An improved methodology for predicting the daily hydrologic response of ungauged catchments

https://doi.org/10.1016/S1364-8152(98)00044-9Get rights and content

Abstract

In order to model fluxes of water from the land surface to the atmosphere, and from one grid cell to another in climate models, predictions of hydrologic response are required for catchments where hydrologic data are not available. A methodology has been presented previously that has the capability of producing estimates of catchment scale hydrologic response for ungauged catchments on a daily timestep (Post and Jakeman, 1998, Ecol. Mod. submitted). In the present paper, it is demonstrated that these daily predictions of hydrologic response can be improved by incorporating information about the hydrologic response of the catchment on a longer timestep. This is because the influence of large scale phenomena such as climate and vegetation may produce a similar water yield in nearby catchments, even though their daily hydrologic response may be different, due for example, to differences in drainage density. Thus, the water yield of an ungauged catchment is inferred on an inter-annual timestep, and this information is used to balance the water budget of a daily timestep rainfall-runoff model. It was found that using tree stocking densities to predict water yields for small experimental catchments in the Maroondah region of Victoria produced better results than those obtained by inferring the water balance parameter of a daily timestep rainfall-runoff model from channel gradient and catchment elongation. Good predictions of inter-annual water yield were also obtained for small experimental catchments in the H. J. Andrews, Hubbard Brook, and Coweeta long term ecological research (LTER) sites in the United States, indicating that it may be possible to produce high quality predictions of daily hydrologic response for ungauged catchments in these regions also.

Introduction

Predictions of the hydrologic response of ungauged catchments over a range of timescales may be required for a number of reasons. For example, construction of a dam may require a prediction to be made of the annual water yield of a river at an ungauged point. Conversely, for an examination of the ecological health of a river system, the frequency and duration of low flows may be of greater importance, requiring hydrologic response data on a daily or even sub-daily timestep. Similarly, the hydrologic component of a climate model may require predictions of catchment scale hydrologic response on a daily timestep. If the catchment being examined is not gauged for streamflow, these estimates must be derived using regionalisation techniques, where the catchment under consideration is assumed to behave similarly to other catchments with similar climatologies and landscape attributes. Defining `similarity' then is something of a key issue. Previous attempts to define similarity have been based on geographical proximity (Mosley, 1981; Hughes, 1987). However, other studies have shown that geographic proximity alone is insufficient and catchment attributes have also been used (Acreman and Sinclair, 1986; Nathan and McMahon, 1990; Zrinji and Burn, 1994). In this study, the catchments under consideration are considered to be sufficiently similar due to their geographic proximity,as well as the similarity of their landscape attributes and hydrologic response.

In general, previous regionalisation studies which have made use of a modelling methodology have met with limited success. Major reasons for this include: (1) the lack of high quality rainfall-runoff and landscape attribute data; (2) the lack of a theoretical framework for defining hydrologically homogeneous regions; (3) uncertainties involved in representing the hydrologic response of a catchment using a rainfall-runoff model; and (4) problems in relating this hydrologic response to landscape attributes using multiple linear regression equations. The third and fourth problems arise primarily due to rainfall-runoff model deficiencies. One major deficiency is that models are typically too complex or overparameterised, meaning that their parameters cannot be estimated unambiguously, and do not have a physical interpretation. Examples include the Stanford Watershed Model (James, 1972); the 18-parameter Sacramento model (Weeks and Ashkanasy, 1985); the 20-parameter HBV3-ETH model (Braun and Renner, 1992); the 19-parameter MODHYDROLOG model (Chiew and McMahon, 1994); and TOPMODEL (Franchini et al., 1996). The use of simpler models offsets some of these problems. However, these models generally do not define the total hydrologic response of the catchment, but rather one aspect, such as low flows (Chang and Boyer, 1977; Gustard et al., 1992; Nathan and McMahon, 1992), or flood frequencies (Patton and Baker, 1976; Reimers, 1990).

It appears that the ideal model for use in regionalisation studies would be one which accounts for the key hydrologic processes occurring in a catchment, but which avoids the problems of overparameterisation. An example of such a model may be a hybrid metric-conceptual model (Wheater et al., 1993). Here, the structure of the model is inferred from the rainfall-runoff data, but is interpretable as a system of storages and flows, which may be considered to represent storages and flows within the catchment. An example of such a model is the IHACRES model (Jakeman et al., 1990; Jakeman and Hornberger, 1993).

Post and Jakeman (1998)presented a regionalisation approach which allowed predictions to be made of the hydrologic response of ungauged catchments in the Maroondah region of Victoria, Australia, based on relationships between the IHACRES model parameters and landscape attributes such as drainage density, catchment slope, catchment area, and channel gradient. In this paper, an extension of this approach will be presented, where the IHACRES model is used to determine the shape of the daily streamflow hydrograph, but the total volume of streamflow is determined by comparing the inter-annual water yield of the catchment to nearby catchments. Thus, water yield information is used to constrain the IHACRES model and improve its predictive capability.

