PLACE IN RETURN Box
to remove this checkout from your record.
TO AVOID FINE return on or before date due.

 

 

MTE DUE

MTE DUE

DATE DUE

 

“”593 30051

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/” COMM.“

 

 

 

 

MULTIF ACTOR CLASSIFICATION OF ECOLOGICAL LAND UNITS
IN NORTHEASTERN LOWER MICHIGAN

By

David Thomas Cleland

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Forestry

1997

ABSTRACT
MULTIFACTOR CLASSIFICATION OF ECOLOGICAL LAND UNITS
IN NORTHEASTERN LOWER MICHIGAN
By

David Thomas Cleland

An integrated classification of ecological land units was developed in northeastern
lower Michigan based on associations of understory and ground-flora species, and key
soil morphological and physical characteristics. A series of complementary multivariate,
graphical, and tabular analyses of floristic and soils data were utilized in the classification
process. The utility of eliminating understory and ground-flora species in the
classification of multifactor ecological land units using a geometric interpretation of a
correspondence analysis was assessed. Species whose squared cosines summed to less
than 45% in the first five dimension of a correspondence analysis were eliminated from
the floristic data set, and a reduced data set was re-analyzed and compared to earlier
results.

Correspondence analysis and detrended correspondence analysis were robust to
the inclusion of low display quality species, and both procedures were effective in
distinguishing ecological land units in commonly occurring and high quality species
space. The use of high quality species improved the consistency in the number of groups
derived fi'om three agglomerative and one divisive clustering procedure, and the
assignment of samples to groups when compared to results based on commonly occurring

species. The utility of eliminating low display quality species in cluster analyses was

corroborated in the reanalysis of data used to formulate a second independently derived
classification by Host (1987).

Principal component analysis of soils data identified important variables used in
differentiating ecological land units, and produced ordination diagrams that grouped
samples into meaningful classes of soil, understory, and ground-flora conditions.
Agglomerative clustering of samples based on soil variables failed to identify the same
groups as ordination space partitioning of samples in soil variable space, or as the
ordination and clustering of samples based on ground-flora.

The continuous distribution of samples ordinated in species or soil variable space
required use of clustering and tabular analyses, and comparisons of floristic and soil
analyses to effectively partition regions containing different ecological land units. The
iterative application of a series of complementary ordination, clustering, graphical, and
tabular analyses was effective in defining multifactor ecological land units for

northeastern lower Michigan.

This is dedicated to my parents, John and Agda Cleland,
and to my wife and daughters Cathy, Alyssa, and Heather.

ACKNOWLEDGMENTS

I would like to express my gratitude to a number of faculty at Michigan State
University. My advisor Dr. James B. Hart encouraged me to enter into a Ph.D. program,
and ensured that my program met the rigor and caliber of research necessary to fulfill the
requirements of the university and the degree. I would like to acknowledge my
committee members, Drs. Don Dickman, Doug Lantagne, and Rique Campa for their
encouragement, patience, and careful review of my dissertation. I would especially like
to thank Dr. Carl Ramm for his assistance and guidance in the appropriate use and
interpretation of multivariate methods. I would also like to thank Drs. Larry Leefers and
Michael Vasievich, who provided emotional support throughout my program.

I would like to acknowledge Dr. Burton Barnes of the University of Michigan for
his research leadership in multifactor ecological land classification. James Jordan of the
Ottawa National Forest translated research concepts into practical classification and
mapping systems that have been applied throughout the Lake States.

I would like to thank a number of Forest Service employees for their assistance
and support in my research program. Drs. Thomas Crow and Forest Stearns provided
constant support and inspiration while demonstrating that qualities of greatness are
measured first by a person’s compassion and humility, then their innate ability, discipline,
and accomplishment. Dr. Eunice Padley, Jeff Bruggink, and Pete Greaney assisted in

ii

field sampling and in interpreting ecological relationships within the study area, and were
great allies in applying ecological land classification in natural resource and ecosystem
management. Richard Watson taught me to question accepted approaches to soil survey
and ecological land classification, and shared his considerable insights into the nature of
forest ecosystems in Michigan.

Finally, I would like to thank my family for their support and sacrifice while I was
taking classes and writing the dissertation. My wife Cathy and daughters Alyssa and
Heather all shared the strain of my program. My father John (Bill) Cleland, who passed
away shortly after my comprehensive written and oral examinations, expressed his
fatherly pride by silently allowing me to concentrate on academic needs instead of family

responsibilities.

iii

TABLE OF CONTENTS

ACKNOWLEDGMENTS ......................................................................................................... ii
LIST OF TABLES ................................................................................................................ vi
LIST OF FIGURES ............................................................................................................... viii
1. INTRODUCTION ........................................................................................................ 1
Research goals ......................................................................................................................... 1
Problem statement ..................................................................................................................... 3
Alternative F orestland Classification Systems .......................................................................... 8
Soil Survey .................................................................................................................... 9
Vegetation Classifications ........................................................................................... 10
Site Index .................................................................................................................... 14
Multifactor Ecological Land Classification ................................................................ 14
Literature review of ecological classification methods ...................................................... l7
Ordination ................................................................................................................ 18
Principal Component Analysis ................................................................................ 20
Correspondence Analysis ..................................... , .................................................. 22
Detrended Correspondence Analysis ...................................................................... 25
Clustering ................................................................................................................ 27

II. ECOLOGICAL LAND UNITS FOR UPLAND FOREST ECOSYSTEMS OF
NORTHEASTERN LOWER MICHIGAN

Introduction .......................................................................................................................... 31
Objectives .......................................................................................................................... 36
Materials and Methods ......................................................................................................... 36
Sample Design and Field Methods ............................................................................ 37

Stand Methods ............................................................................................. 38

Ground-flora Methods ................................................................................. 39

Soil Methods ............................................................................................... 41

Analysis of vegetation ............................................................................................ 43
Analysis of soils data ............................................................................................. 43

iv

Reanalysis of vegetation ........................................................................................ 45

Results ................................................................................................................................. 46
Analysis of Commonly Occurring Understory and Ground-Flora species ................. 46
Analysis of Soils Data ............................................................................................. 54
Analysis of High Quality Understory and Ground-flora Species ............................... 75
Classification of Species ......................................................................................... 84
Final Classification of Samples and Species based upon Combined Ground-flora and
Soil factors ............................................................................................................ 88
Ecological Land Unit Descriptions ............................................................................. 96

Interpretation of Ordination and Classification of Samples Based on Soils Data... 106
Interpretation of Ordinations of Samples in Understory and Ground-flora Species

Space ....................................................................................................................... 1 12
Comparison of Ordination and Clustering Results .............................................................. 120
Comparison of Correspondence Analysis and Detrended Correspondence Analysis
for Ecological Land Units in Northeastern Lower Michigan .................................... 120
Comparison of Correspondence Analysis and Detrended Correspondence Analysis
for Ecological Land Units in Northwestern Lower Michigan .................................. 129
Comparison of Clustering Methods for Ecological Land Units in Northeastern Lower
Michigan .................................................................................................................. 134
Comparison of Clustering Methods for Ecological Land Units in Northwestern
Lower Michigan............ ............................................................................................ 144
Discussion .......................................................................................................................... 156
Management Implications of Ecological Land Classification .............................. 158
III. SUMMARY AND CONCLUSIONS ........................................................................... 165

Summary of Vegetative Analyses in Northeast Lower Michigan - Ordination... l 69
Summary of Vegetative Analyses in Northeast Lower Michigan - Clustering ..... 170
Summary of Analysis of Soils Data in Northeast Lower Michigan - Ordination..172
Summary of Analysis of Soils Data in Northeast Lower Michigan - Clustering. l 74
Summary of Use of High Quality Species in the Classification of Ecological Land

Units in Northwestern Lower Michigan ............................................................... 175
LITERATURE CITED ...................................................................................................... 178
APPENDICES .................................................................................................................... 191

Appendix A ........................................................................................................... 191

LIST OF TABLES

Table 2.1. Cover-abundance classes and ranks used in field sampling and

data analysis ....................................................................................................................... 40
Table 2.2. Field texture codes for soil variables .................................................................. 42
Table 2.3. Soil drainage class codes .................................................................................... 42
Table 2.4. Inertia and Chi-Square decomposition of the first fifteen

principal coordinates ........................................................................................................... 47
Table 2.5. Eigenvalues of the soil correlation matrix .......................................................... 55
Table 2.6. Soil variable names and abbreviations .............................................................. 57

Table 2.7. Soil variables ordered by variable weights in the first dimension of a
PCA .................................................................................................................................... 59

Table 2.8. Soil variables ordered by variable weights in the second dimension of a
PCA ..................................................................................................................................... 60

Table 2.9. Factor patterns for soil variables ordered by variable loadings in the
first dimension of a PCA ......................................................................................................... 61

Table 2.10. Factor patterns for key soil variables ordered by variable loadings in the

second dimension of a PCA .................................................................................................... 62
Table 2.11. Soils table arraying key variables for samples grouped by Ward’s linkage ........ 66
Table 2.12. Soils table arraying key variables for samples grouped by average linkage ....... 70

Table 2.13. Soil table arraying key variables for samples ordered by sample weights
in the first dimension of a PCA of soils data ......................................................................... 73

vi

Table 2.14. Soil table arraying key variables for samples ordered by sample weights

in the first dimension of a CA of vegetation data .................................................................... 74
Table 2.15. Sums of squared cosines in five dimensional subspace, mass, and inertia

for high quality species ............................................................................................................ 76
Table 2.16. Correspondence analysis: inertia and Chi-Square decomposition of the

first fifteen principal coordinates ............................................................................................. 77
Table 2.17. TWINSPAN classification of high quality Species and samples ...................... 86
Table 2.18. Key soil variables for ecological land units ...................................................... 90
Table 2.19. Ecological species groups for ecological land units ........................................ 91
Table 2.20. Overstory composition within ecological land units ....................................... 93

vii

LIST OF FIGURES

Figure 2.1. Study area on the Huron NF in Alcona, Iosco, Oscoda, and
Ogemaw counties .................................................................................................................... 37

Figure 2.2. Correspondence analysis in CA dimensions 1 and 2 labeled by
sample number .................................................................................................................... 46

Figure 2.3. Correspondence analysis in CA dimensions 1 and 2 of commonly
occurring species space labeled by soil texture of upper 30 cm ............................................. 48

Figure 2.4. Correspondence analysis in CA dimensions 1 and 2 of commonly
occurring species space labeled by substrate textures at 100 to 250 cm ............................... 48

Figure 2.5. Exploratory ordination space partitioning of samples in CA dimensions 1
and of commonly occurring species space using average linkage ...................................... 50

Figure 2.6. Exploratory ordination space partitioning of samples in CA dimensions 1
and of commonly occurring species space using flexible beta linkage ................................... 51

Figure 2.7. Exploratory ordination space partitioning of samples in CA dimensions 1
and of commonly occurring species space using Ward’s linkage ...................................... 52

Figure 2.8. Exploratory ordination space partitioning of samples in CA dimensions 1

and of commonly occurring species space using TWINSPAN .............................................. 53
Figure 2.9. Mean and standard deviation 35 soil variables ................................................. 54
Figure 2.10. Loglo of mean and standard deviation for 35 soil variables ........................... 54

Figure 2.11. Samples in PCA dimensions 1 and 2 of soil variable space labeled by texture
of upper 30 cm .................................................................................................................... 63

Figure 2.12. Samples in PCA dimensions 1 and 2 of soil variable space labeled labeled
by texture of upper 100 cm ...................................................................................................... 63

viii

Figure 2.13. Samples in PCA dimensions 1 and 2 of soil variable space labeled labeled

by texture at 150 to 250 cm ..................................................................................................... 64
Figure 2.14. Samples in PCA dimensions 1 and 2 of soil variable space labeled labeled

by drainage class ...................................................................................................................... 64
Figure 2.15. Sample clusters based on 28 soil variables using Ward’s linkage .................. 66
Figure 2.16. Sample clusters based on 28 soil variables using average linkage .................. 69
Figure 2.17. Clustering of samples based on high quality species: beta linkage ................ 79
Figure 2.18. Clustering of samples based on high quality species: Ward’s linkage ........... 80
Figure 2.19. Clustering of samples based on high quality species: average linkage .......... 81

Figure 2.20. Samples in CA dimensions 1 and 2 of high quality species space
partitioned by clusters identified using Ward’s and average linkage ................................... 82

Figure 2.21 .Samples in CA dimensions 1 and 3 of high quality species space

partitioned by clusters identified using Ward’s and average linkage ..................................... 83
Figure 2.22. High quality species in CA dimensions 1 and 2 of sample space .................. 84
Figure 2.23. High quality species CA dimensions 1 and 3 of sample space

partitioned by TWINSPAN clusters ........................................................................................ 87
Figure 2.24. Samples in CA dimensions 1 and 3 of high quality species space

labeled by ELU ........................................................................................................................ 94
Figure 2.25. Samples in PCA dimensions 1 and 2 of soil variable space

labeled by ELU ........................................................................................................................ 95
Figure 2.26. Ecological species groups in CA dimensions 1 and 3 of sample space ........ 105

Figure 2.27. Samples in PCA dimensions 1 and 3 of soil variable space
labeled by ELU ................................................................................................................... 109

Figure 2.28. Sample clusters based on 28 soil variables using Ward’s linkage
labeled by ELU ...................................................................................................................... 110

Figure 2.29. Sample clusters based on 28 soil variables using average linkage
labeled by ELU .................................................................................................................. 1 l 1

ix

Figure 2.30. Samples in dimensions CA 1 and 2 of commonly occurring
Species space labeled by ELU ............................................................................................... 1 13

Figure 2.31. Samples in dimensions CA 1 and 2 of commonly occurring
species space labeled by upper 100 cm. soil texture ............................................................ 113

Figure 2.32. Samples in dimensions CA 1 and 2 of commonly occurring
species space labeled by lower 150 cm. soil texture ............................................................. 113

Figure 2.33. Samples in dimensions CA 1 and 2 of commonly occurring
species space labeled by soil drainage class .......................................................................... 113

Figure 2.34. Correlations of samples labeled by ELU with dimensions 1
and 2 of CA ........................................................................................................................ 115

Figure 2.35. Samples in CA dimensions 1 and 3 of commonly occurring
species space labeled by ELU ............................................................................................... 118

Figure 2.36. Samples in dimensions 1 and 3 of species space labeled by top
100 cm. soil texture ............................................................................................................... 118

Figure 2.37. Samples in dimensions 1 and 3 of species space labeled by lower
150 cm. soil texture ............................................................................................................... 1 18

Figure 2.38. Samples in dimensions 1 and 3 of species Space labeled by upper

30 cm. soil texture ................................................................................................................. 118
Figure 2.39. Tree species in dimensions 1 and 2 of sample space .................................... 119
Figure 2.40. Tree species in dimensions 1 and 3 of sample space ....................................... 119

Figure 2.41. Samples in CA dimensions 1 and 2 of tree species space
labeled by ELU ................................................................................................................... 1 19

Figure 2.42. Samples in CA dimensions 1 and 3 of tree species space
labeled by ELU ................................................................................................................. 119

Figure 2.43. Comparison of CA and DCA for commonly occurring species in
NE MI: dimensions 1 and 2 ................................................................................................ 122

Figure 2.44. Comparison of CA and DCA for commonly occurring species in
NE MI: dimensions 1 and 3 ................................................................................................ 122

Figure 2.45. Comparison of CA and DCA for high quality species in NE MI:
dimensions 1 and 2 ................................................................................................................ 124

Figure 2.46. Comparison of CA and DCA for high quality species in NE MI:
dimensions 1 and 3 ................................................................................................................ 124

Figure 2.47. Comparison of CA of commonly occurring and high quality species
in NE MI: dimensions 1 and 2 ............................................................................................. 125

Figure 2.48. Comparison of CA of commonly occurring and high quality species
in NE MI: dimensions 1 and 3 .............................................................................................. 125

Figure 2.49. Comparison of DCA of commonly occurring and high quality species
inNE MI: dimensions 1 and 2 ......................................................................................... 128

Figure 2.50. Comparison of DCA of commonly occurring and high quality species
in NE MI: dimensions 1 and 3 ........................................................................................... 128

Figure 2.51. CA for commonly occurring species in NW MI: dimensions 1 and 2
labeled by ELU as defined by Host (1987) ........................................................................... 130

Figure 2.52. DCA for commonly occurring species in NW MI: dimensions 1 and 2
labeled by ELU as defined by Host (1987) .......................................................................... 130

Figure 2.53. CA for commonly occurring species in NW MI: dimensions 1 and 3
labeled by ELU as defined by Host (1987) ........................................................................... 130

Figure 2.54. DCA for commonly occurring species in NW MI: dimensions 1 and 3
labeled by ELU as defined by Host (1987) ...................................................................... 130

Figure 2. 55. CA for high quality species in NW MI: dimensions 1 and 2 labeled
by ELU as defined by Host (1987) .......................................................... ; ........................... 133

Figure 2.56. DCA for high quality species in NW MI: dimensions 1 and 2 labeled
by ELU as defined by Host (1987) ....................................................................................... 133

Figure 2.57. CA for high quality species in NW MI: dimensions 1 and 3 labeled
by ELU as defined by Host (1987) ........................................................................................ 133

Figure 2.58. DCA for high quality species in NW MI: dimensions 1 and 3 labeled
by ELU as defined by Host (1987) ................................................................................... 133

xi

Figure 2.59. Clustering of samples labeled by ELU based on high quality species

in NE MI: average linkage .................................................................................................. 136
Figure 2.60. Clustering of samples labeled by ELU based on commonly occurring

species in NE MI: average linkage ........................................................................................ 137
Figure 2.61. Clustering of samples labeled by ELU based on high quality species

in NE MI: Ward’s linkage .................................................................................................... 138
Figure 2.62. Clustering of samples labeled by ELU based on commonly occurring

species in NE MI: Ward’s linkage ................................................................................... 139
Figure 2.63. Clustering of samples labeled by ELU based on high quality species

in NE MI: beta linkage .......................................................................................................... 140
Figure 2.64. Clustering of samples labeled by ELU based on commonly occurring

species in NE MI: beta linkage .......................................................................................... 141
Figure 2.65. TWINSPAN classification of samples based on commonly occurring
species labeled by ELU in NE MI ........................................................................................ 143
Figure 2.66. TWINSPAN classification of samples based on high quality species

labeled by ELU in NE MI ...................................................................................................... 143
Figure 2.67. Classification of 77 samples reported by Host (1987 ................................... 147

Figure 2.68. Reclassification of 77 samples reported by Host (1987) based on
high quality species .............................................................................................................. 147

Figure 2.69. Clustering of 77 samples from NW lower MI based on high quality

species with clusters labeled by letters and ELU’s labeled by respective numbers:
Ward’s linkage ................................................................................................................. 149

Figure 2.70. Clustering of 56 samples from a homogeneous climatic zone in NW
lower MI based on common species with clusters labeled by letters and ELU’s
labeled by respective ntunbers: Ward’s linkage .................................................................... 151

Figure 2.71. Clustering of 56 samples from a homogeneous climatic zone in NW
lower MI based on high quality species with clusters labeled by letters and ELU’s labeled
by respective numbers: Ward’s linkage ................................................................................. 153

Figure 2.72. TWINSPAN on 56 samples from a homogeneous climatic zone in
NW MI based on high quality species .............................................................................. 154

xii

Chapter 1

INTRODUCTION

RESEARCH GOALS

There were three problems related to ecological land classification addressed in
this research. First, multifactor classification of ecological land units had not been
undertaken within the study area of northeastern Lower Michigan. Second, the sequential
application and interpretation of multivariate, graphical, and tabular methods on floristic
and environmental data for fully integrated ecological land classification had not been
previously documented. Third, the utility of using understory and ground-flora species
highly correlated with low-dimensional ordination space (Greenacre 1993, 1984) to
classify multifactor ecological land units and component ecological species groups had
not been investigated.

The primary goal of this research was to develop an integrated classification of
ecological land units in northeastern Lower Michigan based on associations of ground-
flora and understory species and key soil morphological and physical features. Two
secondary goals met while achieving the primary goal were: (i) to utilize and document a
series of complementary multivariate, graphical, and tabular analyses of ground-flora and
soils data in the classification process, and (ii) to determine if the classification of
multifactor ecological land units could be improved by eliminating understory and

ground-flora species that were poorly correlated (r2 < .45) with the first five dimensions

2

of a correspondence analysis, and species that occurred interior to the arch of the first two
dimensions of a correspondence analysis, prior to application of several ordination and
clustering methods.

To meet the first goal, I used ordination, clustering, graphical, and tabular
analyses of understory and ground-flora and soils data in a series of iterations.
Commonly occurring understory and ground-flora species were analyzed initially,
followed by the analysis of soils data, and then an analysis of ground-flora with high
fidelity to the first five dimensions of a correspondence analysis was conducted. After
these data sets were interrogated, data on overstory composition was analyzed using
ordination and tabular methods and related to earlier results. A final set of synthesis
tables was used to derive an integrated multifactor ecological classification.

To meet the second goal, I displayed results at each step of the classification
process and in subsequent discussions. Figures and tables were presented in a sequence
of analyses of floristic and soil data sets. Objective numerical analyses were used in
conjunction with subjective tabular methods. Clustering results were used in initial
ordination space partitioning of samples in both species and soil variable space to ensure
objectivity and enable replication. Subsequent analyses were compared to initial results
using graphs and tables. Steps used in arriving at results could thus be repeated by other
investigators with a minimum of subjective decisions made that affect the final
classification.

To meet the third goal, I explored the use of a geometric interpretation of

correspondence analysis to identify species most strongly correlated with the first five

3

principal coordinates or dimensions. Following initial eigenanalyses and clustering of
commonly occurring ground-flora, a criteria was set for retention of species for
subsequent ordination and clustering procedures. An arbitrary limit was set as other
research indicating appropriate limits was not available for guidance. If forty-five per
cent of the overall variability of species in the forty-five dimensional sample space of the
full data set was not accounted for in the first five dimensions, species were eliminated
from further analyses (personal comm. Dr. Carl Ramm). A reduced data set was
reanalyzed using the same ordination and clustering procedures, and results of the two
sets of analyses were compared using tabular and graphical methods.

These research goals involved both the application and refinement of multivariate
procedures routinely used in ecological classification. In order to assess the utility of
eliminating low fidelity species in the classification process, a classification had to first
be developed. However, assessing the effects of constructing a more parsimonious data
set for ecosystem classification based on the same data set used to develop the initial
classification may be questionable. This is analogous to developing a predictive equation
for a sample and only testing the model on the same sample. For this reason, I not only
compared clustering and ordination results between a larger and reduced data set to meet
the third goal, I also reanalyzed a data set for which results have been published and

tested in field applications (Host 1987).

PROBLEM STATEMENT

Two problems related to ecological land classification were addressed by the

4

goals stated for this research. The first problem was one of classification methodology.
The second problem was the quantification of species’ fidelity to environmental
gradients. These problems, which are interrelated and mutually inclusive, are discussed

briefly.

Problem one: Classification Methodology

Limitations of single factor approaches to the classification of forest land
potentials led to the development of multifactor systems in the latter part of the twentieth
century (Avers and Schlatterer 1991, Padley 1989, Hix 1988, Host et al. 1987, Albert et
al. 1986, Spies and Barnes 1985, Pregitzer and Barnes 1984, Barnes et al. 1982, Jordan
1982). Multifactor land classification uses groups of plants as indicators of important
environmental conditions in conjunction with direct measures of environment to
formulate classification units. ‘

In the development of multifactor ecological land classification systems, the
decision to use a particular ordination method in the analysis of ground-flora data, in
particular correspondence analysis versus detrended correspondence analysis, reportedly
may affect results and interpretations of eigenanalyses (Greenacre 1993, 1984, Palmer
1993, Minchin 1987, Peeteta1.1987, and Hill and Gauch 1982). Furthermore, clustering
methods used to identify “natural groupings” are inherently more subjective than
ordination methods (Jongman et al. 1987, Gauch 1982). Clustering results are influenced
by the choice of variables (Fowlkes et al. 1988), by the transformation or weighting of

these variables (Milligan and Cooper 1988, Jongman et al. 1987, Stoddard 1979), by the

5

choice of polysynthetic divisive or agglomerative procedures (Spies 1985a), and by the
linkage method chosen in the case of agglomerative clustering. Inclusion of superfluous,
or “noisy” variables also confounds clustering results (Fowlkes et al. 1988). These
limitations to any single approach to ordination and classification led to recommendations
to use a series of complementary methods in the ecological classification process (Digby
and Kempton 1987, Pielou 1984, Gauch 1982), yet most ecological classification research
involves the use of a single set of methods, without corroboration with other methods,
followed by reporting of results. In this research, a series of complementary methods
were used and compared to one another to derive the classification of multifactor
ecological land units and their components.

Another methodological problem relates to the definition of indicator plant
communities. In classic plant phytosociology, indicator species are used as indirect
expressions of environmental gradients presumed to be the cause of latent structure in
patterns in composition and relative abundance of ground-flora communities, hence the
reference to ordination of vegetation as indirect gradient analyses (Barton 1994, Backus
1993, Allen et al. 1991 , Allen and Peet 1990). Terms used to describe such species
include species with narrow ecological amplitudes (Pregitzer 1981), species with high
fidelity (Host 1987), and species with an affinity to a given environmental setting. Thus
most community classifications rely upon indirect gradient analyses of species-species
relationships to infer important environmental conditions, and use species-species
relationships to define plant associations (Barton 1994, Parker 1994, Kent and Coker

1992, Pfister et al. 1977, Daubenmire and Daubenmire 1968).

6

A different approach to classifying vegetation relies upon the evaluation of
species-environment relationships (John and Dale 1990), instead of species-species
relationships, to define communities. In ecological land classification, resulting groups
are called “ecological species groups” (Host and Pregitzer 1991, Archambault et al. 1989,
Spies and Barnes 1985b, Pregitzer and Barnes 1982). Yet in multifactor applications, a
range of ecological species groups have been reported as occurring within a given
ecological classification unit (Hix 1988, Host 1987). In essence, this amounts to
retrofitting independently defined component systems by simply conceptually overlaying
classifications. Since “ecological species groups” in a strict sense connote associations of
species based on their relationships to environment (Mueller-Dombois and Ellenberg
1974), this research compared results of analyses of floristic and environmental variables
in a series of iterations to identify congruent changes in ecological species groups and

environmental conditions.

Problem Two: Quantification of Species Fidelity to Environmental Gradients

In the ordination of vegetation, the first few axes generally account for the
majority of the overall variability of species in full-dimensional sample space (Digby and
Kempton 1987, Gauch 1982). Nearly all research on plant-soil relationships report that
the first few axes of an ordination of vegetation are related to environmental gradients,
primarily soil or topographically controlled moisture and to a lesser extent nutrient
availability (Host 1987, Spies and Barnes 1985a, Pregitzer 1981). Thus species with the

greatest linear relationship, or correlation, with the first few principal coordinates would

7

have the highest fidelity, or indicator value, to environmental gradients expressed low
dimensional subspaces provided that other sources of variability such as disturbance were
filtered by a well defined sample frame (van Groenewoud 1992).

The geometric interpretation of the quality of individual points in a given
subspace of a correspondence analysis provides a means of quantifying species’ fidelity to
principal coordinates (Greenacre 1984). In a correspondence analysis, the squared
cosines of the angles between a point and each axis of a set of axes can be added across
any number of dimensions; the sum of squared cosines across all dimension sums to one.
This sum of squared cosines, or display quality, is a measure of how well points are
represented in a particular set of axes, and is mathematically analogous to the correlation
of a species to a given multidimensional subspace (Greenacre 1993). In this research,
groups of species with high correlations or fidelity to low-dimensional subspaces were
used to define ecological species groups associated with underlying environmental
gradients, and to define multifactor ecological land classification units.

The determination of species with high fidelity to low-dimensional subspaces also
facilitated an improved interpretation of ordination diagrams. All species are represented
in ordination diagrams, yet not all species are “indicators” of the environmental or other
gradients expressed in diagrams of 2 or 3 visually observable dimensions (personal
comm. Dr. Carl Ramm). Some species’ profiles will not be accurately represented in low
dimensional subspaces because they lie more along higher planes or within higher
multidimensional subspaces (Greenacre 1993, 1984). In these instances, species that are

actually poorly related to a low-dimensional ordination space are arrayed along side

8

species that are closely related to the two or three dimensional ordination space being
observed. This results in an obscured interpretation of Species and groups of species with
high indicator value, or fidelity, to underlying gradients expressed by a particular
subspace. In this research, superfluous species were filtered prior to interpreting

ordination diagrams.

ALTERNATIVE FOREST LAND CLASSIFICATION SYSTEMS

Two basic types of forest land classification and mapping systems are used in
natural resource research, planning and management. The first classifies and maps
current conditions of land areas, whereas the second classifies and maps potential of land
areas (Jones 1994, Rowe 1980). Inventories of current conditions measure attributes such
as age, stocking, and composition of forest stands, or locations of endangered species.
Inventories of potentials include soil surveys, habitat typing, site index estimates, and
multifactor ecological land classification (Pregitzer and Barnes 1984, Moon 1984). Each
of the latter types of systems have a primary purpose of identifying land areas that have
similar capabilities for management (Jensen et al. 1991). Although the objectives of soil
surveys, habitat typing, site index, and ecological classification strongly overlap, the

assumptions underlying each system are markedly different.

Soil Survey
Information about soil properties is critical to managers. Soil directly influences

plant community composition and productivity, and strongly influences the capacity of

9

the land to produce vegetation (Zak et al. 1989, Bruggink 1988). Maintaining long-term
soil productivity is imperative to sustainable forest management, and knowledge of soil
properties is needed in planning the construction and maintenance of physical structures
such as roads, facilities, and so forth.

In 1893, a new concept of soil was published by V.V. Dokuchaev in Russia
(Glinka 1927, Vilenskii 1957). Soils were conceived to be independent natural bodies
resulting from interactions of climate, vegetation, parent materials, relief, and time. In
1892, E.W. Hilgard extended this concept to soil science in the United States,
emphasizing relationships between soils and climate. Coffey (1912) later articulated that
“the soil is an independent, natural body, a biogeological formation, differing essentially
from the rock that undelies it, although closely related to it . . . . ” As stated in Soil
Taxonomy (1975), "this was a revolutionary concept. The soil scientist did not need to
depend wholly on inferences from the underlying rocks, the climate, or other
environmental factors, considered singly or collectively; rather, he could go directly to the
soil itself and see the integrated expression of all these in its morphology." This concept
of soils assumes that soil morphology fully integrates other important physical and
biological factors.

