266
to decomposition (200- to 1500-year turnover time)
(Parton et aI., 1987). Different vegetation submode1s are used in conjunction with the soil organic
matter submodeI.
The model runs on a monthly time step, although
a hydrologic version under development will run
daily (Parton et al., 1994). Input variables include
monthly mean maximum and minimum air temperature, precipitation, lignin content of plant material,
plant nitrogen, phosphorus, and sulfur content, soil
texture, atmospheric and soil nitrogen inputs, and
initial soil nutrient levels (Metherell et al., 1993). A
gridded version of the model allows spatial representation (Schimel et aI., 1996). Disturbances such
as grazing, fIre, and storms have been incorporated
(Sanford et al., 1991; Holland et al., 1992). The
model is limited for some applications because vegetation type is considered only indirectly and
changes in species composition cannot be predicted.
CENTURY can be joined with a number of other
models to address more comprehensive ecosystem
properties. It has been linked with a mesoscale atmospheric model, RAMS (Pielke et aI., 1992), to
better understand the feedbacks between land surface and atmospheric responses to climate. The
simulation model MCI incorporates biogeochemical processes from CENTURY with vegetation distribution functions from MAPSS (see Section
18.2.7) to represent vegetation and ecosystem
process responses to changing climate (Lenihan et
aI., 1998; Daly et aI., 2000).
18.3.4 CELSS
The CELSS model is a process-based wetland simulation model implemented for a marsh-estuarine
system in southern Louisiana (Sklar et aI., 1985;
Sklar and Costanza, 1986, 1991). Each of nearly
2500 interconnected l-km 2 cells contains a simulation model with eight state variables. Cells are
connected through exchange of water and materials. Each cell is assigned to one of seven possible
habitat types. If environmental conditions change
sufficiently such that the environment is inappropriate for the assigned habitat type, succession occurs by switching habitat-type parameters to a new,
more representative set of parameters. Interactions
among water storage and connectivity (a function
of habitat type, drainage density, waterway orientation, and levee height) at cell boundaries influence sediment deposition and erosion, which are
believed to be important in habitat succession and
productivity of the area (Sklar and Costanza, 1991).
The model simulated habitat changes over the
period 1956 to 1983, correctly predicting 83% of
Ecosystem Structure and Function Modeling
1983 verifIcation data. A range of past and future
climate scenarios and management scenarios was
also modeled (Costanza et aI., 1990). The advantages of using a model of this type are the explicit
incorporation of spatial interactions and the linkage of cause and effect by mechanistic processes.
The relatively coarse spatial and temporal scales
and computational requirements may limit the use
of the model for some applications (Sklar and
Costanza, 1991).
18.4 Validation of Models
Model performance must be evaluated in relation to
its purpose to assess the applicability of the results
to the problem at hand (Grant, 1988; Hunsaker et
aI., 1993). This is particularly important when model
output is used to help solve management problems
(Bart, 1995). Models should be validated by comparing their outputs with relevant data not used in
model construction or parameterization. However,
such data may not be available, particularly for models simulating broad spatial and temporal scales. The
models described in this chapter differ in the extent
to which they have been validated.
The variety of validation tests to determine the
reliability of individual-based plant models has included prediction of independent tree diameter increment data, forestry yield tables, change in
species composition along environmental gradients, and reconstruction of vegetation under past
climates (see Shugart and Smith, 1996, Table 18.1,
for references). A recent version of ZELIG implemented in the western Cascades, Oregon, was
tested by comparing modeled tree species basal
area over a 500-year period with extensive field
data (Hansen et aI., 1995). However, no independent data were available to validate model functions simulating bird species densities in this
model. Validation of the gradient model for Glacier National Park was conducted for overstory
species composition, in which the percent similarity between observed and predicted stand composition was greater than 70% for 91 % of test stands
(Kessell, 1979). In addition, very accurate predictions of fire spread rate and fire perimeters were
obtained in tests of fire behavior for three fires.
Behavior of CRBSUM was validated using
"known" conditions in which a historical vegetation-cover-type map was used as input to run the
model for 100 years under a consumptive demand
management scenario (Keane et aI., 1996a). The
output cover-type map was compared with a current vegetation-cover-type map. As described, the
to decomposition (200- to 1500-year turnover time)
(Parton et aI., 1987). Different vegetation submode1s are used in conjunction with the soil organic
matter submodeI.
