18.5 Conclusion
test produced promising but unsatisfactory results
due in part to errors in the historical and current
vegetation maps and consumptive demand disturbance probabilities. Predicted runoff from the
MAPSS model was validated over the conterminous United States and was compared to two
runoff maps constructed for the globe and Asia
(Neilson and Marks, 1994; Neilson, 1995). A bias
toward underprediction of runoff by the model, detected in all tests, was ascribed to underestimation
of precipitation in mountainous terrain and increased observed runoff resulting from land cover
conversions (Neilson and Marks, 1994). Vegetation distribution predicted by MAPSS was partially validated over the entire globe using a global
existing vegetation data set (Olson et aI., 1983) and
output from another biogeographic model, BlOME
(Prentice et aI., 1992).
Regional evapotranspiration and net photosynthesis estimates produced by FOREST-BGC were
difficult to validate directly (Running et aI., 1989).
Observed and predicted evapotranspiration were
similar in two watersheds in the western Montana
study area, and comparable water balance simulations were successfully validated against standlevel field data (e.g., timing and magnitude of seasonal leaf water potential and seasonal soil
moisture depletion) (Donner and Running, 1986;
Nemani and Running, 1989; Running et aI., 1989).
Comparison of predicted net photosynthesis with
observed stem volume growth in six sites near the
study area yielded R2 = 0.94 (McLeod and Running, 1988). The CENTURY model was validated
by simulating steady-state soil carbon and nitrogen
levels and aboveground plant production for 24
sites in the Great Plains and then comparing simulation values with mapped plant production and
soil carbon and nitrogen levels at these sites (Parton et aI., 1987). The model adequately represented
the effects of soil texture and climate on soil carbon and nitrogen in the Great Plains and was judged
to have done an excellent job of simulating aboveground plant production. It was not possible to validate the CELSS model because of its highly aggregated and generalized implementation (Sklar et
aI., 1985).
18.5 Conclusion
The models considered in this paper differ in spatial and temporal extent and resolution as well as
in the primary entities simulated (Table 18.1).
Transition matrix models have been constructed for
a wide variety of spatial and temporal scales. Most
267
other ecosystem structure models have been implemented at finer spatial extents and resolutions
than most of the ecosystem function models considered; the ecosystem structure model MAPSS is
a notable exception. Determination of the appropriate variables and hierarchical level of organization to be modeled is critical for successful use of
a simulation model in an ecological assessment
(Allen and Starr, 1982; Sklar and Costanza, 1991).
It may not be straightforward or even possible to
implement a model at a different resolution than
that for which it was developed (Turner et aI.,
1995). Therefore, the level of resolution of the
model should closely match the resolution required
to address the objectives of the assessment, and
modeling results should be explicitly interpreted in
view of the model scale (Baker, 1989a; Hunsaker
et aI., 1993). Models should be selected not only
for appropriate scale, but also for appropriate level
of complexity. The models discussed differ in the
complexity with which processes are represented
and in the incorporation of spatial feedbacks among
model cells (Sklar and Costanza, 1991). For example, successional dynamics can be simulated
with cell to cell interactions using LANDIS or the
individual-based plant model ZELIG (Urban,
1990) or without such spatial interactions using
CRBSUM (Keane et aI., 1996a). Models should be
selected for use in an ecological assessment that
contain just enough complexity to adequately meet
assessment objectives (Holt et aI., 1995).
Lack of necessary data needed to parameterize a
model or provide initial conditions may be an important factor limiting applicability of a model to
a particular problem. When existing data are not
available, implementation of a model may depend
on whether users can afford to conduct studies or
make measurements required to test and run the
model (Turner et aI., 1995). In addition, consideration should be given to the problem of propagating uncertainty in existing data through their use as
model inputs (Hunsaker et aI., 1993). Mapped data
used in spatially explicit models are often generated by interpolation methods based on a finite
number of observations, potentially leading to increased uncertainty in model output (Hunsaker et
aI., 1993).