Soil surveys are conducted to provide land managers information for many uses
such as agricultural crop production, building and road construction, septic system
installation, forest management, and so on. Traditional soil surveys have been found
inadequate for predicting timber production and forest land management potentials,

(Carmean 1979), and the need to tailor soil surveys for forest management purposes has

10

been well documented ( Grigal 1984, Moon 1984, Zanner and Hannah 1972, Gysel and
Arend 1956). Moreover, the criteria used in soil classification, and many of the
laboratory analyses used to characterize soils, are based on agricultural needs and
calibrations. Alternate criteria and laboratory methods based on forest rather than

agricultural crop response are needed (Stone 1975).

Vegetation Classifications

Information about vegetation and vegetative potentials is essential to forest and
wildlife managers (Arno et al. 1986, Coffman et al. 1983, Pregitzer and Barnes 1982).
The earliest scientific classification of vegetation is generally credited to Alexander von
Humbolt (1807) who systematically classified areas dominated by plants into vegetation
types based on plant physiognomy (McIntosh 1978, Whittaker 1973). Other early
influential plant ecologists (Show 1822, Kemer 1895, Giesbach 1872, Drude 1890,
Warming 1895 and 1909, and Schimper 1903) also largely described vegetation based on
similarities in physiognomy. In Finland, Cajander (1926) identified forest types based on
combinations of tree species and ground-flora communities, and demonstrated that the
growth rate of a given tree species varied depending on forest type.

This early work led to the development of “habitat types” in the United States.
Daubenmire (1952) defined a habitat type as all the land capable of supporting a
particular plant association at climax. Climax plant associations are described by Kotar
and Coffman (1984) as late stages in succession "whereby the plant species are apparently

self-regenerating in predictable patterns and have long been free of disturbance by fire,

11
grazing, logging, etc."

Habitat type classification systems typically rely solely on climax vegetation, even
though many landscapes support plant communities in a variety of successional stages
due to anthropogenic and natural disturbance (Dzwonko 1993, Franklin et al.1993, Foster
1992, Delcourt et al. 1983, Pfister et al. 1977). Climax vegetation is classified in
relatively undisturbed conditions to describe habitat types, then habitat types are
identified in disturbed lands based on remnants of typical climax plant associations
(Kotar et al. 1988, Kotar and Coffinan 1984).

Classifications based on vegetation made similar presumptions as soil surveys
with respect to the integrative power of plant indicators. Plant phytosociology has
assumed that "the climax community reflects the most meaningful integration of
environmental factors affecting vegetation" and that "each habitat type represents a
relatively narrow segment of environmental variation" . . . "that uses the plant community
as an integrated indicator of environmental factors (Kotar et al. 1988)."

In their classic work of 1968, Daubenmire and Daubenmire state "it is clear . . .
that no useful correlation exists between vegetation types as described herein and (soil)
profile types distinguished on the basis of color, texture, structure, depth, sequence or

horizon, etc. Many studies, however, have shown that plants, particularly ground-flora
species, can be used to indicate specific soil properties (McIntyre 1994, Boeye and
Verheyen 1994, Franklin and Merlin 1992, Whitney 1991, Haase 1990, Pregitzer and
Barnes 1982, Grigal and Arneman 1970, Waring and Major 1964). Ground-flora species

groups have also been associated with overstory species composition, glacial

12

physiography, and nutrient cycling rates (Host and Pregitzer 1992, Padley 1989, Zak et
al.1986, Spies and Barnes 1985b, Barnes 1984, Moon 1984, Pregitzer and Barnes 1984,
Peet and Loucks 1977, Curtis 1959). In Canada, Rowe (1971), and Rowe and Sheard
(1981) have described similar associations.

There are several advantages to the use of vegetation in determining land
capability. Vegetation is easily observed, whereas observing soil conditions requires
labor-intensive methods (Bekele 1994, Pregitzer and Barnes 1984). In many instances
there are predictable relationships between ecologically important soil factors (e. g., pH,
organic matter content, soil texture, and nutrient availability) and the distribution of
ground-flora (Haase 1990, Odland 1990, Prentice and Cramer 1990, Padley 1989, Zak et
al. 1986, Spies and Barnes 1985a, Barnes 1984, Jones 1984, Moon 1984, Pregitzer and
Barnes 1982). Habitat type systems provide significant information about plant
community composition, and provide a basis for grouping observations to study processes
such as forest growth or succession and extrapolate results to similar sites (Arno 1986).

The major drawback of habitat typing is that it is based on one component of the
ecosystem, namely vegetation, and is not linked to the more stable abiotic factors
(Pregitzer and Barnes 1984). Because the habitat typing system uses only vegetation to
determine the classification of a land area, it cannot differentiate among ecosystems
where disturbance has eliminated or changed the composition and relative abundance of
plant indicators (Cook 1996, Meier et al. 1995, Gilliam et al. 1995,. Roberts and
Christensen 1987, Marquis and Brenneman 1981), where deep-lying soil characteristics

affect overstory growth (Host 1987, Hannah and Zahner 1970), or where compensating

13

factors have produced similar plant communities. Rowe (1984) discussed the limitations
of the use of vegetation as the prime indicator of forest land potential, contending the
chief deficiency of vegetation is that it is just one part of the ecosystem.

The single factor approach of habitat typing for land classification is analogous to
soil classification, using climax vegetation types instead of soil morphology as the basis
for identifying environmental conditions influencing ecological systems. Yet plant
ecologists do not agree on the assumption that there is a fixed and predictable endpoint in
the successional pathways in plant community composition through time (Cook 1996,
Abrams 1985).

“Ecological species groups” is another approach to Vegetation classification. This
approach relies upon species-environment relationships instead of species-species
relationships to distinguish communities (Host and Pregitzer 1994, Archambault et al.
1989, Spies and Barnes 1985b, Pregitzer and Barnes 1982). The method is based on the
premise that communities are combinations of species whose composition is dependent
upon local environment (Rydgren 1994, Archambault et al 1989, Mueller-Dombois and
Ellenberg 1974).

Mueller-Dombois and Ellenberg (1974) recommend establishing species groups
using any of three methods. The first is to conduct field investigations and subjectively
identify groups of species with similar ecological relationships. The second is to study
the ecological behavior of individual species in relation to single site factors until
relationships can be expressed in figures. The third is to use tabular methods or

correlative studies for species groups by identifying which of the differential or highly

14

correlated species form ecological groups.

The major drawback of ecological species groups is that it is based on one factor,
vegetation, although communities are linked to the more stable abiotic factors, as opposed
to habitat typing. The omission of environmental factors in this single factor approach
precludes development of several research and management applications including
studies of whole systems for research on processes such as carbon and nutrient cycling,
and management information needs such as limitations for road construction, equipment

operability, and other physically constrained management options.

Site Index

Site index is the most common and widely applied method of predicting the
growth potential of forests in the Lake States. Site index is a direct measure of height
growth at a base age (Grigal 1984, Carmean 1979), and its use depends upon finding trees
whose height growth has not been suppressed, or damaged by ice or wind. Height growth
is usually well related to timber volume production and site index is widely used in yield
tables and computer simulations.

Site index alone yields no information about plant community composition,
succession, or soil (Carmean 1975, Jones 1969). It is primarily a tool for estimating the
potential for timber production in normally stocked, even-aged stands, and cannot by

itself provide the information needed for integrated resource management (Pregitzer

1981).

15

Multifactor Ecological Land Classification

Studies comparing classification systems have shown that combinations of
physiography, soil, and vegetation data provided better classifications than any single
component used separately (Palmer 1990, Spies and Barnes 1985a, Pregitzer and Barnes
1984). The basic goal of multi-factor ecological land classification is to identify land
areas with different management potentials by discerning congruent changes in
community and environment and defining class limits for each component based on
mutual relationships (Jensen et al. 1991, Hart 1988). This goal is accomplished by the
simultaneous sampling, analysis, and mapping of biotic and abiotic variables (Host 1987).
In this approach, both species-species relationships and species-environmental
relationships are identified. Changes in species composition are related to underlying
environmental gradients to produce an integrated classification of land units.

The concepts underlying multifactor ecosystem classification began to emerge as
early as 1789, when Alexander von Humboldt wrote that, "All natural bodies are
interrelated. Find a certain type of soil and a certain type of plant and you will find a
certain type of rock." In the century following Alexander von Humboldt’s observations,
Gutrovich (1894) developed a system in Russia that classified different kinds of forests
based on species composition and landscape topography. During the 1920's, a study of
site relationships was conducted in West Germany, eventually leading to the classification
and inventory of the state of Baden-Wurttemburg (Barnes 1984). The German system
employed a regional hierarchy based upon climatic and physiographic influences, and a

local hierarchy based upon topography, soil and ground-flora.

16

In the 1940's, G.A. Hills developed a site classification system in Ontario that
included vegetation and physiography after finding that soil surveys or habitat types alone
did not adequately classify land for both agricultural and forestry uses. Stan Rowe
continued this work throughout the 1960's and 1970's, describing "landscape units" based
on geomorphology, incorporating principles of both community ecology and habitat
typing, and emphasizing practical aspects of classification and inventory.

In the 1980's and 1990's, a number of researchers applied multifactor concepts in
the evaluation and classification of forests and forested landscapes in Michigan and
Wisconsin. This includes work by Barnes et al. (1982), Jordan (1982), Pregitzer and
Barnes (1984), Spies and Barnes (1985a), Albert et al. (1986), Host et al. (1987), Hix
(1988), Archambault (1989), and Padley (1989).

Multifactor land classification uses groups of indicator plants as indirect
expressions of important environmental conditions in conjunction with direct measures of
environment (e. g., climate, landforms, and soils) to define taxa. Ecological units are then
mapped using both vegetative and environmental criteria identified in the classification
process, with interpretations made possible given spatially explicit information (Johnson
1992, Brenner and Jordan 1991). The modern approach to multifactor ecological
classification builds upon systems developed by numerous researchers throughout the
twentieth century (Cajander 1926, Koppen 1931, Fenneman 1938, 1950, Hills 1952,
Whitaker 1960, Kuchler 1964, Daubenmire and Daubenmire 1968, Wertz and Arnold
1972, Corliss 1974, Rowe 1980, Eyre 1980, Jordan 1982, Barnes et al. 1982, Bailey 1983,

Jones et al. 1983, Driscoll et al. 1984, Smalley 1986, McNab 1987, Omernik 1987,

17

Jensen et al. 1991, Cleland et al. 1992, 1996).

LITERATURE REVIEW of ECOLOGICAL CLASSIFICATION METHODS

Ecological classifications in various forms have been developed using relevees
(species lists ordered by sampling quadrants), tabular analyses, and more recently by
various multivariate numerical methods (Host and Pregitzer 1991, Retuerto and
Carballeira 1991, Spies and Barnes 1985a, Gauch 1984, Mueller-Dombois and Ellenberg
1974). Braun-Blanquet's procedure of 1921 used tabular methods in successive
approximations to identify groups of species occurring in similar samples, and to identify
samples with similar species composition. These early tabular classification techniques
were informal and inherently subjective (Whittaker 1960, Mueller-Dombois and
Ellenberg 1974), and thus the recognition of differential species groups and groups of
similar samples depended on the individual investigator's understanding of species-
species and species-environmental relationships within a study area.

In the past few decades, more objective multivariate procedures have been applied
in ecological classification research (Digby and Kempton 1987). Research problems are
often approached in an overall program using ordination for indirect and direct gradient
analyses, clustering to group samples and variables, and tabular synthesis of results to
corroborate patterns detected through ordination and clustering (Gauch 1982).

In these procedures, explorative data analyses are used to detect intercorrelations
among variables, to check assumptions of data structure underlying particular analyses

and suggest appropriate transformations, and to identify sample outliers using descriptive

18

statistics and graphical displays of raw data. Ordination is also used in an exploratory
data analysis sense to detect outliers, and in a multivariate sense to screen variables,
reduce dimensionality, and summarize community and environmental patterns (Gauch
1982).

Ordination is often followed by clustering to identify natural groupings of samples
and species. Results of ordination and clustering may be compared, and subsets of data
may be interrogated to further elucidate relationships. Several complementary analysis
techniques may be applied to the same data set, with the analysis progressing by
successive refinement. Community patterns may be compared with environmental
patterns to produce an integrated interpretation of ordination and classification results.
The communication of results is promoted, however, by employing a moderate number of
commonly used, relatively standardized methods (Pielou 1984, Romesburg 1979).
Finally, hypothesis testing methods may be used a posteriori to assess ecosystem-level
differences in processes such as productivity (Host et al. 1988), successional pathways

(Host et al. 1987) and nutrient cycling (Zak et al. 1986, 1990).

Ordination

Ordination, a primary tool for examining continuous change in ground-flora
distributions, is a matrix approximation technique used to identify important variables,
smnmarize data in a scatter diagram (Bray 1957), and reduce the dimensionality of a
multivariate data set (Causton 1988, Gauch 1982). Important variables are those with the

highest weights and loadings, or correlations, with the first few eigenvectors. The

19

reduction of dimensionality aims to provide low-dimensional summaries of high-
dimensional data sets that ideally account for a large part of the total variability lying in
higher dimensional space ( Morzuch and Ruark 1991, Krzanowski 1988). In ecological
studies, ordination is also used to discover latent structure of vegetation data due to
species' responses to underlying environmental gradients (Prentice 1977).

Principal component analysis (PCA; Morrison 1976), correspondence analysis
(CA; Hill 1974, Greenacre 1984, 1993), and detrended correspondence analysis (DCA;
Hill and Gauch 1980) are among the most commonly used ordination techniques in
modern ecological studies. These are variance-maximization methods that involve
rotating the axes of a multidimensional cloud of points in multivariate space. This
maximization of variance is equivalent to minimizing the variance of the projection
distances from the axis. The first axis, or eigenvector, is in the direction that accounts for
as much variance as possible along the ordination axis. A second axis is then found that
is orthogonal to the first axis that accounts for the maximal remaining variance, and so
on, for as many axes as desired.

Geometrically, the general intentions of PCA, CA, and DCA are identical; a
multidimensional cloud of points is projected efficiently in fewer dimensions which
maximally account for the structure of a cloud of points in multidimensional space
(Oksanen and Huttunen 1989, Gauch 1982). The methods of projection differ, however,
due to the selection of particular distance measures, selection of the weights attributed to
the points, and the positioning of the origin (Greenacre 1984). Centered PCA uses

Euclidean distances, equal weights for points, and location of the origin at the centroid;

20

non-centered options are also available. CA and DCA use chi-squared distances, weights
for sample or species points proportional to the total for the species or sample,
respectively, and an origin at the centroid (Liebart et al. 1984).

The eigenstructure decomposition used in these methods is based on an assumed
underlying linear model (Morrison 1976). Thus a data set must meet several
assumptions, primarily that the variables have normal distributions and have uncorrelated
errors (Dale 1976, Hotelling 193 3). For descriptive purposes not involving formal tests
of hypotheses, departures from multivariate normalcy are acceptable in the use of PCA

(Gauch 1982).

Principal Component Analysis

PCA is used to reduce a data set with a relatively large number of correlated
variables by transforming linear combinations of the original data to a data set with fewer
uncorrelated variables (Newcomer 1984) that retain most of the information content of
the original data in a reduced subspace (Morzuch and Ruark 1991, Jongman et al. 1987).
The first component accounts for a certain portion of the generalized variance present in
the original data set; successive components account for decreasing proportions of the
variance while remaining uncorrelated with previous components.

The relative amount of the total variation explained by each principal component
is calculated by examining its variance (i.e., the eigen roots) relative to the overall
variability (Morrison 1976). If a large proportion of the overall variability in a data set is

accounted for in the first few principal components, these components can be used to

21

summarize the whole of the variability and covariability of the original variates (Morzuch
and Ruark 1991). In the end, however, PCA results must be assessed in terms of
ecological utility; mere percentage of variance accounted for has not been found a reliable
indicator of the quality of results (Gauch 1982).

PCA requires continuous data as it operates on a variance-covariance matrix or
correlation matrix. The correlation matrix is commonly used when variables have
dissimilar scales and ranges, unequal variances, or violate other assumptions underlying
the model (Padley 1989). Data with dissimilar scales and unequal variances will
inevitably emphasize variables with large means and variances, and such emphasis may
be unwarranted. The use of the correlation matrix avoids rather than solving this
problem, but is considered satisfactory if all the variables used are of similar importance
(Chatfield and Collins 1980, Morrison 1976).

Examining the coefficients, also termed weights, for each variable permits
identification of the variables that are most important, or, which account for a large
portion of the variation and would therefore be best for use in predictive models (Padley
1989, Pregitzer and Barnes 1984). Important variables may also be identified by
examining the correlations, also called factor loadings or factor patterns, of the raw
variables with the PC's; factor loadings should correspond with factor weights. A
correlation indicates the importance of a variable, so that those with high correlations are

those which express a large amount of variation in the dimension being considered.

22

Correspondence Analysis

Correspondence analysis (CA) was developed to analyze counts or measures of
species abundance (Causton 1988, Host 1987). CA produces an ordination of species and
samples. Species scores, when graphed, reflect the relative positions of individual
species across all samples in response to one or more environmental or temporal
gradients (Hauser and Mucina 1981). Sample scores can be used to produce another
ordination, with Similar samples plotting close together and dissimilar samples plotting
far apart. CA also provides information on the importance of individual species and
sample points, and the relationship of species and samples to principal axes, termed
coordinates, in a reduced subspace using a geometric interpretation (Greenacre 1984).

A cross-tabulation is converted into a contingency table in CA by dividing profiles
of variable and sample frequencies by variable and sample totals. These sets of relative
frequencies add up to a constant of one or 100, depending on whether the data are
proportions or percentages, respectively. The geometry of such data has special
properties that allow a geometric interpretation of the data structure. Rather than simply
representing species in sample space, or samples in species space, a vector going from
the origin to the sample points could represent samples or species (Greenacre 1993,
1984). It is then possible to describe much of the information about sample or species
relationships in terms of the angles between pairs of vectors because the angles can be
expressed by their cosines which has statistical meaning because the cosine is identical to
the correlation coefficient between the two samples or species. Cosines of angles

between vectors for sample pairs are analogous to a samples-by-samples dissimilarity

23

matrix for which the dissimilarity measure used is the correlation coefficient after
subtraction from one (Greenacre 1993).

CA uses the chi-squared statistic as a measure of the discrepancy between the
observed frequencies in a contingency table and the expected frequencies. The chi-
squared statistic measures how far row or column profiles are from their average profiles.
The chi-squared distance fimction is similar to the Euclidean distance in physical space,
except that each squared difference between coordinates is divided by the corresponding
element of the average profile (Liebart et al. 1984).

In a CA, each row or column profile of relative frequencies has a unique weight
associated with it, called a mass, which is proportional to the column or row sum,
respectively, in the original cross-tabulation. The average profile is the sum of each
profile weighted by that profile's mass.

The total variability, or inertia of a contingency table is the chi-squared statistic
divided by the total of the table. Total inertia can be decomposed into principal inertias
associated with each principal axis, into inertia associated with each element in a
contingency table, and into the partial contributions of each element to each respective
principal inertia (Greenacre 1993).

That component of inertia along a principal axis, called principal inertia, is the
inertia of the row points (or column points) projected onto the axis (Greenacre 1993,
1984). Hence the first principal inertia is the total inertia of a set of projected points lying
on the straight line along the first principal coordinate. Second and higher axis principal

inertias are determined using the same process.

24

Principal inertia can be further broken down into row and column components
along individual principal axes, termed partial contributions (Greenacre 1993, 1984).
These partial contributions to inertia for either sample or species sum to one in each
dimension, and are a measure of the contribution of each entity to the inertia of that
principal axis.

The relative contributions of each principal axis to a point's inertia are also
determined in CA. In the geometric interpretation of these relative contributions, the
cosine of an angle between a point in multidimensional space is determined by projecting
that point perpendicularly onto the principal axis. A vector is then extended to a point
intersecting a circle in standard position with a circle of radius one. This point has a pair
of coordinates that can be described by cosine and sine fimctions; the cosine of the angle
is equal to the ratio of the derived vectors. The square of this ratio is the contribution of
the respective principal axis to the sample or variable inertia. Relative contributions can
be thought can be being equivalent to squared correlations between the profile points and
the principal axes (Greenacre 1993, 1984).

The decomposition of each principal inertia over the samples or rows is a measure
of the contributions of each principal coordinate to the inertia of these profiles. This
decomposition indicates which points are best explained by the combination of axes
forming subspaces by examining the squared cosines for each axis (Greenacre 1993,
1984)

A particularly useful feature of CA is the geometric interpretation of the quality of

individual points in a given subspace (Greenacre 1984). The squared cosines of the

25

angles between a point and each axis of a set of axes can be added across any number of
dimensions; the sum of squared cosines across all dimension sums to one. This sum of
squared cosines, or quality, is a measure of how well points are represented in a particular
set of axes. An examination of the quality of display of each point helps to diagnose
which points are far off the plane and whose positions are thus inaccurate in the map.

The accuracy of display, or display quality is the percentage of inertia accounted for in a
specific subspace (Greenacre 1993, 1984).

Points are often distributed along an arch in CA. Some points, however, may be
interior, or even central to the arch, and other points may be on the periphery of the arch.
Interior points indicate species with especially broad or undiscriminating (weedy)
distributions, and similarly, samples with a broad or mixed collection of species.
Peripheral points indicate species with especially narrow distributions, having strong
indicator value, and suggest samples with a simple collection of similar species (Gauch

1982)

Detrended Correspondence Analysis

In vegetation analyses, PCA and CA assume that species abundances change
linearly along environmental gradients. When nonlinearities occur, CA produces an arch
effect due to the compression of first axis ends ( Peet et a1. 1987, Hill and Gauch 1980 ).
Moreover, a vegetation data matrix generally consists of some measure of species
presence or abundance arranged by species and samples. Since Species are often absent

from many sample plots, data matrixes often contain a high proportion of zeros. For

 

26

reasons inherent to the mathematics of the technique, these high proportions of zeros also
tend to distort the second and subsequent principle axes (Host and Pregitzer 1991).

Hill (1979) argues that the arch effect is a high-order dependence of the second
and higher eigenvectors on the first eigenvector, so that a given distance of separation in
the ordination does not carry a consistent meaning in terms of implied differences
between the samples or species. If these derived orthogonal and linearly independent
eigenvectors are related by a quadratic or higher-order relation, information on important
secondary gradients in the data may be deferred to higher axes. The interpretation of
results is thus made more difficult because spurious axes must be distinguished from
valid axes, and higher axes in higher dimensions may have to be explored to detect
meaningful gradients (Hill and Gauch 1980).

This limitation of traditional ordination methods led to the development of DCA
(Hill 1979). The detrending procedure aims to remove the quadratic dependence of the
second axis on the first axis while extracting a second axis. In detrending, the first axis is
divided into segments, and within each segment, the sample scores are readjusted to have
a zero average. The readjustment of scores results in a set of detrended scores that are
used to derive a new axis based on residual variation.

Wartenberg et al. (1987) are skeptical of the value of detrending. They assert that
the archlike curvature is an inherent property of successive-replacement data that results
from the partial overlap in the distribution of species along a one-dimensional
environmental gradient. They suggest DCA can hide the real data structure, and even

introduce new distortions. Similarly, Greenacre (1984) criticized detrending because the

 

27

control over geometry is lost. Okasen (1988) found that when the two first eigenvalues of
CA are close to each other, their order can be reversed due to random variation in the

data. In DCA, the second axis is detrended with respect to the first and therefore very
variable configurations result when the orientation of the first axis in the plane is

changed. This can lead to situations where the detrended solutions are very unstable
under random variation and therefore can only be casually interpreted. These findings
suggest additional limitations in the use of DCA.

Van Groenewoud (1992) recommended that the only way to get reasonable results
using CA or DCA is to restrict the sampling mainly to one gradient, or pre-stratify the
samples before analysis to represent mainly one gradient at a time. Concluding that all
ordination methods are influenced by data curvature and scaling, Greenacre (1993)

recommends reporting the arch unsealed in two dimensions.

Clustering

The objective of clustering procedures, commonly called classification, is to
identify naturally occurring groups based on all variables in a multivariate data set (Host
et al. 1993). Both the process of classification and the choice among classification
techniques are more complex and more subjective than those of ordination (Gauch 1982).
The utility of a given technique is therefore judged in relation to others, and often several
classification techniques are applied to the same data sets with results compared
afterwards.

The most commonly used classification methods in ecological land classification

28

include polythetic hierarchical agglomerative clustering and polythetic hierarchical
divisive clustering. The term polythetic means that information on all variables is used
to assign observations to a cluster, as opposed to earlier monothetic methods that used
single variables in a non-multivariate analysis (Gauch 1982).

Polythetic agglomerative clustering has two steps. In ecological studies, the
species-by-species data matrix is first used to compute a samples-by-samples dissimilarity
matrix using any of several distance measures such as Euclidean distance or percent
dissimilarity (Jongman et al. 1987). Second, an agglomeration procedure is applied
successively to build up a hierarchy of increasingly large clusters, starting with clusters
consisting of a single member, and agglomerating these hierarchically until finally a
single cluster contains all the species (Gauch 1982).

In agglomerative clustering methods, the groups that closest resemble each other
are always fused. The definition of closest, or dissimilarity, between groups differs
among methods; the linkage method selected determines the dissimilarity, or distance
between two clusters (Jongman et al. 1987). The most common linkage methods used in
ecological studies include complete-linkage (or furthest-neighbor), single-linkage (or
nearest neighbor), average-linkage, and Ward's method (or minimum-variance) clustering
(Digby and Kempton 1987). Flexible-beta linkage has also been found to recover
underlying cluster structures efficiently (personal comm. Dr. Carl Ramm).

The simplest polythetic divisive classification is subjective ordination space
partitioning (Hill 1977). Sample points are positioned in low-dimensional ordination

space and partitions are placed subjectively by drawing lines through sparse regions of the

29

cloud of sample points.

Two-way indicator species analysis (TWINSPAN; Hills 1979) is another
polythetic divisive clustering technique. TWINSPAN begins with all species or samples
(depending on the objectives) in a single cluster, and divides these into smaller clusters by
first ordinating data by CA. Species characterizing the CA axis extremes are emphasized
to polarize the samples, and the samples are then divided into two clusters by breaking
the ordination axis near its middle. This procedure is repeated until each cluster has no
more than a chosen minimum number of members. A corresponding species
classification is produced, and the samples and species hierarchical classifications are
used together to produce an arranged data matrix similar to a Braun-Blauquet table.

Van Groenewoud (1992) asserts that TWINSPAN does not perform a cluster
analysis, and thus does not discern spatial vegetative patterns. He concludes that the
usefulness and reliability of TWINSPAN depends on how well CA extracts axes that
have ecological meaning, how well the CA axes are divided into meaningful segments,
and how closely species are associated with certain segments of the multivariate
coenoplane.

A methodological problem in applied clustering involves the decision of whether
or not to standardize the input variables prior to the computation of a Euclidean distance
or other dissimilarity measure (Milligan and Cooper, 1988, Stoddard 1979). A number of
transformations are commonly used, including logarithmic transformations, square-root
transformations, exponential transformations, range transformations, standardization to

mean of zero and unit variance, and transformation to an ordinal scale (Jongman et al.

30

1987). Cooper (1988) found that transformation by the range of a variable gave
consistently superior recovery of the underlying cluster structure.

Standard clustering algorithms can fail to identify clear cluster structure if that
structure is confined to a subset of variables (F owlkes et al. 1988). The inclusion of
unnecessary variables in cluster analysis can therefore mask cluster structure, resulting in
mixed assignments of observations. Consequently, superfluous and redundant variables
are often screened prior to clustering through exploratory data analyses including

descriptive statistics and ordination, regression analysis, or other methods.

Chapter II.

ECOLOGICAL LAND UNITS FOR UPLAN D FOREST ECOSYSTEMS OF
NORTHEASTERN LOWER MICHIGAN
INTRODUCTION

Multifactor ecological classification systems have been developed in areas of
Michigan and neighboring Wisconsin at regional down to local scales. At local scales,
these efforts identified soil-plant relationships (Hix 1988, Host 1987, Spies and Barnes
1985a, Pregitzer and Barnes 1984), and studied stand-level processes such as forest
growth (Host et al. 1988, Greaney 1987), succession (Host et al. 1987, 1988), and
regeneration (Bruggink 1988), and soil nutrient cycling (Padley 1989, Zak et al. 1989).

At landscape- and regional levels, patterns in biomass (Host et al. 1988), ground-flora and
overstory composition (Host and Pregitzer 1992), nutrient cycling (Zak and Pregitzer
1990, Zak et al. 1989), surficial geology-plant-soil relationships (Albert et al. 1986), and
climatic gradients and classification (Denton and Barnes 1988) have also been
documented.

These research efforts applied concepts of regionalization and spatial hierarchies
(Bailey 1991, Gallant et al. 1989, Albert et a1. 1986, O’Neil et al. 1986, Barnes et al.
1982, Rowe 1980) in which ecosystems are conceived as occurring in a nested geographic
arrangement, with smaller ecosystems embedded in larger ones (Allen and Starr 1982).
Regionalization is based on an understanding that the structure and function of

ecosystems are largely regulated along energy, moisture, nutrient, and disturbance

31

32

gradients (Cleland et al. 1985). These gradients are affected by numerous environmental
and biological factors including climate, geology, soils, flora, fire, and wind, and these
factors vary at different spatial and temporal scales ( Cleland et al. 1996, Spies and
Barnes 1985a, Barnes at al. 1982, Jordan 1982).

At global, continental, and regional scales, ecosystem patterns correspond with
climatic regions, which change mainly due to latitudinal, orographic, and maritime
influences (Denton and Barnes 1988, Bailey 1987). Within climatic regions, landforms
modify macroclimate (Bailey 1987, Smalley 1986, Rowe 1984), and affect the movement
of organisms, the flow and orientation of watersheds, and the frequency and spatial
pattern of disturbance by fire and wind (Swanson et al. 1988). Within climatic -
geomorphic regions, water, plants, animals, soils, and topography interact to form
ecosystems at local scales (Pregitzer and Barnes 1984, Major 1969). Thus ecological
systems exist at many spatial scales, from the global ecosphere down to regions of
microbial activity.

While the association of multiple factors is all important for understanding
ecological systems, not all factors are equally important at all spatial scales (Cleland et al.
1992). The integration of important factors at relevant spatial scales provides the basis
for ecological land classification. At local scales, ground-flora, understory, topographic,
and soils are commonly used as a basis for classification and mapping ecological land
units. The challenge of ecosystem classification is to distinguish natural associations of
ecological factors at different spatial scales, and to define and map ecological land units

that reflect these different levels of organization (Cleland et al. 1996).

