Info

Source: Adapted from Lauer and Sussman (1979).

Source: Adapted from Lauer and Sussman (1979).

stressed that these results pertain only to the configuration of these models as they existed at the time of the tests, and these results are only valid for the particular test case run. Different results might be obtained using more recent configurations of the same models or with other test scenarios and geometries, or even with other measurement standards or host computers.

The cumulative accuracy measures were also computed (Table 11.3). These results support the intuitive notion that model accuracy improves with increasing model complexity and computational intensity. Accuracy alone, however, should not be the sole evaluation criterion. A broader view of the model evaluation process is presented in Section 11.6.

11.6 Evaluation guidelines

Model evaluation entails the systematic gathering and promulgation of information about models to determine model limitations and domains of applicability. There is no evaluation procedure appropriate for all models. The procedures must be tailored on the basis of model structure, documentation and other information available to the evaluators. Criteria for model evaluation can be segregated into five basic categories (US General Accounting Office, 1979): documentation, verification, validation, maintainability and usability.

11.6.1 Documentation

Documentation of computer models is important for two reasons: (1) to ensure that the model is thoroughly understood and can be operated and maintained in the present as well as in the future; and (2) to facilitate independent evaluation of the model, particularly by someone other than the model developer or initial user.

Table 11.2 Means (/x) and standard deviations (a) of the differences between the FFP propagation model and candidate model results for one particular case over direct path (DP), bottom bounce (BB) and convergence zone (CZ) regions

Model Range interval

Table 11.2 Means (/x) and standard deviations (a) of the differences between the FFP propagation model and candidate model results for one particular case over direct path (DP), bottom bounce (BB) and convergence zone (CZ) regions

Model Range interval

0 0

Post a comment