3.7 Definition and Prediction of Ecosystem Properties
ecosystem over its known range of distribution.
These assumptions may not be met.
The biotic component of an ecosystem is characterized by the link between distribution patterns
of individual species. their occurrence in landscape
features. and the distribution of landscape features
(see discussion in Bourgeron et aI.. 1994a). Therefore. the distribution pattern of landscape features
constrains the frequency and pattern of species distributions in the landscape. Attempts to characterize the distribution of species by using ecological
classifications and mapping units can be successful only if the pattern of gradual change in individual species distributions is correlated with gradual changes in the environment at relevant scales.
Ecosystem characterization should not assume that
all occurrences of an ecological unit have identical
properties with respect to a species requirement.
Species distributions should be established with
data that span the range of environmental variability over which they are distributed. Use of landscape classifications to predict species distributions
include the Cherrill et al. (1995) model that explicitly orders ecological information into a hierarchical series of matrices representing the relationships between successive levels of spatial
resolution and ecological organization. Despite
limitations (e.g .• poor prediction of scarce species.
difficulties in testing landscape-scale models). this
approach was successful for common species and
provided insight into the consequences of land-use
change. He et aI. (1998) used a probabilistic algorithm to assign information from forest inventory
sampling points and ecoregional boundaries to a
satellite-based land cover classification map for regional forest assessment. New maps of tree species
distributions and stand age were derived that reflected differences at the ecoregional scale. The inventory data provided important secondary information on age class and secondary species not
available from the remotely sensed data.
Current theory regarding species diversity patterns has limited predictive power. Empirical relationships have been developed at one of three
scales: global (Whittaker. 1972). regional (Pielou.
1979; Brown. 1984). or local (Grime. 1979; Woodward. 1987). Models relating diversity to disturbance (e.g., Huston. 1979. 1994) do not have sitespecific predictive power. The cumulative impact
of niche relations. habitat diversity. mass effects
(the flow of individuals from favorable to unfavorable areas), and ecological equivalency (the fact
that different species may be ecologically equivalent to each other) on diversity has been summarized in a muItiscale context (Shmida and Wilson.
47
1985). Recent work has focused on relationships
among hierarchical levels (e.g., Ricklefs. 1987; Caley and Schluter. 1997; Angermeier and Winston.
1998). The dependence of local diversity on regional patterns has been shown for some biota
(Ricklefs. 1987). but not for others (Jackson and
Harvey. 1989). Diversity at one scale may have
complex relationships to structure, processes, and
disturbances at other scales. Neilson et aI. (1989)
suggested that the prediction of local diversity patterns should be rooted in an understanding of the
hierarchy of constraints imposed by regional and
local factors. as well as of their interactions. The
spatial and temporal context provided by muItiscaled ecosystem characterization should be useful
for analysis of biotic diversity (Whittaker. 1972;
Hoover and Parker. 1991).
Progress has been made in developing empirical
relationships between diversity and environment
using predictive statistical models. Useful relationships (Margules et al .• 1987; Nicholls. 1991a. b;
Austin et aI.. 1996) generally involve more than
one environmental variable. These relationships
should be derived from survey data and probably
cannot be extended beyond the bounds of the data
(Margules et aI.. 1987). However. site-specific predictions can be made for particular study areas. The
use of hierarchical ecosystem characterization
schemes could provide the appropriate stratification for extrapolating site-specific results.
Problems of scale interactions and model generalization beset predictions of ecosystem primary
production. Spatially explicit models of biogeochemistry (see Chapter 18) have been developed
using a combination of geographic information systems (GIS) and terrestrial regional ecosystem models (Houghton et aI., 1983; Burke et aI.. 1990; Band
et aI.. 1993), but their widespread use is limited by
gaps in soil, climate, and vegetation databases
(Stewart et aI.. 1989). Models like CENTURY
(Parton et aI.. 1987) and BIOME-BGC (Running
and Hunt, 1993; Hunt et aI., 1996) explicitly link
abiotic and biogeochemical factors with primary
production and carbon storage (see also Schimel et
aI., 1990). The problem of predicting ecosystem
processes is an active research field (Schimel et aI..
1997). For example, VEMAP members (1995)
compared simulations of net primary production
(NPP) produced by three biogeochemistry models
(CENTURY. BIOME-BCG. and TEM) for the conterminous United States under current climate conditions and a range of climate change scenarios.
The models estimated similar continental-scale
NPP values under current conditions. Schimel et al.