33

Regionalization is also important in understanding the emergent properties of
ecosystems. The concept of emergent properties in ecological systems was formally
introduced by Salt (1979), who first expressed that “An emergent property of an
ecological unit is one which is wholly unpredictable from observation of the components
of that unit.” Cleland et al. (1996) stated that “the conditions and processes occurring
across larger ecosystems affect and often override those of smaller ecosystems, and the
properties of smaller ecosystems emerge in the context of larger systems.” In northeast
lower Michigan, Padley (1989) hypothesized that landscape-level fire disturbance
regimes altered overstory and ground-flora succession to include both pyrophilic and
mesophilic species in morainal landforms nested within or lying adjacent to fire-prone
outwash landforms. Host and Pregitzer (1992) later suggested that similar relationships
existed among landforms and local ecosystems in northwest lower Michigan. In these
cases, conditions within a local ecosystem could only be explained by examining patterns
and processes at a broader spatial scale, and relating local conditions to a broader spatial
context.

This research involved the classification of fine scale ecological land units in
northeastern Lower Michigan. The classification was nested within Districts and
Subdistricts as defined by Albert et a1. (1986) to accommodate broader scale influences
and reduce the variability in climate and landforms affecting soil and plant relationships
that were used as a basis for classification. Applications of hierarchy theory were
employed in defining the sample frame at the Subdistrict scale.

The classification of ecological land units based on local soil-plant relationships

34

was necessary in northeastern lower Michigan because classifications developed in other
regions of Michigan may not be applicable in this area. The indicator value of ecological
species groups varies across climatic and physiographic regions (Pregitzer and Barnes
1982), and species assemblages are likely to change with changes in macroclimate or
regional physiography (Host and Pregitzer 1991). Analysis of climatic data indicated that
temperatures in northeastern lower Michigan were significantly colder, and growing
seasons significantly shorter, than those of northwestern Michigan (Cleland et al. 1994).
Host (1987) found many of the same species described by Spies and Barnes (1985b) in
the Sylvania Recreation Area in the Upper Peninsula present in northwest lower
Michigan, but in different assemblages that represented responses to different
environmental conditions. He reported that the usefiilness of particular species groups
defined for northwest lower Michigan was unknown for the finer-textured and more
calcareous landforms of the northeastern Lower Peninsula (Host and Pregitzer 1992).

Multivariate methods were used in each of the research studies conducted in
Michigan and Wisconsin. Pregitzer (1981) used four agglomerative clustering techniques
to classify species groups, two agglomerative clustering techniques to classify samples
based on environmental variables, PCA to examine relationships between sample plots
and physiographic and soil variables, and discriminant analysis to test the robustness of
the classification. He did not ordinate samples based on species, did not explore use of
divisive clustering (TWINSPAN), and used commonly occurring species in clustering
procedures.

Spies and Barnes (1985a) used TWINSPAN and DCA to classify species and

35

samples. They reported that agglomerative clustering failed to identify meaningful
species groups due to bad fusions at the initial stage of clustering. They used commonly
occurring species in ordination and clustering procedures.

Host (1987) used TWINSPAN to classify Species and samples, DCA to ordinate
species and samples, and tabular methods to arrive at the final classification. He did not
use multivariate methods to analyze soils data, used commonly occurring species in
ordination and clustering procedures, and did not describe how the classification of
ecological species groups was integrated with analyses of soils data to arrive at the final
classification.

Hix (1988) used two agglomerative clustering techniques to classify samples
based on various combinations of soils, physiographic, and vegetative variables, and
TWINSPAN followed by tabular arrangements of species abundance classes to formulate
ecological species groups. He did not use ordination methods to analyze soils data, used
commonly occurring species in ordination and clustering procedures, and did not describe
how the classification of ecological Species groups was integrated with analyses of soils
data to arrive at the final classification.

In this research, a series of complementary methods were used and compared to
one another to derive the classification of multifactor ecological land units and their
components. This contrasts in several respects with the past use of multivariate methods

in research including that conducted in Michigan and Wisconsin.

36

OBJECTIVES

The overall objectives for this research were to classify samples and species
groups in northeast lower Michigan using a combination of indirect and direct gradient
analyses, clustering techniques, and graphical and tabular methods. The specific
objectives for this research were to:

(1) Develop an integrated classification of local land units based on ground-flora,
and soils conditions in northeast lower Michigan;

(2) Document the use of a series of complementary methods and compare results
to one another to derive the final classification of multifactor ecological land units and
their components; and

(3) Determine if the integrated ecological land units classified using ordination
and clustering methods could be improved by eliminating ubiquitously or superfluously
distributed understory and ground-flora species based on their sums-of-squared cosines in
five-dimensional subspace and position within the arch of two-dimensional subspaces of

a correspondence analysis.

MATERIALS AND METHODS
Sample Design and Field Methods

A random, stratified sampling design was used to locate forty-seven late
successional forest stands on upland sites throughout the Huron National Forest,
Michigan (Figure 2.1). Stratification was based on general landform classes (Host 1987).

Only well-stocked stands with minimal evidence of recent disturbance (windthrow,

37

logging, firewood cutting) were selected for sampling. The sample unit was defined as a
relatively homogeneous area of at least one hectare (2.5 acres) within a randomly selected
Forest Service stand. Four random sample points were established within each

homogeneous area, and the sample points were averaged to produce summary data for

analysis. Sampling methods follow those recommended by Host et al. (1993)

 

 

0‘.) r a d, \
."1 ,.
r,» l r
L ‘ 9
\ "\
'1
i ,_ '\
"1" I v\(
i i A} "I ‘1 l 1 l i .( -
‘ f ,. ‘ ' l
k“ 77' I: ' .
3 it“; I
a l 1
, j- e , -.
‘ l l l
‘/ ‘ "> 1 .
r l,_ ., l _, I , A R
" ' l l . .
K i 5 l r 1| - r \1‘
‘1 l l 1 ‘1 1 ,1? \
. , i
.,,l- i— —l ‘ 1‘ (“L ,‘4, _ I l
i i . 11—: i .
\ 1 i _ i I l 4 ‘ ,1
\I I I ‘1 \ ‘ i l.
x ‘ i ‘ 7 ~
\ I l ' l a t
‘\ J i L ' '1‘ _ ‘ l \K
‘\ i ‘ II ‘ ‘ J]
x i ‘ ‘1 ‘ l 1‘ ‘1 9‘
l 1 1
""‘rl" 1“ I . 12“,
L l j l l l 1 H X’
r ’ 1‘ E \L - ,1 I(
I} _ |,__ , i . 4 I ’7‘.
f r 7' T l' l l ’/
/ i '- '. l 1
F l l ,1. I , .1
I” .7 7‘ #_ W 7 77’ 7, _ T \ 7
I/ 1 l ,r
I
"‘ l l 1 a r
/' fi__,__“_, 7 r L_L_-_ _‘ ‘ LflJ‘A ((((( A A” 7

L~-—"—”T’

Figure 2.1 Study area on the Huron NF in

Alcona, Iosco, Oscoda, and Ogemaw
counties.

Selection criteria were developed and used in locating sampling locations. These
criteria included:

(1) the overstory must be at least 40 years old;

(2) the canopy must be closed, as far as site conditions allow;

38

(3) density must be uniform throughout the stand, extensive open areas or wide

age distributions (range over 10 years) are cause for rejection;

(4) stand composition must be relatively uniform;

(5) the topography must represent upland conditions;

(6) the soils must be moderately or more well-drained; mottling in the upper 40

cm. is cause for rejection;

(7) aspen must not comprise more than 30% of total stand basal area;

(8) more than 30% of the overstory consisting of multiple stems (stump sprouts) is

cause for rejection; and

(9) evidence of cutting, thinning, underplanting or other disturbance within the

past 40 years is cause for rejection.

Photo-interpretation of color infrared (CIR) photographs in conjunction with
landform maps and Forest Service compartment maps were used to identify candidate
sampling locations. Candidate sampling locations were randomly selected from each
strata. Stands from each stratum were visited in the field and evaluated using the
rej ection/acceptance criteria. If the stand passed the acceptance criteria, a homogeneous

area (minimum size one ha) was delineated and defined as the research site.

Stand Methods
Stand measurements were made on four randomly located variable radius plots
using a 10 BAF prism. All live tally trees were measured for diameter at breast height

(dbh), species, total height, crown ratio, and crown class. Merchantable height was

39

estimated for all live tally trees with dbh greater than the minimum merchantable dbh
(dbh over 4 inches). All borderline trees were measured to determine whether they should
be tallied. Dbh was measured using a diameter tape and recorded to the nearest 0.1 inch.
Total height and merchantable height were recorded to the nearest foot using a Relaskop.
For cubic foot volume, merchantable height was measured from a one foot stump height
to a 4-inch top diameter, or to where merchantability was limited by the bole. The
Relaskop's optical dendrometer was used to determine the upper stem diameter and
height.

Two dominant, uninjured, and free growing trees on each stand sample point were
bored to determine average age at breast height (1.37 m). This was converted to total age
and with total height was used to estimate site index. Trees off the sample point but
within the homogeneous area were also measured to provide ancillary data on site index.

Secondary stand data processing included construction of stand and stock tables.
Gross cubic foot volume was calculated for the preliminary sample summaries using
Beers' version of the Gevorkiantz formula, a composite volume equation that has shown
good performance in the Lake States (Martin 1984). Site index was estimated using the

equations developed by Hahn and Carmean 1982.

Ground-flora Methods
Ground-flora composition and abundance were sampled using four five by thirty
meter plots centered on the overstory sample points. In flat terrain, plots were oriented on

north-south axes. In hilly terrain, plots were generally oriented perpendicular to the slope

40

to avoid sampling across different physiographic conditions.

Average percent ground cover was determined for all moss, herbaceous, and
woody species in the plot using a modified Braun-Blanquet cover-abundance rank
(Mueller-Dombois and Ellenberg 1974; Table 2.1). Abundance values were determined
by traversing the plot several times to record the species present, and then assigning
abundance values after the species list was compiled. Relative frequencies for
ground-flora species were determined by recording species presence/absence in six one
m2 frequency frames located at five m intervals along the long axis of each plot. These
plots were also used to determine seedling densities in a subset of sample plots (Host et
al. 1987). Nomenclature for vascular plants followed Gleason (1952); nomenclature for
bryophytes followed Crum and Anderson (1981). Samples of unknown species were
collected, pressed, and identified at the Michigan State University Beal-Darlington

Herbarium.

Table 2.1 Cover-abundance classes and ranks used in field sampling and data analysis.

Class midpoint Range of cover Rank
r Trace - 0.1 l
+ 0.1 — 1.0 2
2 1 — 2 3
10 2 - 15 4
25 15 - 33 5
50 33 - 66 6
8O 66 — 100 7

41

Soil Methods

The forest floor, soil solum, substrata and groundwater characteristics were
sampled at four points in each research site. Information at each point was collected from
soil pits extending into C horizons, and to depths up to 4.5 m using a 9 cm. diameter
bucket auger. Forest floor and soil horizons were described based on color, texture,
coarse fragments, mottles, structure, soil reaction, clay skins, uncoated particle
differences, soil reaction and rooting. Soil textures estimated in the field were converted
into categorical variables for numerical analyses corresponding with the centroid of the
textural triangle (Table 2.2; USDA Soil Taxonomy, Padley 1989). Drainage classes were
also coded into categorical variables (Table 2.3). Soil depth was defined according to
Soil Taxonomy (1975) as the lower limit of biological activity, which coincides with the
rooting depth of native perennial plants. The presence of deep lying textural bands has
been shown to have a significant influence on tree growth and species composition (Hart
et al. 1969, Hannah and Zahner 1970, Cleland et al. 1985, Host et al. 1988). It was
therefore important to record the presence, absence, or intensity of deep-lying textural
bands and data were collected for the upper 4.5 m in soils with stratified sandy and loamy
sand upper sola.

A Banding Intensity and Continuity variable (BIC) representing substratum
characteristics was described categorically for data analysis. BIC values assigned to
sub-plot locations were: 0 for sands with no textural bands; 1 for loamy sand to sandy
loam bands less than 5 cm. thick; 2 for sands with sandy loam bands 5 to 15 cm.

cumulative thickness; 3 for sands with sandy loam bands greater than 15 cm. in

42

thickness; and 4 for sands with sandy clay loam or finer bands greater than 10 cm. thick.
Substratums were not sampled when sola of sandy clay loam or finer textures greater than

30 centimeter were encountered.

Table 2.2. Field texture codes for soil variables.

Texture Sand% Silt% Clay% Silt+Clay%
s 90 5 5 10
£8 88 6 6 12
vfs 86 7 7 14
ls 83 12 5 17
lfs 81 13 6 19
$1 65 25 10 35
fsl 63 26 11 37
l 42 4O 18 58
scl 57 15 28 43
cl 33 35 32 67
sil 20 65 15 80
si 10 85 5 90
sic 6 47 47 94
sicl 10 55 35 90
c 20 2O 6O 80
fscl 55 16 29 45
vfsl 6O 28 12 40

Table 2.3. Soil drainage class codes.

CODE DRAINAGE CLASS

Excessively well drained
Somewhat excessively drained
Well drained

Moderately well drained
Somewhat poorly drained
Poorly drained

Very poorly drained

\ImU'lubWNl-J

 

43

Analysis of Vegetation

Vegetative data were interrogated using ordination, agglomerative and divisive
clustering, graphical and tabular methods. Vegetative data transformed based on a
modified Braun-Blanquet cover-abundance rank (Table 2.1; Host et al. 1993) were
ordinated using correspondence analysis (CA). Graphical displays of the first five CA
axes were examined to detect sample outliers. The data set was reduced as necessary, and
CA was used again to produce graphical displays of samples in species space, and species
in sample space.

Samples were classified based on commonly occurring species using both
agglomerative and divisive hierarchical clustering procedures. For agglomerative
clustering, Ward's, flexible beta using a -0.25 beta value, and average linkage methods
were used. Divisive hierarchical clustering methods included ordination space
partitioning and TWINSPAN. A subjective decision that a minimum cluster size of three
samples was used in interpreting dendograms. Results of these four clustering methods
were compared to one another using graphical and tabular methods. Results were also
used in exploratory ordination space partitioning of samples in species space based on

CA.

Analysis of Soils Data
Soils data were interrogated using principal components analysis, agglomerative
clustering, graphical and tabular methods. Explorative data analysis was conducted on

soil variables to test for assumptions underlying PCA, and to detect intercorrelated

44

variables for data reduction purposes. Descriptive statistics and histograms were
examined, and means and variances were correlated to test the assumptions of normality,
homogeneity of variance and uncorrelated errors underlying the PCA model (Appendix
A).

Soils data were standardized to a mean of zero and unit variance and ordinated
using PCA. Correlations among soil variables, variable weights, and factor loadings were
examined to eliminate redundant or unimportant variables in clustering procedures.
Variables with low weights and low loadings were removed, and the reduced soils data
set was ordinated again using PCA to reassess data reduction needs.

Prior to clustering, raw data for the reduced number of soil variables were
transformed using a range transformation in which individual values were divided by the
maximum minus minimum values recorded for respective variables. Range transformed
soils data were classified using agglomerative clustering procedures and Wards, flexible
beta with a -0.25 beta value, and average linkage methods.

Tabular methods were used to compare and synthesize results of the analyses of
vegetation and soils data. Relationships between soil morphological and physical
characteristics and ground-flora composition within and across samples were subjectively
examined, and ordinal placement of samples was adjusted. Results from tabular
comparisons suggesting the need for filrther subdivision of sample groups were used to

subset samples occurring within heterogeneous clusters.

 

45

Reanalysis of Vegetation

Analyses of ground-flora data were repeated in a second iteration on a data set for
which species with low fidelity to five-dimensional ordination space were removed. The
mass, inertia, and quality of display for species in each of the first five dimensions were
examined, and species with less than 45% of their total variability, or sum-of-squared-
cosines, accounted for in five dimensions were removed to create a more parsimonious
data set for subsequent clustering procedures. The reduced vegetative data set was
reclassified using ordination and clustering procedures, and results of each method were
compared to one another using graphical and tabular methods. Samples were classified
using Ward's, flexible beta (-0.25 beta value) and average linkage methods. Divisive
hierarchical clustering methods included ordination space partitioning and TWINSPAN.

Results of these four clustering methods were compared to one another using
graphical and tabular methods. Results were also used in ordination space partitioning of
samples in species space based on CA.

A final set of synthesis tables was constructed based on high quality species and
important soil variables. Samples were grouped according to interpretation of ordination
and clustering results. Samples were then subjectively reordered using tabular methods to
arrive at final groupings. Ecological unit descriptions were based on these final

groupings.

46

RESULTS
Analysis of Commonly Occurring Understory and Ground-flora Species

Table 2.4 displays the decomposition of inertia, or variability, for the first fifteen
principal coordinates of a CA of forty-six samples and 104 commonly occurring
understory and ground-flora species. F ifty-one per cent of the total variability in firll
dimensional space was accounted for by the first five principal coordinates.

Figure 2.2 arrays samples in the first and second dimensions of commonly
occurring species space based on CA. Based on initial evaluation of ground-flora

composition and soil textures, the first axis is interpreted to represent a decreasing

 

 

 

 

1.5
26
1
3'9
1
1
05 9q1 2‘96
- e 4a
443
5
N 7
5 o 35’ 7 4 23
are 13.2 33
2" 2,6 as
-O.5 14 as
24 13 23
.1
-1.5
.1.5 -1 -o.5 o 0.5 1 1.5
CA1

Figure 2.2. Samples in CA dimensions 1 and 2 of
commonly occurring species space labeled by sample
number.

47

moisture gradient with increasing sample scores. The second axis appears to ordinate
samples along a successional stability gradient. The interpretation of gradients expressed

along principal coordinates is offered in the discussion section of this chapter.

Table 2.4. Inertia and Chi-Square decomposition of the first fifteen principal coordinates.

Singular Principal Chi-

Values Inertias Squares Percents
0.78191 0.61138 1181.65 25.42%
0.51640 0.26666 515.40 11.09%
0.37222 0.13855 267.78 5.76%
0.35271 0.12441 240.44 5.17%
0.32661 0.10667 206.18 4.44%
0.30059 0.09036 174.63 3.76%
0.29766 0.08860 171.25 3.68%
0.27453 0.07537 145.66 3.13%
0.26933 0.07254 140.20 3.02%
0.25966 0.06742 130.31 2.80%
0.23066 0.05321 102.83 2.21%
0.22441 0.05036 97.33 2.09%
0.22113 0.04890 94.51 2.03%
0.21610 0.04670 90.26 1.94%
0.20878 0.04359 84.25 1.81%

Figures 2.3 and 2.4 display the same ordination as Figure 2.2, samples in the first
and second dimensions of commonly occurring species space, with samples labeled with
soil textures in the upper 30 cm., and in substrates between 150 and 250 cm.,
respectively. These figures show edaphic gradients in silt plus clay content in the upper
30 cm. and in the substrates of soil profiles. Silt plus clay content decreased with
increasing sample scores along the first axis, with samples grading from mesic to xeric

due to differences in soil textures and moisture regimes.

48

1.5

 

 

 

 

§
’ §
LS
S
S S SS
0.: s
m.
" L
3 List.
LS SS
0 Fs’IISL 5
§
SS
sis Lis 3
LS EL LES
S
.o,‘ S 15
L S é{S
-1
4.5 or 4.5 o 0.5 1 1.

Figure 2.3. Samples in CA dimensions 1
and 2 of commonly occurring species space
labeled by soil texture of upper 30 cm.

"5 .. l- w .13-“. _. .. l 1.-“ ”1...-

 

§
1
E§
SCI.
S
Sgtl‘ SE‘S
0.5 LS
N L
g “fat
SCI. 3.3
o SCSCI. fl}
89a sxs
LSS‘IEL SCL
SL EEL SCSI.
-0.8 SI SCL §EL
SIC%ICL SICL

 

 

 

0 1.!

CA1

Figure 2.4. Samples in CA dimensions 1 and
2 of commonly occurring species space
labeled by substrate textures at 100 to 250
cm.

Figures 2.5, 2.6, 2.7, and 2.8 display the partitioning of the first two dimensions of

samples ordinated in species space based on results of three agglomerative clustering

linkage methods (average, Ward’s, and flexible beta linkage), and the third level of

TWINSPAN divisive clustering using cut levels of 0, 0.5, 1.5, and 3.0 for rank ground-

flora abundance. Results from the three agglomerative linkage methods were relatively

consistent, with six clusters identified. Arrows indicating separation of sample groups at

respective levels of each dendogram have been added to facilitate reader’s interpretation.

Average linkage placed samples in the

same clusters as flexible beta linkage

method with the exception of sample 18, which was an outlier in the former. Average

linkage placed samples in the same clusters as Ward’s with the exception of sample 20.

 

49

Flexible beta linkage placed samples in the same clusters as Ward’s linkage method with
the exception of samples 14 and 48. TWINSPAN identified 5 clusters, with 3 sets of
outliers consisting of l or 2 samples identified. Thus TWINSPAN results were somewhat
different from results of agglomerative clustering procedures. Further interpretation of

the classification of samples within clusters was delayed until soils data were analyzed.

50

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ovméam nnwénm
m U z ¢ a. m H a

 

 

 

 

 

 

CA1

Figure 2.5. Exploratory ordination space partitioning of

samples in CA dimensions 1 and 2 of commonly occurring
species space based on clustering results using average

linkage.

51

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

"n.nunn .«o.vvun huh.voh onv.oon nun‘
I O I 1 u. a u n

 

 

 

 

 

 

 

CA1

Figure 2.6 . Exploratory ordination space partitioning of
samples in CA dimensions 1 and 2 of commonly occurring

species space based on clustering using flexible beta

linkage.

52

 

 

 

 

 

 

 

 

 

 

D

ho.nonu

P
4

P

ooh.o.hn

P
4

 

 

 

 

P F

a...nonn

 

«Ho.noo «Nu.u
MUZ‘HmHO

 

 

 

 

 

 

(TA

Figure 2.7. Exploratory ordination space partitioning of

samples in CA dimensions 1 and 2 of commonly occurring
species space based on clustering using Ward’s linkage.

 

34 l l
83 890345121

00 000000000
00 000000000
00 111111111

-1

Figure 2.8. Exploratory ordination space partitioning of samples in
CA dimensions 1 and 2 of commonly occurring species space based

224222 11
4926786739

0000000000
1111111111
0000000000

H

I—‘O

53

132234 3334444
452304 4236157

111111 1111111
000000 0000000
000000 1111111

124334433
210198967

111111111
111111111
000000000

22
05

11
ll
11

 

 

 

 

 

-10, -1

on clustering using TWINSPAN.

Sample
number

TWINPAN
division

level 1
level 2
level 3

54

Analysis of Soils Data.

Soils data was interrogated using PCA, agglomerative clustering, graphical, and
tabular methods. Principal component analysis is based on an assumed underlying linear
model (Morrison 1976), and thus data were tested to determine if variables were normally
distributed with uncorrelated means and errors. Results indicate that most variables were
not normally distributed (appendix A). Means and errors were correlated (r2 = .416, p <
.01; Figures 2.9 and 2.10). The correlation matrix rather than the variance-covariance

matrix was therefore used in principal component analyses of soils data

 

 

’5. ‘ g 5“ . u
it . i - ‘6 f“
a. ‘ . - 3" .~ -

 

 

 

. ' 1:1. .'

coo woo 1m "0 1 2 a 4
Man

 

 

Lon or non
Figure 2.9. Mean and standard deviation Figure 110- Log”, 0f mean and standard
for 35 soil variables. deviation for 35 soil variables.

Principal Component Analysis of Soils Data
In the decomposition of variability of the soils data using PCA (Table 2.5), sixty-

eight percent of the overall variability set was accounted for in the first five dimensions,

55

with thirty-seven percent accounted for in the first dimension, twelve percent in the
second, eight percent in the third, six percent in the fourth, and five percent in the fifth.
The soils data set was not summarized well in low dimensional subspaces, and patterns in
the third and higher dimensions did not assist in the interpretation of samples in variable

space or variables in sample space.

Table 2.5. Eigenvalues of the soil correlation matrix.

Eigenvalue Difference Proportion Cumulative
PRINl 13.0212 8.83728 0.372035 0.37203
PRINZ 4.1839 1.40512 0.119541 0.49158
PRIN3 2.7788 0.77655 0.079395 0.57097
PRIN4 2.0023 0.15331 0.057208 0.62818
PRINS 1.8490 0.41870 0.052828 0.68101
PRIN6 1.4303 0.20308 0.040865 0.72187
PRIN7 1.2272 0.12913 0.035062 0.75693
PRIN8 1.0981 0.18705 0.031373 0.78831
PRIN9 0.9110 0.03433 0.026029 0.81434
PRIN10 0.8767 0.08809 0.025048 0.83938
PRINll 0.7886 0.06999 0.022531 0.86191
PRIN12 0.7186 0.10785 0.020531 0.88245

Table 2.6 arrays soil variable names and abbreviations. Table 2.7 arrays the first
six eigenvectors of a PCA with soils variables ordered by variable weights in the first
dimension. Important variables based on high positive first dimensional weights included
soil textures in substrata, then textures in sola. Important variables based on high
negative first dimensional weights included variables describing depths to textural
discontinuities, thickness of sandy Bs horizons, and depths to mottles and watertables.

Important variables based on high positive second dimensional weights (Table

2.8) included thickness of 02 and E horizons, E horizon development code, depths to

56

mottles, dominant texture of sola and substrates, dominant texture of discontinuities, and
average texture of discontinuities. Important variables based on high negative second
dimensional weights included B horizon thickness, depth to C horizons, A1 horizon
thickness, depth to accumulation of varves, and drainage class.

The most important soil variables based on high positive first dimensional factor
loadings, or variable correlations with variates derived in the first principal component,
included a coded variable describing substrate textures, textures between 100 and 150
cm., and sola textures (Table 2.9). The most important soil variables based on negative
factor loadings included depth to heavy textured horizons, depths to mottles and water
tables, and Bs horizon thickness.

The most important soil variables based on high positive second dimensional
factor loadings included depths to mottles, dominant texture of sola and substrates,
textures between 100 and 250 cm., and depth to watertables (Table 2.10). The most
important soil variables based on high negative second dimensional factor loadings
included drainage class code, texture of the top 30 cm., and a coded variable describing

substrata textures.

 

57

Table 2.6. Soil variable names and abbreviations.

VARIABLE

Average banding intensity code
Texture of heaviest band
Texture of SCL or heavier bands
Texture of 150—450 cm.

B horizon texture

Texture of upper 150 cm.
Average texture to 450 cm.
Dominant banding texture
Drainage class code

Number of textural discontinuities to 150 cm.

Texture to 30 cm.

Depth to B horizon

Depth to C horizon

B horizon thickness

E horizon thickness

B intensity code1

A horizon thickness
Physiographic form

B horizon development code2
Aspect

02 horizon thickness

Depth to vfs accumulations > 15 cm.
Elevation

Slope

Depth of major rooting zone

A1 horizon value

Depth to LS accumulation > 15 cm.
B horizon value

Gross water table depth

Bs horizon thickness

Depth to mottles

Depth to till

Depth to SCL

Depth to SL accumulation > 15 cm.
Depth to SCL accumulation > 15 cm.

1. B intensity code: 1 = Bw, 2 = Bs, 3 = Bh

ABBREVIATION

BIC8
MAXSIC
TXSICL
LTSICL
BSICL
UTSICL
SICL
DOMSICL
DCL
LD150
T3OSICL
BDEP
CDEP
BTH
ETH
BIN
AlTH
PF
EDC
ASP
02TH
ACV
ELEV
SLOP
EFFDE
AlV
ACLS
BVA
GWTDE
BSTH
MOTDE
TILLD
DEPSCL
ACSL
ACSCL

2. B horizon code: 1 = mixed, 2 = discontinuous under <50% of pedon, 3 = discontinuous

under >50% of pedon, 4 = continuous

 

58

Variable weights in the first dimension and factor loadings in the first and second
dimensions identified the same important variables, and these variables primarily related
to soil textures and drainage classes. These factors directly affect moisture and nutrient
regimes and hence were ecologically important. Soil textures directly affect water
holding capacity, and also chemical equilibria and nutrient availability. Drainage class
codes, depths to mottles, and depths to water tables are measures of free water available
for tree, understory, and to a lesser extent ground-flora consmnption. Variable weights in
the second dimension identified 02, E, B, and A1 horizon thicknesses and development
codes as important variables. In addition, variable weights in the second dimension
identified depth to mottles, dominant banding textures, depth to varves, and drainage
class as important variables, similar to weights in the first dimension and factor loadings

in the first and second dimensions.

59

Table 2.7. Soil variables ordered by variable weights in the first dimension of a PCA.

VAR

BIC8
MAXSIC
TXSICL
LTSICL
BSICL
UTSICL
SICL
DOMSICL
DCL
LDlSO
T3OSICL
BDEP
CDEP
BTH
ETH
BIN
AlTH
PF

EDC
ASP
02TH
ACV
ELEV
SLOP
EFFDE
AlV
ACLS
BVA
GWTDE
BSTH
MOTDE
TILLD
DEPSCL
ACSL
ACSCL

lllll
OOOOOOOOOOOOOOOOOOOOOOOOOO

PRINl

.26088
.25424
.23284
.23064
.22443
.22027
.21581
.21210
.20888
.20632
.19464
.14638
.13233
.12711
.12375
.08624
.07510
.06737
.06550
.00905
.00305
.01725
.02741
.03647
.06888
.11272
.12209
.12257
.13729
.14648
.15452
.23641
.24134
.24897
.25080

000000

PRINZ

.07806
.00516
.18292
.16174
.02141
.08720
.20936
.22086
.16096
.10888
.11219
.02743
.33843
.35015
.27175
.09360
.30910
.04727
.26104
.17523
.33550
.20019
.09300
.07952
.03730
.03344
-O.
-0.

0.
.00589
.25778
.02516
.00472
.09003
.05912

07776
01335
10503

000

l
0

00000000000000

PRIN3

.00927
.03327
.00540
.05627
.05929
.02322
.06652
.10450
.12965
.09067
.00707
.34369
.11880
.08038
.18260
.02232
.31817
.05953
.11120
.43895
.00020
.00001
.08468
.29356
.32810
.39302
.21874
.15128
.17595
.01535
.02719
.08757
.03789
.01380
.03693

PRIN4

.06334
.03571
.00074
.04761
.07258
.10159
-0.
.08686
.15876
.05569
.02122
.21280
.06834
.13397
.06375
.24810
.13256
.17168
.27689
.10630
.32585
.03233
.56457
.33608
.16249
.16352
.10589
.09865
.13386
-0.
.14794
.00935
.09022
.07362
.05425

05099

00637

-0

I
O

OOOOOOOOOOOOOOO

PRINS

.03260
.11722
-0.
-0.
-0.
.13309
.08636
.02732
.02534
.01330
.27743
.26421
.12624
.13021
.26667
.14818
.13066
.25538
.28118
.07609
.00052
.24902
.05477
.02467
.04336
.12842
.16088
.23272
.29512
.44384
.12145
.03171
.02468
.01980
.00285

07531
00824
21899

OOOOOOOOOOO

PRIN6

.09904
.08593
.08200
.06962
.20384
.13282
.01356
.03963
.00887
.05059
.00884
.09225
.17767
.13616
.00151
.44605
.02937
.52302
.07200
.09471
.02747
.27480
.25036
.00813
.07803
.12214
.03920
.30379
.29367
.02500
.05268
.04143
.10390
.06411
.03631

60

Table 2.8. Soil variables ordered by variable weights in the second dimension of a PCA.

