Global sensitivity analysis in wastewater applications: A comprehensive comparison of different methods
- Autori: Cosenza, A.; Mannina, G.; Vanrolleghem, P.; Neumann, M.
- Anno di pubblicazione: 2013
- Tipologia: Articolo in rivista (Articolo in rivista)
Three global sensitivity analysis (GSA) methods are applied and compared to assess the most relevant processes occurring in wastewater treatment systems. In particular, the Standardised Regression Coefficients, Morris Screening and Extended-FAST methods are applied to a complex integrated membrane bioreactor (MBR) model considering 21 model outputs and 79 model factors. The three methods are applied with numerical settings as suggested in literature. The main objective considered is to classify important factors (factors prioritisation) as well as non-influential factors (factors fixing). The performance is assessed by comparing the most reliable method (Extended-FAST), by means of proposed criteria, with the two other methods. In particular, similarity to results obtained from Extended-FAST is assessed for sensitivity indices, for the ranking of sensitivity indices, for the classification into important/non-influential factors and for the method's ability to detect interaction among factors and to provide results in a reasonable time. It was found that the computationally less expensive SRC method was applied outside its range of applicability (R2) = (0.3–0.6) < 0.7. Still, the SRC produced a ranking of important factors similar to Extended-FAST. For some variables significant interactions among the factors were revealed by computing the total effect indices STi using Extended-FAST. This means that to obtain reliable variance decomposition and to detect and quantify interactions among the factors, the use of the Extended-FAST is recommended. Regarding the comparison between Morris screening and Extended-FAST a poor agreement was found. In particular, the Morris screening overestimated the number of both important and non-influential factors compared to Extended-FAST for the analysed case study.