Comparison of two different approaches of sensitivity analysis

https://doi.org/10.1016/S1474-7065(02)00049-9Get rights and content

Abstract

Due to spatial variability, budget constraints or access difficulties model input parameters always are uncertain to some extent. Therefore the knowledge of sensitive input parameters is beneficial for model development and application. It can lead to a better understanding and to better estimated values and thus reduced uncertainty.

In the present paper two simple approaches of sensitivity analysis are compared by the use of the physically based, continuous time hydrological model SWAT. In both approaches, one parameter is varied at a time while holding the others fixed, but the way of defining the range of variation is different. Similar results are obtained suggesting that parameter sensitivity may be determined without the results being influenced by the chosen method. Most sensitive parameters for hydrology and water quality are the physical soil properties such as bulk density, available water capacity or hydraulic conductivity. Plant specific parameters like maximum stomatal conductance or maximum leaf area index as well as slope length, slope steepness, and curve number also show a high sensitivity. Both approaches can be considered as equivalent, as they provide the same overall ranking into more and less sensitive parameters. An identification of the sensitive parameters is possible independently from the chosen variation range.

Introduction

Physically based models are used to simulate a wide range of complex aspects. The purpose of using a model is to establish baseline characteristics whenever data is not available. Furthermore, there are long-term impacts that are difficult to calculate, especially in ecological modeling. Using long data series, process-based deterministic models can compute the great number of calculations required to describe the complexity of a system (lake, river, watershed, etc.). They can provide reliable information on the behavior of the system.

Due to spatial variability, budget constraints or access difficulties model input parameters always contain uncertainty to some extent. However, a model user has to assign values to each parameter. The model is then calibrated against measured data to adjust the parameter values according to certain criteria. This implies that the modeler has a clear understanding of all the parameters used as input to the model and of the processes represented in the model. Parameters that are not well understood may be left unchanged even though they are sensitive or are adjusted to implausible values. Not knowing the sensitivity of parameters can also result in time being uselessly spent on non-sensitive ones. Focus on sensitive parameters can lead to a better understanding and to better estimated values and thus reduced uncertainty.

Therefore sensitivity analysis as an instrument for the assessment of the input parameters with respect to their impact on model output is useful not only for model development, but also for model validation and reduction of uncertainty (Hamby, 1994).

There are many different methods of sensitivity analysis (Beven, 2001; Hamby, 1994). Yet, do they yield equivalent results? Are the same parameter sensitivities identified, regardless of the chosen method?

Section snippets

The hydrologic model

In this article, the results of two different, relatively simple approaches are compared using the example of the hydrological model SWAT-G (Eckhardt et al., 2002b, in press). SWAT-G is a river basin scale model operating on a daily time step. It is a derivative of soil and water assessment tool (SWAT; Arnold et al., 1998), which was developed to predict the impact of land management practices in meso- to macroscale basins. It is physically based. Major model components describe processes

Results

Table 2, Table 3 show the results of both approaches for sensitivity analysis. In general the obtained sensitivities are consistent with results determined in other studies (e.g. Baffaut, 2001, submitted for publication).

Regarding the general watershed attributes, such as surface runoff lag time, residue decomposition factor or peak rate factor for tributary channels, (listed in the SWAT input file .BSN) some of these parameters which according to variant A are sensitive show no or lower

Discussion

Both sensitivity analysis approaches provide approximately similar results and hence can be considered as equivalent. Though in individual cases differing results are possible, the overall ranking into more and less sensitive parameters is the same. Thus an identification of the sensitive parameters is possible, independently from the chosen variation range.

Because of methodical limitations the results of the present study can only provide an orientation. Nevertheless the obtained results can

Acknowledgements

This study has been supported by the Deutsche Forschungsgemeinschaft within the scope of the Collaborative Research Centre (SFB) 299.

References (18)

  • J.G Arnold et al.

    Large area hydrologic modeling and assessment. Part I: Model development

    Journal of the American Water Resources Association

    (1998)
  • Baffaut, C., 2001. Reliability analysis and modeling with SWAT. Journal of American Water Resources Association,...
  • K.J Beven
  • Breuer L., Eckhardt K., Frede, H.-G., 2002. Plant parameter values for ecological approaches in temperate climates....
  • K Eckhardt et al.

    Ein methodischer ansatz zur analyse der sensitivität komplexer modellsysteme

  • Eckhardt, K., Friedrich, C., Frede, H.-G. 2002a. Unterscheidbarkeit von Böden in der hydrologischen Modellierung....
  • Eckhardt, K., Haverkamp, S., Fohrer, N., Frede, H.-G., 2002b. SWAT-G, a version of SWAT99.2 modified for application to...
  • D.M Hamby

    A review of techniques for parameter sensitivity analysis of environmental models

    Environmental Monitoring and Assessment

    (1994)
  • G.M Hornberger et al.

    An approach to the preliminary analysis of environmental systems

    Journal of Environmental Management

    (1981)
There are more references available in the full text version of this article.

Cited by (0)

View full text