VAR PRINI PRINZ PRIN3 PRIN4 PRINS PRIN6
02TH 0.0030 0.3355 0.0002 0.3258 0.0005 -0.0275
ETH 0.1237 0.2717 0.1826 -0.0637 0.2667 -0.0015
EDC 0.0655 0.2610 -0.1112 0.2769 0.2812 -0.0720
MOTDE -0.1545 0.2578 0.0272 -0.1479 -0.1215 0.0527
DOMSIC 0.2121 0.2209 0.1045 —0.0869 0.0273 —0.0396
SICL 0.2158 0.2094 0.0665 -0.0510 -0.0864 -0.0136
TXSICL 0.2328 0.1829 0.0054 -0.0007 -0.0753 0.0820
ASP 0.0091 0.1752 0.4390 0.1063 -0.0761 -0.0947
LTSICL 0.2306 0.1617 -0.0563 0.0476 -0.0082 0.0696
LDlSO 0.2063 0.1089 -0.0907 -0.0557 0.0133 -0.0506
GWTDE -0.1373 0.1050 0.1760 0.1339 -0.2951 —0.2937
BIN 0.0862 0.0936 0.0223 —0.2481 0.1482 0.4461
ACSL -0.2490 0.0900 -0.0138 -0.0736 0.0198 0.0641
UTSICL 0.2203 0.0872 0.0232 -0.1016 -0.1331 0.1328
SLOP -0.0365 0.0795 0.2936 0.3361 -0.0247 -0.0081
ACSCL -0.2508 0.0591 -0.0369 -0.0543 0.0028 0.0363
EFFDE -0.0689 0.0373 0.3281 0.1625 -0.0434 0.0780
BDEP 0.1464 0.0274 0.3437 -0.2128 0.2642 -0.0923
TILLD -0.2364 0.0252 -0.0876 0.0094 0.0317 0.0414
BSTH -0.1465 0.0059 -0.0153 -0.0064 0.4438 -0.0250
MAXSIC 0.2542 0.0052 0.0333 -0.0357 -0.1172 0.0859
DEPSCL -0.2413 0.0047 0.0379 -0.0902 0.0247 0.1039
BVA -0.1226 -0.0134 0.1513 0.0986 -0.2327 0.3038
BSICL 0.2244 -0.0214 0.0593 -0.0726 -0.2190 0.2038
AlV —0.1127 -0.0334 0.3930 -0.l635 0.1284 0.1221
PF 0.0674 -0.0473 —0.0595 0.1717 0.2554 0.5230
ACLS -0.1221 -0.0778 0.2187 -0.1059 -0.1609 0.0392
BIC8 0.2609 -0.0781 0.0093 0.0633 0.0326 -0.0990
ELEV -0.0274 -0.0930 0.0847 0.5646 0.0548 0.2504
T3OSICL 0.1946 -0.1122 0.0071 0.0212 -0.2774 -0.0088
DCL 0.2089 -0.l610 -0.1296 0.1588 0.0253 -0.0089
ACV —0.0172 -0.2002 0.0000 0.0323 -0.2490 0.2748
AlTH 0.0751 -0.3091 0.3182 -0.1326 0.1307 0.0294
CDEP 0.1323 -0.3384 0.1188 0.0683 0.1262 -0.1777
BTH 0.1271 -0.3502 0.0804 0.1340 0.1302 -0.1362

61

Table 2.9. Factor patterns for soil variables ordered by variable loadings in the first
dimension of a PCA.

Var

BIC8
MAXSIC
TXSICL
LTSICL
BSICL
UTSICL
SICL
DOMSICL
DCL
LD150
T3OSICL
BDEP
CDEP
BTH
ETH
BIN
AlTH
PF

EDC
ASP
02TH
ACV
ELEV
SLOP
EFFDE
AlV
ACLS
BVA
GWTDE
BSTH
MOTDE
TILLD
DEPSCL
ACSL
ACSCL

FACTORl

.9414
.9174
.8402
.8323
.8099
.7949
.7788
.7654
.7538
.7445
.7023
.5282
.4775
.4587
.4465
.3112
.2710
.2431
.2364
.0327
.0110
-0.0622
-0.0989
-0.1316
-0.2485
-0.4068
-0.4406
-0.4423
-0.4954
-0.5286
-0.5576
-0.8531
-0.8709
-0.8984
-0.9050

OOOOOOOOOOOOOOOOOOOOO

FACTORZ

-0.1597
0.0106
0.3742
0.3308

-0.0438
0.1784
0.4282
0.4518

-0.3292
0.2227

-0.2295
0.0561

-0.6922

-0.7l62
0.5559
0.1915

-0.6323

-0.0967
0.5340
0.3584
0.6863

-0.4095

-0.l902

0.1627,

0.0763
-0.0684
-0.1591
-0.0273

0.2148

0.0121

0.5273
0.0515
0.0097
0.1841
0.1209

FACTOR3

.0155
.0555
.0090
-0.0938

0.0988
.0387
.1109
.1742
.2161
.1512
.0118
.5729
.1980
.1340
.3044
.0372
.5304
.0992
.1854
.7317
.0003
.0000
.1412
.4894
.5469
.6552
.3646
.2522
.2933
-0.0256

0.0453
-0.1460

0.0632
-0.0230
-0.0616

GOO

COD

1
O

1
00000000

I
O

OOOOOOOOOOO

FACTOR4

0.0896
-0.0505
-0.0011

0.0674
-0.1027
-0.1438
-0.0722
-0.1229

0.2246
-0.0788

0.0300
-0.3011

0.0967

0.1896
-0.0902
-0.3511
.1876
.2429
.3918
.1504
.4611
.0457
.7989
.4756
.2299
.2314
.1498
.1396
.1894
-0.0090
-0.2093

0.0132
-0.1277
-0.1042
-0.0768

OOOOOOOOO

II
0000

FACTORS

0.0443
-0.1594
-0.1024
-0.0112
-0.2978
-0.1810
-0.ll74
0.0372
0.0345
0.0181
0.3773
0.3593
0.1717
0.1771
0.3626
0
0
0
0

.2015
.1777
.3473
.3823
-0.1035
0.0007
-0.3386
0.0745
-0.0336
-0.0590
0.1746
-0.2188
-0.3164
-0.4013
.6035
.1652
.0431
.0336
.0269
.0039

000000

FACTOR6

0.1184
0.1028
0.0981
0.0833
0.2438
0.1589
0.0162
-0.0474
-0.0106
-0.0605
-0.0106
-0.1103
-0.2125
-0.l628
-0.0018

0.5335

0.0351

0.6255
-0.0861
-0.ll33
-0.0329

0.3286

0.2994
-0.0097

0.0933

0.1461

0.0469

0.3633
-0.3512
0.0299
0.0630
0.0496
0.1243
0.0767
0.0434

62

Table 2.10. Factor patterns for key soil variables ordered by variable loadings in the
second dimension of a PCA.

Variable FACTORZ
MOTDE 0.52728
DOMSICL 0.45176
LTSICL 0.33082
GWTDE 0.21483
ACSL 0.18414
UTSICL 0.17836
ACSCL 0.12092
DEPSCL 0.00966
BSICL -0.04380
BIC8 -0.15967
TBOSICL -0.22947
DCL -0.32923

The first PC axis ordinated samples along a textural and moisture gradient.
Textures of the upper 30 cm. and 100 cm. graded from sands to sandy loams with
increasing sample weights along the first axis (Figures 2.11 and 2.12). Textures of the
lower 150 cm. graded from sands to sandy clay loam with increasing sample weights
along the first axis (Figure 2.13). Soil drainage graded from excessively well drained to
well drained with increasing sample weights along the first axis (Figure 2.14).

The second axis ordinated samples along a substrate silt content gradient (Figure
2.13), and along a drainage class gradient in which samples containing moderately well
drained soils were separated from well drained soils (Figure 2.14). Silt content of the soil
horizons at depths of 100 to 250 cm. increased somewhat with increasing sample weights
along the second axis (Figure 2.13). Soil drainage graded from well drained to
moderately well drained with decreasing sample weights along the second axis, although

the gradient was not as distinct as that expressed along the first axis (Figure 2.14)

63

 

 

 

 

3
L8
2
S S s SLSL
1 S
S LS LSS L
S
2 s 3 LS LS SL
g0 3:35 3 L5 3 F§L
3 LS LSLFSLLS
1 1'3 SL
EELFSI.
'2 LS
SL
-3
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
PCA1

Figure 2.11. Samples in PCA dimensions 1 and 2 of
soil variable space labeled by texture of upper 30
cm.

 

 

 

 

3 1
L5
2
S S s SLSL
1 s
3 L8 L83 L
3 LS
2 s s 3 LS s
8 0 SS 5 LS F§L L
S 5'3 L8
5 Ls LFSL
1 1‘3 SL
stLFSL
2 Ls
SL
-3
-2 -1.5 -1 -o.s o 0.5 1 1.5 2
PCA1

Figure 2.12. Samples in PCA dimensions 1 and 2 of
soil variable space labeled by texture of upper 100
cm.

64

 

 

 

 

 

 

 

 

3 3
SICL WD
2 2
I SIC wo we
1 SICL 1 WO
8 SL SGI. s|c|_ EXD WDWD WDD Mwo WD
SL
2 s s SCL SCL 3'“ 2 E xo wo wo wo
0 ° gEllis 3'- SCL 33511 5'“ 8 ° DEXD WD m
‘L s sc W
5 set LSCL D we quowo
558 121%
-1 L SCL -1 MWD
SIBSet We
-2 SCL '2 MWD
scr. MWD
-3 3
.2 1 5 .1 .o 5 o 0.5 1 15 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
PCA1 PCA 1

Figure 2.13. Samples in PCA dimensions 1 Figure 2.14. Samples in dimensions 1 and 2
and 2 of soil variable space labeled by of soil variable space labeled by drainage
texture at 100 to 250 cm. class.

Clustering of Samples Based on Soil Variables

Seven soil variables were removed for clustering analyses because of
intercorrelations and low weights and loadings based on PCA. These variables included
texture of heaviest band, texture of bands of sandy clay loam or heavier textures,
physiographic form, aspect, elevation, slope, and depth to till.

Clustering of samples using twenty-eight soil variables based on Ward’s linkage
identified seven groups of samples (Figure 2.15, Table 2.11). The cluster labeled 1
consisted of samples composed of deep sandy soils. The cluster labeled 2 consisted of

samples with sandy sola overlying loamy sand or sandy loam substrata. The cluster

65

labeled 3 consisted of samples with sandy sola overlying fine loamy substrata of silty
loam, sandy clay loam, or clay loam textures. The cluster labeled 4 consisted of samples
with loamy sand sola overlying sandy clay loam substrata; one sample had sandy loam
sola. The cluster labeled 5 consisted of samples with heterogeneously textured sola; four
samples had sandy loam sola and three samples had loamy sand sola. All samples in
cluster 5 had sandy clay loam substrata. The cluster labeled 6 was heterogeneous, with
sola ranging from loamy sand to fine sandy loam and substrata ranging from sandy loam
to silt loam to sandy clay loam. The cluster labeled 7 consisted of samples with
heterogeneous sola textures and silty clay loam substrata.

Clustering of samples using twenty-eight soil variables based on average linkage
identified eight groups of samples (Figure 2.16, Table 2.12). The cluster labeled 1
consisted of samples with loamy sand sola overlying sandy clay loam substrata. The
cluster labeled 2 consisted of samples with sandy loam and fine sandy loam sola
overlying silty clay loam substrata. The cluster labeled 3 consisted of samples with
loamy sand sola overlying heterogeneous substrata with sandy loam, silt loam, and sandy
clay loam textures. The cluster labeled 4 consisted of samples with sand loam sola
overlying sandy clay loam substrata. The cluster labeled 5 consisted of sandy loam sola
overlying silty clay loam substrata; one sample was loamy sand over sandy loam. The
cluster labeled 6 consisted of heterogeneous sola textures and silty clay loam substrata.
The cluster labeled 7 consisted of samples with sandy sola and sandy clay loam substrata.

The cluster labeled 8 consisted of samples with sand sola and substrata.

0.000

group

sample

665

5.000

12-——
+—
41- ||
3l-II

+..
36-- L
25—-—+
39-?
49—+

+—
48--I

+
4o---

+-||

 

 

‘Iv----+

 

 

 

+
42—--—
24--—- l

+_—
29---- l

+

23---- I
+..

2--—I ’
+—
3o---

+_

44 -----

\rq‘rqqq‘1\I\IqmmmmmmmmmwmmmmhbnnwwwwNNNNNI-IHHHHHHHI—a
U1
1
I

Figure 2.15. Sample clusters based on 28 soil variables using Ward’s linkage.

67

Table 2.11. Soils table arraying key variables for samples grouped by Ward’s linkage.

CHNDUP STA T30 UTEX LTEX BTEX DPSCL ACSL ACSCL DOMB DCL MOTDE GWTDE
1 12 S S S S 500 500 500 S WD 290 500
l 41 S S S S 500 500 500 S EXD 500 500
l 31 S S S S 500 500 500 S EXD 500 500
l 36 S S S S 500 500 500 S EXD 500 500
l 25 S S S S 500 500 500 S EWD 500 500
1 39 S S S S 500 500 500 S EXD 500 500
l 49 S S S S 500 500 500 S EXD 500 500
1 48 S S LS S 500 500 500 LS EXD 500 500
l 40 S S S S 500 500 500 S EXD 442 500
2 21 S S S S 500 500 500 LS SED 500 500
2 37 S S LS LS 500 500 500 LS EWD 500 500
2 20 S S SL S 500 240 500 SL EXD 500 500
2 46 LS S SL LS 244 500 500 SL WD 500 500
2 45 S LS SL LS 183 335 500 SL WD 500 500
3 33 S S SCL L 46 290 79 SICL EXD 500 500
3 34 S S SIL S 259 274 500 SIL EXD 290 500
3 14 S SCL SCL S 106 500 111 SCL WD 380 500
3 32 S S CL S 87 154 205 SICL EXD 500 500
4 47 L8 S SCL SL 37 185 52 SCL WD 500 500
4 35 LS LS SCL SL 32 60 500 SCL WD 500 500
4 1 LS SL SCL LS 45 70 50 SCL WD 500 500
4 26 SL SCL SCL SCL 40 18 59 SCL WD 500 500
5 10 SL SL SCL L 88 30 93 SCL MWD 45 82
5 5 SL SL SCL SL 15 15 98 SCL MWD 36 500
5 9 SL L SCL L 52 18 57 SCL MWD 23 500
5 ll LS SL SCL L 116 61 143 SCL MWD 55 500
5 8 SL SL SCL SL 105 17 137 SCL WD 168 500
5 19 L8 SL SCL SCL 45 38 69 CL WD 161 161
5 l3 LS SCL SCL SL 81 17 86 SCL WD 234 366
6 4 LS LS SIL LS 179 142 185 SCL MWD 94 203
6 3 LS LS SCL SL 118 15 155 SCL WD 45 500
6 7 FSL SCL SCL SL 74 57 79 SCL WD 129 305
6 6 SL SCL SCL L 138 21 76 SCL WD 167 366
6 22 LS SL SCL SL 133 136 268 SCL MWD 171 309
6 28 LS S SL LS 125 198 244 SCL WD 135 194
6 38 FSL SCL SCL SCL 171 173 187 SCL WD 153 411
7 18 S L SICL S 133 134 140 SICL WD 500 500
7 43 SL SICL SICL L 21 110 40 SIL WD 374 389
7 27 SL SICL SICL SICL 54 34 72 SICL WD 500 500
7 42 L8 SICL SICL SL 165 37 183 SICL WD 500 500
7 24 L SICL SICL SICL 48 22 60 SICL WD 111 500
7 29 SL SICL SICL SICL 32 15 47 SICL WD 106 500
7 23 S SCL SICL L 100 48 115 SICL MWD 90 145
7 2 FSL FSL SICL SL 43 25 49 SCL MWD 24 500
7 30 SL SCL SICL SCL 41 26 57 SCL MWD 120 155
7 44 L8 SICL SICL SCL 28 50 44 SCL MWD 18 101

68

Clustering results between the two linkage methods were not fully consistent.
Cluster one of Ward’s linkage was equivalent to cluster seven of average linkage. Cluster
three of Ward’s linkage was equivalent to cluster six of average linkage. Cluster seven of
Ward’s linkage was equivalent to average linkage clusters two and five combined. Four
of seven samples in cluster five of Ward’s linkage were the same as four of six samples in
average linkage cluster four. Four of seven samples in cluster six of Ward’s linkage were
the same as four of six samples in average linkage cluster one.

Interpretation of clustering results indicate that soil groupings were highly
variable, including a number of combinations of possible soil textures. The ecological

significance of these clusters is discussed at the end of this chapter.

 

69

DISTANCES

0.000 1.000
group sample

 

 

 

 

 

 

db—-—-——

 

 

 

 

mmmmmmmmmmmq\Iqmoxmmmmmmmmaaasoan.»wwuuwwmmmpppppH

N

b
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I

+—

I
I

Figure 2.16. Sample clusters based on 28 soil variables using average linkage.

70

Table 2.12. Soils table arraying key variables for samples grouped by average linkage.

GROUP STA T30 UTEX LTEX BTEX DPSCL ACSL ACSCL DOMB DCL MOTDE GWTDE

11 LS SL SCL L 116 61 143 SCL MWD 55 500
13 LS SCL SCL SL 81 17 86 SCL WD 234 366
6 SL SCL SCL L 138 21 76 SCL WD 167 366
7 FSL SCL SCL SL 74 57 79 SCL WD 129 305
3 LS LS SCL SL 118 15 155 SCL WD 45 500
4 LS LS SIL LS 179 142 185 SCL MWD 94 203
23 S SCL SICL L 100 48 115 SICL MWD 90 145
2 FSL FSL SICL SL 43 25 49 SCL MWD 24 500

30 SL SCL SICL SCL 41 26 57 SCL MWD 120 155
38 FSL SCL SCL SCL 171 173 187 SCL WD 153 411
22 LS SL SCL SL 133 136 268 SCL MWD 171 309

1

1

1

1

1

l

2

2

2

3

3

3 28 LS S SL LS 125 198 244 SCL WD 135 194
3 47 LS S SCL SL 37 185 52 SCL WD 500 500
3 35 LS LS SCL SL 32 60 500 SCL WD 500 500
3 1 LS SL SCL LS 45 70 50 SCL WD 500 500
4 26 SL SCL SCL SCL 40 18 59 SCL WD 500 500
4 9 SL L SCL L 52 18 57 SCL MWD 23 500
4 5 SL SL SCL SL 15 15 98 SCL MWD 36 500
4 10 SL SL SCL L 88 30 93 SCL MWD 45 82
4 8 SL SL SCL SL 105 17 137 SCL WD 168 500
4 19 LS SL SCL SCL 45 38 69 CL WD 161 161
4 44 LS SICL SICL SCL 28 50 44 SCL MWD 18 101
5 29 SL SICL SICL SICL 32 15 47 SICL WD 106 500
5 24 L SICL SICL SICL 48 22 60 SICL WD 111 500
5 42 LS SICL SICL SL 165 37 183 SICL WD 500 500
5 27 SL SICL SICL SICL 54 34 72 SICL WD 500 500
5 43 SL SICL SICL L 21 110 40 SIL WD 374 389
5 18 S L SICL S 133 134 140 SICL WD 500 500
6 33 S S SCL L 46 290 79 SICL XD 500 500
6 34 S S SIL S 259 274 500 SIL EXD 290 500
6 14 S SCL SCL S 106 500 111 SCL WD 380 500
6 32 S S CL S 87 154 205 SICL EXD 500 500
7 45 S LS SL LS 183 335 500 SL WD 500 500
7 46 LS S SL LS 244 500 500 SL WD 500 500
7 20 S S SL S 500 240 500 SL EXD 500 500
8 37 S S LS LS 500 500 500 LS EXD 500 500
8 21 S S S S 500 500 500 LS SED 500 500
8 39 S S S S 500 500 500 S EXD 500 500
8 49 S S S S 500 500 500 S EXD 500 500
8 48 S S LS S 500 500 500 LS EXD 500 500
8 41 S S S S 500 500 500 S EXD 500 500
8 31 S S S S 500 500 500 S EXD 500 500
8 36 S S S S 500 500 500 S EXD 500 500
8 12 S S S S 500 500 500 S WD 290 500
8 40 S S S S 500 500 500 S EXD 442 500
8 25 S S S S 500 500 500 S EXD 500 500

 

71

Tables 2.13 and 2.14 were constructed for synthesizing patterns in soil variables
based on samples ordered by PCI and CA 1. The ordering of samples according to
results of PCA and CA led to different groups of observations based on soil variables.

Samples ordered by PC] (Table 2.13) showed an interesting pattern. Samples
with silty clay loam substrates occurred together in a sequence corresponding to changes
in sola textures, with sandy loam sola preceding loamy sands and sands. Samples with
sandy clay loam substrates followed those with silty clay loam substrates in a sequence
corresponding to sola textures, with sandy loam sola preceding loamy sands and sands.
Samples with sandy loam substrates followed those with sandy clay loam substrates.
Finally, samples composed of deep sandy soils were grouped together. This ordering of
samples along the first PC according to substrata textures then sola textures corresponded
with variable weights and factor patterns for the first PC.

Samples ordered by CA1 (Table 2.14) showed a different pattern. The first group
of samples sequenced in this ordination was composed of sandy loam sola overlying
sandy clay loam substrates. Next samples with loamy sand and sandy loam sola over
silty clay loam substrates were grouped together. Samples with sandy sola overlying finer
textured substrates of sandy loam, sandy clay loam, silty clay loam, and clay loam
substrates were then ordinated together. The final discernible group consisted of deep
sandy soils.

Correspondence analysis of samples in species space reflected a moisture and
nutrient gradient that was indirectly expressed by ground-flora composition. Groups of

samples ordered based on CA of vegetation appeared more ecologically meaningful than

72

the ordination of samples based on soil variables using PCA. The first group composed
of loamy sands and sandy loams over sandy clay loam substrata would provide an
optimum rooting zone, with water held above and within the fine loamy substrate. Seven
of the first nine samples were moderately well drained, indicating an additional source of
water beyond soil moisture holding capacities. The next group had equivalent sola, with
silty clay loam substrates. The first group of samples with silty substrates were well
drained, followed by samples that were moderately well drained. The next group
consisted of sandy soils underlain by textural discontinuities. Host et al. (1987) and
Cleland et al. (1985) reported that these types of ecosystems were more productive than
ecosystems underlain by soils only containing deep sands. The final group of deep sands
were typical of xeric outwash systems in terms of soil characteristics and ground-flora
composition.

The ordination of samples in species space seemed to result in a more ecologically
meaningful sequence of observations based on soil characteristics than the ordination of
samples in soil variable space. Soil variables used in this research were principally
categorical variables, and may not have adequately described subtle edaphic conditions
affecting biota. Ground-flora composition and relative abundance appeared to effectively
serve as phytometers of important soil conditions from a biological standpoint, moreso

than soil variables themselves.

 

 

73

Table 2.13. Soil table arraying key variables for samples ordered by sample weights in the
first dimension of a PCA of soils data.

STA T30 UTEX LTEX BTEX DPSCL ACSL ACSCL DOMB DCL MOTDE GWTDE

29 SL SICL SICL SICL 32 15 47 SICL WD 106 500
24 L SICL SICL SICL 48 22 60 SICL WD 111 500
27 SL SICL SICL SICL 54 34 72 SICL WD 500 500
43 SL SICL SICL L 21 110 40 SIL WD 374 389
42 LS SICL SICL SL 165 37 183 SICL WD 500 500
30 SL SCL SICL SCL 41 26 57 SCL MWD 120 155
44 LS SICL SICL SCL 28 50 44 SCL MWD 18 101
23 S SCL SICL L 100 48 115 SICL MWD 90 145
2 FSL FSL SICL SL 43 25 49 SCL MWD 24 500
19 LS SL SCL SCL 45 38 69 CL WD 161 161
10 SL SL SCL L 88 30 93 SCL MWD 45 82
9 SL L SCL L 52 18 57 SCL MWD 23 500
38 FSL SCL SCL SCL 171 173 187 SCL WD 153 411
7 FSL SCL SCL SL 74 57 79 SCL WD 129 305
18 S L SICL S 133 134 140 SICL WD 500 500
11 LS SL SCL L 116 61 143 SCL MWD 55 500
6 SL SCL SCL L 138 21 76 SCL WD 167 366
5 SL SL SCL SL 15 15 98 SCL MWD 36 500
8 SL SL SCL SL 105 17 137 SCL WD 168 500
33 S S SCL L 46 290 79 SICL EXD 500 500
3 LS LS SCL SL 118 15 155 SCL WD 45 500
26 SL SCL SCL SCL 40 18 59 SCL WD 500 500
4 LS LS SIL LS 179 142 185 SCL MWD 94 203
13 LS SCL SCL SL 81 17 86 SCL WD 234 366
22 LS SL SCL SL 133 136 268 SCL MWD 171 309
28 LS S SL LS 125 198 244 SCL WD 135 194
47 LS S SCL SL 37 185 52 SCL WD 500 500
1 LS SL SCL LS 45 70 50 SCL WD 500 500
14 S SCL SCL S 106 500 111 SCL WD 380 500
32 S S CL S 87 154 205 SICL EXD 500 500
35 LS LS SCL SL 32 60 500 SCL WD 500 500
34 S S SIL S 259 274 500 SIL EXD 290 500
45 S LS SL LS 183 335 500 SL WD 500 500
46 LS S SL LS 244 500 500 SL WD 500 500
20 S 8 SL S 500 240 500 SL EXD 500 500
37 S S LS LS 500 500 500 LS EXD 500 500
40 S S S S 500 500 500 S EXD 442 500
12 S S S S 500 500 500 S WD 290 500
48 S S LS S 500 500 500 LS EXD 500 500
21 S S S S 500 500 500 LS SED 500 500
25 S S S S 500 500 500 S EXD 500 500
41 S S S S 500 500 500 S EXD 500 500
49 S S S S 500 500 500 S EXD 500 500
31 S S S S 500 500 500 S EXD 500 500
36 S S S S 500 500 500 S EXD 500 500
39 S S S S 500 500 500 S EXD 500 500

74

Table 2.14. Soil table arraying key variables for samples ordered by sample weights in

the first dimension of a CA of vegetation data.

STD

T30

SL
SL
LS
SL
FSL
LS
LS
SL
SL
FSL

FSL

mmmmmmmmmmmmm

UTEX

L

SL
SL
SL
FSL
SL
LS
SICL
SL
SCL
LS
SCL
SICL
SCL

SL
SICL
SCL
SCL
SICL
SICL

SL
SCL
SICL
SCL
SCL

LS

mmmmmmmmmmmmmmmm

LTEX

SCL
SCL
SCL
SCL
SICL
SCL
SIL
SICL
SCL
SCL
SCL
SCL
SICL
SCL
SL
SCL
SICL
SCL
SCL
SICL
SICL
SICL
SCL
SICL
SICL
SCL
SICL
SCL
SIL
SCL
SL

S

SL
SCL
S

CL

S

S

LS
LS

(f)
(DCDCDCDCDU

BTEX

L

L

L

SL
SL
LS
LS

L

SL
SCL
SL
SL
SICL
SCL
LS
SCL
SICL
SL

SICL
SL

SL
SCL
SCL

r' F‘U) U)
mmmt-‘mt-‘t-‘m

mmmmmmmmmmmmr‘

DEPSCL ACSL ACSCL

52
88
116
105
43
45
179
21
15
171
118
74
32
40
125
45
54
81
138
48
165
133
133
41
28
106
100
37
259
32
244
500
183
46
500
87
500
500
500
500
500
500
500
500
500
500

18
30
61
17
25
70
142
110
15
173
15
57
15
18
198
38
34
17
21
22
37
134
136
26
50
500
48
185
274
60
500
500
335
290
500
154
500
500
500
500
500
500
500
500
500
500

57
93
143
137
49
50
185
40
98
187
155
79
47
59
244
69
72
86
76
60
183
140
268
57
44
111
115
52
500
500
500
500
500
79
500
205
500
500
500
500
500
500
500
500
500
500

DOMB

SCL
SCL
SCL
SCL
SCL
SCL
SCL
SIL
SCL
SCL
SCL
SCL
SICL
SCL
SCL
CL
SICL
SCL
SCL
SICL
SICL
SICL
SCL
SCL
SCL
SCL
SICL
SCL
SIL
SCL
SL

S

SL
SICL
S
SICL
S

LS
LS
LS
SL
LS

S
S
S
S

DCL MOTDE GWTDE

MWD
MWD
MWD
WD
MWD
WD
MWD
WD
MWD
W0
W0
W0
WD
WD
WD
WD
WD
WD
WD
WD
WD
WD
MWD
MWD
MWD
WD
MWD
WD
EXD
WD
WD
EXD
WD
EXD
WD
EXD
EXD
SED
EXD
EXD
EXD
EXD
EXD
EXD
EXD
EXD

23

45

55
168

24
500

94
374

36
153

45
129
106
500
135
161
500
234
167
111
500
500
171
500

18
380

90
500
290
500
500
442
500
500
290
175
500
500
500
500
500
500
500
500
500
500

500

82
500
500
500
500
203
389
500
411
500
305
500
500
194
161
500
366
366
500
500
500
309
500
101
500
145
500
500
500
500
500
500
500
500
500
500
500
500
500
500
500
500
500
500
500

75

Analysis of High Quality Understory and Ground-flora Species

Table 2.15 arrays species retained for additional analyses based on their the sums
of squared cosines in five dimensional subspace. These were species with greater than
forty-five per cent of their total variability in full dimensional sample space accounted for
in the first five dimensions. Allium tricoccum and Uvularia perfoliata were also retained
despite their relatively low five-dimensional display qualities of thirty-eight and thirty-
two percent, respectively, because of their high indicator value of fine textured sola or
substrates for mesic hardwood ecosystems in northwest lower Michigan (Host 1997),
and corroborating field observations in northeast lower Michigan.

