Agriculture Reference
In-Depth Information
10.2.5.1 Model Calibration
The calibration process involves the adjustment of selected model parameters within
an expected range so as to minimize the discrepancies between observed and pre-
dicted values. Calibration of the model is typically accomplished by adjusting input
parameters and rate constants so that output values are as close as possible to
measured values in the calibration data set.
This approach to calibrating a model has the disadvantage of requiring a calibra-
tion data set. In addition, the approach can be very time consuming, especially if
the range of values explored for each parameter and rate constant is relatively large.
Furthermore, since the process adjusts several input parameters and rate constants at
the same time, inadequacies in the modeling of one process may be coincidentally
compensated for when the input parameter or rate constant is changed for another
process. Consequently, this approach may lead to good simulation for the calibra-
tion data set, but will not necessarily guarantee good simulations for independent
data sets.
Another approach to calibration is to conduct laboratory studies to determine the
rate constants and input parameters for each of the important processes in the model,
and then use those rate constants and parameters to validate the model.
10.2.5.2 Model Verification/Validation
Validation is essentially an independent test of the model, where the model pre-
dictions are compared with data not used in the calibration testing. The goal of
verification is to quantify confidence in the predictive capacity of the model by
comparison with experimental data.
In contrast to traditional experiments, validation experiments are performed to
generate high quality data for the purpose of assessing the accuracy of a model
prediction. To qualify as a validation test, the specimen geometry, initial condi-
tions, boundary conditions, and all other model input parameters must be prescribed
accurately.
Evaluation of model performance should include both statistical criteria and
graphical display. A model is a good representation of reality only if it can predict
an observable phenomenon with acceptable accuracy and precision.
10.2.6 Statistical Indicators for Model Performance Evaluation
One of the most common issues raised concerning model usage is that of reliabil-
ity. There is a pressing need to assess model validity in quantitative terms, e.g.,
statistical limits of model performance against the prototype system. The follow-
Search WWH ::
Custom Search