2.4. Sources of Model Errors
39
porary disturbances in those currents or cycles, such as EI Niiio effects, are
neglected. To reduce errors from spatial boundary assumptions, spatial modelers often draw spatial boundaries a bit larger than they know they
actually need or set up special rules for the processes along the boundary
that mimic the interactions between the system that is explicitly modeled
with what lies outside that system.
Errors of inappropriate spatial and temporal resolution can be related to
each other. For example, a model used to trace the movement of individuals in a population may subdivide the area into adjacent grid cells and then
specify the decision rules by wh ich movement from one cell to the next occurs. If the population is highly mobile , movement in a given time step can
be farther than the resolution of the space, and errors of spatial resolution result.
To run a computer model requires that initial conditions and parameter
values are specified for a point in time. Those initial conditions and parameter values are often derived on the basis of field or laboratory measurements, and are typically fraught with errors. Good empirical work should
report confidence intervals for the measurements, and a good model
should explore a system's dynamics at least within the reported range of
confidence intervals to minimize errors of model inputs.
Once the model is specified, the difference equations are solved by the
computer in a specific orde r. To see in which order the equations in your
STELLA model are executed, navigate with the downward-pointing triangle
to the model 's equation window , then choose "Equation Prefs . . . " from
the Equation pull-down menu and select "Order of Execution." There is
nothing you can do to influence this order within STELLA once all your
equations are defined, but notice that the choice of order may introduce errors of an inappropriate order of execution. For example, if in a spatial
model of migrating individuals the death rate is a function of population
density and den sity is computed before migration occurs, then death rates
will be different from the case in which density is calculated after migration.
Keep this in mind when you specify your model , and if necessary introduce
time-lags to achieve the desired order in calculation s.
Your modeling effort should start with a clear que stion in mind . The
choices of system components that you wish to model , spatial and temporal resolution, data source s, solution method and DT should be driven by
that question. Avoid having these choices be driven by the answer that you
expect and wish to generate .
At some point in your modeling career, you may find that you are so excited by your model results that you overextend the conclus ions, for example by describing the dynamics you see with words such as never or always. Even if you have done all you can to base your model on the best
available knowledge, the discussion of the various sources of errors presented above should highlight the danger of making errors of drawing
inappropriate conclusions.
39
porary disturbances in those currents or cycles, such as EI Niiio effects, are
neglected. To reduce errors from spatial boundary assumptions, spatial modelers often draw spatial boundaries a bit larger than they know they
actually need or set up special rules for the processes along the boundary
that mimic the interactions between the system that is explicitly modeled
with what lies outside that system.
Errors of inappropriate spatial and temporal resolution can be related to
each other. For example, a model used to trace the movement of individuals in a population may subdivide the area into adjacent grid cells and then
specify the decision rules by wh ich movement from one cell to the next occurs. If the population is highly mobile , movement in a given time step can
be farther than the resolution of the space, and errors of spatial resolution result.
To run a computer model requires that initial conditions and parameter
values are specified for a point in time. Those initial conditions and parameter values are often derived on the basis of field or laboratory measurements, and are typically fraught with errors. Good empirical work should
report confidence intervals for the measurements, and a good model
should explore a system's dynamics at least within the reported range of
confidence intervals to minimize errors of model inputs.
Once the model is specified, the difference equations are solved by the
computer in a specific orde r. To see in which order the equations in your
STELLA model are executed, navigate with the downward-pointing triangle
to the model 's equation window , then choose "Equation Prefs . . . " from
the Equation pull-down menu and select "Order of Execution." There is
nothing you can do to influence this order within STELLA once all your
equations are defined, but notice that the choice of order may introduce errors of an inappropriate order of execution. For example, if in a spatial
model of migrating individuals the death rate is a function of population
density and den sity is computed before migration occurs, then death rates
will be different from the case in which density is calculated after migration.
Keep this in mind when you specify your model , and if necessary introduce
time-lags to achieve the desired order in calculation s.
Your modeling effort should start with a clear que stion in mind . The
choices of system components that you wish to model , spatial and temporal resolution, data source s, solution method and DT should be driven by
that question. Avoid having these choices be driven by the answer that you
expect and wish to generate .
At some point in your modeling career, you may find that you are so excited by your model results that you overextend the conclus ions, for example by describing the dynamics you see with words such as never or always. Even if you have done all you can to base your model on the best
available knowledge, the discussion of the various sources of errors presented above should highlight the danger of making errors of drawing
inappropriate conclusions.