(1991) summarized current activities and stressed
ecosystem over its known range of distribution.
These assumptions may not be met.
The biotic component of an ecosystem is characterized by the link between distribution patterns
of individual species. their occurrence in landscape
features. and the distribution of landscape features
(see discussion in Bourgeron et aI.. 1994a). Therefore. the distribution pattern of landscape features
constrains the frequency and pattern of species distributions in the landscape. Attempts to characterize the distribution of species by using ecological
classifications and mapping units can be successful only if the pattern of gradual change in individual species distributions is correlated with gradual changes in the environment at relevant scales.
Ecosystem characterization should not assume that
all occurrences of an ecological unit have identical
properties with respect to a species requirement.
Species distributions should be established with
data that span the range of environmental variability over which they are distributed. Use of landscape classifications to predict species distributions
include the Cherrill et al. (1995) model that explicitly orders ecological information into a hierarchical series of matrices representing the relationships between successive levels of spatial
resolution and ecological organization. Despite
limitations (e.g .• poor prediction of scarce species.
difficulties in testing landscape-scale models). this
approach was successful for common species and
provided insight into the consequences of land-use
change. He et aI. (1998) used a probabilistic algorithm to assign information from forest inventory
sampling points and ecoregional boundaries to a
satellite-based land cover classification map for regional forest assessment. New maps of tree species
distributions and stand age were derived that reflected differences at the ecoregional scale. The inventory data provided important secondary information on age class and secondary species not
available from the remotely sensed data.
Current theory regarding species diversity patterns has limited predictive power. Empirical relationships have been developed at one of three
scales: global (Whittaker. 1972). regional (Pielou.
1979; Brown. 1984). or local (Grime. 1979; Woodward. 1987). Models relating diversity to disturbance (e.g., Huston. 1979. 1994) do not have sitespecific predictive power. The cumulative impact
of niche relations. habitat diversity. mass effects
(the flow of individuals from favorable to unfavorable areas), and ecological equivalency (the fact
that different species may be ecologically equivalent to each other) on diversity has been summarized in a muItiscale context (Shmida and Wilson.
47
1985). Recent work has focused on relationships
among hierarchical levels (e.g., Ricklefs. 1987; Caley and Schluter. 1997; Angermeier and Winston.
1998). The dependence of local diversity on regional patterns has been shown for some biota
(Ricklefs. 1987). but not for others (Jackson and
Harvey. 1989). Diversity at one scale may have
complex relationships to structure, processes, and
disturbances at other scales. Neilson et aI. (1989)
suggested that the prediction of local diversity patterns should be rooted in an understanding of the
hierarchy of constraints imposed by regional and
local factors. as well as of their interactions. The
spatial and temporal context provided by muItiscaled ecosystem characterization should be useful
for analysis of biotic diversity (Whittaker. 1972;
Hoover and Parker. 1991).
Progress has been made in developing empirical
relationships between diversity and environment
using predictive statistical models. Useful relationships (Margules et al .• 1987; Nicholls. 1991a. b;
Austin et aI.. 1996) generally involve more than
one environmental variable. These relationships
should be derived from survey data and probably
cannot be extended beyond the bounds of the data
(Margules et aI.. 1987). However. site-specific predictions can be made for particular study areas. The
use of hierarchical ecosystem characterization
schemes could provide the appropriate stratification for extrapolating site-specific results.
Problems of scale interactions and model generalization beset predictions of ecosystem primary
production. Spatially explicit models of biogeochemistry (see Chapter 18) have been developed
using a combination of geographic information systems (GIS) and terrestrial regional ecosystem models (Houghton et aI., 1983; Burke et aI.. 1990; Band
et aI.. 1993), but their widespread use is limited by
gaps in soil, climate, and vegetation databases
(Stewart et aI.. 1989). Models like CENTURY
(Parton et aI.. 1987) and BIOME-BGC (Running
and Hunt, 1993; Hunt et aI., 1996) explicitly link
abiotic and biogeochemical factors with primary
production and carbon storage (see also Schimel et
aI., 1990). The problem of predicting ecosystem
processes is an active research field (Schimel et aI..
1997). For example, VEMAP members (1995)
compared simulations of net primary production
(NPP) produced by three biogeochemistry models
(CENTURY. BIOME-BCG. and TEM) for the conterminous United States under current climate conditions and a range of climate change scenarios.
The models estimated similar continental-scale
NPP values under current conditions. Schimel et al.
(1991) summarized current activities and stressed