Multivariate analyses on assemblage data aid in showing the user where distinctions between classes may occur. However, subjectivity in drawing distinctions
among classes is unavoidable, as investigators must decide on the appropriate level
of similarity at which to consider sites within the same class [84]. The final decision
is made as a compromise between including as many reference classes as possible,
while still including enough replicate sites within classes to adequately represent
within-class assemblage variability among sites. Bowman and Somers [85] recommend “a minimum of 20, but preferably 30–50 reference sites per group,” though
this may be overly optimistic given the data constraints experienced in many
studies.
2.3.3 Conclusions Regarding Site Classification
Hawkins et al. [43] and Melles et al. [86] draw distinctions between geographybased methods in which reference classes can be clearly delineated within discrete
spatial units and geography-independent methods driven by patterns in assemblage
variation regardless of physical location. Typological classifications which include
map-based delineations of classes are geography dependent, whereas biotic classifications, focused on patterns of assemblage variation, are geography-independent.
However, the most effective classifications consider both geography-dependent and
geography-independent factors, for example, limitations on the spatial scale over
which biotic classifications are developed can increase their classification strength
[87]. While biotic classifications provide precise descriptions of the patterns of
variability with respect to the assemblage of interest, the resulting classifications
may not be applicable to other assemblages. Inclusion of geography-dependent
variables that implicitly encompass a wide range of environmental factors provides
a more comprehensive classification of sites [88]. A priori typological classification
based on large-scale variables (e.g., ecoregions) provides useful, convenient, and
easily communicated initial classifications of sites, though classifications are often
improved when supplemented by smaller-scale variables that are not spatially
discrete (e.g., flow regime; [89]) or not associated with geography (e.g., sampling
date; [75]). Good scientific practice requires that the effectiveness of a priori
approaches be evaluated with a posteriori evaluations of relationships between
classes and biota [64–67, 88].
3 Predictive Modeling of Aquatic Assemblages
The objective of predictive modeling for bioassessment is to control for the
confounding effects of natural environmental variables so that the effects of
stressors on metrics can be clearly evaluated. The methods used to meet this
objective are as diverse and varied as the assemblages themselves. As an introduction to the core concepts in predictive modeling, we outline the basic steps of the
244
A.L. Garey and L.A. Smock
among classes is unavoidable, as investigators must decide on the appropriate level
of similarity at which to consider sites within the same class [84]. The final decision
is made as a compromise between including as many reference classes as possible,
while still including enough replicate sites within classes to adequately represent
within-class assemblage variability among sites. Bowman and Somers [85] recommend “a minimum of 20, but preferably 30–50 reference sites per group,” though
this may be overly optimistic given the data constraints experienced in many
studies.
2.3.3 Conclusions Regarding Site Classification
Hawkins et al. [43] and Melles et al. [86] draw distinctions between geographybased methods in which reference classes can be clearly delineated within discrete
spatial units and geography-independent methods driven by patterns in assemblage
variation regardless of physical location. Typological classifications which include
map-based delineations of classes are geography dependent, whereas biotic classifications, focused on patterns of assemblage variation, are geography-independent.
However, the most effective classifications consider both geography-dependent and
geography-independent factors, for example, limitations on the spatial scale over
which biotic classifications are developed can increase their classification strength
[87]. While biotic classifications provide precise descriptions of the patterns of
variability with respect to the assemblage of interest, the resulting classifications
may not be applicable to other assemblages. Inclusion of geography-dependent
variables that implicitly encompass a wide range of environmental factors provides
a more comprehensive classification of sites [88]. A priori typological classification
based on large-scale variables (e.g., ecoregions) provides useful, convenient, and
easily communicated initial classifications of sites, though classifications are often
improved when supplemented by smaller-scale variables that are not spatially
discrete (e.g., flow regime; [89]) or not associated with geography (e.g., sampling
date; [75]). Good scientific practice requires that the effectiveness of a priori
approaches be evaluated with a posteriori evaluations of relationships between
classes and biota [64–67, 88].
3 Predictive Modeling of Aquatic Assemblages
The objective of predictive modeling for bioassessment is to control for the
confounding effects of natural environmental variables so that the effects of
stressors on metrics can be clearly evaluated. The methods used to meet this
objective are as diverse and varied as the assemblages themselves. As an introduction to the core concepts in predictive modeling, we outline the basic steps of the
244
A.L. Garey and L.A. Smock