2.2.5 Model Evaluation
Every model should be evaluated with respect to transparency and robustness.
Transparency refers to how easy it is to understand a model and robustness refers
to how closely the predictions of the model match with observed outputs from
the system. However, after the above-mentioned steps in constructing a model, the
model can be run and its output can be compared with measurements from the
system. A test of the model output against observations is especially necessary
when a model is to be used in an application mode. The users of the model need to
be given some notion of situations in which the model has proven useful, with a
disclaimer for reliability in any other situations (Sinclair and Seligman 1996).
Model evaluation should be done based on predefined criteria established in the
objective-definition stage. If the predictions are reasonably matched with the
measurements from the system, based on predefined criteria, the modeling process
is complete and the model is ready to use. If the results of the model evaluation
stage are not satisfactory, it is necessary to reconsider the hypotheses, equations,
and the quantitative methods use dot construct the model. If alternative methods
and equations are identified, then the model evaluation should be repeated. If
predictions and the output are not still satisfactory, it is likely that there is some
basic problem with the hypotheses. More experimental investigation is required
and the modeling process should be set aside until observations allow improved
hypotheses (Soltani and Sinclair 2012).
There are two critical criteria in evaluating the suitability of a model: transparency and robustness. Transparency means that the model parameters, flow
diagrams, and code can be readily understood by those that were not involved in its
development. As much as possible, the functions are stand-alone descriptions of
processes in the plant and crop. Transparency is facilitated by a minimum number
of empirical coefficients, and these coefficients can be independently observed and
measured. Robustness means that the model produces simulation results that
compare favorably with observation. The judgment of ‘‘favorable’’ will depend
directly on the original objectives for the model (Teh 2006).
2.3 Types of Models
Mathematical modeling can be divided in two principal groups: mechanistic (white
box) and phenomenological (black box) models. The white box models are
deterministic and use physical modeling, thus they are explicative about the
modeled system. Black box models, also called identification models, are direct
descriptions of the data. Black box models have the disadvantage of not giving an
explanation of the subjacent mechanisms. A combination of these two models
results in gray box models.
56
M. A. Vázquez-Cruz et al.
Every model should be evaluated with respect to transparency and robustness.
Transparency refers to how easy it is to understand a model and robustness refers
to how closely the predictions of the model match with observed outputs from
the system. However, after the above-mentioned steps in constructing a model, the
model can be run and its output can be compared with measurements from the
system. A test of the model output against observations is especially necessary
when a model is to be used in an application mode. The users of the model need to
be given some notion of situations in which the model has proven useful, with a
disclaimer for reliability in any other situations (Sinclair and Seligman 1996).
Model evaluation should be done based on predefined criteria established in the
objective-definition stage. If the predictions are reasonably matched with the
measurements from the system, based on predefined criteria, the modeling process
is complete and the model is ready to use. If the results of the model evaluation
stage are not satisfactory, it is necessary to reconsider the hypotheses, equations,
and the quantitative methods use dot construct the model. If alternative methods
and equations are identified, then the model evaluation should be repeated. If
predictions and the output are not still satisfactory, it is likely that there is some
basic problem with the hypotheses. More experimental investigation is required
and the modeling process should be set aside until observations allow improved
hypotheses (Soltani and Sinclair 2012).
There are two critical criteria in evaluating the suitability of a model: transparency and robustness. Transparency means that the model parameters, flow
diagrams, and code can be readily understood by those that were not involved in its
development. As much as possible, the functions are stand-alone descriptions of
processes in the plant and crop. Transparency is facilitated by a minimum number
of empirical coefficients, and these coefficients can be independently observed and
measured. Robustness means that the model produces simulation results that
compare favorably with observation. The judgment of ‘‘favorable’’ will depend
directly on the original objectives for the model (Teh 2006).
2.3 Types of Models
Mathematical modeling can be divided in two principal groups: mechanistic (white
box) and phenomenological (black box) models. The white box models are
deterministic and use physical modeling, thus they are explicative about the
modeled system. Black box models, also called identification models, are direct
descriptions of the data. Black box models have the disadvantage of not giving an
explanation of the subjacent mechanisms. A combination of these two models
results in gray box models.
56
M. A. Vázquez-Cruz et al.