jonasthu 0 Posted July 4, 2016 Report Share Posted July 4, 2016 Hi, I've been trying to calibrate my transient model using PEST, however I'm experiencing some difficulties. It executes the estimation process and finishes the parameter estimation but it fail's to give me any new values for the chosen parameters. This only happens when I try to calibrate it as a transient model. When I run it in steady state there are no problems and it estimates new values for my parameters. When I run it as a transient model there is a message in the PSLAVES windows that pops which states that the "Total sum of squared, weighted residuals: NaN", and I think the problems lies here. However I'm not sure how to fix it. If anyone has any idea what's causing this and could help me with a solution I would be very grateful. Jonas Quote Link to post Share on other sites
woodward 23 Posted July 4, 2016 Report Share Posted July 4, 2016 I haven't used PEST with GMS, only with Groundwater Vistas (GV), but it looks like the problem is that the model is not returning a valid answer. Could be because of (1) invalid calibration data (2) parameter values that result in invalid model output (3) MODFLOW crash. Could also be a problem if a parameter value that is log scaled is 0 or less. In GV there's a switch that will cause MODFLOW always to terminate even if there's an error, so you should set that. Also put bounds on the parameters to make sure you'll always get a valid simulation. Quote Link to post Share on other sites
jonasthu 0 Posted July 5, 2016 Author Report Share Posted July 5, 2016 Thanks for the replay. I also thought that it might be problems with the calibration data, so I tried running a simplified model with only 2 stress periods and 1 observation point which contained observation data that I made up and knew would be +/- 1 m from the final solution. However I still get the same message of "weighted.. residuals: NaN". The parameters values are the same that I found using the steady state model, which runs nicely and gives me an OK fit between computed and observed head values. So I don't think that is the problem. The problem might be with the MODFLOW simulation. After I run MODFLOW, the same message ("Total sum of squared weighted residuals: NaN") appears in the window that pops up. However, the MODFLOW simulation finishes and ends in a proper way, and I'm able to view the solution. I'm grateful for any other inputs on what might be causing this. Jonas Quote Link to post Share on other sites
Alan Lemon 3 Posted July 5, 2016 Report Share Posted July 5, 2016 One thing that I have seen before is models that use the CHD package where the value assigned to the CHD is 0.0. There is a bug in MODFLOW where this can produce NaN. If this is the case change the value to something that is nonzero (1e-10). Quote Link to post Share on other sites
woodward 23 Posted July 5, 2016 Report Share Posted July 5, 2016 So if you only run MODFLOW transient you still get the message? Then it is not caused by PEST. Which version of MODFLOW? Have you got the parameter estimation routine of MODFLOW2000 turned on by mistake? Quote Link to post Share on other sites
jonasthu 0 Posted July 6, 2016 Author Report Share Posted July 6, 2016 Finally I works. It was the bug that Alan mentioned, with CHD value assigned was 0. I changed it to 1e-10 and now it works. Thanks for the help all! Quote Link to post Share on other sites
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.