In conclusion, simulation models may be the
only means for investigating some ecological phenomena (Baker, 1989a). In such cases, careful selection of appropriately scaled, robust (verified and
validated) models is needed to meet the objective
of an ecological assessment and to allow the consequences of alternative management or conservation strategies to be identified.
test produced promising but unsatisfactory results
due in part to errors in the historical and current
vegetation maps and consumptive demand disturbance probabilities. Predicted runoff from the
MAPSS model was validated over the conterminous United States and was compared to two
runoff maps constructed for the globe and Asia
(Neilson and Marks, 1994; Neilson, 1995). A bias
toward underprediction of runoff by the model, detected in all tests, was ascribed to underestimation
of precipitation in mountainous terrain and increased observed runoff resulting from land cover
conversions (Neilson and Marks, 1994). Vegetation distribution predicted by MAPSS was partially validated over the entire globe using a global
existing vegetation data set (Olson et aI., 1983) and
output from another biogeographic model, BlOME
(Prentice et aI., 1992).
Regional evapotranspiration and net photosynthesis estimates produced by FOREST-BGC were
difficult to validate directly (Running et aI., 1989).
Observed and predicted evapotranspiration were
similar in two watersheds in the western Montana
study area, and comparable water balance simulations were successfully validated against standlevel field data (e.g., timing and magnitude of seasonal leaf water potential and seasonal soil
moisture depletion) (Donner and Running, 1986;
Nemani and Running, 1989; Running et aI., 1989).
Comparison of predicted net photosynthesis with
observed stem volume growth in six sites near the
study area yielded R2 = 0.94 (McLeod and Running, 1988). The CENTURY model was validated
by simulating steady-state soil carbon and nitrogen
levels and aboveground plant production for 24
sites in the Great Plains and then comparing simulation values with mapped plant production and
soil carbon and nitrogen levels at these sites (Parton et aI., 1987). The model adequately represented
the effects of soil texture and climate on soil carbon and nitrogen in the Great Plains and was judged
to have done an excellent job of simulating aboveground plant production. It was not possible to validate the CELSS model because of its highly aggregated and generalized implementation (Sklar et
aI., 1985).
18.5 Conclusion
The models considered in this paper differ in spatial and temporal extent and resolution as well as
in the primary entities simulated (Table 18.1).
Transition matrix models have been constructed for
a wide variety of spatial and temporal scales. Most
267
other ecosystem structure models have been implemented at finer spatial extents and resolutions
than most of the ecosystem function models considered; the ecosystem structure model MAPSS is
a notable exception. Determination of the appropriate variables and hierarchical level of organization to be modeled is critical for successful use of
a simulation model in an ecological assessment
(Allen and Starr, 1982; Sklar and Costanza, 1991).
It may not be straightforward or even possible to
implement a model at a different resolution than
that for which it was developed (Turner et aI.,
1995). Therefore, the level of resolution of the
model should closely match the resolution required
to address the objectives of the assessment, and
modeling results should be explicitly interpreted in
view of the model scale (Baker, 1989a; Hunsaker
et aI., 1993). Models should be selected not only
for appropriate scale, but also for appropriate level
of complexity. The models discussed differ in the
complexity with which processes are represented
and in the incorporation of spatial feedbacks among
model cells (Sklar and Costanza, 1991). For example, successional dynamics can be simulated
with cell to cell interactions using LANDIS or the
individual-based plant model ZELIG (Urban,
1990) or without such spatial interactions using
CRBSUM (Keane et aI., 1996a). Models should be
selected for use in an ecological assessment that
contain just enough complexity to adequately meet
assessment objectives (Holt et aI., 1995).
Lack of necessary data needed to parameterize a
model or provide initial conditions may be an important factor limiting applicability of a model to
a particular problem. When existing data are not
available, implementation of a model may depend
on whether users can afford to conduct studies or
make measurements required to test and run the
model (Turner et aI., 1995). In addition, consideration should be given to the problem of propagating uncertainty in existing data through their use as
model inputs (Hunsaker et aI., 1993). Mapped data
used in spatially explicit models are often generated by interpolation methods based on a finite
number of observations, potentially leading to increased uncertainty in model output (Hunsaker et
aI., 1993).
In conclusion, simulation models may be the
only means for investigating some ecological phenomena (Baker, 1989a). In such cases, careful selection of appropriately scaled, robust (verified and
validated) models is needed to meet the objective
of an ecological assessment and to allow the consequences of alternative management or conservation strategies to be identified.