The major problem in such an approach is that high quality hydrologic and climatologic data are required for a number of `similar' catchments before predictions can be made of the hydrologic response of an ungauged catchment in a region. In order to solve this problem, a study is currently being carried out which attempts to identify the underlying controls on hydrologic response across a range of regions. If these controls are similar or vary in some systematic way between regions, it should be possible to derive relationships for ungauged catchments in regions which do not have a large number (or potentially any) gauged catchments within them.

Section snippets

The IHACRES model

This model consists of two modules, a non-linear loss module to convert rainfall to effective rainfall, and a linear module to route this effective rainfall to streamflow. Effective rainfall is defined as that rainfall which eventually leaves the catchment as streamflow. As a result, all of the losses of water occur in the non-linear module. In the non-linear module, an antecedent precipitation index, sk is calculated at each timestep as an internal state variable. The index sk is given bysk=rkc

Maroondah catchments

The Maroondah catchments are located in the central highlands of Victoria in south-eastern Australia. Fig. 2 shows the location of this region to the north-east of Melbourne. There are 17 monitored catchments, ranging in size from 4 to 65 ha. The elevations of the weirs range from 232 to 835 m. Mean monthly temperatures vary from 4°C in July to 17°C in January. Average annual precipitation is 1250 mm, with a slight winter maximum, 75% falling in the 8 months from April to November. Snowpack

Application to LTER catchments

Having determined that this technique can be used to predict the daily streamflow of ungauged catchments in the Maroondah region of Victoria, Australia, a study is now being carried out which attempts to extend these results to catchments in other regions. By understanding the controls on hydrologic response in a variety of regions, it is hoped to derive relationships which can be used to predict the hydrologic response of ungauged catchments, even if nearby gauged catchments are not available.

Conclusions

A methodology was presented in Post and Jakeman (1998)for making predictions of daily streamflow for catchments in a region by relating the parameters of the IHACRES rainfall-runoff model to measurable landscape attributes. In this paper, this methodology has been extended by demonstrating that more accurate predictions of streamflow are obtained by predicting the maximum value of the non-linear store in the model by inferring the inter-annual water yield of the catchment under examination from

Acknowledgements

The authors would like to thank the Melbourne Water Corporation, as well as the H.J. Andrews, Coweeta, and Hubbard Brook LTER sites for contributing data and useful advice to this study. Funding for this study was provided in part by NSF grants DEB-9526987 (LTER Intersite Hydrology Research) and the H.J. Andrews, Coweeta and Hubbard Brook LTER grants.

References (23)

  • Gustard, A., Bullock, A., Dixon, J. M., 1992. Low flow estimation in the United Kingdom. Institute of Hydrology Report...
  • Cited by (51)

    • Toward a self-calibrated and independent SM2RAIN rainfall product

      2021, Journal of Hydrology
      Citation Excerpt :

      We finally aim to assess whether the self-calibrated SM2RAIN product performances are in line with those of the calibrated product. Multiple descriptors are considered for the estimation of SM2RAIN algorithm parameter values through a regression-based regionalization approach (Jakeman et al., 1992; Post et al., 1998; Sefton and Howarth, 1998; Seibert, 1999; Wagener et al., 2004): several datasets are selected describing climatic (rainfall and evapotranspiration) and land (soil texture and soil type, SM, soil temperature, topography and vegetation cover) characteristics. The datasets have been selected for different reasons, including their relation with soil state and their availability worldwide.

    • The efficacy of calibrating hydrologic model using remotely sensed evapotranspiration and soil moisture for streamflow prediction

      2016, Journal of Hydrology
      Citation Excerpt :

      Calibrating hydrologic models with ET and SM has important implications for streamflow prediction in ungauged catchments. A widely used method to predict streamflow at ungauged or sparsely gauged catchments is the regionalization of parameters or flow indices (Abdulla and Lettenmaier, 1997; Jakeman et al., 1992; Parajka et al., 2007; Post et al., 1998; Sefton and Howarth, 1998; Viney et al., 2009; Wagener and Wheater, 2006; Wagener et al., 2004; Zhang et al., 2011; Yadav et al., 2007). Many studies described the limitations of parameter regionalization approaches.

    • Hydrological Modeling

      2011, Treatise on Water Science
    • Predicting daily streamflow using rainfall forecasts, a simple loss module and unit hydrographs: Two Brazilian catchments

      2007, Environmental Modelling and Software
      Citation Excerpt :

      Parameter C in the loss module is calculated to give a water balance between effective rainfall and observed streamflow over the model calibration period, which should therefore start and finish at low flows of similar magnitude to minimise the change in the amount of water stored in the catchment over that period. Conceptually, 1/C can be considered (Post et al., 1998) to be the depth, e.g. in millimetres, of a catchment wetness store. The DRCs τ(q) and τ(s) are exponential decay time constants for separate quick and slow response UHs, respectively.

    View all citing articles on Scopus
    View full text