Table 2.16 displays the decomposition of inertia for the first fifteen principal
coordinates of a CA of forty-six samples and forty-nine high quality understory and
ground-flora species. Sixty-eight per cent of the total variability in forty-five dimensional
space was accounted for within the first five dimensions, with thirty-seven percent of
inertia accounted for by the first principal coordinate and fourteen percent accounted for
by the second principal coordinate.

A higher proportion of the total inertia was decomposed along the first five
principal coordinates for the high quality ground-flora data set than for the commonly
occurring ground-flora data set, for which fifty-one per cent of the total variability in full

dimensional space was accounted for by the first five principal coordinates.

 

76

Table 2.15. Sums of squared cosines in five dimensional subspace, mass, and inertia for
high quality species.

Species Spp Code SDQuality Mass Inertia

Actaea rubra ACPA 0.504244 0.007768 0.015550
Acer rubrum ACRU 0.816312 0.052685 0.017437
Acer saccharum ACSA 0.929969 0.057920 0.049944
Allium tricoccum ALTR 0.386129 0.002026 0.011941
Amelanchier Spp. AMAR 0.633853 0.026005 0.024220
Apocynum androsaemifolium APAN 0.736796 0.005572 0.013508
Arctostaphylos uva—ursi ARUV 0.635972 0.003546 0.018517
Aster macrophyllum ASMA 0.578214 0.030564 0.012989
Athyrium felix-femina ATFE 0.481367 0.002702 0.016195
Athyrium Spp. ATSP 0.459617 0.004222 0.009350
Carex annectens CAAN 0.845810 0.012327 0.022018
Carex deweyana CADE 0.648147 0.007092 0.042612
Carex pennsylvanica CAPE 0.541322 0.057413 0.025903
Carex plantaginea CAPL 0.519425 0.008274 0.016297
Caulophyllum thalictroides CATH 0.565080 0.005066 0.015425
Cladina spp. CLRA 0.730446 0.002871 0.011827
Comptonia peregrina COPE 0.927055 0.013509 0.040373
Cretaegus Spp. CRSP 0.511502 0.005066 0.010778
Dicranum polysetum DIPO 0.636293 0.006417 0.016617
Dryqpteris spinulosa DRSP 0.590485 0.002702 0.008627
Fagus grandifolia FAGR 0.713320 0.017562 0.022915
Fraxinus americana FRAM 0.804374 0.054880 0.026993
Gaylussacia baccata GABA 0.764684 0.020601 0.040150
Galium triflorum GATR 0.718202 0.020432 0.018011
Hamamelis Virginiana HAVI 0.614521 0.022627 0.023242
Hepatica acutiloba HEAC 0.606066 0.010469 0.017788
Lonicera canadense LOCA 0.484476 0.013340 0.013203
Melampyrum lineare MELI 0.778551 0.009287 0.021834
Mitella diphylla MIDI 0.662048 0.011820 0.015562
Oryzopsis asperifolia ORAS 0.646829 0.047957 0.015541
Osmorhiza chilensis OSCH 0.748104 0.021783 0.023910
Ostrya virginiana OSVI 0.718115 0.032759 0.026034
Prenanthes alba PRAL 0.526168 0.020432 0.013604
Pteridium aquilinium PTAQ 0.753352 0.066532 0.031056
Quercus alba QUAL 0.745493 0.028875 0.024651
Quercus rubra QURU 0.603897 0.036981 0.012520
Quercus velutina QUVE 0.888014 0.017899 0.031063
Ribes cynobasti RICY 0.619363 0.015535 0.017356
Rubus allegheniensis RUAL 0.691570 0.008443 0.026516
Sanicula trifoliata SATR 0.591941 0.011651 0.017899
Solidago hispida SOHI 0.512137 0.014691 0.017461
Tilia americana TIAM 0.728322 0.018744 0.016035
Trillium grandiflorum TRGR 0.447832 0.011651 0.009081
Uvularia perfoliata UVPE 0.337519 0.008105 0.010621
Vaccinium angustifoliu VAAN 0.904778 0.045424 0.029523
Viburnum acerifolium VIAC 0.611664 0.031577 0.019663
Viola spp. VISP 0.783282 0.027356 0.015867

77

Gains in inertia accounted for by CA of the high quality ground-flora data set
when compared with the CA of commonly occurring species were twelve percent for the
first principal coordinate, three percent for the second principal coordinate, one percent
for the third principal coordinate, and less than one percent for the fourth and fifth
principal coordinates, totaling a seventeen percent increase in total variability accounted
for within the first five dimensions. This increase in the overall variability accounted for
in the first five dimensions was due to the inclusion of species with relatively high
correlations, or fidelity, to this subspace and elimination of species more strongly

associated with higher dimensions.

Table 2.16. Correspondence analysis: inertia and Chi-Square decomposition of the first
fifteen principal coordinates.

Singular Principal Chi-

Values Inertias Squares Percents 7 14 21 28 35
0.81798 0.66909 990.586 37.37% ***************************
0.50238 0.25239 373.658 14.10% **********

0.34035 0.11584 171.496 6.47% *****

0.31220 0.09747 144.300 5.44% ****

0.29561 0.08738 129.371 4.88% ***

0.24810 0.06155 91.130 3.44% **

0.23386 0.05469 80.967 3.05% **

0.21129 0.04464 66.096 2.49% **

0.20517 0.04209 62.320 2.35% **

0.19614 0.03847 56.958 2.15% **

0.18304 0.03350 49.602 1.87% *

0.17885 0.03199 47.357 1.79% *

0.17158 0.02944 43.585 1.64% *

0.16319 0.02663 39.428 1.49% *

0.15866 0.02517 37.267 1.41% *

78

Figures 2.17, 2.18, and 2.19 display results of sample groups agglomerated based
on high quality species using three linkage methods. There was close agreement in
clusters defined and sample assignment to clusters among all clustering procedures.
Clustering using beta linkage identified the same clusters and samples within clusters as
clustering using average and Ward’s linkage methods with the exception of samples 34,
35, and 37 that formed a discrete cluster, and sample 21. Clustering using average
linkage method identified the same clusters and samples within clusters as clustering
using Ward’s linkage with the exception of sample 18, that was identified as an outlier in
the second level. Interpreting the assignment of samples 34, 35, and 37 at one lower level
of agglomeration using flexible beta linkage, however, gave identical results as the other
two agglomerative linkage methods. Clustering based on TWINSPAN gave the same
results as agglomerative clustering with the exception of two samples, samples 44 and 48.

The consistent recovery of the same cluster structure by three different
agglomerative linkage methods and divisive clustering indicated that the derived
classifications were valid. The interpretation of these groupings and final classification
of samples and species was made by tabularly examining patterns in soil characteristics.

This interpretation is offered in the final section of this chapter.

79

1.719 291.391 581.063 870.734 1160.406

 

 

 

 

 

 

 

 

 

 

Figure 2.17. Clustering of samples based on high quality species: beta linkage.

1986.090

1489.997

993.904

80

497.812
|-------+-------+-------+-------+-------+-------+-------+-------+

1.719

 

 

 

 

 

 

 

. -

- -

. -

- -

- -

— - lllllllllllll

. . a -

— - a _

. - u -

_ _ llllllllll -

. . — . .

- _ - - -

. lllllllllll _ . .

. u . _ - —

- - . llllll - -

- - Illllllll . u c -

- - - . . - u u

- . IIIIIIIIIIIII . . u — .
llllll — - . . . — u -
u - . - - _ lullllllllll a - .
— I'll - . - . _ - — - -
- - - u - u - Ill-I'll - - — -
n - — - - . - _ . - - Ill -
II - - - - - - . - . - - - -
- - - - - . IIIIIII - - - - - - - -
- - - - - - u . ll Illl - c n - - -
. - - - - Ill - - - c n _ IIII— - . - _
a u - - - _ a n . - - - — . . . . .lirlllll -
.ll . - - . . - u - u . - . n - u - - - III
- . _|I|||I - ll. . u - u u - - - .Il- . u - - -
- u — - ll. - u - - u - - .lIl - — — - - - u u . n -
u - .Il- - c - - . IIIIIII - - - — - u u . - . . n u - III. - -
- - . _ c — . _ . -||.|. . ll. — - - . _ . u . .lulul- . - — u .
n . - n - - . - - - u - - — . .Il. - - u n u - - II- III-
. - - . u . .ll‘ . . _ .III . c - . llllllll. . . n - . - II. II

1 0 3?96898238420152836?457014203459169
1213458916712122234412144434344333221232424333

 

 

Figure 2.18. Clustering of samples based on high quality species: Ward’s linkage.

 

81

351.072 525.749 700.426

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2.19. Clustering of samples based on high quality species: average linkage.

82

 

 

 

 

 

Figure 2.20. Samples in CA dimensions 1 and 2 of high quality
species space partitioned by clusters identified using Ward’s and
average linkage methods.

Figure 2.20 shows samples in the first and second dimensions of high quality
species space partitioned by clusters identified using Ward’s and average linkage
methods. Two clusters could not be partitioned due to intermingled placement of
samples.

Figure 2.21 shows samples in the first and third dimension of high quality species

space partitioned by clusters identified using Ward’s and average linkage methods. All

 

83

clusters could be partitioned in these two dimensions, suggesting that latent structure of
the vegetative data emerged in the first and third dimensions of a CA as opposed to the

first and second dimensions.

 

   

 

 

 

Figure 2.21. Samples in CA dimensions 1 and 3 of high quality species space
partitioned by clusters identified using Ward’s and average linkage methods.

84
Classification of Species
Ordination of species in the first two dimensions of sample space identified two
species, Carex pensylvanica Lam. and Athyrium spp., that occurred far within the arch
(Figure 2.22). Pteridium aquilinium, T rientalis borealis, and Acer rubrum were also

somewhat interior to the arch. Interior points in the first two dimensions of a CA indicate

 

 

 

 

2
ARUV
CLRA
cops
15
MEU
1 UVE
“853%.! m
OATH
ATSP DIPO
CAAN
APAN
as EIFE
RIY CAPE
N SCH
5 PM VAAN
R
M
0 PTAQUAL
ACPA AMAR
TRGR RUAL
PRAL TRBO ACRU
05 UVPE GURU
" onsuA HAVI
vmc LO A CRSP
gATR
4
~t5
05 1 L5 2
.1.5 .1 -o.5 o
CA1

Figure 2.22. High quality species in CA dimensions 1 and 2 of sample
space.

85

species with especially broad or undiscriminating distributions (Gauch 1982).
TWINSPAN classified Carex pensylvanica Lam. and Athyrium spp. as a separate group at
the forth level of division (Table 2.17). Inspection of the TWINSPAN synthesis table
corroborated a ubiquitous distribution of these species, and Rubus allegheniensis, and
Quercus alba, hence these seven species were removed from further consideration for
use in formulating ecological species groups.

Classification of species by TWINSPAN identified eight groups and two outliers
(Figure 2.23). Groups consisted of species assemblages grading from highly mesophilic
species typifying northern hardwood ecosystems to pyrophilic xeric species typifying oak
and pine-oak ecosystems. Interpretation of these groups, made by comparing species
assemblages to sample clusters and soil variables, are offered in the final classification
and synthesis of results sections of this chapter.

Classification of species using agglomerative clustering procedures produced
inconsistent results. The number of clusters and species’ assignment to groups varied
among agglomerative procedures, and were not comparable to groups identified by
TWINSPAN and ordination diagram partitioning. Data reduction based on subjective
interpretation of the TWINSPAN synthesis table and subsequent agglomerative clustering
also failed to identify consistent clusters. Consequently agglomerative clustering was not

further used to identify ecological species groups.

86

Table 2.17. TWINSPAN classification of high quality species and samples.

Species

Group

cooooomooooooooq\I\ImmmmmmmoomwmmmbbanbwwwwwNMNNNNNNI—IHHHH

ALTR
CATH
DRSP
CAAN
CADE
HEAC
RICY
ACSA
FAGR
OSVI
SOHI
TIAM
CAPL
GATR
MIDI
OSCH
ACPA
ATFE
FRAM
PRAL
UVPE
TRGR
VISP
VIAC
ASMA
ORAS
QURU
TRBO
CAPE
ATSP
PTAQ
ACRU
LOCA
SATR
CRSP
HAVI
RUAL

QUAL

APAN
DIPO
GABA
MELI
QUVE
ARUV
CLRA
COPE

1 1 22 1 12234241122344443343331243423432
1189023456873697983428423041572364572101859960
1122-2 ----------------------------------------
—22232—22 ------- 2 -----------------------------
12122-22 --------------------------------------
332332323--332 --------------------------------
4—333 ------- 3 ---------------------------------
—23223223 ------- 233-2 ----- 2 -------------------
213333223--3123-332-2 ----- 2 ———————————————————

444444444334343334433 -------------------------
3223-3333332333222-—-—1 -----------------------
323432433324243334323-4 -----------------------
133-2223322313—332-—-1 ------------------------
22332233332332—223322 -------------------------
-2233----222—--1232-2 -------------------------
-2333-233-—2-311343333-2-—3 ---------- 1 ————————
--233-22212—-2—2332321-2 ----------------------
244333132113--—2333323 ------- 11 ---------------
----- 31--21--22233221-—21-------—---—---------
--—2 ------------ 1232 --------------------------
334443444334344334444--3334 -------- 1 ----------
—-2-22222323323333—322-332—-—212—-21--1—1—-—12
-—--22122212-3---223-2-2222 -------------------
1-3221222-—-—222221212—1322--—1—-12 -----------
32433334422322-333—31212222—-111-1 ------------
~—2—2-3~3-23341331-3-—4434421122223332 ————————

-1 ------ 222233-3331333233332-2222333-2122---2-
2-3---32421334343-233344344334333333222-2---2-
2 ------ 22333332333333-3333333322332333322 -----

- ---------- 22312-2—222231--22-22----2222-2----
-42232333--2-2-23-323-2-4133334244344-4444444—
---------------- 22212------—--1-2-—--2-2-2-2--
-------- 2—-233—--—-334343333444443444434444344
--------- 123332322233-343433444444433433332123
—--1--2--—2-2—11—-—22-23233212111-3211 ------- 2
- --------------- 3-—213--322--2222332 ----- 2—---
------------------------ 212---222231---—---12-
---------- 1-—22----2--13233333433423-—-33--——-
----------------------- 3-—2-----1344-2----1-3-
-------------------- 2—--232333343433-33233—323
---------------------- 13313323443-333233333222
------------------- 1--333333334333443444444443

------------------------ 2---—----322---2121-22
--------------- 1-----------21-12----2223-2221-
--------------------------- 43231—2-32—2443442-
------------------------ 1—----2--122—1-2332233
——————————————————————— 2-—-1-122-23323-3343334
----------------------------------- 2---2-3222-
------------------------------------- 1-22222-1
----------------------------------- 3—2-4243433

0000000000000000000000111111111111111111111111
0000000001111111111111000000000000001111111111
0000011110000000111111000001111111110001111111
01111 001111100000101111000000111 0000001

Sample number

00000
00000
00000
00001
00001
0001
0001
001000
001000
001000
001000
001001
00101
00101
00101
00101
0011
0011
010
010
010
011
011
1000
1001
1001
1001
1001
101
1100
1100
11010
11010
11010
11011
11011
11011
11100
11100
11100
11101
11110
11110
111110
111110
111111
111111
111111

87

 

1.5

 

 

 

 

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
CA 1

Figure 2.23. High quality species in CA dimensions 1 and 3 of sample space partitioned
by TWINSPAN clusters.

88

Final Classification of Samples and Species based on Combined Ground-flora and Soil
Factors

Seven ecological land classification units (ELU’s) and ecological species groups
were defined based on a series of complementary analyses of ground-flora and soils
conditions. Comparisons of soils data arrayed in Tables 2.12 and 2.13 to clusters of
samples defined by high quality species revealed two distinctly different soil-substrata
conditions within one cluster that included samples consisting of deep sands and of sands
underlain by fine loamy substrates. This cluster was subdivided into two ecological land
units as a consequence. Placement of sample 38 was also changed due to similarity of
soil textures to a different cluster. Sample 38 had sandy loam soil textures to a 30 cm
depth overlaying sandy clay loam textures between 30 cm and 150 cm, but was originally
grouped with samples containing loamy sand textures to depths of 100 cm. This large
difference is soil textures was used as a basis to subjectively assign sample 38 to a group
composed of equivalent soil textures. The remaining groupings of samples based on high
quality understory and ground-flora species showed consistent patterns in soil
characteristics. Examination of overstory composition within combined soil-ground-flora
units corroborated the meaningful formulation of seven ecological land units based on
congruent changes in ground-flora and soil variables. Table 2.18 arrays key soil
variables, Table 2.19 arrays key species, and Table 2.20 arrays overstory composition for
these classes.

The identification of seven ecological land units by comparison of clustering and

tabular results was corroborated by partitioning ordinations of samples in high quality

 

89

understory and ground-flora species space (Figure 2.24) and samples in soil variable
space (Figure 2.25). The CA of high quality species and PCA of twenty-eight soil
variables placed similar samples adjacent to one another, and dissimilar samples away
from one another, with equivalent sample placement within respective ordination space
groupings. These units are described in the Ecological Land Units (ELU) section of this

chapter.

Table 2.18 Key soil variables for ecological land units.

ELU STA

HIAIAFJFJHIHIAIA

ONQONO‘QO) 01010101010101 Abbbbb wwwww NNNNNNN

\1\J\’\]\J\l

HOKOGDUIAWNH

H H

l3
19
26
38
28

24
27
29
42
43

14
18
22
23
30
44

32
33
34
35
45
46
47

12
21
37
40
41
48

20
25
31
36
39
49

T30

LS
FSL
LS
LS
SL
SL
SL
SL
LS

SL
FSL
LS
LS
SL
FSL
LS

SL
SL
LS
SL

CDUJU)

LS

LS
LS

(DMCDUDUJCD

mmmmmm

UTEX LTEX BTEX

SL
FSL
LS
LS
SL
SL
L
SL
SL

SCL
SCL
SCL
SL
SCL
SCL
S

SICL
SICL
SICL
SICL
SICL

SCL

SL

SCL

SCL
SICL

(DUJU)

LS
LS

(DU)

(DCDCDUJCOU)

mmmmmm

SCL
SICL
SCL
SIL
SCL
SCL
SCL
SCL
SCL

SCL
SCL
SCL
SCL
SCL
SCL
SL

SICL
SICL
SICL
SICL
SICL

SCL
SICL
SCL
SICL
SICL
SICL

CL
SCL
SIL
SCL
SL
SL
SCL

mmmmmm

LS
SL
SL
LS
SL
SL
L
L
L

L
SL
SL
SCL
SCL
SCL
LS

SICL

SICL

SICL
SL

SL

SCL
SCL

CDC/DU)

mmmmmm

9O

DEPSCL ACSL

45
43
118
179
15
105
52
88
116

138
74
81
45
40
171
125

48
54
32
165
21

106
133
133
100
41
28

87
46
259
32
183
244
37

500
500
500
500
500
500

500
500
500
500
500
500

70
25
15
142
15
17
18
30
61

21
57
17
38
18
173
198

22
34
15
37
110

500
134
136
48
26
50

154
290
274
60

335
500
185

500
500
500
500
500
500

240
500
500
500
500
500

ACSCL

50
49
155
185
98
137
57
93
143

76
79
86
69
59
187
244

60
72
47
183
40

111
140
268
115
57
44

205
79

500
500
500
500
52

500
500
500
500
500
500

500
500
500
500
500
500

DOMB BIC8
SCL 4.25
SCL 5.00
SCL 4.50
SCL 5.00
SCL 4.50
SCL 5.00
SCL 5.00
SCL 5.00
SCL 4.75
SCL 5.00
SCL 4.75
SCL 3.75

CL 5.00
SCL 5.00
SCL 5.00

SL 4.00

SICL 5.00
SICL 5.00
SICL 5.00
SICL 4.00
SIL 4.50
SCL 2.75
SICL 5.00
SCL 4.50
SICL 3.75
SCL 5.00
SCL 5.00
SICL 3.25
SICL 4.50
SIL 1.50
SCL 2.00
SL 2.25
SL 1.25
SCL 2.50

S 1.00
LS 0.75
LS 1.00

S 0.00

S 0.25
LS 0.75
SL 1.00

S 0.25

S 0.00

S 0.00

S 0.00

S 0.00

DCL MOTDE GWTDE

WD
WD
WD
WD
WD
WD
WD

WD
WD
WD
WD
WD

WD
WD
WD
WD
WD
WD
WD

WD

SED
EXD
EXD
EXD
EXD

EXD
EWD
EXD
EXD
EXD
EXD

500
24
45
94
36
168
23
45
55

167
129
234
161
500
153
135

111
500
106
500
374

380
500
171
90
120
18

500
500
290
500
500
500
500

290
500
500
442
500
500

500
500
500
500
500
500

500
500
500
203
500
500
500

82
500

366
305
366
161
500
411
194

500
500
500
500
389

500
500
309
145
155
101

500
500
500
500
500
500
500

500
500
500
500
500
500

500
500
500
500
500
500

4 4 4 4 4 4
3 1 4 3 3 3
- 4 5 2 -

4
—----2

1-----
-l____
:g::::
1-2—--
5---..-

3 3 3 3 3 3
_ _ _ 3 _ _
4 1 4 5 4 -
5 2 5 4 4 -
4 - - 5 3 -
- 2 - 5 5 2
3 - - 3 3 -
5 5 4 4 6 6
5 5 2 5 6 5
2 3 5 5 5 3
3 3 5 5 5 4

91
222222
__4___
245-—-
344-23
445443
551—33
3514--
434445
665455
662443
2-—-13
--11

- 4

1 1 1 1 1 1 1 1 1
- 3 2 - 3 2 5

2 - 2 3 - 1 1 3 3
1 3 - - - 1 3 2 -
- - - - - 6 4 4 5
5 3 4 3 4 5 4 4 5
4 5 4 6 6 5 6 5 5
- 3 3 3 3 - 2 - 2
4 3 3 4 5 1 4 - 3
l 4 2 3 5 4 5 5 5
FAGR 2 4 5 4 4 4 3 5 -
6 6 6 6 6 6 6 6 6
2 3 6 5 4 5 5 6 5
- - 2 3 3 - 2 4 5
6 5 1 4 3 3 6 4 5

2 4

ELU
DRSP
ALTR
CADE
CAAN
VISP
PRAL
SOHI
RICY
ACSA
OSVI
MIDI
OSCH

Table 2.19. Ecological species groups for ecological land units.
CATH

.2. .5. 2236. 65346
...... 33352 6421
...1.. 34355 44525
...32_ 21454 65. .6
...... ..1.. 61. .6
....4. 243.4 ..4.5
453355 21.6. .._.2
354465 24464 1....
124553 .3554 4.4.4
3. .344 54.465 433.5
.32144 .1162 ..1.5
..1414 .3455 5. . .6
._31.3 1.353 21. .1
..22.4 2.254 ....2
...21. .2.64 12. .4
.....4 ..455 4. . .5
..2.35 2.464 5.. .4
...244 42364 63. .6
.35.52 23562 2....
235.54 .456. .....
.53.54 .4662 3. . .5
.....2 .155. ....3
.4. .45 22663 4.. .6
.2. .55 22663 ....3
.4.145 12463 4. . .4
.4.4.2 31553 .....
.33.23 ..44_ .....
ECLARM ERPWL CIRPS
P.Anrp.T D.Gns M auvmiqum
TEACAI VRI IAAR
AHCAGT UTVFP VHSCO

- - _ 2 _ _
_ _ _ 1 - -

5 5 5 5 5 1
5 5 4 5 5 4
1 3 3 - - -

4 3 5 - 4 3
5 4 4 3 5 5
- - 3 1 - 2

1-—-3--—-
--—223---
_ _ 3 _ - _ - 1 _

ASMA
QURU
LOCA
APAN
MELI
DIPO
GABA
QUVE

COPE
Values are cover rank abundance arrayed in Table 2.1.

 

Table 2.19 (cont). Ecological species groups for ecological land units.

VIAC
HAVI
SATR
CRSP
ORAS

ASMA
QURU
LOCA
AMAR
VAAN

ARUV
CLRA
COPE
APAN
MELI
DIPO
GABA
QUVE

5 5 5 5 5 5 5

A

Lfibl—‘UTI

6 6 6 6 6 6

01

I

I

I

I
AHNbb

I

UL)
U1
11>
LA)
L4)

Ul I HUII
LIT-56001.5
GUI l U'Il—l
(DWI-4.5.1:.
OWDLULDUI
(huh I NU)

NN-bll
HQUJII
ILAJAII
UWILAHILOHI
UWU'IILAJIbblN
U'IO'NItbl—‘LUNI

92

7 7 7 7 7 7

I
(3.1:.le
I

(”WI
mm

UNI
01.wa

U'Isbl—‘LD-UJUW I vb
mmbuDNON-D-N
UTONNbe-‘U'INN

I IU'ldell—‘l
Amwwlmmm
mmwmbmwm

0)

Values are cover rank abundance arrayed in Table 2.1.

93

Table 2.20. Overstory composition within ecological land units.

STA l 2 3 4 5 11 8 9 10 6 7 13 19 26 28 38

ELU 1 1 1 1 l 1 1 1 1 2 2 2 2 2 2 2

SM 103 75 60 40 63 80 50 75 80 13 43 18 5 15 25 45
BW 15 30 48 80 45 38 30 18 28 5 10 0 0 80 20 45
WA 0 20 18 5 0 8 5 8 0 13 8 0 0 45 25 30
BE 0 3 5 3 0 10 20 13 0 10 0 13 10 5 5 5
1W 0 0 0 3 0 5 0 0 0 0 0 0 0 5 0 10
PB 0 0 3 0 0 0 0 0 0 8 3 15 5 0 0 15
R0 0 0 0 0 0 10 0 0 0 8 33 60 70 10 80 0
RM 0 0 0 0 0 0 0 0 0 23 17 10 20 0 0 0
BA 0 13 13 23 18 0 15 5 13 18 18 2 25 10 0 5
W0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
BP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

STA 27 42 24 29 43 18 14 22 23 30 44 35 34 46 47 33

ELU 3 3 3 3 3 4 4 4 4 4 4 5 5 5 5 5
SM 0 40 30 10 80 10 0 0 0 0 0 0 0 0 0 0
BW 35 0 70 15 50 0 0 0 0 0 0 0 0 0 0 0
WA 10 0 10 35 0 0 0 0 0 15 0 0 0 0 0 0
BE 0 0 10 0 0 0 0 0 0 O 0 0 0 0 0 0
IW 0 5 10 0 0 0 0 0 0 0 0 0 0 0 0 0
PB 50 5 5 5 15 0 0 10 0 0 40 0 0 0 0 0
R0 65 60 35 105 15 60 83 105 55 120 45 45 100 95 63 45
RM 5 25 0 0 0 30 28 35 25 10 50 5 30 10 20 15
BA 0 0 0 0 0 5 3 0 0 5 0 0 5 10 0 0
W0 0 0 0 0 0 25 3 20 25 0 10 60 5 15 35 40
BP 0 0 0 0 0 0 0 10 0 0 0 15 0 0 0 25

STA 12 41 40 21 37 48 36 31 39 20 49 25

ELU 6 6 6 6 6 6 7 7 7 7 7 7
SM 0 0 0 0 0 0 0 0 0 0 0 0
BW O 0 0 O 0 0 0 0 0 0 0 0
WA 0 0 0 0 0 0 O 0 0 0 0 0
BE 0 0 0 0 0 O 0 0 0 0 0 O
IW 0 0 0 0 0 0 0 0 0 0 0 0
PB 0 0 0 0 0 0 0 0 0 0 0 0
R0 53 80 80 15 23 10 3 5 0 O 0 0
RM 30 15 10 8 0 5 0 10 0 0 0 0
BA 3 0 0 8 5 10 O 0 0 13 5 0
W0 0 25 30 8 23 55 0 2 2 3 0 5
BP 0 0 0 83 15 35 55 48 65 43 70 50

Values are basal area in fiz.

94

1.5

 

0.5

 

CA3
o

 

-0.5

 

 

 

 

 

 

 

-1.5 -1 -0.5 0 0.5 1 1.5
CA 1

Figure 2.24. Samples in CA dimensions 1 and 3 of high quality species space labeled by
ELU.

95

 

 

 

 

3
1‘ 3 ‘
_ ‘1‘ I
5 '- 1 3
5 5 ‘, ‘1‘ 3 l
1 H p 4
{I I“? I 5 " 3 ‘1
N 1 ’6 6 4 4 1 ‘3‘ II
< I " . - —- .- ‘3
0 0 ) 5 6 I 7 5 1.1414 7
0' 1 7 I 2 " \:
1’. 77 . {"1 2 2 1
4,6 1 I 2 I
_ 2 7 I
_1 1 1,4}
I 1 [,4
-2
.1 l
-3
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

PCA 1

Figure 2.25. Samples in PCA dimensions 1 and 2 of soil variable space labeled by ELU.

96

Ecological Land Unit Descriptions

ELU 1 consists of the sugar maple-basswood-blue cohosh plant association on
moderately well-drained sandy loams over sandy clay loam substratums on moraines and
ground moraines; some pedons have loamy sand caps. These soils classify as coarse
loamy over fine loamy, mixed, Typic Eutroboralfs and F ragiboralfs at the Subgroup level.
The representative horizon sequence is O-AE-E-Bw-Bt—E&B-C 1 -C2. The forest floor
typically consists of a 2 cm. layer of sugar maple, basswood, and white ash litter. The
black A horizon averages 10 cm. in thickness. Gray E horizons average 2.8 cm. in
thickness. Bhs horizons are absent. Reddish brown (5 YR 4/3) loamy sand and sandy
loam Bw horizons overlie fine loamy C horizons. Mottles indicating perched watertables
generally occur within the upper 50 cm.

The ground-flora of sites sampled for ELU 1 consists of Caulophyllum
thalictroides, Dryopteris spinulosa, Allium tricoccum, Carex deweyana, and Carex
annectens. These species form the Caulophyllum species group.

The overstories of sites sampled for ELU 1, in decreasing order of basal area, are
dominantly composed of sugar maple, basswood, white ash, and beech. The average
stand in this ELU was 63 years old (1988); basal area averaged 133 ft2 per acre. Sugar
maple contributed 69 ii2 of BA per acre, basswood 37 ft2 , white ash 7 112, and beech 7
ftz. Stand MAI averaged 49 fi’lac./yr., with a standard error of 3.6.

ELU 1 is similar to ELU 2 and ELU 3. ELU 1 differs from ELU 2 and 3
floristically by having Caulophyllum thalictroides, Dryopteris spinulosa, Allium

tricoccum, Carex deweyana, and Carex annectens in the ground-flora. The soils of ELU

 

97

l differ from ELU 2 by having dominantly sandy loam textures in the upper 100 cm., as
opposed to sandy clay loam, and a moderately well drained drainage class. The soils of
ELU 1 differ from ELU 3 by having dominantly sandy loam textures in the upper 100
cm., as opposed to silty clay loam, and having sandy clay loam substrates at depths
greater than 100 cm. as opposed to silty clay loam substrates. The overstory of ELU 1 is
similar in composition and productivity to ELU 2, but contains higher basal areas of sugar
maple and lower basal areas of red oak, red maple, and paper birch.

