Questions for Discussion:
1. According to Freyberg, what are the two basic problems that always arise in model calibration?
2. List the four approaches that have been suggested to help reduce these problems.
3. Briefly describe the problem presented to the students. Was the ground-water system similar in complexity to most natural systems? In what way(s) would a natural system be different?
4. What is the usual goal of calibration? Is the end objective usually parameter identification?
5. What did the students need to predict with their “calibrated” model?
6. What data were given to the students? How would this compare to a real problem where a natural system was to be modeled? In general, would there be more or less information available?
7. How would the success of the calibration be measured? What is RMSE?
8. Why weren’t the students measured on the success of the calibrated fit to the observed head data in the well cells?
9. Why did the water levels in the aquifer near the river cells rise after the modification to the system?
10. What is shown in the histogram of Fig. 6? Which group did the “best” overall? Was this group “much” better than the others or were several of the groups up near them?
11. Why are the error measures of recharge and hydraulic conductivity so well correlated, yet there is virtually no correlation between aquifer depth and the prediction error for the pumping well cells?
12. Why is the RMSE consistently smaller for the overall prediction than for the prediction at the six pumping well cells?
13. “The ability of a calibrated parameter set to reproduce the observed data was clearly not an indicator of the ability of that parameter set to predict the system response under modified conditions.” Interestingly, the group with the “best” calibrated heads at the observation wells had the worst RMSE for the predicted heads at the production wells after the system was modified! This emphasizes that “the quality of the parameter set cannot be measured by the quality of the calibration on the observed heads.”
14. Explain why all of the groups doing the calibration/prediction exercise finished with a RMSE for the entire head field that always exceeded the RMSE for calibrated head at the 16 observation well cells.
15. What calibration strategy provided the “best” results? What
provided
the worst results?