Validation

Once a model has been programmed and appears to be giving reasonable results, it must be validated to ensure that there are no subtle errors that affect the results significantly but not so much as to make them obviously unrealistic. A common way to initiate the validation process is to set parameters in the model in such a way that the model duplicates a situation for which there is an analytical solution. For example, if d0/dt, u, and Q are set to zero in Equation (11.12) and w is assumed to be downward and to decrease linearly with depth, the model should reproduce the Robin (1955) solution (Equation (6.24)). In flow models, the deformation of an infinite slab of ice on a uniform slope (Equation (10.42)) is a good choice. Of course, once these comparisons have been made, there is still the question of whether coding of some of the terms neglected in the test, such as u(d0/dx) in Equation (11.12), is correct. The modeler will have to be more imaginative to find independent ways to test these algorithms. One possibility is to compare the output of similar models developed totally independently, as discussed next.

0 0

Post a comment