ELU 2 consists of the red oak-sugar maple-viola plant association on well-drained
sandy loams over sandy clay loam substratums on moraines and ground moraines; some
pedons have loamy sand caps. These soils classify as coarse loamy over fine loamy Typic
Eutroboralfs and Alfic Haplorthods at the Subgroup level. The representative horizon
sequence is O-AE-E-Bs-Bt-E&B-C1-C2. The forest floor typically consists of a 2 cm.
layer of red oak, sugar maple, basswood, and white ash litter. The black A horizon
averages 6.6 cm. in thickness. Gray E horizons average 7.0 cm. in thickness. Bhs
horizons are absent. Reddish brown (5 YR 4/3) loamy sand and sandy loam Bw and Bt
horizons overlie fine loamy C horizons.

The ground-flora of sites sampled for ELU 2 consists of Viola species including
Viola pubescens and Viola canadensis, Prenanthes alba, Ribes cynobasti, Solidago
hispida, and a diverse assemblage of tree seedlings including Acer saccarum, Tilia
americana, Fagus grandifolia, and Ostrya virginiana. These species form the Viola
species group.

The overstories of sites sampled for ELU 2, in decreasing order of basal area, are

98

composed of red oak, sugar maple, basswood, white ash, red maple, beech, paper birch,
and ironwood. The average stand in this ELU was 68 years old (1988); basal area
averaged 137 ft2 per acre. Red oak contributed 37 ft2 of BA per acre, sugar maple
contributed 23 fiz, basswood 23 ftz, white ash 17 ftz, red maple 10 ftz, and beech 7 ftz.
Stand MAI averaged 50 fi3/ac./yr., with a standard error of 4.7.

ELU 2 is similar to ELU 1 and ELU 3. EL 2 differs from ELU l floristically by
lacking Caulophyllum thalictroides, Dryopteris spinulosa, Allium tricoccum, Carex
deweyana, and Carex annectens. EL 2 differs from ELU 3 floristically by lacking
Athyriumfelix-femina and Hepatica acutiloba, with Mitella diphylla and Osmorhiza
chilensis only occasionally present in ELU 2. The soils of ELU 2 differ from ELU 3 by
having dominantly sandy clay loam textures in the upper 100 cm., as opposed to sandy
loam, and a well drained drainage class. The soils of ELU 2 differ from ELU 3 by having
dominantly sandy clay loam textures in the upper 100 cm., as opposed to silty clay loam,
and having sandy clay loam substrates at depths greater than 100 cm. as opposed to silty
clay loam. The overstory of ELU 2 is similar in composition and productivity to ELU 1
and ELU 3, but contains lower basal areas of sugar maple and higher basal areas of red
oak, red maple, and paper birch than ELU l, and contains lower basal areas of red oak
than ELU 3.

ELU 3 consists of the red oak-basswood- miterwort plant association on
well-drained sandy loams over silty clay loam substratums on moraines and ground
moraines; some pedons have loamy sand caps. These soils classify as Typic Eutroboralfs

at the Subgroup level. The representative horizon sequence is O-AE-E-Bw-Bt-Cl. The

 

99

forest floor typically consists of a 2 cm. layer of red oak, sugar maple, basswood, red
maple and white ash litter. The black A horizon averages 4.6 cm. in thickness. Gray E
horizons average 10.2 cm. in thickness. Bhs horizons are absent. Reddish brown (5 YR
4/3) sandy loam Bw horizons overlie silty clay loam B and C horizons.

The ground-flora of sites sampled for ELU 3 consists of Mitella diphylla,
Osmorhiza chilensis, Athyrium felix-femina, Hepatica acutiloba, Carex plantaginea,
Actaea rubra, Galium triflorum, and Tilia americana. These species are form the Mitella
species group.

The overstories of sites sampled for ELU 3 are composed of red oak, basswood,
sugar maple, paper birch, white ash, red maple, ironwood and beech. The average stand
in this ELU was 68 years old (1988); basal area averaged 144 ft2 per acre. Red oak
contributed 56 ft2 of BA per acre, basswood 34 ftz, sugar maple 32 ftz, paper birch l6 ftz,
white ash 11 fiz, red maple 6 ftz, and beech 2 fiz. Stand MAI averaged 57 fi’/ac./yr., with
a standard error of 6.8.

ELU 3 is similar to ELU 2 and ELU 3. ELU 3 differs from ELU 1 floristically by
having Athyriumfelix-femina present, and by lacking Caulophyllum thalictroides,
Dryopteris spinulosa, Allium tricoccum, Carex deweyana, and Carex annectens in the
ground-flora. ELU 3 differs from ELU 2 by having Athyriumfelix-femina and Hepatica
acutiloba present, and by having Osmorhiza chilensis, Carex plantaginea, Actaea rubra,
and Galium triflorum present in higher coverages and frequencies. The soils of ELU 3
differ from ELU 1 by having dominantly silty clay loam textures in the upper 100 cm., as

opposed to sandy clay loam, and a well drained drainage class. The soils of ELU 3 differ

100

from ELU 2 by having dominantly silty clay loam in the upper 100 cm. and lower 100
cm., as opposed to sandy clay loam textures. The overstory of ELU 3 is similar in
composition and productivity to ELU 1, but contains higher basal areas red oak and
paper birch and lower basal areas of sugar maple. The overstory of ELU 3 is similar in
composition and productivity to ELU 2, but contains higher basal areas red oak and paper
birch.

ELU 4 consists of the red oak-red maple-bellwort plant association on well
drained sands with sandy B3 to coarse to fine loamy B2t horizons overlying fine loamy
substratums on ice-contact topography. These soils classify as Alfic and Entic
Haplorthods. The representative horizon sequence is O-E-le-Bt2-C1-C2. The forest
floor typically consists of a 2 cm. layer of red oak and red maple litter. The black A
horizon averages 4.6 cm. in thickness. Gray E horizons average 5.9 cm. in thickness.
Bhs horizons are absent. Reddish brown (7.5 YR 4/) sandy Bs and dark reddish brown (5
YR 4/3) sandy loam to sandy clay loam Bt horizons overlie fine loamy C1 or C2
horizons.

The ground-flora of sites sampled for ELU 4 consists of Uvularia perfoliata,
Trillium grandiflorum, Viola spp., Fraxinus americana, and Prenanthes alba. These
species form the Uvularia species group. The ground-flora of ELU 4 contains elements
of both northern hardwood and oak ecosystems, including species within the Uvularia
species group, and Viburnum acerifolium, Hamamelis virginiana, and Vaccinium
angustifolium.

The overstories of sites sampled for ELU 4, in decreasing order of basal area, are

101

composed of red oak, red maple, white oak, paper birch, and white ash. The average
stand in this ELU was 68 years old (1988); basal areas averaged 144 ft2 per acre. Red
oak contributed 78 ft2 of BA per acre, red maple 30 1’9, white oak l4 fiz, paper birch 8 ftz,
and white ash 3 ftz. Stand MAI averaged 45 fiJ/ac./yr., with a standard error of 5.7.

ELU 4 is similar to ELU 5. ELU 4 differs from ELU 5 floristically by having
species typifying northern hardwood ecosystems present including Uvularia perfoliata,
Trillium grandiflorum, Viola spp., Fraxinus americana, in addition to species
characterizing ELU 5. The soils of ELU 4 differ from ELU 5 by having shallower depths
to sandy clay loam textures in Bt and C horizons. The overstory of ELU 3 is similar in
composition and productivity to ELU 4.

ELU 5 consists of the Redoak-red maple-Viburnum plant association on well
drained sands with sandy Bs overlying coarse to fine loamy substratums on ice-contact
topography. These soils classify as Entic Haplorthods. The representative horizon
sequence is O-E-le -B52-C1-C2. The forest floor typically consists of a 2 cm. layer of
red and white oak, and red maple litter. The black A horizon averages 2.9 cm. in
thickness. Gray E horizons average 5.1 cm. in thickness. Bhs horizons are absent.
Reddish brown (7.5 YR 4/4) and brown (10 YR 4/4) sandy Bs horizons overlie dark
reddish brown (5 YR 4/3) sandy loam to sandy clay loam C1 or C2 horizons.

The ground-flora of sites sampled for ELU 5 consists of Viburnum acerifolium,
Hamamelis virginiana, Sanicula trifoliata, Cretaegus Spp., Amelanchier spp.,and
Oryzopsis asperifolia. These species form the Viburnum species group.

The overstories of sites sampled for ELU 5, in decreasing order of basal area, are

102

composed of red oak, white oak, black oak, and red maple. The average stand in this
ELU was 79 years old (1988); basal areas averaged 131 ft2 per acre. Red oak contributed
75 ft2 of BA per acre, white oak 26 ftz, red maple 19ft2, and black oak 19ft2. Stand MAI
averaged 40 ftJ/ac./yr., with a standard error of 3.5.

ELU 5 is similar to ELU 4 and ELU 6. ELU 5 differs from ELU 4 floristically by
lacking species typifying northern hardwood ecosystems including Uvularia perfoliata,
Trillium grandiflorum, Viola spp., Fraxinus americana. ELU 5 differs from ELU 6
floristically by having high frequencies and coverages of Sanicula trifoliata and
Cretaegus Spp. The soils of ELU 5 differ from ELU 4 by having greater depths to sandy
clay loam textures in substrates. The soils of ELU 5 differ from ELU 6 by having deep
lying coarse and fine loamy substrates, the principal criteria differentiating these two
ELU’s.

ELU 6 consists of the mixed oak-red maple-big leaf aster plant association on
excessively well-drained sands of ice contact topography. Soils in ELU 6 classify Entic
Haplorthods at the Subgroup level. Soil development is medial. Horizon sequences are
O-A-AE-Bs-C. The forest floor is frequently a thin layer of oak litter. The black (10 YR
2/1) A horizon averages 4.1 cm. in thickness. Gray E horizons average 2.2 cm., and are
absent in some pedons. Yellowish brown (7.5 YR and 10 YR 5/4) Bs horizons have
weak spodic development, and overlie sandy substratums.

The ground-flora of sites sampled for ELU 6 consists of Aster macrophyllum,
Quercus rubra, Lonicera canadense, and Vaccinium angustifolium. These species form

the Aster macrophyllum species group.

103

The overstories of reference sites sampled for ELU 6 are composed of red, black,
and white oak, and red maples. The average stand in this ELU was 70 years old (1988),
with a basal area of 73 112. Red oak contributed 37 ft2 of basal area, white oak averaged
34 fiz, black oak averaged 19 ft2 , and red maple averaged 11 ftz. Stand MAI averaged 28
ft3/ac/yr, with a standard error of 2.7.

ELU 6 is similar to ELU 5. ELU 6 differs from ELU 5 floristically by having high
coverages of Sanicula trifoliata and Cretaegus spp., and differs in soil characteristics by
lacking coarse to fine loamy substratums. The overstory of ELU 6 is similar in
composition to ELU 5, but is significantly less productive based on an ANOVA and
Duncan’s multiple range test (p < .05).

ELU 7 consists of the Northern pin oak-white oak-bearberry plant association on
excessively well-drained sands of outwash plains. Soils in ELU 7 classify Typic
Udipsamments at the Subgroup level. Soil development is minimal. Horizon sequences
are O-A-Bw-C. The forest floor is frequently a thin layer of oak litter. The black (10 YR
2/1) A horizon averages 4.1 cm. in thickness. Gray E horizons average 2.2 cm., and are
absent in some pedons. Yellowish brown (10 YR 5/6) Bw horizons lack spodic
development, and overlie sandy substratums.

The ground-flora of sites sampled for ELU 7 consists of Arctostaphylos uva-ursi,
C ladina spp., C omptonia peregrina, Apocynum androsaemifoliumspr, Melampyrum
lineare, Dicranum polysetum, Gaylussacia baccata, and Quercus velutina. These
species form the Arctostaphylos species group.

The overstories of reference sites sampled for ELU 7 are composed of upland pin,

104

black, and white oak. The average stand in this ELU was 70 years old (1988), with a
basal area of 73 fiz. Pin and black oak contributed 55 ft2 of basal area, and white oak
averaged 2 fiz. Red maple was absent. Stand MAI averaged 14 ft3/ac/yr, with a standard
error of 2.1.

ELU 7 is similar to ELU 6 in terms of coarse sandy soil textures. ELU 7 differs
from ELU 6 by lacking red maple in the overstory, and by having high coverages of
Arctostaphylos uva-ursi, Cladina spp., Comptonia peregrina, Apocynum
androsaemifoliumspr. The productivity of ELU is significantly less than all other ELU’s
(p < .05).

Figure 2.26 displays high quality species comprising the seven ecological species
groups defined in this research based on direct relationships to environmental conditions.
Ecological species groups 1, 2, 4, and 8 are identical to groups 1, 2, 4, and 8, respectively,
identified by TWINSPAN based on high quality ground-flora (Table 2.17, Figure 2.23).
TWINSPAN identified eight species groups, whereas tabular comparisons of ground-flora
distributions to samples grouped according to soil and floristic similarities only identified
seven meaningful groups. Because TWINSPAN did not differentiate two important
ecological land units from one another, units 5 and 6, it artificially separated species
comprising these two units into three groups. Moreover, tabular inspection of the
distribution of Pteridium aquilinium indicated that this species was too widely distributed
to be of indicator value of a particular environmental condition, hence ecological unit,

and was subsequently removed as a member of any ecological species group.

 

105

 

1.5

 

0.5

CA3

-0.5

 

 

 

 

-2 -1.5 -1 -0.5 0 0.5 1 1.5
CA 1

Figure 2.26. Ecological species groups in CA dimensions 1 and 3 of sample space.

 

106

Interpretation of Ordination and Classification of Samples Based on Soils Data

Principal component analysis of twenty-eight soil variables identified important
soil characteristics that differentiated multifactor ecological land units and also produced
ordination diagrams that grouped samples into meaningful classes of soil, ground-flora,
and overstory conditions (Figure 2.25). The first PC axis ordinated samples along a
textural gradient, (Figures 2.11, 2.12, and 2.13), with solas grading from sands to sandy
loams with increasing sample weights, and substrates grading from sands to sandy clay
loam with increasing sample weights. Soil drainage classes also graded from excessively
well drained to well drained with increasing sample weights along the first axis (Figure
2.14). The second axis ordinated samples along a substrate silt content gradient (Figure
2.13), and along a drainage class gradient in which samples containing moderately well
drained soils were separated from well drained soils (Figure 2.14). Silt content of soil
horizons at depths of 100 to 250 cm. increased with increasing sample weights along the
second axis (Figure 2.13). Soil drainage graded from well drained to moderately well
drained with decreasing sample weights along the second axis (Figure 2.14).

The most important soil variables based on high positive first dimensional factor
loadings, or variable correlations with variates derived in the first principal component,
included a coded variable describing substrate textures, textures between 100 and 150
cm., and sola textures (Table 2.9). The most important soil variables based on negative
factor loadings included depth to heavy textured horizons, and depths to mottles and
water tables. The most important soil variables based on high positive second

dimensional factor loadings included depths to mottles, dominant texture of sola and

107

substrates, textures between 100 and 250 cm., and depth to watertables (Table 2.10). The
most important soil variables based on high negative second dimensional factor loadings
included drainage class code, texture of the top 30 cm., and a coded variable describing
substrata textures.

Table 2.18 arrays key soil variables used in defining ecological land units that
include sola and substrate textures, and depths to lithologic discontinuities and
watertables. These were the key differentiating soil variables in the final classification
that was based on all factors.

Ordinations of samples in soil variable space in the first two dimensions of a PCA
distinguished all ecological land units (52.24), with only a few outliers not grouping
within respective classes. With the exceptions of ELU’s 6 and 7, ordinations of samples
in variable space in the first and third dimensions (See Figure 2.27) failed to group,
hence distinguish ELU’s, however.

Ordinations of samples in soil variable space in the first two dimensions of a PCA
(Figure 2.25) placed ecological land unit 3, that contained high silt content in the upper
100 cm. and in substrates, in regions with high positive sample weights in both the first
and second dimensions. Ecological land units 1 and 2 differed pedologically in terms of
textures in the upper 100 cm., with unit 2 containing sandy clay loam textures and unit 1
sandy loam textures; these units occurred immediately adjacent to one another in the
ordination diagram. Ecological land units 4 and 5 differed from each other in depths of
sandy sola overlying fine textured substrates; substrates were shallower in ecological land

unit 4. These units also occurred immediately adjacent to one another in the ordination

108

diagram. Ecological land units 6 and 7 differed from each other in terms of B horizon
development, with unit 7 classifying as a udipsamment and unit 6 classifying as a
spodosol. B horizon values and intensity codes were not identified as important variables
in PCA, but the nevertheless the ordination of samples in variable space distinguished
these units.

Agglomerative clustering of samples based on twenty-eight soil variables using
Ward’s and average linkage methods (Figures 2.28 and 2.29) failed to identify the same
groups as ordination space partitioning of samples in soil variable space, or as the
ordination and clustering of samples based on ground-flora. Clustering results between
the two linkage methods were not consistent. Interpretation of clustering results indicate
that soil groupings were highly variable, and simply included all possible combinations of
soil textures in sola and substrata. The ecological significance of these clusters are highly
limited, and are not corroborated by other ordination, graphical, and tabular analyses of
soils or ground-flora data, or clustering of ground-flora.

Analyses of soils data using PCA identified key differentiating variables and
assisted in defining final units. Ordination space partitioning of samples in soil variable
space was effective in identifying all final units. The placement of boundaries required
comparison with understory and ground-flora analyses, however. Clustering led to

inconsistent results and thus had to be disregarded.

109

 

 

 

 

3
2 5 1
7 5
3
2 l 3
2
1 6 11
m 7 6
6 4 3
D4 1 2 4
1
6
O 6
2
7
6
76 5 3
-1 5 4 1
5 4
5 l 2
4
4
-2
—2 -15 -1 --05 0 05 l 15 2
PCA1

Figure 2.27. Samples in PCA dimensions 1 and 3 of soil variable space labeled by ELU.

110

0._000 5.000

 

 

 

 

 

+----+

 

 

 

 

AaHnwuwwwaNaNNNr—aI—INNHHI—‘HI-swwmmmammmmqmmmm414144141mm
H
o
I
I
I
I
I

gilgfire 2.28. Sample clusters based on 28 soil variables using Ward’s linkage labeled by

 

111

0.009 1.000

 

 

 

 

 

 

 

 

 

 

 

\Immxlxlmo‘xlsJONm\IU'IU'IU'I-bU'ImbwwwwwANO—‘HHHNHLDUINANbHAD—‘I—JNNNH
N
A
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I

Figure 2.29. Sample clusters based on 28 soil variables using average linkage labeled by
ELU.

112

Interpretations of ordinations of samples in understory and ground-flora species space

The first dimension of a CA of ground-flora ordinated samples along a moisture
and nutrient gradient; this gradient is evident in the soils synthesis table with samples
ordered by weights of the first principal coordinate (Table 2.14) and in the final soils
synthesis table (Table 2.18). Figures 2.30, 2.31, 2.32, and 2.33 array samples in
commonly occurring species space, with samples labeled based on soil textures in the
upper 100 cm., soil textures in the lower 150 cm., and soil drainage class. These
diagrams further illustrate the moisture-nutrient gradient expressed along the first axis of
an ordination of vegetation.

Ground-flora integrates environmental conditions resulting from multiple soil
characteristics, hence the interpretation soil characteristics in understory and ground-flora
species space must consider multiple soil characteristics. ELU’s graded from mesic to
xeric along the first axis as sample weights graded from negative to positive. ELU 1
consisted of moderately well drained sandy loam solas to depths of 100 cm. overlying
sandy clay loam substrates. ELU 2 consisted of well drained sandy loam solas to depths
of 30 cm. overlying sandy clay loam substrates. ELU 3 consisted of sandy loam solas to
depths of 30 cm. overlying silty clay loam substrates. ELU 4 consisted of shallow sandy
caps overlying fine loamy substrates. ELU 5 consisted of deep sandy caps overlying fine
loamy substrates at depths greater than 100 cm. ELU 6 consisted of deep sandy soils with
moderate spodic horizon development. ELU 7 consisted of deep sandy soils lacking

spodic horizon development.

 

 

 

 

1.5
7
1
f7
1 1
0.5 111 6
1
N ?1
25 o 32 2 5
2 $2 3 ?
5
~05 a 4 14 ’
4
-1
-1 .5 ._...., w-.. _ .. -.._._ _ . _ ...._..___ _._._..-.._‘._ __
-15 -1 .05 8A1 0.5 1 1.5

Figure 2.30. Samples in CA dimensions 1
and 2 of commonly occurring species space
labeled by ELU.

 

 

 

 

1.5
§
1 §
SCL
SSQFL 31.3
“05 an; L8
L
<
o E 33
0 L
5391; s
[SQEL SCL
~05 SCL §6L
SICBICL Qfi}
-1
-1 5 1 0.5 o o 5 1 1 5
CA 1

Figure 2.32. Samples in dimensions 1 and
2 of species space labeled by lower 150 cm.
soil textures.

113

 

 

 

 

1.5 {mm-- ~-~
§
1 §
i5LSL
0.5
N $ L S
5
o s L §
fiesta L?
s
L $
, SEE
-0.5: SCL
SICLL £1} Lg
-1 -—~ J
-15 1 05 0.5 1 15

Figure 2.31. Samples in CA dimensions 1
and 2 of commonly occurring species space
labeled by upper 100 cm. soil texture .

 

 

 

 

1.5
E199
‘ £219
WD
1W 5%”
0'5 m EXD
N
( “AWE
0 WM) SEUJ
° Wow 5%
é o
\M) V“)
“B0 V“)
-o.5 WD . v5
Mm MVlD
-1 .-
-1 5 -1 O 5 0 0 5 1 1 5
CA1

Figure 2.33. Samples in CA dimensions 1
and 2 of commonly occurring species space
labeled by soil drainage class.

114

This gradation in soil texture-drainage class combinations of samples in
understory and ground-flora species space appears to accurately represent a decreasing
moisture-nutrient gradient with increasing sample weights frequently reported for indirect
gradient analyses of ground-flora, corroborated with actual direct measures of soil
properties.

Interpretation of the meaning of the second principal coordinate of a CA of
ground-flora was less straightforward than interpretation of the first principal coordinate.
Six of the seven derived ecological land unit classifications could be partitioned into
discrete groupings in the first two dimensions, although sample placement in ordination
space changed continuously in a horseshoe pattern (Figure 2.30). Ecological land units 2
and 3 were intermingled in this two-dimensional plane.

It appears that the second dimension separates samples based on a successional
stability gradient. The idea of a successional gradient along this axis is corroborated by
interpretation of the squared cosines of samples in species space. Samples classified as
ecological land units 1, 6, and 7 were most strongly correlated with the first axis, whereas
both ecological land units 4 and 7 were more closely associated with axis 2 (Figure 2.34).

Ecological land unit 4 consists of ground-flora typifying both mesic northern
hardwood and dry-mesic and xeric oak flora. The ground-flora of sites sampled for
ecological land unit 4 consists of Uvularia perfoliata, Trillium grandiflorum, Viola spp.,
F raxinus americana, and Prenanthes alba, but also includes high coverages of Aster
macrophyllum, Lonicera canadense, Hamamelis virginiana, Amelanchier Spp, and

Vaccinium augustifolium, which are understory and ground-flora species that typify more

115

xeric ecosystems. Based on ground-flora composition and soil textures, this system has
the potential to succeed to mesophilic sugar maple-white ash ecosystems, or dry-mesic
oak-red maple ecosystems, depending on natural or anthropogenic disturbance regimes.
Padley (1989) discussed the effects of landscape context and exogenous
disturbance through wildfire on ground-flora and overstory composition for this area of
Michigan, and Host (1987) reported similar influences on overstory and ground-flora
composition for northwestern lower Michigan. In these studies, local ecosystems nested
within a larger landscape matrix composed of landforms and forests susceptible to fire
were compositionally different from equivalent adjacent local ecosystems based on soil

physical and chemical properties, and this was hypothesized as being due to fire history.

 

 

 

 

0.5
4
0.4
4
4 7
No.3 7
0
e 4 7
8
0.2 7
3 5 1
35 1 6 1
7
0.144 551 1
2 1 1
3 3 5 1
2
0 2 s 2 M—
0 0.1 0.2 03 0.4 0.5 0.6

Figure 2.34. Correlation of samples labeled by ELU

Coéine 1

with dimensions 1 and 2 of CA.

116

Therefore, ecological land unit 4 may be a fire disclimax and successionally unstable.

Ecological land unit 7 consists of well-washed outwash sands and xeric forest
types and ground-flora. This ecosystem is approaching an oak savanna condition, and is
known to contain prairie remnants in parts of northeastern lower Michigan including
rough fescue (F estula scabrella) and Hill’s thistle (Circuin hilli). This ecosystem may
have the potential to revert to savanna or prairie conditions following a significant
disturbance event such as wildfire, and could also be regarded as successionally unstable.

Figures 2.35, 2.36, 2.37, and 2.38 illustrate environmental gradients expressed by
the third dimension of a CA of understory and ground-flora species. The third dimension
of a CA of ground-flora ordinated samples primarily along a soil textural and possibly
secondary moisture gradient in which samples containing high silt content in sola (Figure
2.36) and substrates (Figure 2.37) were generally separated from samples containing
sandy clay loam sola and substrates; this separation distinguished ecological land unit 2
from unit 3. Other key soil variables were not ordinated along this axis. The plane
defined by dimensions 1 and 3 accentuated the separation of ecological land units,
enabling all units to be differentiated from one another in ordination space.

Figures 2.39 and 2.40 display tree species in sample space based on CA. Along
the first axis tree species grade from mesic northern hardwoods to dry-mesic red oak-red
maple-white pine to xeric black and pin oak—pine assemblages.

Figures 2.41 and 2.42 display samples labeled by ecological land unit in tree
species space. Along the first axis samples grade from the most mesic ecological unit,

unit 1, to the most xeric unit, unit 7. Samples of ecological unit 1 group together in the

117

first and third dimensions; these units are composed of high basal areas of sugar maple
and basswood. Ecological units 2 and 3 group together in the first and third dimension;
these units are composed of sugar maple, red oak, white ash, and red maple. Ecological
units 4, 5, and 6 group together in the first and third dimension; these units are composed
of red oak, red maple, and white oak. Ecological unit 7 group together in the first and

third dimension; these units are composed of black and pin oak.

118

 

 

 

 

 

 

 

1
1 s
SCL
0.5 0.5 1 f s
n L SCL s S S
( o 5 '0 SL 505 SICL S S S
o 1 3 o Lgégfisgsmt 5131. s
5101. S1§$L s
SICL
-0.5 -05 s 5
L3 8
- s
1 -1
'1'5 ’1 ‘ ' CA1 ' ' .15 .1 05 0 0,5 1 1.5

Figure 2.35, Samples in CA dimensions 1 Figure 2.36. Samples in CA dimensions 1
and 3 of commonly occurring species space and 3 0f commonly occurring Species Space

 

 

 

 

 

 

 

labeled by ELU , labeled by top 100 cm. soil texture.
15 ...__..._ ..-.-..__....z w..- .-. z, . ..--.. - V -_”‘..,...,._w.. — . _ . -.-....._...._ . _.. . ..... .. . 7. .. 7 .. , 15 .._ _ .-... .. ....w , 7.. v .2 ‘
1 s 1 s
SCL s s
0.5 i S 0.5 §: 5
L
selCL SCL SCLCE S FSLS SE LS SS 3
SCL SICL SL LS n 5L LS LS s
g o sgggfififlsmt §I‘é't 5 o SLg'Sng‘sf $.16 SL 155;
SICL $99k S s L E SL Lg s s
SICL L
.05 S1CL 511‘ SL -05 s S s
SCL 5 L5 5
L8 5
-1 -1
-1.5 -15 ~
-1.5 1 -05 o 0.5 1 -15 -1 -05 0 0.5 1 1.5
CA 1 CA 1

Figure 2.37. Samples in CA dimensions 1
and 3 of commonly occurring species space
labeled by lower 150 cm. soil texture .

Figure 2.3 8. Samples in CA dimensions 1
and 3 of species space labeled by upper
30 cm. soil texture.

 

 

119

 

 

 

 

 

 

 

 

 

 

 

 

3 .... ~ - ~ ~ ~ - —~ -! 3 3;“ ~ “1
JP 1
l RP
1 8P 1 wo
N 1 n
31 RP i 31
1 1 8°
1% 1 1 BA
QA
BA 1 3M
0 '- o
2 “1° 1 1'11; \18”
pa BC 1 BP
1 am WC 1
_1 1 0A _.__..1 ,1 1- PB -
-1 0 1 2 3 -1 0 1 3
CA 1 CA 1
Figure 2.39. Tree species in dimensions 1 Figure 2.40 Tree species in dimensions 1
and 2 of samples space. and 3 of sample space.
2.5 2.5
2 2 a
777 6
1.5 -, 7 1.5
g I1 1 ’1
°‘ 1 1 n 1 5 6.11 1 7‘.
I: c ’ l
a a ’2 .1 ,1
50.5 3 50.5 11‘. '1 g 5 1
.§ .§ ‘1 1’11 '4 .’ 1
o 72 o . 3 2 . \ 5 ,
2 6 1 :2 3 2 ’ ' \.
as 5 2 5 . l 6 1
.05 2 5 05 31144 , 7 7 7.,
6 s
-1 5 -1 7
.15 ~15
4.5 -1 -O.5 O 0.5 1 1.5 2 2.5 -1.5 -1 -O.5 O 0.5 1.5 2 2.5
Dimension 1 Dimension 1

Figure 2.41. Samples in CA dimensions 1

Figure 2.42. Samples in CA dimensions 1
and 2 of tree species space labeled by ELU. and 3 of tree species space labeled by ELU.

120

COMPARISON OF ORDINATION AND CLUSTERING RESULTS

Multifactor ecological units classified in this research were used in the following
sections to compare results of CA to DCA using commonly occurring and high quality
ground-flora, and compare results of agglomerative clustering and TWINSPAN. To
further explore differences in results of CA to DCA, and differences in clustering results,
using commonly occurring and high quality ground-flora species, units published using a
multifactor approach fiom northwest lower Michigan (Host 1987) were used as a basis
for comparison. Host (1987) identified nine ecological land classification units using
multivariate classification and ordination methods for upland forests of northwestern
Lower Michigan based on similarities in ground-flora and overstory composition and soil
characteristics among 76 sample stands. Two-way indicator species analysis
(TWINSPAN) was used in conjunction with ordination space partitioning (Gauch 1982)
to identify ecological species groups, and relationships between species groups and
significant environmental parameters were considered to derive a multifactor ecological

land classification.

