofmann et al., 2009, Mertens et al., 2002).

Physically based models and conceptual models are two major types of hydrologic models, which also vary in their spatial resolution. As past observations for model calibration are scarce in ungauged catchments, a sound theoretical basis of the applied model is indispensable for predictions in ungauged basins, which makes conceptual models inappropriate for this kind of application (Lee et al., 2007). In this study we therefore use the physically based, distributed model CATFLOW (Maurer, 1997, Zehe et al., 2001), which was developed at a rural catchment with temperate climate and has been applied successfully for simulations of the water balance in several studies (Graeff et al., 2009, Lee et al., 2007, Lindenmaier et al., 2005, Reusser et al., 2009, Zehe et al., 2001, Zehe et al., 2005) of which most were conducted at catchments with temperate climate and rural land-use, similar to the catchment investigated in this study. Moreover, CATFLOW gives the opportunity to check, if latent heat flux and soil water content are simulated consistently in the hydrologic model (CATFLOW) and the meteorological model (COSMO-DE-A).

The aims of our study are (i) to test the accuracy of the new COSMO-DE-A data by comparison with interpolated station data and to apply and test bias-correction functions to account for systematic errors in COSMO-DE-A fields, (ii) to determine and regionalize soil hydraulic properties in a data-sparse area for parameterizing the hydrologic model, and (iii) to evaluate the hydrologic response (i.e. stream flow, soil moisture and latent heat flux) of a hydrologic prediction system that is driven by COSMO-DE-A fields and operated in a small ungauged catchment at various temporal scales. This can serve as a basis for future simulations of the hydrologic response of small, ungauged catchments.

In this study, we evaluate for the first time a new meteorological data set with high spatial and temporal resolution generated by an operationally applied numerical weather forecast model in which meteorological observations are assimilated (COSMO-DE-A) for its suitability for different time scales of hydrologic modeling with the process-oriented CATFLOW model. An extensive field measurement campaign was done during this study to collect soil physical properties and was accompanied by using a factorial model to regionalize the point measurements in space in order to parameterize the CATFLOW model.

Section snippets

Bedrock and relief

The Gailsbach catchment (10.3 km2) is located in the Upper Middle Rhine Valley, west Germany. It is part of the Variscan Rhenish Massif that mainly consists of Lower Devonian sedimentary rocks (Meyer and Stets, 1996, Landesamt für Geologie, 2005). The Rhenish Massif was peneplained under warm and humid climates during the Cretaceous and Paleogene and the silica-rich schist was the basis to form a strongly weathered mantle consisting of saprolite (Felix-Henningsen, 1990). The Miocene landscape

Hourly meteorological station data

Hourly station data of precipitation (P), global radiation (G), temperature (T), relative humidity (rH), and wind speed (u) (observation height: 2 m above ground) for the period 2004/01/01 to 2008/11/30 were obtained from the German Weather Service (DWD) and Agrarmeteorologie Rheinland-Pfalz (AM-RLP). Both institutions maintain automatic weather stations at Bacharach, Rüdesheim-Presberg, and Wahlbach, which are located 3.5 km, 9.5 km and 11.5 km away from the center of the Gailsbach catchment.

Comparison of observed and simulated meteorological data

The residuals of hourly global radiation, temperature, and precipitation between COSMO-DE-A and REF-data are analyzed with respect to their distribution and their autocorrelation (Fig. 3). Residuals for temperature range from −8.4 K to 4.5 K, whereas the mean temperature of COSMO-DE-A is ca. 1 K cooler as compared to the REF-data (Table 7). The range of residuals for global radiation is relatively small (Fig. 3A), compared to the magnitude of mean daily global radiation.

There is no significant

Conclusion

In this paper we assessed for the first time a newly produced hourly meteorological data set that originates from an operationally applied weather forecast model with respect to its applicability for small-scale hydrologic modeling.

We first analyzed error components of hourly COSMO-DE-A fields with respect to spatially interpolated data from meteorological stations. Uncorrected COSMO-DE-A precipitation is significantly higher (rE = 50%) than reference precipitation. Moderate differences are

Acknowledgments

This study was jointly granted by the Forschungsfonds 2008 and the Earth System Science Research Center of the Johannes Gutenberg University Mainz, Germany. We are grateful to Erwin Zehe (TU München, Germany) for providing the code of CATFLOW and to Theresa Blume (GFZ German Research Centre for Geosciences, Potsdam, Germany) for her valuable support. Thanks go to the German Weather Service (DWD) for providing the COSMO-DE analysis data. We thank Heini Wernli (ETH Zurich, Switzerland) and Erwin

References (44)

  • A. Amoozegar

    A compact constant-head permeameter for measuring saturated hydraulic conductivity of the vadose zone

    Soil Sci. Soc. Am. J.

    (1989)
  • J. Boé et al.

    Projected changes in components of the hydrological cycle in French river basins during the 21st century

    Water Resour. Res.

    (2009)
  • G. Doms et al.

    A Description of the Nonhydrostatic Regional Model LM. Part I: Dynamics and Numerics

    (2002)
  • H. Ellenberg

    Vegetation Mitteleuropas mit den Alpen

    (1996)
  • P. Felix-Henningsen

    Die mesozoisch-tertiäre Verwitterungsdecke (MTV) im Rheinischen Schiefergebirge

    (1990)
  • Görgen, K., Beersma, J., Brahmer, G., Buiteveld, H., Carambia, M., de Keizer, O., Krahe, P., Nilson, E., Lammersen, R.,...
  • T. Graeff et al.

    Process identification through rejection of model structures in a mid-mountainous rural catchment: observations of rainfall–runoff response, geophysical conditions and model inter-comparison

    Hydrol. Process.

    (2009)
  • M. Hofmann et al.

    Regionalizing soil properties in a catchment of the Bavarian Alps

    Eur. J. Forest Res.

    (2009)
  • M. Iqbal

    An Introduction to Solar Radiation

    (1983)
  • IUSS Working Group WRB

    World Reference Base for Soil Resources

    (2006)
  • D. Jacob et al.

    An inter-comparison of regional climate models for Europe: model performance in present-day climate

    Clim. Change

    (2007)
  • H.-B. Kleeberg

    Zum Problem der Regionalisierung in der Hydrologie

  • View full text