176
John R. Richardson and Cory W. Berish
usually the most stringent. To calibrate flows and storages, the data
must be relevant and accurate. There is a need to know more than
just zero-order flow information. Flow values relative to the interacting processes are needed to develop models that are nonlinear.
Usually, the bounds of all of the flows are not well known. Events that
occur on longer time scales and at larger spatial extents than those
measured may have significant consequences on the model. Generalized information or modeled data from a larger domain can be used
to help set the bounds of the model.
Verification checks that the model behaves correctly with the calibration data. This analysis may require time-series data to adequately
verify that the model performs as it was designed. The data set used
to verify the model is usually the same as the data used to calibrate
it. This practice is used to ensure that the model structure is correct
and behaves in the same manner as the system being modeled.
Validation determines that the model behaves in the same manner
with an independent data set. It is bad modeling practice to validate
the model with the same data used to calibrate and verify the model.
A second, parallel data set is required to independently validate the
model. Often, when data are collected, the data set is split, and half is
used to calibrate and verify, and the other half is used to validate. Validation data sets can also come from other studies of similar systems.
Ensuring that the temporal and spatial characteristics of the validation data set match the model and the previously used data may be
important to prevent domain errors. Validation is not always required
for a model to be used.
Prediction/analysis develops data and information that can be used
to support management decisions. The data derived from the model
must meet the same temporal and spatial conditions as the calibration data. Validation requires that the data be within the same data
frame as the input data; predictive results often go beyond the sampling area or time. As long as the model and calibration data reflect
the appropriate temporal and spatial domain, the predictions with the
model will have relevance. Predictions outside the time and spatial
domain of the input data may not accurately reflect conditions in the
real world.
GIS maps are models!
Rarely does GIS information reflect a direct relationship to what is
on the ground. It is mapped through a model and has uncertainty in
its spatial representation and uncertainty in its categorical data. Maps
have a powerful visual impact even though they may not accurately
represent the reality of what is on the ground.
John R. Richardson and Cory W. Berish
usually the most stringent. To calibrate flows and storages, the data
must be relevant and accurate. There is a need to know more than
just zero-order flow information. Flow values relative to the interacting processes are needed to develop models that are nonlinear.
Usually, the bounds of all of the flows are not well known. Events that
occur on longer time scales and at larger spatial extents than those
measured may have significant consequences on the model. Generalized information or modeled data from a larger domain can be used
to help set the bounds of the model.
Verification checks that the model behaves correctly with the calibration data. This analysis may require time-series data to adequately
verify that the model performs as it was designed. The data set used
to verify the model is usually the same as the data used to calibrate
it. This practice is used to ensure that the model structure is correct
and behaves in the same manner as the system being modeled.
Validation determines that the model behaves in the same manner
with an independent data set. It is bad modeling practice to validate
the model with the same data used to calibrate and verify the model.
A second, parallel data set is required to independently validate the
model. Often, when data are collected, the data set is split, and half is
used to calibrate and verify, and the other half is used to validate. Validation data sets can also come from other studies of similar systems.
Ensuring that the temporal and spatial characteristics of the validation data set match the model and the previously used data may be
important to prevent domain errors. Validation is not always required
for a model to be used.
Prediction/analysis develops data and information that can be used
to support management decisions. The data derived from the model
must meet the same temporal and spatial conditions as the calibration data. Validation requires that the data be within the same data
frame as the input data; predictive results often go beyond the sampling area or time. As long as the model and calibration data reflect
the appropriate temporal and spatial domain, the predictions with the
model will have relevance. Predictions outside the time and spatial
domain of the input data may not accurately reflect conditions in the
real world.
GIS maps are models!
Rarely does GIS information reflect a direct relationship to what is
on the ground. It is mapped through a model and has uncertainty in
its spatial representation and uncertainty in its categorical data. Maps
have a powerful visual impact even though they may not accurately
represent the reality of what is on the ground.