Sensitivity analysis based on regional splits and regression trees (SARS-RT)

摘要: A global sensitivity analysis with regional properties is introduced. This method demonstrated on two synthetic and one hydraulic example. It can be shown that an uncertainty based one-dimensional scatter plots correlation analyses such as the Spearman Rank Correlation coefficient lead to misinterpretations of any model results. The which has been proposed in this paper multiple regression trees (so called Random Forests). splits at each node tree are sampled from a probability distribution. Several criteria enforced level splitting ensure positive information gain also distinguish between behavioural non-behavioural representations. latter distinction applied generalized likelihood estimation (GLUE) (RSA) framework analyse results used here derive (model) structures. Two methods used: first total achieved by parameter evaluated. In second parameters sets permuted error rate computed. compared values without permutation. allows evaluation combinations thus gives insight into structure response surface. examples demonstrate capability methodology stress importance application analysis.