Comparison of Correspondence Analysis and Detrended Correspondence Analysis for
Ecological Land Units in Northeastern Lower Michigan

Figure 2.43 compares CA and DCA results for commonly occurring species in the
first and second dimensions of an ordination diagram. CA displays the classic arch
effect, and fails to differentiate ecological land units 2 and 3 that are intermingled. All

other units are distinguished by CA. DCA differentiates all units in the first two

121

dimensions of commonly occurring species space.

Figure 2.44 compares CA and DCA results for commonly occurring ground-flora
species in the first and third dimensions of an ordination diagram. CA differentiates all
units in this two-dimensional subspace. Similarly, DCA differentiates all units in this
two-dimensional subspace of commonly occurring ground-flora species space.

Comparing CA for commonly occurring ground-flora species in the first and third
dimensions of ordination space (Figure 2.44) to DCA results for commonly occurring
ground-flora species in the first and second dimensions (Figure 2.43) reveals a nearly
identical configuration of samples, with only the scale differing between ordination
diagrams. Detrending removed the curved configuration of the first and second
dimensions of CA, but CA revealed the same patterns in the first and third dimensions as

DCA revealed in the first and second.

122

 

 

 

 

 

 

 

 

 

 

 

 

 

1 1
0.5 0.5.
2 5
N m 5
< z < ..
o 0 o 0 ;
-O.5 -051 5
-1.5 4‘ -1.52
-1.5 -1 -o.5 0.5 1 1.5 -1.5 -1 -o.5 o 0.5 1 1.5
8A 1 CA 1
200 1 200
150 150

   

 

 

 

 

 

 

 

   

N no
< <
0100 0100
D D
7
‘4
4
50 4 50
4 .
4
0 O {.5
0 400 0 100 0 300 400
6 A1 6&1

Figure 2.43. Comparison of CA and DCA Figure 2.44. Comparison of CA and DCA
for commonly occurring species in NE M1: for commonly occurring species in NE MI:
dimensions 1 and 2. dimensions 1 and 3.

123

Figure 2.45 compares CA and DCA results for high quality ground-flora species
in the first and second dimensions of an ordination diagram. CA displays the classic arch
effect, and fails to differentiate ecological land units 2 and 3 that are intermingled; all
other units are distinguished. DCA also fails to differentiate ecological land units 2 and 3
that are intermingled, but differentiates all other units.

Figure 2.46 compares CA and DCA results for high quality ground-flora species
in the first and third dimensions of an ordination diagram. Both CA and DCA
differentiate all units in this two-dimensional subspace. The configuration of samples in
this two-dimensional ordination are very similar, although the sign of scores in the third
dimension are opposite for equivalent samples for CA as opposed to DCA, with samples
and units arranged in reverse order along the third axis.

Figure 2.47 compares CA results for commonly occurring and high quality
ground-flora species in the first and second dimensions of an ordination diagram. Both
diagrams display the classic arch effect, and both ordinations fail to differentiate
ecological land units 2 and 3 that are intermingled. All other units are distinguished by
both ordinations.

Figure 2.48 compares CA results for commonly occurring ground-flora species
and high quality ground-flora species in the first and third dimensions of an ordination
diagram. CA differentiates all units in this two-dimensional subspace based on both data
sets. CA on high quality ground-flora species identifies one sample in ecological land
unit 1 as an outlier, and also places samples closer to one another, producing tighter

groupings, for units 4, 5, 6, and 7.

124

 

 

 

 

 

 

 

 

 

 

1.5 1.5
1 1
0.5 0.5
N 0‘)
' <
5 0 o o
2
-o.5 ‘ 4 14 1 -o.5
4
-1 1 1
1
1 5 1 -0.5 0 O 5 1 1 5 -1.5 1 0 5 0.5 1 1.5
8A1
CA 1
150
5

DEA 2
O
‘5:
0|
0}
\l

 

 

 

50 g

 

 

 

 

 

 

 

 

0 1 00 (#00 400

200 DCA

Figure 2.45. Comparison of CA and DCA

for high quality species in NE MI:
dimensions 1 and 2.

500 o 100 am 400 50

200 DCA

Figure 2.46. Comparison of CA and DCA
for high quality species in NE MI:
dimensions 1 and 3.

 

 

125

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 f 1 fa
1
0.5 1 1 05; (36
E s.
N . 1 a) W 55 6
< < <
-o.5 -o.5 ' 5 1
-1 -1 16
1
-1.5 .1.5g
-1.5 -1 -o.5 0.5 1 1.5 4.5 -1 -o.5 o 0.5 1 1.5
8A1 CA1
1.5 1.5 1
1 1
0.5 0.5
N 0
< <
o 0 o o
2
-o.5 i -o.5
4
-1 _1
-1.5 -1.5
-1.5 -1 -o.5 o 0.5 1 1.5 4.5 -1 -o.5 8A1 0.5 1 1.5

CA1

Figure 2.47. Comparison of CA of
commonly occurring and high quality
species in NE MI: dimensions 1 and 2.

Figure 2.48. Comparison of CA of
commonly occurring and high quality
species in NE MI: dimensions 1 and 3.

126

Figure 2.49 compares DCA results for commonly occurring and high quality
ground-flora species in the first and second dimensions of an ordination diagram. DCA
on commonly occurring ground-flora species identifies all units, whereas DCA on high
quality ground-flora fails to differentiate ecological land units 2 and 3 that are
intermingled. All other units are distinguished. Clusters of the same units are slightly
more compact in high-quality ground-flora species space for all units.

Figure 2.50 compares DCA results for commonly occurring ground—flora species
and high quality ground-flora species in the first and third dimensions of an ordination
diagram. DCA differentiates all units in this two-dimensional subspace based on both
data sets. DCA based on high quality ground-flora species places samples closer to one
another than DCA on commonly occurring ground-flora species. Samples scores are
reversed along the third axis for the two data sets, with equivalent samples and units
arranges in reverse order along the third axis.

In summary, CA and DCA were both effective in distinguishing ecological land
units classified in northeastern Lower Michigan in commonly occurring and high quality
species space. Detrending removed the curved configuration of the first and second
dimensions of CA, but CA revealed the same patterns in the first and third dimensions as
DCA revealed in the first and second. The implication is that either method may be used
to discern the underlying structure of vegetative data sets, but investigators need to
interpret the first and third as opposed to the first and second dimensions of CA to arrive
at the same results as DCA. Apparently information on secondary gradients expressed in

DCA axis two was deferred to higher axes in CA. However, an earlier interpretation of

127

the second principal coordinate of CA was that this axis represented a successional
stability gradient. This interpretation could not be made using DCA.

The use of high quality as opposed to commonly occurring ground-flora species in
DCA or CA ordination procedures provided little if any advantage based on this sample.
Clusters of the same units were slightly more compact in high-quality ground-flora
species space for several units for both DCA and CA. For DCA, there was a loss of
efficiency in identifying latent structure of ground-flora data using high quality species,
with two units that were distinguished in the first and second dimension using the
commonly occurring data set intermingled using the high quality data set. Information
on important secondary gradients expressed in DCA axis 2 was apparently deferred to the
third axis in high quality species space.

A benefit of using CA instead of DCA lies in the geometric interpretation of a
CA. Species sums of squared cosines can be added across any number of dimensions to
determine the correlation of a species to that particular low dimension subspace. The
fidelity of a species to reduced subspaces can be quantified and interpreted in ordination,
or used to condition data for subsequent clustering procedures by eliminating

ubiquitously or spuriously distributed individuals.

128

 

 

200 200 ,

 

 

 

 

 

 

[X3A.l l)CV\l

 

 

200 1 200g

ISO

 

 

 

 

 

 

 

 

 

 

 

 

6 7
9
0 100 200 300 400 500 200 300 500
l)CfiLl [XJA I
Figure 2.49. Comparison of DCA of Figure 2.50. Comparison of DCA of
commonly occurring and high quality commonly occurring and high quality

species in NE MI: dimensions 1 and 2. species in NE Ml: dimensions 1 and 3.

129

Comparison of Correspondence Analysis and Detrended Correspondence Analaysis for
Ecological Land Units in Northwestern Lower Michigan

Figures 2.51 to 2.58 display samples in commonly occurring and high quality
species space labeled by ecological units as defined by Host (1987). Figure 2.51 displays
samples in dimensions 1 and 2 of commonly occurring species space based on CA. The
ordination diagram displays the classic arch effect. Units 1, 2, 5, and 9 can be partitioned
discretely, whereas units 3 and 4, and units 7 and 8 are intermingled in this ordination
space.

Figure 2.52 displays samples labeled by ecological land unit in dimensions 1 and
2 of commonly occurring species space based on DCA. Units 1, 2, 4, 5, 6, and 9 can be
partitioned discretely, whereas samples of unit 3 are scattered above and below the
region occupied by unit 4, and units 7 and 8 are intermingled in this ordination space.

Figure 2.53 displays samples labeled by ecological land unit in dimensions 1 and
3 of commonly occurring species space based on CA. As in Figure 2.52, units 1, 2, 4, 5,
6, and 9 can be partitioned discretely, whereas samples of unit 3 are scattered above and
below the region occupied by unit 4, and units 7 and 8 are intermingled in this ordination
space.

Figure 2.54 displays samples labeled by ecological land unit in dimensions 1 and
3 of commonly occurring species space based on DCA. Units 4, 5, 6, and 9 can be
partitioned discretely, whereas samples of unit 3 are scattered above and below the
region occupied by unit 4. Units 1 and 2, and units 7 and 8 are intermingled in this

ordination space.

130

 

 

 

 

1.5 1
1 6
.- ,.» LB» . I
55 556
-.2 5 ‘ , 7
0.5 ' ..
13 ‘, r 1 88
1 1,3 ‘72-.. , a 7778
2 o (2 2/
o 122? // 9 1 g 7
1 1/ J ,9.
1
-o.5 / !9 9
1 2 1" i. 9 9-
21I
,1 1
-1 {111
-1'5 _..., ....W. W, .....»“...—.--..mMHVr- . 2
-1 -O.5 0 0.5 1 1.5 2

CA1

Figure 2.51. CA for commonly occurring
species in NW MI: dimensions 1 and 2
labeled by ELU as defined by Host (1987).

 

 

 

 

 

 

 

 

200 1
2 3
1' 1‘1 3
2 2
150 ‘1 3 3
2‘1 -- -.
‘. . . 5 -
4- 4‘4 . I 18
i, 2 7s 45:5 I‘I ‘1
2 3 ‘1 5‘ 1' ‘1
N .75;- 1 1 , a
300 i 1 , 19
C) 5 f 117 a 7 1 _
1' 1 1
3 1C 1 7 '1’
,I a 7 9
121» 1 19”,} ‘9 '1
50 2‘1 ".6 a i
1 .
1
I
1
1
1
1 I
0 F—LL— — ——— .
O 100 200 300 400
DCA 1

Figure 2.52. DCA for commonly occurring
species in NW MI: dimensions 1 and 2
labeled by ELU as defined by Host (1987).

 

 

 

 

2 1 1601 ~-r ~ ‘8
7 14o 1
1.5 1 ‘1\ 1
8 1 120 7‘1 1 9-
" a 1 8 7 1 1
1 i 1
1 100 a 1 ,1
n 7 1 n 7 1 11
' ‘ 1 < so ”””- 8 1
50.5 1 7 _,° 7 1 8 m1 32‘23 1 7 8 7 1 1
1,111 i ‘3 1 I 2% .3 243' [=5 1“ if
lé‘ 2, 9 1 60 " V» ' 3'1 '3‘ 3‘1 5 1516‘ i 1
0 1 “1; ’9 _ 55 ‘ 1 1‘ It's 1'6 e 77‘ 9
1? 3 j‘ is 6: 7 .- 91 4014 $ 8 1
. :1 2° 1 11‘ .119 1
9 9 1 1 9 j
-1 1L1”— * i o 1 2 ~.- 1.91
-1 -o_5 0 0,5 1 1,5 2 O 100 200 300 400
CA 1 DCA 1
Figure 2.53. CA for commonly occurring Figure 2.54. DCA for commonly occurring
species in NW MI: dimensions 1 and 3 species in NW MI: dimensions 1 and 3
labeled by ELU as defined by Host (1987). labeled by ELU as defined by Host (1987).

131

Figure 2.54 displays samples labeled by ecological land unit in dimensions 1 and
2 of high quality species space based on CA. The ordination diagram displays the classic
arch effect. Units 1, 2, 5, 6, 7, 8, and 9 can be partitioned discretely, whereas units 3 and
4 are intermingled in this ordination space.

Figure 2.56 displays samples labeled by ecological land unit in dimensions 1 and
2 of high quality species space based on DCA. Units 1, 2, 5, 6, 7, and 9 can be
partitioned discretely, whereas units 3 and 4 are intermingled in this ordination space.

Figure 2.57 displays samples labeled by ecological land unit in dimensions 1 and
3 of high quality species space based on CA. Units 1, 2, 5, 6, and 9 can be partitioned
discretely, whereas units 3 and 4, and units 7 and 8 are intermingled in this ordination
space.

Figure 2.58 displays samples labeled by ecological land unit in dimensions 1 and
3 of high quality species space based on DCA. Units 1, 2 5, 6, and 9 can be partitioned
discretely, whereas units 3 and 4, and units 7 and 8 are intermingled in this ordination
space.

In summary, CA and DCA were both effective in distinguishing most, but not all
ecological units classified by Host (1987) in northwestern Lower Michigan in commonly
occurring and high quality species space. Detrending removed the curved configuration
of the first and second dimensions of CA, but CA revealed the same patterns in the first
and third dimensions as DCA revealed in the first and second dimensions for both the
commonly occurring and high quality ground-flora data sets; only the scales and

compactness of groups differed. In contrast to comparisons of ordinations in northeastern

132

Michigan, use of high quality species improved the discriminating power of ordination
space partitioning for both CA and DCA by distinguishing unit 7 from unit 8, which were
intermingled in all dimensions of commonly occurring species space in CA and DCA.
Use of high quality species as opposed to commonly occurring species reversed the signs
of sample scores along both the second and third dimensions. Use of high quality species

also produced tighter groupings of samples for both CA and DCA.

 

 

 

M
l l
gm. .- $9
0.52 \1': @9 '-
2 7‘
f2 \ l
\ l
N 4 \ 3 3 7
< 3i) » p 7 i 7
0 4,- g, --., is l f r
‘ :55 i [x "I \ 77
.0'5 ‘ 5 1“ x“ \\ h 8
l
(6 l
-1 1 1
1
6 {I
_6
-15 ~
1 05 0 05 1 15
CAs1

Figure 2.55. CA for high quality species in
NW MI: dimensions 1 and 2 labeled by ELU

as defined by Host (1987).

 

 

 

 

 

1 .
67 6 1: $9
0.5 I," '
/’i"H i6, ...-5r i
.1 \\.__ ,
l ‘ 7 .
5 55 [I
0 , X5 ‘ i
,1 ‘” f 3
(3] __ _ 9 l a j
‘. (r V. I!
o '1 :.
7
1 8 8 i
77 I?
-15 =. g";
-2
1 -05 O 05 1 15

CA1

Figure 2.57. CA for high quality species

133

 

 

 

200 5.2.“, ...- ..,,-.........~..,.-- ..-. 2.2..--“ a... ...--...-._,. -...--W.-." .. -_-._ .... ....m..__~_,.,._w_._ 2.2m»...
1
1
11 3
150 '-.
‘ l
2 .i
N . *7-.-»- ._\. .
3100 2 6. ' a “$7 7 7‘19)
o . i i ‘~ 99
{s 2 “x ) l \ 7 \ ’l .7 .
f iii ! y ,’ 1‘
i 52F 3 66, 6" 8
,‘ l .. «a
50 2‘ 4 5 \
2 {4 ./,« l
2i 3 I' '
. ., 5,’
a 4'5?
l f
2’ 4 J
0 223- - -_
o 100 200 300 400
DCA1

Figure 2.56. DCA for high quality species
in NW MI: dimensions 1 and 2 labeled by

ELU as defined by Host (1987).

160

 

140

 

O 100 200 300
DCA\1

~4afi

 

Figure 2.58. DCA for high quality species

in NW MI: dimensions 1 and 3 labeled by in NW MI: dimensions 1 and 3 labeled by ELU

ELU as defined by Host (1987).

as defined by Host (1987).

134

Comparison of Clustering Methods for Ecological Land Units in Northeastern Lower
Michigan

Figures 2.59 to 2.66 display clustering results for samples based on commonly
occurring and high quality ground-flora species using three agglomerative clustering
techniques and TWINSPAN divisive clustering. Agglomerative clustering of samples
using average linkage (Figure 2.59) and Ward’s linkage (Figure 2.61) methods for the
high quality ground-flora data set produced identical results, and differed in four respects
from equivalent clustering of samples based on the commonly occurring ground-flora
data set (Figures 2.59 and 2.61 , respectively). For both linkage methods, the high quality
data set differentiated ecological land units 2 and 3 from one another, whereas the
commonly occurring ground-flora data set failed to separate these units. The high quality
data set also placed stand 42 adjacent to other samples that were classified as ecological
land unit 2, and placed stand 20 in a cluster with other samples that were accurately
classified as ecological land unit 7. Commonly occurring ground-flora identified two
clusters composed of mixtures of ecological land units 5 and 6, whereas the high quality
data set only identified one mixed cluster.

Agglomerative clustering of samples using flexible-beta linkage method for the
high quality ground-flora data set (Figure 2.63) differed in five respects from clustering of
the commonly occurring ground-flora data set (Figure 2.64). The high quality data set
difi‘erentiated ecological land units 2 and 3 from one another, placed stand 42 with other
samples that classified as ecological land unit 2, placed stand 20 with other samples that

classified as ecological land unit 7, and misclassified one less sample in the cluster

135

composed of ecological land unit 7. Commonly occurring ground-flora identified two
clusters composed of mixtures of ecological land units 5 and 6, whereas the high quality

data set only identified one mixed cluster.

136

D I S T A N C E

525.749 700.426

176.395 351.072

1.719

+-------+-------+-------+-------+----—--+-------+

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_ -

_ -

_ -

_ .

- .
- Illllllll
- . .
— . .
- - -
illlllllllll - -
- . . -
. - _ _
- - lllllll|l|l|l .
n - . u .
_ - . - o
. - llllllllll - c
u _ - - - -
IlllI . - - . -
— . - - - n
- llllllll . - . .
- . . - - . -
c - - - c . .
. - . IIIII - . -
_ - . - . . - . .
. - . - - Ill - .
— _ - . . - - - u
- _ _ Illllll - - . . —
llllll - _ . . — . _ - -
- . — - - - . - _ . -
- - - lllllll - - - — u _ .
— - - . u _ . . — . . _
g - u . - llllllll. _ u . u _
u - - . - . _ - - - _ - -
. - - . I'll - . - . - _ - -
- _ - _ . - - . - . — _
- IIIII - . - — - . . - u
- _ _ - Illll - . — . — — c
- - - - u - . . - - . - —
I'll . — . — - - - _ . . . —
_ _ ll. - _ - - - - - -
_ - - . _ - III - . — -
- — - . - _ - _ - _ - -
_ - — - . u _ - -||l| . u
. - - - - _ lllllll - . . u
- Ill . _ - . . - - . -
. _ Illlll . - c - - — -
_ _ c u . - . - — Ill -
IIIIII - _ . - . - - — -
. n . - Ill-I. . - - p -
ll . _ - _ - _ . — -
. — c . - _ . . ll - -
- . u - - Il— - . . . - -
- . u - _ - a - - ll _ u -
- c 'l- - u . - p . n - - _
- . c - . _ u — Ill — - - . - -
— - - — - Ill - _ - - _ - - u u - -
- - - . . . - - u - - — - . - all g n _
c u __ - - - — u - - g - - - _ - _ .llll —
u . - .llll - - a u . . c - _ u _ g - . -
u . Ill - - a . - - n - u c - u - - _ — .IIII- _
u - n . g u c - - . — - - .ll - c - u - - - u n n
n - u - u - - _ c . - u - - c - - - — n - - — n . .Il-
- . - a - - .ll— - - ullln— -||ll.||l|- . - - - - _ - - . u u
1 3796 23498 031567284570142034591698
12134589 671212 44223 4344443433322 23242433 1
11111111 222322 33332 6565556655676444447??? 4

 

Figure 2.59. Clustering of samples labeled by ELU based on high quality species in NE

MI: average linkage.

1016.955

764.247

137
511.540

D I S T A N C E

258.833

6.125

Il'lllllllllI'I-IIIIIIII'IIIIII-Illlalllll

 

______ |--——---+------—+—-——-—-+-------+-----—-+--—---—+--——---+---—---+

_ _
_ _
_ .
_ _
- IIIIIIIII
_ _ _
_ _ .
. lllllllll .
_ . _ _
_ _ . _
llllllllllll . . _
_ _ _ _ _
_ _ llllllll _ .
_ _ _ . _ .
_ . . _ _ _
_ . . I I _ .
_ _ _ _ . _ _
_ _ . _ . I I I .
_ _ . _ . _ _ _
. lllllllll _ _ _ _ _ .
_ _ . _ _ . _ . .
llllll . _ _ . . _ _ .
. _ . _ . . _ _ _ _
. _ _ '''''' _ _ _ _ _ .
_ _ _ . _ _ . _ _ . _
_ _ _ d _ IIIIIIII _ . . _ _
. _ . . _ _ . _ . _ . _
_ _ . . _ llllll _ _ . _ _ _
. _ . _ _ _ _ _ . _ . _ _
_ llllll _ . _ _ . . _ _ . _ _
_ _ _ _ . _ _ _ . . . _ . .
_ _ _ _ _ . . _ . _ I I _ _ .
_ _ _ _ lllllll . _ _ . _ _ . _ . _
. . _ _ . _ _ _ _ . _ _ _ . _ .
. _ _ _ . _ _ _ _ . . _ . _ I I I .
_ lllll _ _ _ . _ _ _ _ _ _ . _ _ _ .
_ _ _ _ _ . . _ . _ _ _ _ . _ _ _ .
_ I I . _ _ _ . _ _ _ _ . _ _ . _ _ _
_ _ _ . _ I I . . _ . . . _ _ _ _ _ _ _
_ _ _ . _ _ _ _ _ . . . _ _ _ . . . . .
_ _ _ . _ . _ _ I I I _ . _ _ _ _ _ _ . _ _
_ _ _ . _ _ _ _ . _ _ . _ _ . . _ _ _ I I .
_ _ _ _ _ . _ I I I _ . . _ . _ _ _ _ _ _ . _ .
_ _ _ _ _ _ _ _ . _ . . . _ . _ _ _ _ . . . _
. _ _ I I I _ . _ _ _ . . _ . _ _ _ . . _ _ . . .
. _ _ . . _ . . _ _ _ _ . . _ I I I I _ . . _ . . . .
_ . _ . _ _ _ . _ _ _ _ . . I I _ _ . . _ _ _ . _ _
_ . . _ . . _ _ _ _ . _ _ IIIIII . _ _ _ _ . _ _ _ _ . .
_ _ _ _ . _ _ . _ I I I I _ _ . _ _ _ _ _ _ . . _ _ _ . . .
_ _ _ I I I . _ _ _ _ _ _ . _ _ _ . _ _ _ . _ . _ _ _ . . _
_ _ . _ . _ . _ . _ . _ _ . . _ . _ _ _ _ _ _ . . . — _ .
_ _ . . _ . _ . _ _ _ _ . _ _ I I I I _ _ . _ . . _ _ _ . . . .
_ _ _ _ . I I _ _ _ _ _ . . _ _ . _ . _ _ _ _ . _ _ . . _ . .
_ . . _ _ _ . _ _ . _ . I I . _ _ _ _ . _ _ _ _ _ . . _ . _ . .
_ . _ _ . _ _ _ _ _ _ _ _ . . _ _ . _ . _ _ _ . . . . . I I . . _
. _ _ _ _ _ . _ _ . _ _ _ . _ _ . I I I _ _ . _ I I I I _ _ . _ . . _ . . _
. _ _ _ _ _ _ _ _ _ _ _ _ _ I I _ _ _ . _ _ _ _ . _ _ _ . _ _ _ . . _ _
. _ _ _ . _ . _ _ . _ I I _ . . _ _ . _ _ — . _ _ _ _ . _ . _ . . . . . .
_ _ _ _ . . _ _ _ _ _ . . _ _ . . _ . _ _ I I _ . _ . . . _ _ _ _ . _ . . _
. ~ _ I I . _ _ . . _ _ _ . . _ _ _ _ _ . _ _ . _ _ . _ . . . _ _ _ . - - . .
. . . _ . . . . _ . . _ _ . _ _ . . _ . _ _ _ . . . _ I I _ _ _ . _ _ _ _ — . _
. _ _ _ _ . _ . _ _ _ _ _ . . . . . . . I I I I _ . . _ . _ _ . _ _ . _ _ _ . _ _ .
. _ _ _ - - . . _ _ . _ . I I . . . _ . . _ _ _ . . I I I I _ _ . I I . _ . . . _ _ . _ . _
O
1 0 3 9 r0 8 7 2 9 8 3 2 0 1 5 3 6 7 2 8 0 1 5 9 9 6 1 4. 5 7 4. 2 O 3 4. 8
1 2 1 3 4. 5 8 9 l 6 7 l 1 2 2 2 4 2 3 4. l 4. 4. 4. 3 4 4. 3 4. 2 2 2 3 4. 3 3 3 3 3 1 2 3 2 4. 1
1 1 1 1 1.. 1 1 1 1 2 2 2 2 2 2 3 3 3 2 3 6 6 6 5 5 5 5 6 6 7 6 7 7 7 7 7 5 5 6 4. 4. 4. 4. 4. 4.

 

Figure 2.60. Clustering of samples labeled by ELU based on commonly occurring species

in NE MI: average linkage.

138

D I S T A N C E

993.904 1489.997 1986.090
-------+-------+-------+-------+-------+-------+-------+-------+

497.812

1.719

 

 

- .
- -
. -
- u
- .
- - 'lllllllll
- — - -
— . _ n
. _ - -
. - llllllll .
- . u — .
- - n c -
— lllllll'l _ — —
- . — - . -
— . - llllll . u
u - lllllll - . - -
u . _ _ - - — .
_ - llllulnllll . - - . -
lllllll - — _ _ - — c u
. - - - _ lllllll _ . -
Illl - - - _ . . — - .
- — - - - _ lllllll - - - .
c _ . . - _ . - - - III .
I - - . - - — . - u _ _ - .
- u . . - III . . — - — - - -
- — - - — - — ll I'll - - - _ . a
_ . c - lllll . - _ u - - I'll. - . - .
- - u - — . u - . u - — - - - — .IIII -
Il.|||| — u - — c - _ . n — — - — . c _ - - |||
- .lll - ll. - . - . a _ - - —|.ln|- - - u - -
— n - Illu- - - - . - - . .lllll . - — u — - — u u . -
- .ll- - u c g - II - - . a - . u u n a a . - - II - -
. . u - c u - c .l'. . .IIIII- _ - c . a - - - .ll- — p - - u
u - - - u g - u - . o - — - .ll- - u — n - - c c . - .llll— .IIIII-
- . n - - -|I- . - a .ll. - - . . .Illllllll- - - . - - - . - —I|- Ill
1 0 3796 98238 201528367457014203 59169
21345891671212 23441 14443434433322.1232 24333
11111111222322 32334 6665665555567 4444 77777

 

 

Figure 2.61. Clustering of samples labeled by ELU based on high quality species in NE

MI: Ward’s linkage.

139

DISTANCE

1790.786 2385.673

1195.899

601.012
|-------+-------+------—+-------+—------+-------+-------+-------+

6.125

 

 

 

 

 

. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
.
.
.
.
.
.
. _
. .
. .
_ .
. .
_ . lllllllllll
. . . .
_ . . .
. . . .
. . . .
. . |||||||| .
. . . _ .
. . . . _
. . . .
. lllllllll . . .
. . . . . .
_ . . lllllll _ .
. . . . . . .
. . '''''' . . . .
. . lllllll . . . . .
llllll . . . . . . . .
. _ . . . . . . _ .
. . . . . . . . _ 999
. . . . . . . . . . .
. 9999 . . . . . . . . .
. . . . . . 99 . . . . .
. . . . . . . lllll . . .999
99 . . . . . . . . .. .
.. . . . . . . . . .. .
.. . . . 999 _ . 999 . . _. .
.. . . 9.999 . . . . . . _ .. .
.99 . . 99. . . . . . . . . .. .
... . .999 .. . . . . . . . 9999.. .
... . . . .99 . . . . . . . . ... .
... . . . ... . . . 99.9 . . . . ... 9
...999. . ... . . . . . . . 99 . ... .
.... .. . ... 9.9 . . . . . . ..9999... .
.... .99. ..99.. . ._ . . . 99 ... .... .
.... .... ...... 99.__. . . .. ... ...99.
...99.... ...—.99.... 999999 .99..999......
1- c
21.31.589.167
111.11.111.22

 

 

Figure 2.62. Clustering of samples labeled by ELU based on commonly occurring species

in NE MI: Ward’s linkage.

1160.406

144)
581.063 870.734

0 I S T A N C E

291.391

1.719

 

 

 

+ 999999999999999999999999999999

. . .

. . .

. . .

. . .

. . .

. . .

. . .

IT . .

. . .

. . .

. . .

. . .

. . .

. . .

. . .

+ 999999999999 .

. . . .

. . . .

. . . .

. . . .

. . . 99999999999

. . . . .

. . . . .

IT . . . .

. . . 99999999 .

. . . . . .

. . . . . .

. . 999999999 . . .

. . . . . . .

. . . . . . .

. . . . . . 99999

+ . . . . . . .

. . . 9999999 . . . .

. . . . . . . . .

. 999999 . . . . . . .

. . . . . . . . .

. . . . . 999999 . . .

. . . 999999 . . . . . .

. . . . . . . . . . .

IT. . . . . . . . 999 . .

. 99 . . . . . . . . . . .

u . . . u . . . . . . . .

. . 9999 . . . . . . . . . .

. . . . . . . . . 99 . . . .

. . . . . . . 99 . . . . . . .

. . . . . . 999 . . . . . . . . .

. . . . . . . . . . . . . . . . .

IT . . . . . . . . . 999999 . . . . . .

. . . . . 999 . . . . 9999 . . . . . . .

. 99 . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .999 . .

. . . . . . . . . . . . . . . . . . . . . 999

. . . . . 99. . . . . . . . . . . . . . . . .

. . .999 . . . . . . . . . . . . . 99. . . . . .

