Sensitivity Analysis of Hydrological Model Using a Hybrid of Variance Based and Latin Hypercube Sampling Method.

Abstract:

Numerous existing hydrological models are different perspectives of the system, and inevitably, are imperfect representations of reality. Irrespective of the choice of models, the major source of error in any hydrological modeling is due to the uncertainty in the determination of model parameters, owing to a mismatch between model complexity and the level of data which is available to parameterize, initialize, and calibrate such models. A sensitivity analysis (SA) to determine the possible values assigned to the parameters and the qualitative and or quantitative variations in the output of a model associated should be an integral part of any hydrological modelling study. SA methods help to identify the parameters that have a strong impact on the model outputs and hence influence the model response.Various methods are available to perform sensitivity analysis and the perturbation technique varies from approach to approach. In this study the applicability of a more complex sensitivity analysis approach, variance based method, is studied. However, since, variance based methods considers the entire range of parameter space and their possible interactions, hence making it computationally intensive; therefore, Latin hypercube sampling method is employed to screen the sample space. To check the feasibility of integrating Variance based method with Latin hypercube sampling, this hybrid approach is applied on the outputs generated by Variable Infiltration Capacity Model (VIC), which is a semi-distributed model having the feature of sub grid variability. Mahanadi basin (India), draining an area of around 1,41,600sq km, has been selected as the study area, as it is repetitively facing the adverse hydro-meteorological conditions such as floods, droughts and cyclones in the recent times. The developed methodology, in addition to some relaxation over computational time, provide more rigorous analysis of model parameters by capturing the influence of the full range of variation of all model parameters on model output.