dwelter/pestpp

Negative Sobol Indices

Closed this issue · 2 comments

hamiddashti commented

Hi, I'm running PEST++ GSA with Sobol and I'm getting small negative numbers with respect to objective function (figure below) for some parameters. I tried different sampling size up to 1000 and the results are almost similar. I cannot do more than 1000 due to computation resources issue.

image

Is PEST++ returns confidence interval on indices for each variables? I mean if the confidence interval includes zero and since these negative numbers are small, I assume I can safely turn this negative numbers into zero.
Thanks

jtwhite79 commented

@hamiddashti - the USGS version of the GSA utility ( PESTPP-SEN https://github.com/usgs/pestpp as of version 4.2.16) has been reworked so that the sobol indices for a simple test problem are the same as those yielded by SALib. Both GSA and PESTPP-SEN only report mean indices (not confidence intervals). PESTPP-SEN can be extended to yield confidence intervals on the indices, just havent gotten around to it yet.

hamiddashti commented

Git it! Thank you. I'll go ahead and close this issue.