I have some results that show an estimate for a particular zone may be either more or less sensitive to observations than other zones. Pest++ output only gives composite sensitivities (combined sensitivity to all observations) - I am looking for a more detailed breakdown, but one I can easily read. Any additional tips on sensitivity analysis of my Pest++ results are welcome.
I highly recommend that you use either of the following two statistical packages that I use to perform sensitivity studies:
1) Minitab
2) JMP
You can download trial versions of these statistical packages at no cost and try to apply either of them to you perform the sensitivity study that you cited.
Hope this helps answer your question.
Professor Yehia Khalil
Yale University
USA
Does this webpage help: http://www.pesthomepage.org/PEST_Utilities.php
you probably have consulted it, but just in case
Thank you both! I did have to go back to the pest utilities page - I was there once - first time did not find right ones - turns out I need utilities that address parameter identifiability. Dr. Khalil - those programs you mention look very useful as well - I hope to look into them.
OK I found 2 PEST utilities named IDENTPAR and GENLINPRED which perform some kind of identifiability analysis - and tried to run in a directory with my PEST++ output. I am getting an error message saying that my jacobian output file is not compatible with my pest control file. That at least one obs/adj par is different between them. (they have identical timestamps - so this does not make sense). So I am not sure these utilities are compatible with PEST++ output. Does anyone know? In the PEST++ manual it says authors tried to keep "many" of outputs the same just for this reason - so utilities would work.
well before running GENLINPRED you need to transfer the parameters to the original pest file using PARREP utility (parrep where xxx your original pst file which ) fix the no of iteration to -2 so that it creates .jco file for you with just single set of parameters run. then run GENLINPRED. if your concern is parameter uncertainty you need to create param.unc file with against each parameters you have to define the standard deviation of each parameters.