3.5 Characterization by Agglomeration or Subdivision: Bottom-up or Top-down Approach?
43
compositional and structural differences lying on
either side of the boundaries (Rowe, 1996). Differences in interpretation of features can be traced
to the hypotheses, which can be tested (e.g., Uhlig
and Jordan, 1996; see Section 3.4).
3.4 Boundaries
Boundaries are recognized by perceived changes in
the relationships among attributes (e.g., vegetation,
landform, climate) of ecological units. Boundary
recognition may be difficult because of two problems: (1) the continuous or semicontinuous nature
of most ecological attributes, which may be expressed as gradients, and (2) sampling considerations. The first problem leads to indistinct boundaries between ecological units; indeed, such
boundaries are often termedjUzzy. Therefore, clear
rules need to be formulated for the delineation of
lines on a map derived from attributes. Three questions should be addressed (after Zonneveld, 1989):
1. What attributes determine the identity of the
unit?
2. What attributes determine a unit's relevance for
a particular purpose?
3. What attributes determine the patterns of spatial
variability in the unit?
The first question should be answered by precisely defining the objectives of the characterization, which influences the choice of the criteria for
classification and ecological unit mapping (see Section 3.6). Various analytical techniques can then be
used to quantify the explanatory power of a map
with respect to the patterns and processes of interest (see discussion in Chapter 22). The second
question is answered by correlating a specific purpose with one or more of the attributes. For example, if assessment of erosion potential is the purpose of a project, slope is an important attribute.
The third question deals with spatial heterogeneity
among ecological units and becomes more important the higher the unit is in a hierarchy (Zonneveld,
1989, 1994).
The second problem of boundary recognition is
rooted in the fact that ecosystem components (e.g.,
species) generally are not randomly distributed. For
example, if an area contains n different biophysical environments that have sharp boundaries, each
supporting a distinct set of plant species, a sufficiently large sample of randomly distributed adequately sized plots will generate n vegetation
groupings after analysis, each characterized by a
specific species combination (after Zonneveld,
1994). However, in such landscapes, which lack intermediate biophysical environments (ecotones), a
purely random sampling design would be likely to
include plots that overlap the boundary between
two biophysical environments, describing a transitional condition that does not in fact exist. The size
and shape of the plots sampled influence the degree to which this phenomenon affects ecosystem
characterization. The pattern of distribution of
ecosystems and their components affects the sampling design (see Chapters 6 and 7 for complementary views on sampling strategies for ecological assessments).
3.5 Characterization by
Agglomeration or
Subdivision: Bottom-up
or Top-down Approach?
An important decision to be made when conducting ecosystem characterization is whether information should be integrated by agglomeration (also
called typification or bottom-up) or by subdivision
(also called top-down) of the data (Figure 3.1). The
choice of the approach is important because the relationships between levels of ecological hierarchies
are not symmetrical (see Chapter 2). Emergent
properties of higher levels in the hierarchy cannot
be predicted from the properties of lower levels.
This phenomenon must be considered while attempting to extrapolate across scales within an ecological hierarchy (see discussion in Perera et aI.,
1996).
The bottom-up approach is nonspatial (Le., independent from spatial constraints, which would
not necessarily be identified; see Wessman, 1992).
Bottom-up approaches are traditionally used for
taxonomic classification systems and are inductive
(Figure 3.1). There is no relationship to map scale.
In a top-down approach, the context is developed
(i.e., patterns are analyzed), and then function is predicted (i.e., process is inferred; Perera et al., 1996).
Top-down approaches take into consideration spatial patterns and hierarchy (Zonneveld 1989, 1994)
and identify those factors that constrain sublevels
(Wessman, 1992). These approaches are traditionally used for biogeoclimatic classifications (see
Chapter 22) and are deductive. The units defined
may be complex mosaics of ecosystems (Figure 3.1).
