Assessment of Parametric Sensitivity Analysis Methods Based on A Quasi Two-Dimensional Groundwater Model

Abstract

Parametric sensitivity analysis (SA) aims to select the sensitive parameters that most significantly affect the model output variables, which helps to improve model optimization efficiency by adjusting a small number of sensitive parameters instead of all adjustable parameters. The qualitative and quantitative SA methods have been commonly used to quantify the sensitive parameters of the models. However, the response surface model based quantitative SA method was rarely used. Taking the simulation of a quasi twodimensional (quasi-2D) groundwater model as an example, this study systematically assess eight SA methods divided into three categories (qualitative SA, quantitative SA, and the response surface model-based quantitative SA). The study validates the effectiveness of these methods by comparing the parameter sensitivity results, and also demonstrates the efficiency of these methods by determining the minimum sample size required. Using the minimum samples means the least number of model runs. The results show that P1 and P2 are the most sensitive parameters of the quasi-2D model for simulating groundwater table elevation. Except for local method, four global qualitative SA methods obtain reasonable parameter sensitivity rankings using 200 samples, but the parameter sensitivity scores fail. For obtaining accurate sensitivity scores, at least 2000 samples are required by the quantitative SA methods. However, for the response surface model-based quantitative SA method, 60 samples are sufficient to obtain accurate sensitivity scores, demonstrating that the method is an effective and highly efficient, and should be recommended as the primary parametric SA method, especially for the complex models with large computational demand.