Methods and guidelines for effective model calibration

EXAMPLE FIELD APPLICATIONS AND SYNTHETIC TEST CASES
The nonlinear regression methods, diagnostic and inferential statistics, and guidelines de-
scribed above have been used successfully in a number of applications. Because nonlinear regres-
sion is a useful but imperfect tool for model calibration, its application is not always 
straightforward. Thus, it can be helpful to consider other applications when designing and report-
ing a modeling study using nonlinear regression. References describing applications using the 
methods described in this report include Cooley (1979, 1983a), Cooley and others (1986), Chris-
tiansen and others (1995), McKenna and Poeter (1995), Anderman and others (1996), Barlebo and 
others(1996), D’Agnese and others (1996, 1998, in press), Tiedeman and others (1997), and Eberts 
and others (in press). Similar approaches were used by Gailey and others (1991), Tiedeman and 
Gorelick (1993), Yager (1993), Kuiper(1994), Olsthoorn (1995), Christensen (1997), and Chris-
tensen and others (in press). In addition, Hill and others (1998) discuss the calibration of a complex 
synthetic test case using the methods discussed above, and Poeter and McKenna (1995) present a 
synthetic ground-water transport test case that evaluates stochastically generated zonations using 
nonlinear regression methods. The synthetic test cases provide the opportunity to conclusively 
evaluate the accuracy of the models calibrated using the methods described in this report, and in 
both studies better models determined as discussed in this report produced more accurate predic-
tions. Poeter and Hill (1997) demonstrate many of the ideas presented in this report using simple 
examples. 

---

68
USE OF THE GUIDELINES WITH DIFFERENT INVERSE MODELS
The methods and guidelines presented above were used to design UCODE and MOD-
FLOWP. Many aspects of the methods and guidelines, however, are broadly applicable to other 
inverse models, with some modifications. Likely differences between UCODE and MODFLOWP 
and other inverse models are categorized below, with a few words about how the guidelines would 
be adapted.