The model runs on a monthly time step, although
a hydrologic version under development will run
daily (Parton et al., 1994). Input variables include
monthly mean maximum and minimum air temperature, precipitation, lignin content of plant material,
plant nitrogen, phosphorus, and sulfur content, soil
texture, atmospheric and soil nitrogen inputs, and
initial soil nutrient levels (Metherell et al., 1993). A
gridded version of the model allows spatial representation (Schimel et aI., 1996). Disturbances such
as grazing, fIre, and storms have been incorporated
(Sanford et al., 1991; Holland et al., 1992). The
model is limited for some applications because vegetation type is considered only indirectly and
changes in species composition cannot be predicted.
CENTURY can be joined with a number of other
models to address more comprehensive ecosystem
properties. It has been linked with a mesoscale atmospheric model, RAMS (Pielke et aI., 1992), to
better understand the feedbacks between land surface and atmospheric responses to climate. The
simulation model MCI incorporates biogeochemical processes from CENTURY with vegetation distribution functions from MAPSS (see Section
18.2.7) to represent vegetation and ecosystem
process responses to changing climate (Lenihan et
aI., 1998; Daly et aI., 2000).
18.3.4 CELSS
The CELSS model is a process-based wetland simulation model implemented for a marsh-estuarine
system in southern Louisiana (Sklar et aI., 1985;
Sklar and Costanza, 1986, 1991). Each of nearly
2500 interconnected l-km 2 cells contains a simulation model with eight state variables. Cells are
connected through exchange of water and materials. Each cell is assigned to one of seven possible
habitat types. If environmental conditions change
sufficiently such that the environment is inappropriate for the assigned habitat type, succession occurs by switching habitat-type parameters to a new,
more representative set of parameters. Interactions
among water storage and connectivity (a function
of habitat type, drainage density, waterway orientation, and levee height) at cell boundaries influence sediment deposition and erosion, which are
believed to be important in habitat succession and
productivity of the area (Sklar and Costanza, 1991).
The model simulated habitat changes over the
period 1956 to 1983, correctly predicting 83% of
Ecosystem Structure and Function Modeling
1983 verifIcation data. A range of past and future
climate scenarios and management scenarios was
also modeled (Costanza et aI., 1990). The advantages of using a model of this type are the explicit
incorporation of spatial interactions and the linkage of cause and effect by mechanistic processes.
The relatively coarse spatial and temporal scales
and computational requirements may limit the use
of the model for some applications (Sklar and
Costanza, 1991).
18.4 Validation of Models
Model performance must be evaluated in relation to
its purpose to assess the applicability of the results
to the problem at hand (Grant, 1988; Hunsaker et
aI., 1993). This is particularly important when model
output is used to help solve management problems
(Bart, 1995). Models should be validated by comparing their outputs with relevant data not used in
model construction or parameterization. However,
such data may not be available, particularly for models simulating broad spatial and temporal scales. The
models described in this chapter differ in the extent
to which they have been validated.
The variety of validation tests to determine the
reliability of individual-based plant models has included prediction of independent tree diameter increment data, forestry yield tables, change in
species composition along environmental gradients, and reconstruction of vegetation under past
climates (see Shugart and Smith, 1996, Table 18.1,
for references). A recent version of ZELIG implemented in the western Cascades, Oregon, was
tested by comparing modeled tree species basal
area over a 500-year period with extensive field
data (Hansen et aI., 1995). However, no independent data were available to validate model functions simulating bird species densities in this
model. Validation of the gradient model for Glacier National Park was conducted for overstory
species composition, in which the percent similarity between observed and predicted stand composition was greater than 70% for 91 % of test stands
(Kessell, 1979). In addition, very accurate predictions of fire spread rate and fire perimeters were
obtained in tests of fire behavior for three fires.
Behavior of CRBSUM was validated using
"known" conditions in which a historical vegetation-cover-type map was used as input to run the
model for 100 years under a consumptive demand
management scenario (Keane et aI., 1996a). The
output cover-type map was compared with a current vegetation-cover-type map. As described, the