Global sensitivity analysis in hydrological modeling: Review of concepts, methods, theoretical framework, and applications

Abstract

s u m m a r y Sensitivity analysis (SA) aims to identify the key parameters that affect model performance and it plays important roles in model parameterization, calibration, optimization, and uncertainty quantification. However, the increasing complexity of hydrological models means that a large number of parameters need to be estimated. To better understand how these complex models work, efficient SA methods should be applied before the application of hydrological modeling. This study provides a comprehensive review of global SA methods in the field of hydrological modeling. The common definitions of SA and the typical categories of SA methods are described. A wide variety of global SA methods have been introduced to provide a more efficient evaluation framework for hydrological modeling. We review, analyze, and categorize research into global SA methods and their applications, with an emphasis on the research accomplished in the hydrological modeling field. The advantages and disadvantages are also discussed and summarized. An application framework and the typical practical steps involved in SA for hydrological modeling are outlined. Further discussions cover several important and often overlooked topics, including the relationship between parameter identification, uncertainty analysis, and optimization in hydrological modeling, how to deal with correlated parameters, and time-varying SA. Finally, some conclusions and guidance recommendations on SA in hydrological modeling are provided, as well as a list of important future research directions that may facilitate more robust analyses when assessing hydrological modeling performance.

Exploring the impact of forcing error characteristics on physically based snow simulations within a global sensitivity analysis framework

“…These cases were considered because they sampled realistic errors (NB and NB_gauge) and minimum errors (NB_lab). We expected that the error ranges exerted a major control on model uncertainty and sensitivity, as demonstrated in several prior sensitivity analyses (see review of Song et al, 2015).…”

“…It is unclear how (1) different error types (bias vs. random errors), (2) different error distributions, and (3) different error magnitudes across all forcings affect model output. The impact of forcing errors on models can be tested by corrupting forcings with specified characteristics (e.g., artificial biases and random errors) and quantifying the impact on model outputs (e.g., Oudin et al, 2006;Spank et al, 2013), but we are unaware of any detailed studies that have done this type of experiment for all meteorological forcings commonly required for physically based snow models. We hypothesize that (1) model outputs are more sensitive to biases than random errors in forcing variables, (2) the assumed probability distribution for biases will alter the relative ranking of importance in forcing errors, and (3) the magnitude of forcing biases will have a strong influence on which forcing errors are most important.…”

“…In their recent review of global sensitivity analysis applications in hydrological modeling, Song et al (2015) identified the selection of probability distributions (this section) and ranges (Sect. 3.2.3) as among the most important considerations.…”

“…Numerous approaches that explore uncertainty in numerical models have been developed in the literature of statistics (Christopher Frey and Patil, 2002), environmental modeling (Matott et al, 2009), and optimization/calibration of hydrology and earth systems models (Beven and Binley, 1992;Duan et al, 1992;Kavetski et al, 2002Kavetski et al, , 2006aKuczera et al, 2010;Razavi and Gupta, 2015;Song et al, 2015;Vrugt et al, 2009Vrugt et al, , 2008. Among these, global sensitivity analysis is an elegant platform for testing the impact of input uncertainty on model outputs and for ranking the relative importance of inputs while considering coexisting sources of uncertainty.…”

A comprehensive numerical analysis of the hydraulic behavior of a permeable pavement

“…Nossent et al (2011) reported that for most parameters, less than 5000 samples were sufficient to reach a stable solution. An extensive review of the GSA in hydrological models is reported in Song et al (2015). Here, we report the number of model runs for each GSA performed, together with the type of GSA, the number of parameters of the model, and the objective function used.…”