A very general definition of sensitivity analysis, on which most modellers would agree, is that SA looks at how a given model, generally a computational one, responds to variation in input parameters. There are naturally several aspects of the response of interest. Some typical questions, which SA attempts to answer are: 1) Is there some region in the space of the input parameters for which the model variation is maximum or diverges?; 2) To which parameters is the model most sensitive (locally)?; 3) To which parameters is the model most sensitive (globally)?; 3a) could a simple regression model for the system be realised on a subset of parameters?; 3b) could a subset of parameters be identified which would account for 90% of the variance of my model prediction?; 3c) for system with hundred of parameters: how do i identify the controlling factors with a minimum number of numerical experiments?.

SCOTT Marian;  SALTELLI Andrea; 

1997-02-03

JRC14405

https://publications.jrc.ec.europa.eu/repository/handle/JRC14405,