Comment on “Two statistics for evaluating parameter identifiability and error reduction” by John Doherty and Randall J. Hunt

Journal of Hydrology
By:

Links

Abstract

Doherty and Hunt (2009) present important ideas for first-order-second moment sensitivity analysis, but five issues are discussed in this comment. First, considering the composite-scaled sensitivity (CSS) jointly with parameter correlation coefficients (PCC) in a CSS/PCC analysis addresses the difficulties with CSS mentioned in the introduction. Second, their new parameter identifiability statistic actually is likely to do a poor job of parameter identifiability in common situations. The statistic instead performs the very useful role of showing how model parameters are included in the estimated singular value decomposition (SVD) parameters. Its close relation to CSS is shown. Third, the idea from p. 125 that a suitable truncation point for SVD parameters can be identified using the prediction variance is challenged using results from Moore and Doherty (2005). Fourth, the relative error reduction statistic of Doherty and Hunt is shown to belong to an emerging set of statistics here named perturbed calculated variance statistics. Finally, the perturbed calculated variance statistics OPR and PPR mentioned on p. 121 are shown to explicitly include the parameter null-space component of uncertainty. Indeed, OPR and PPR results that account for null-space uncertainty have appeared in the literature since 2000.

Publication type Article
Publication Subtype Journal Article
Title Comment on “Two statistics for evaluating parameter identifiability and error reduction” by John Doherty and Randall J. Hunt
Series title Journal of Hydrology
DOI 10.1016/j.jhydrol.2009.10.011
Volume 380
Issue 3-4
Year Published 2010
Language English
Publisher Elsevier
Contributing office(s) National Research Program - Central Branch
Description 8 p.
First page 481
Last page 488
Google Analytic Metrics Metrics page
Additional publication details