Global sensitivity analysis in ASM applications: comparison of the SRC and Extended-FAST method for a UCT-MBR model

Abstract

In this study global sensitivity analysis is performed to identify influential as well as non-influential parameters in a model of a University Cape Town Membrane Bioreactor (UCT-MBR). In particular, the Standardised Regression Coefficients (SRC) and Extended-FAST sensitivity analysis methods are applied. The sensitivity of model variables towards parameter variation is analysed for CODTOT, SNH4, SNO3, SPO, and MLSS along five reactor compartments. Both methods indicate that the parameters identified as being influential differ from section to section due to the different processes involved. Moreover, the relevant influence of the membrane filtration parameters is detected in the first plant section due to the influence of the recycled sludge. It is found that the computationally less expensive SRC method is applied outside its range of applicability with R2 = (0.3-0.6) < 0.7. Nevertheless, the ranking obtained with the SRC method for the influential parameters is very similar to that of the Extended-FAST method, except for MLSS. However, to obtain reliable quantitative information on variance decomposition and to detect and quantify (in some cases considerable) interactions present among parameters the use of the computationally more expensive Extended-FAST is found to be necessary in this case study.