Starting from a site-specific ecosystem, the bottomup approach leads to increasingly higher level taxonomic classes. In contrast, the top-down approach
starts from large, ecologically heterogeneous areas
43
compositional and structural differences lying on
either side of the boundaries (Rowe, 1996). Differences in interpretation of features can be traced
to the hypotheses, which can be tested (e.g., Uhlig
and Jordan, 1996; see Section 3.4).
3.4 Boundaries
Boundaries are recognized by perceived changes in
the relationships among attributes (e.g., vegetation,
landform, climate) of ecological units. Boundary
recognition may be difficult because of two problems: (1) the continuous or semicontinuous nature
of most ecological attributes, which may be expressed as gradients, and (2) sampling considerations. The first problem leads to indistinct boundaries between ecological units; indeed, such
boundaries are often termedjUzzy. Therefore, clear
rules need to be formulated for the delineation of
lines on a map derived from attributes. Three questions should be addressed (after Zonneveld, 1989):
1. What attributes determine the identity of the
unit?
2. What attributes determine a unit's relevance for
a particular purpose?
3. What attributes determine the patterns of spatial
variability in the unit?
The first question should be answered by precisely defining the objectives of the characterization, which influences the choice of the criteria for
classification and ecological unit mapping (see Section 3.6). Various analytical techniques can then be
used to quantify the explanatory power of a map
with respect to the patterns and processes of interest (see discussion in Chapter 22). The second
question is answered by correlating a specific purpose with one or more of the attributes. For example, if assessment of erosion potential is the purpose of a project, slope is an important attribute.
The third question deals with spatial heterogeneity
among ecological units and becomes more important the higher the unit is in a hierarchy (Zonneveld,
1989, 1994).
The second problem of boundary recognition is
rooted in the fact that ecosystem components (e.g.,
species) generally are not randomly distributed. For
example, if an area contains n different biophysical environments that have sharp boundaries, each
supporting a distinct set of plant species, a sufficiently large sample of randomly distributed adequately sized plots will generate n vegetation
groupings after analysis, each characterized by a
specific species combination (after Zonneveld,
1994). However, in such landscapes, which lack intermediate biophysical environments (ecotones), a
purely random sampling design would be likely to
include plots that overlap the boundary between
two biophysical environments, describing a transitional condition that does not in fact exist. The size
and shape of the plots sampled influence the degree to which this phenomenon affects ecosystem
characterization. The pattern of distribution of
ecosystems and their components affects the sampling design (see Chapters 6 and 7 for complementary views on sampling strategies for ecological assessments).
3.5 Characterization by
Agglomeration or
Subdivision: Bottom-up
or Top-down Approach?
An important decision to be made when conducting ecosystem characterization is whether information should be integrated by agglomeration (also
called typification or bottom-up) or by subdivision
(also called top-down) of the data (Figure 3.1). The
choice of the approach is important because the relationships between levels of ecological hierarchies
are not symmetrical (see Chapter 2). Emergent
properties of higher levels in the hierarchy cannot
be predicted from the properties of lower levels.
This phenomenon must be considered while attempting to extrapolate across scales within an ecological hierarchy (see discussion in Perera et aI.,
1996).
The bottom-up approach is nonspatial (Le., independent from spatial constraints, which would
not necessarily be identified; see Wessman, 1992).
Bottom-up approaches are traditionally used for
taxonomic classification systems and are inductive
(Figure 3.1). There is no relationship to map scale.
In a top-down approach, the context is developed
(i.e., patterns are analyzed), and then function is predicted (i.e., process is inferred; Perera et al., 1996).
Top-down approaches take into consideration spatial patterns and hierarchy (Zonneveld 1989, 1994)
and identify those factors that constrain sublevels
(Wessman, 1992). These approaches are traditionally used for biogeoclimatic classifications (see
Chapter 22) and are deductive. The units defined
may be complex mosaics of ecosystems (Figure 3.1).
Starting from a site-specific ecosystem, the bottomup approach leads to increasingly higher level taxonomic classes. In contrast, the top-down approach
starts from large, ecologically heterogeneous areas