Comparison of variance-based and moment-independent global sensitivity analysis approaches by application to the SWAT model

Abstract

Link to publication record in Explore Bristol Research PDF-document This is the author accepted manuscript (AAM). The final published version (version of record) is available online via Elsevier at http://www.sciencedirect.com/science/article/pii/S1364815217301159. Please refer to any applicable terms of use of the publisher. University of Bristol -Explore Bristol Research General rightsThis document is made available in accordance with publisher policies. Please cite only the published version using the reference above. Full terms of use are available: http://www.bristol.ac.uk/pure/about/ebr-terms AbstractGlobal Sensitivity Analysis (GSA) is an essential technique to support the calibration of environmental models by identifying the influential parameters (screening) and ranking them.In this paper, the widely-used variance-based method (Sobol') and the recently proposed moment-independent PAWN method for GSA are applied to the Soil and Water Assessment Tool (SWAT), and compared in terms of ranking and screening results of 26 SWAT parameters. In order to set a threshold for parameter screening, we propose the use of a "dummy parameter", which has no influence on the model output. The sensitivity index of the dummy parameter is calculated from sampled data, without changing the model equations. We find that Sobol' and PAWN identify the same 12 influential parameters but rank them differently, and discuss how this result may be related to the limitations of the Sobol' method when the output distribution is asymmetric.

A sensitivity analysis of the PAWN sensitivity index

“…S1). This corroborated previous observations on the faster convergence rate of PAWN compared to Sobol' indices [1,6]. By defining N ∼ U(200, 2000) we set our study in a scenario where the uncertainty with regards to the required sample size needed to compute robust PAWN indices is moderate.…”

“…obtained by fixing the i-th parameter to j = 1, 2, ..., n values or intervals within its uncertainty range. The difference between Y U and Y Cij is assessed via the Kolmogorov-Smirnov test, although other distance-based tests, such as the Anderson-Darling's, may also be used [6]. The final PAWN index for a given parameter is obtained by calculating the mean, the median, the maximum or any other summary statistic over all the KS values computed between Y U and Y Cij .…”

“…Based on trials with the Liu et al [3] function, the Ishigami and Homma [7] function, the SWAT model [6] or a wind-energy converter model [8], it has been observed that PAWN might reach convergence much faster than Sobol' indices. A key question, however, is to know how the selection of N, n, the sampling of Y U or the summary statistic affects the accuracy of PAWN.…”

Multicriteria assessment framework of flood events simulated with vertically mixed runoff model in semiarid catchments in the middle Yellow River

“…C3. The observed peak flows should fall within 1 standard deviation (σ ) of the mean (approximately 68.3 % uncertainty interval) peak flow estimated by the hydrologic uncertainty processor (HUP), one component of the Bayesian forecasting system detailed in Krzysztofowicz (1999) and Biondi et al (2010).…”

A comprehensive sensitivity and uncertainty analysis for discharge and nitrate-nitrogen loads involving multiple discrete model inputs under future changing conditions

“…Thus, PAWN is applicable to discrete model inputs. Further, PAWN is moment-independent and was found to be a robust measure for sensitivity of non-symmetrically distributed outputs of environmental models (Pianosi and Wagener, 2015;Zadeh et al, 2017).…”