. . . . . . . . . . . . . .999 . . . . . . . . . . .

IT . .9999. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . 99. . .

. . . . . . . . . . . 99 99. . . . . . . . . . . . . . . . .

. . . . . . . . . .99. . . . . . . . . . . . . . . . . .99 . .

. . . . . . . . . . . . . . . .99. . . . . . . . . .99. . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .999.

. . . . . . .99. . . . . . . . . . . . . . 99 . . . . . . . . . . .99. .

. . . . . . . . . . . . . . 99. . . . . .9999. 9999. . . . . . . . . . . . .99

1 0 3796 98238 20153672 4.574203401591169

11213458916749.2412 23441 14443443 333123242224333
111111111222322 32334 66655556 55644444767777?

 

 

Figure 2.63. Clustering of samples labeled by ELU based on high quality species in NE

MI: beta linkage.

141

D I 3 1' A N C I
6.125 385.426 764.727 1144.028 1523.328

A A
‘T T

1'

4+-
«0-
4+

T

 

A
T

 

p

 

 

 

HHHHHi-IHH
UQUwa-‘NH
I
I

 

 

 

 

 

NNNUUNNN
b
N
I
I
I
I

 

 

 

 

 

 

 

 

 

 

 

O

.

O
I
I
I
I
I
I
I
I

OOQQQQQ
U
0
I
I
I
I
I
I

 

 

 

 

 

 

 

 

 

Figure 2.64. Clustering of samples labeled by ELU based on commonly occurring species
in NE MI: beta linkage.

142

Results of divisive clustering of samples using TWINSPAN for the commonly
occurring ground—flora data set (Figure 2.65) differed in five respects from divisive
clustering of the high quality ground-flora data set (Figure 2.66). TWINSPAN clustering
of samples based on high quality ground-flora species identified all seven ecological land
units, whereas clustering based on commonly occurring ground-flora species only
identified three units correctly, with units 2 and 3 mixed in one cluster, and units 6 and 7
mixed in another. TWINSPAN clustering of samples based on high quality ground-flora
species misclassified two samples consisting of ecological unit 6 in the cluster composed
of unit 5, whereas clustering of samples based on commonly occurring ground-flora
species only misclassified one sample in this cluster. TWINSPAN clustering of samples
based on high quality ground-flora species misclassified stand 18 as unit 3, whereas
clustering of samples based on commonly occurring ground-flora species identified this
unit as an outlier. TWINSPAN clustering of samples based on high quality ground-flora
species did not erroneously cluster samples 38 and 43 as a separate cluster, although stand
38 was misclassified into the cluster composed of unit 3. TWINSPAN clustering of
samples based on high quality ground-flora species misclassified a total of six samples in
the seven groups identified, whereas clustering of samples based on commonly occurring
ground-flora species misclassified a total of two samples in the three groups correctly

identified.

143
23 111111111 3332322222 4 454444 5555655 666777776 77 ELU

34 l 1 224222 11 1 132234 3334444 124334433 22 Stand
83 890345121 4926786739 8 452304 4236157 210198967 05

00 000000000 0000000000 0 111111 1111111 111111111 11 level I
00 000000000 1111111111 1 000000 0000000 111111111 11 level II
00 111111111 0000000000 1 000000 1111111 000000000 11 level III

Figure 2.65. TWINSPAN classification of samples based on commonly occurring
species labeled by ELU in NE MI.

111111111 2222223 323334 44444 655555556 666 7677777 ELU

1 l 22 1 12 234241 12234 444334333 124 3423432 Stand
118902345 6873697 983428 42304 157236457 210 1859960

000000000 0000000 000000 11111 111111111 111 1111111 level I
000000000 1111111 111111 00000 000000000 111 1111111 level II
000001111 0000000 111111 00000 111111111 000 1111111 level III

Figure 2.66. TWINSPAN classification of samples based on high quality species labeled
by ELU in NE MI.

In surmnary, use of high quality ground-flora in each of the three agglomerative
clustering procedures resulted in consistent classification of sample groups, ordering of
groups, and assignment of samples to respective groups; six of seven units were
accurately distinguished. Use of commonly occurring ground-flora also resulted in
relatively consistent classification of sample groups, ordering of groups, and assignment
of samples to respective groups for each of the three agglomerative clustering procedures;
five of seven units were distinguished accurately. The only differences among results of
the three linkage methods for commonly occurring ground-flora species was one less
accurate assignment of a sample to unit 4 for beta linkage, the inclusion of one sample of

unit 6 in the cluster for composed of samples of unit 7 for Ward’s linkage, the inclusion

144

of two samples of unit 6 in the cluster composed of samples of unit 7 for beta linkage,
and the identification of one sample, stand 18, as an outlier by average linkage.
Agglomerative clustering of commonly occurring ground-flora identified two clusters that
were composed of mixtures of ecological land units 5 and 6 with each method, and was
less effective in differentiating ecological land units 2 and 3 from one another. None of
the agglomerative procedures for either data set were effective in distinguishing between
ecological units 5 and 6.

TWINSPAN divisive clustering of samples based on commonly occurring
ground-flora species (Figure 2.65) only distinguished three ecological land units, units 1,
4, and 5, with all other clusters composed of mixed groupings of ecological land units.
TWINSPAN divisive clustering of samples based on high quality ground-flora species
(Figure 2.66) distinguished all seven ecological land units, although there were six
samples misclassified in total. In contrast, each of the three agglomerative techniques
only misclassified two samples among the six clusters identified using high quality
species, although samples for two ecological units, units 5 and 6, were grouped together,

hence undifferentiated from one another.

Comparison of Clustering Methods for Ecological Land Units in Northwestern Lower
Michigan

The data sets and nine ecological land classification units defined by Host (1987)
were used as a basis for comparing agglomerative and TWINSPAN clustering results on

commonly occurring and high quality ground-flora species in this section.

145

Figure 2.67 displays the classification as published by Host and Pregitzer (1991)
based on ninety-three commonly occurring species using TWINSPAN, with samples
labeled by ecological land unit. Six groups of samples were identified. The first cluster
identified unit 1, with three misclassified samples consisting of unit 2. The second
cluster was a mixture of units 2, 3, and 4. The third cluster was a mixture of units 3, 4,
and 5. The fourth cluster was a mixture of units 5 and 6. The fifth cluster was a mixture
of units 7 and 8. The sixth cluster contained only unit 9.

Figure 2.68 displays a reclassification of the same samples as used by Host based
on fifty-two high quality species whose sums of squared cosines were equal to or greater
than forty-five per cent in a five—dimensional subspace. The third and fourth levels of
division were selected subjectively based on differences in unit membership within
clusters using TWINSPAN. Samples were labeled by ecological land unit as defined by
Host (1987). The first cluster grouped seven samples of unit 1, with one misclassified
sample of unit 2. The second cluster grouped four samples of unit 2, with two
misclassified samples of unit 1. The third cluster grouped six samples of unit 2, with
three misclassified samples of unit 1, 2, and 3. The fourth cluster was a mixture of units
2, 3, and 4. The fifth cluster grouped five samples of unit 4, with one misclassified
sample of unit 3. The sixth cluster grouped five samples of unit 5, with three
misclassified samples of units 3 and 4. The seventh cluster grouped five samples of unit
6, and three samples of unit 5. The eighth cluster grouped three samples of unit 7, with
one misclassified sample of unit 8. The ninth cluster grouped five samples of unit 8, with

one misclassified sample of unit 7. The tenth cluster grouped six samples of unit 9, with

146

two misclassified samples of unit 7.

TWINSPAN on high quality species effectively grouped samples into unit 1, 2, 4,
7, 8, and 9, and less effectively grouped samples into unit 6. There was one
misclassification in the group of seven samples comprising unit 1, five misclassifications
in the group of fifteen comprising unit 2, one misclassification in the group of six
samples comprising unit 4, three misclassifications in the group of eight samples
comprising unit 6, one misclassification in the group of four samples comprising unit 7,
one misclassification in the group of six samples comprising unit 8, and two
misclassification in the group of eight samples comprising wit 9. The use of ground-
flora as an indirect measure of environment for multifactor units reported by Host (1987)
that included criteria for soil characteristics was improved using high quality species

based on these clustering results.

.86on bmfizvéwfi no woman Ammo: 60$ .3 wotomou moEEwm E. .8 coumofimmfloom .woN 83mm...

 

 

m Hm>mH HHooo HHHooo HHooooo HHHHHo HHHHHH ooooooooooooo HHooooooo Hooooo HHHHHHo
v Hm>mH HHooooo HHHHHH o HHHH o HHHHHHHo HHoooooo HHHHHH HHHHHHHHHHHHH ooooooooo HHHHHH ooooooo
m Hm>mH HHHHHHHo HHHHHH H 0000 o HHHHHHHH oooooooo HHHHHH HHHHHHHHHHHHH HHHHHHHHH 000000 0000000
m Hm>mH HHHHHHHH 000000 o 0000 o HHHHHHHH HHHHHHHH oooooo ooooooooooooo ooooooooo oooooo ooooooo
H Hm>mH HHHHHHHH HHHHHH H HHHH H 00000000 00000000 000000 ooooooooooooo ooooooooo 000000 0000000
aHm mhmmnmmm mmmnmm n nbnm m omowmomm «mmmvmmm mvvvvv mammmmvvmqmmm qmmNmHmmm NHNHNN HHNHHHH
nonmmvom mmmoHH n son v movomvvn omsonoma mmommm emomqmmmHHmmH nmmHmmoom mmHmmm mvmmmmo
ecmom mmmmvm m mmmvmm mmvm h HQVH on moHvaHH mnnmm mbmvvaH SH mHhhopmmN Nnmmmm mmmmqvm
7
4
1
Hume: mom .3 3:82 815% R .8 538.385 Hod 2:me
aHm mommmm mnpmmnnmmnmnmb moooommm 4mmmmvvmmm vvmqvmmmmmmmqvqNNNNNNNNNNQmH HNHNHNHHHH
smmsvo oommHonvoman vvmvmoov sonomoomms mmHmmommmmmmmHmvmmmHmHmmmnmm mmvammomm
ncmom mmmmm vammmthqmmm ow HoHH vaHHmbmmp nommmm nvvaHH mmm mbmmsmHn hmmmvmmmvm
_ _ _ _ _ _
zucHz _ ZucHz _ _ _ onmz Hoe _ oHcmx _
con _ 20H _ _ _
_ _ onoz _ _

 

 

 

mUOOZUHmm cumcuuoz xmo

148

Figure 2.69 displays agglomerative clustering of the samples classified by Host
based on high quality ground-flora species using Ward’s linkage method. Nine groups
were identified, and were labeled A through I for discussion purposes.

Cluster A grouped eight samples of unit 2, one sample of unit 3, and one sample
of unit 4. Cluster B was heterogeneous, and grouped two samples of unit 1, three
samples of unit 2, one sample of unit 3, and one sample of unit 4. Cluster C grouped four
samples of unit 4, three samples of unit 3, and three samples of unit 2. Cluster D grouped
four samples of unit 4, and three samples of unit 3. Cluster B grouped seven samples of
unit 1, and one sample of unit 2. Cluster F grouped five samples of unit 6 and two
samples of unit 5. Cluster G grouped three samples of unit 3, and two samples of unit 4.
Cluster H grouped five samples of unit 9. Cluster I grouped seven samples of unit 7, and
six samples of unit 8.

Agglomerative clustering of 76 samples based on high quality species was only
effective in grouping units 1, 2, 6, and 9. The procedure segregated cluster D from cluster
G, however, although both were composed of units 4 and 5. Cluster D contained
samples from the northern half of the Manistee National Forest, whereas cluster G
contained samples from the southern half of the Manistee National Forest. Cleland et al.
(1993) identified extensive climatically and physiographically uniform regions in the
northern portions of Lower Michigan that distinguished the southern half of the Manistee
National Forest from the northern half based on growing season. Additional analyses on
data sets for commonly occurring and high quality ground-flora species that eliminated

samples from the southern half of the National Forests were conducted as a consequence.

149

 

2
2 n “
3+ _ _
. _ .
m. _ _
1. _ _
_ . _
3. . .
_ . _
+ _ _
_ _ _
_ . _
. _ _
u _ _
_ _
fi. _ _
_ . _
7.+ _ _
l. _ _
6 _ _ _
3 _ _ _
_ _ _
2. lllll I ll _
_ _ _ _
_ _ _ _
+ _ _ _
. _ . _
_ _ . _
_ lllllllllllllllllllll _ _
_ _ _ . _
_ . _ _ _
om. . _ . _
7. _ _ _ _
.+ _ _ _ lllllllllllll
. _ . . _ .
:7o_ . . _ _ _
5. lllllllllllllllllll _ _ _ _
l. _ _ . _ _ _
. _ _ _ _ _ _
_ . . _ lllllll _ .
_ _ _ _ . _ _ _
+ _ _ _ . lllllll _ .
_ lllllllllll . _ . _ . _ _
_ . _ _ llll . _ . _ .
_ . _ ||||||||||| . _ _ _ _ _ _
_ lllll _ _ . _ _ . _ _ . .
l. _ _ _ _ _ _ _ _ _ lll _ _
_ _ _ lllll . . . _ _ _ _ . _ lllllllllll
fi. _ . _ . _ _ . _ _ _ _ . . llllllllll .
.+ _ _ _ _ . . _ . _ lll _ . _ _ _ _
9. . . _ _ _ _ llll. _ . _ _ _ llll. _ _
8. _ _ lll _ _ _ _ . _ lll. . _ _ _ _ _ _ _
7. lllll _ . . . _ . _ _ _ _ _ _ _ _ _ _ . _ llllll _
_ _ .llll _ _ _ lllllll . . _ . _ _ _ _ lll. _ _ . _ _ _
_ . _ _ . _ _ _ _ _ . lll . _ ll. _ _ _ _ _ _ _ _ _ _ _
_ _ _ _ . . _ _ _ _ llll. _ _ _ . _ _ _ ll . _ _ . . _ lllll _ _
_ _ . _ _ . _ . _ _ _ _ _ _ _ ll _ _ _ll_ _ _ _ _lll _ _ _ _ _ _
+lll _ _ _ _ _ . _ . _ _ _ _ _ _ _ . _ . _ _ _ _lll. _ . _ _ ll _ _ _ _
_ . _ _ _ _ . . _ _ . llll. _ ll. _ _ _ . _ _ _ _ _ _ _ . . _ . _ _ _ lllll _ _ ll
. _ ll _ . . . _ _ llll _ _ _ _ ll. _ . . _ _ . _ _ _ _ _ . _ . _ _ ll. _ _ . _ _ lll. .
_ _ _ _ _ . _ lll. _ _ _ llll _ _ _ _ _ . . _ . _ . _ _ _ _ _ _ _ _ . _ ll. _ _ _ _ _ lllll _ _ _ _
._ _. _ll ll. ._ lll. _ ll . lll _________ ...—.... ______.__ll_ _ __._ ___
4._ll_ll__ll_.ll_llll_ llllll _ll_ lllll ...-....ll__.....ll___________ lllll _._ll___
9 _
_
r3.l 1805691538 8582838 196725463 07126485949039 4.4 00270097 4.277 311351660657324
3 15134627766936562373634451377821111145457555446516768767622237132224234535437
w 223232222242214322122334432344.554.4551111..1112665665644455999997388887877787796
E
r AAAAAAAAAAABBBBBBBBCCCCCCCCCCDDDDDDDEEEEBEBEFFFFFFFGGGGGHHHHHIIIIIIIIIIIIIIII
e
t
s
u
l
C

uality species
ard’s inkage.

Figure 2.69. Clustering of 77 samdples from NW lower MI based on high (iv
with clusters labeled by letters an ELU’s labeled by respective numbers:

150

Figure 2.70 displays agglomerative clustering of 56 samples from the northern
Manistee National Forest classified by Host based on commonly occurring ground-flora
species using Ward’s linkage method. Eight groups were identified, and were labeled A
through H for discussion purposes.

Cluster A grouped eight samples of unit ELU 2, two samples of unit 3 and one
sample of unit 4. Cluster B was heterogeneous, with one sample of unit 2, one sample of
unit 3, three samples of unit 5, and four samples of unit 4. Cluster C grouped three
samples of unit 3 and one sample of unit 1. Cluster D grouped four samples of unit 6,
and two samples of unit 5. Cluster B grouped six samples of unit 1, and one sample of
unit 2. Cluster F grouped four samples of unit 9. Cluster G grouped four samples of unit
7 and six samples of unit 8. Cluster H grouped seven samples of unit 7, and six samples
of unit 8 Agglomerative clustering of 56 samples within a climatically homogeneous area
based on commonly occurring species was only effective in grouping units 1, 2, and 9.

Figure 2.71 displays agglomerative clustering of the samples from the northern
Manistee National Forest classified by Host based on high quality ground-flora species
using Ward’s linkage method. Seven groups were identified, and were labeled A
through G for discussion purposes.

Cluster A grouped seven samples of unit ELU 2, two samples of unit 3 and one
sample of unit 4. Cluster B was heterogeneous, with two samples of unit 2, one sample
of unit 4, and four samples of unit 3. Cluster C grouped four samples of unit 5 and two
samples of unit 4. Cluster D grouped four samples of unit 6, and one sample of unit 5.

Cluster B grouped six samples of unit 1, and one sample of unit 2. Cluster F grouped five

151

1301.772 1950.345 2598.918

653.198
—-—---—+-—--———+----—--+---——--+——-—--—+—--——--+-—----—+-—-—-——+

DISTANCE

6O
10
48
55
49
54
50
53
59
6

22
24
27
7

25

922H3132H2024n2922H2424:34H3:55.4132H3136roa6595cu1.1111.1112n9099n91,8H/968n6968H/7,8n/7HIQJ
AAAAAAAAAAABBBBBBBBBCCCCDDDDDDEEEEEEEFFFFGGGGGGGGGGGHHHH

Figure 2.70. Clustering of 56 samples from a homogeneous climatic zone in NW lower
MI based on commonly occurring species with clusters labeled by letters and ELU’s

labeled by respective numbers: Ward’s linkage.

152

samples of unit 9. Cluster G grouped six samples of unit 7, seven samples of unit 8, and
one sample of unit 9.

Agglomerative clustering of 56 samples within a climatically homogeneous area
based on high quality ground-flora species was only effective in grouping units 1, 2, 6,
and 9. Results of clustering samples based on high quality species differed from the same
clustering on commonly occurring species in three respects. First, unit 6 was effectively
identified using high quality species. Second, five samples of unit 9 were grouped
together, as opposed to four samples using commonly occurring species. Third, four
samples of unit 5 were grouped together in a cluster of six that contained two samples of
unit 4.

Figure 2.72 displays TWINSPAN clustering of the samples from the northern
Manistee National Forest classified by Host based on high quality ground-flora. Eight
groups of three or more samples were identified, with one group of two outliers also
separated. The first group consisted of six samples of unit 1 and one sample of unit 2.
The second group consisted of seven samples of unit 2, one sample of unit 1, and one
sample of unit 2. The third group was heterogeneous, with two samples of unit 2, four
samples of unit 3, and four samples of unit 4. The fourth group consisted of five samples
of unit 5. The sixth group consisted of four samples of unit 6. The consisted of three
samples of unit 7. The seventh group consisted of seven samples of unit 8, and two
samples of unit 7. The eight group consisted of six samples of unit 9, and one sample of

unit 7.

153

2283.181

1143.434 1713.308

573.561
|—-——--—+---——--+--——-—-+—---—--+-------+-------+-———---+-—-—-—-+

3.688

1

51
18
30
45
9

38
29
58
28
33

22323222422233343245544556666511111129999978878888787779
AAAAAAAAAAABBBBBBBBCCCCCCDDDDDEEEEEEEFFFFFGGGGGGGGGGGGGH

Figure 2.71. Clustering of 56 samples from a homogeneous climatic zone in NW lower

MI based on high quality species with clusters labeled by letters and ELU’s labeled by

respective numbers: Ward’s linkage.

154

TWINSPAN clustering of the samples from the northern Manistee National Forest

based on high quality ground-flora was effective in grouping samples in units 1, 2, 5, 6, 7,

8, and 9. Only one cluster contained a heterogeneous mix of samples. Apparently

TWINSPAN was efficient in identifying multifactor units classified by Host (1987) that

were differentiated by combinations of soil, substrate, and floristic criteria in the original

classification. The ability of ground-flora communities to serve as phytometers of

important environmental conditions was improved by eliminating species with low

fidelity, or species that are poorly correlated, with the first few dimensions of a CA, and

retaining species most closely associated with underlying environmental gradients

expressed in the first few dimensions of a CA.

1111112

4554555
8059439

0000000
0000000
0000000
0000011

222222132

3 2233513
098938154

000000000
000000000
111111111
000000000

2233444433

145 1514
1316282885

0000000000
0000000000
1111111111
1111111111

34

34
97

00
11
00
00

55555

11216
67040

00000
11111
00000
00011

6666

14
4504

0000
1111
1111

777

445
016

111
000
000
000

888788887

222023335
356711567

111111111
000000000
000000111
111111

9999799

32234 2
2277364

1111111
1111111
0000011

ELU

Sample

Cluster
level 1
level 2
level 3
Level 4

Figure 2.72. TWINSPAN on 56 samples from a homogenous climatic zone in NW MI
based on high quality species.

CUT LEVELS: 0 .00

1.00 2.00 3.00 4.00

155

DISCUSSION

A series of complementary multivariate, graphical, and tabular analyses was
effective in classifying multifactor ELU’s in northeast lower Michigan. Results of
clustering of samples based on species aided in the exploratory partitioning of samples in
species space. The analysis of soils data and comparison of results to ordination and
clustering of samples based on species enabled an interpretation of gradients expressed
along principal coordinates and the separation of samples grouped within heterogeneous
clusters. The elimination of species poorly correlated with low—dimensional sample
subspaces through a geometric interpretation of a CA was effective in conditioning
ground-flora data sets for clustering procedures, although ordination results were not
substantially improved. While clustering results of samples based on ground-flora were
improved in terms of consistency in number of groups identified and assignment of
samples to groups using the high quality ground-flora data set, clustering alone did not
sufficiently identify final ELU’s.

The utility of eliminating ground-flora species poorly correlated with the first five
dimensions of a CA was validated in the analysis of a second independently derived
classification by Host (1987). Samples from a different climatic region were
distinguished in the clustering of 76 samples. Elimination of the samples from a different
climatic region and reanalysis of 56 six samples within a homogeneous climatic zone
enabled TWINSPAN to effectively identify all nine ecological units previously classified
without the use of soils data that was critical to the original classification.

The determination of species with high fidelity to low-dimensional subspaces also

156

facilitated an improved interpretation of ordination diagrams. Species that were
“indicators” of the important environmental factors were included in final ordination
partitions of species in sample space and in ecological species groups.

Ordination space partitioning of samples in the first and third dimensions of
commonly occurring or high quality species space of a CA was effective in identifying all
seven ecological land units classified in this research; these units differed in terms of both
floristic and soil characteristics. Ordination space partitioning of samples in the first and
second dimensions of commonly occurring or high quality species space of a DCA was
also effective in identifying all seven ecological land units classified in this research.
Ordination space partitioning of samples in the first and second dimensions of soil
variable space was also effective in identifying ecological land units. Partitioning of
continuously distributed samples in species or soil variable space required use of
clustering results and tabular comparisons within and between floristic and soil data sets
to locate where regions of ecological units changed. Ordination space partitioning of the
seven regions occupied by ecological land unit clarifications based on either the ground-
flora or soil data sets would not have been possible without the availability and analysis
of both data sets.

CA and DCA were robust to the inclusion of ubiquitously distributed ground-flora
species, and the use of high quality species in ordination procedures had little if any
advantage over use of commonly occurring species for data collected in northeast lower
Michigan, and actually obscured results for DCA the first and second dimensions. Use of

high quality species in DCA for the data set from northwestern Lower Michigan

157

improved the discriminating power of the analysis in the first and second dimension,
however.

Important soil morphologic and physical characteristics were identified using
PCA. The examination of factor weights and factor patterns identified soil variables
primarily controlling in situ moisture and nutrient gradients that differentiated multifactor
ecological land units. Ordinations of samples in soil variable space in the first two
dimensions of a PCA distinguished all ecological land units, with only a few outliers not
grouping within respective classes. The soils data set was not summarized well in low
dimensional subspaces, and patterns in the third and higher dimensions did not assist in
the interpretation of samples in variable space or variables in sample space.

The clustering of samples based on twenty-eight soil variables failed to group
samples according to ecological units as defined in this research. Nonetheless, data on
soil characteristics was essential in formulating ecological units in several regards. First,
tabular inspection of soil variables revealed textural and drainage patterns responsible for
patterns in ground-flora assemblages. Second, soil-plant relationships were used as a
basis for formulating ecological species groups. Third, differences in soil characteristics
were used as a basis for partitioning continuously distributed samples in species space.
Fourth, differences in soil characteristics were used as a basis for subdividing one large
cluster of samples objectively grouped by agglomerative and divisive clustering
techniques. F ifih, the ordination space partitioning of seven groups of samples in soil
variable space that matched the ordination space partitioning of samples in species space

corroborated the classification of samples based on ground-flora.

158

Overstory composition was not a basis of classification in this research, but
nonetheless units had relatively consistent composition of overstory species. These
findings served as a validation of the units derived based on ground-flora and soil

variables.

Management Implications of Ecological Land Classification

The units formulated in this research and the procedures applied should be useful
for management in several respects. These classification units can aid in the development
of map unit legends and in the interpretation of soil-plant relationships for respective
units. ELU’s can also serve as a basis for testing hypotheses in follow up research
regarding key management concerns. Forest management information needs including
rates of forest growth (Host et al. 1988), regeneration (Bruggink 1988), nutrient cycling
(Zak et al 1989, Padley 1989), maintenance of long term soil productivity, and other
interpretations such as wildlife forage values can be studied within the study area based
on this ecological classification framework. Given positive results, findings can be
extrapolated to similar areas based inventories of existing vegetation and ELU’s.

The procedures used in the classification of ELU’s in northeast lower Michigan
can aid scientists and managers involved in multifactor ecological land classification. In
particular, investigators may want to eliminate species poorly correlated with low
dimensional subspaces in agglomerative clustering or TWINSPAN analysis of samples
based on species, and in the interpretation of ordination diagrams of species and species’

synthesis tables. Furthermore, the results of this research show that no single multivariate

159

analysis is sufficient for multifactor ecological land classification. Iterating between
environmental and floristic analyses, and careful, concurrent interpretation of ordination
and clustering results is necessary.

The overall program of ecological land classification by numerous researchers in
Michigan (Padley 1989, Host 1988, Spies and Barnes 1985a, Pregitzer 1981) can
contribute to a larger need of understanding the basic nature and distribution of forest
ecosystems. The results of research addressed in this effort, for example, corroborated
the theory that plant distributions are regulated along energy, moisture, nutrient, and
disturbance gradients. The separation of samples from different climatic zones in
northwest lower Michigan (Albert et al. 1986) for equivalent ELU’s as defined by Host
(1987) based on clustering using high quality species (Figure 2.69) substantiated that
even minor differences in climate can affect the composition or relative abundance of
ground-flora associated with ELU’s.

In this research, a random stratified sampling scheme was used to reduce the
variability in vegetation due to climatic or disturbance effects. Only well stocked,
minimally disturbed late successional forest stands in upland, moderately or more well
drained soils were sampled. This was one of several possible sampling frames that could
have been employed. Openlands, young regenerating forests, wetlands, or other
combinations of forest types and soils conditions could have also comprised the sampling
frame. Opportunities for quantifying relationships and conditions in these various
sampling frames that affect patterns in forest ecosystem structure and function is another

contribution of ecological land classification efforts.

160

Figure 2.73 displays a conceptual model of spatial and temporal sources of
variability affecting forest ecosystem structure and function (Cleland et al. 1994). This
figure shows spatial variations in environment measured at local and regional scales, and
temporal variations in age and succession measured in years and centuries.

Within a local area, particular locations are wetter or drier, or more or less fertile
than other locations because of differences in soil properties or hydrology. Each of these
conditions supports certain assemblages of plants and animals. In Michigan’s forests, this

environmental-biotic gradient could include oak savannas, xeric jack pine, dry-mesic

 

 

 

 

 

(<1 climax
,\\Q mid- .
successuonal
o
L” L. //
.1 -° / old
< //
O mature
(D (D //
C /
:: dry wet YOUHQ

 

 

 

SPACE

Figure 2.73. Environmental and temporal sources of variability in forest
ecosystem structure and function.

161

pine-oak, mesic northern hardwood, and hydric hardwood or conifer communities. Fine
scale ecological land units are designed to classify these general forest type - environment
combinations into more precise classes. At broader spatial scales, temperature and
moisture gradients vary with latitude, elevation and proximity to major bodies of water,
each of which affects macroclimate. These changes at macroscales represent spatial
sources of environmental variability that affect regional, landscape and local ecosystem
structure and function. Meso- and macroscale ecological tmits are designed to classify
and regionalize these regional and landscape level ecological units (Cleland et a1. 1996,
Albert et al. 1986).

Figure 2.73 also shows changes occurring through time. At temporal scales
measured in years, a given ecosystem may be supporting vegetation that is young, mature,
or old growth. These changes at finer time scales represent temporal variations that affect
local ecosystem structure and function. At temporal scales measured in centuries,
ecosystems undergo change such as succession. Successional developments affect the
nature and complexity of food webs, ratios of net primary production to respiration, and
rates of nutrient and carbon cycling (Odum 1969). From a landscape perspective, these
changes form a shifting mosaic of local ecosystems at different successional stages
(Borman and Likens 1979). Changes occurring at these time scales represent temporal
variations affecting landscape and local ecosystem structure and function.

Figure 2.74 displays a CA of 123 samples in dimensions 1 and 3 of commonly

occurring species space for a combined data set from both northwest and northeast lower

162

 

 

 

 

 

1.5
M
M
1 W
M
M u
M
M M M"
M M M
n M
" r
H m "H”
H H
H
H
H
-1.5
-1 -0.5 0 0.5 1 1.5 2
CA 1

Figure 2.74. CA of 123 samples in dimensions 1 and 3 of commonly
occurring species space for a combined data set from northwest and
northeast lower Michigan labeled by National Forest: M denotes Manistee
NF, H denotes Huron NF.

Michigan. Samples grade from xeric to mesic along the first principal coordinate with
increasing sample weights. The third dimension effectively separates samples from these
two different climatic zones, suggesting that subtle differences in species relative
abundance reflect climatic gradients expressed along the third principal coordinate.
Figure 2.75 displays a CA of 88 samples in dimensions 1 and 2 of commonly
occurring species space labeled by forest type. These samples consist of the 46 samples
from northeast lower Michigan used in this research, and an additional 42 samples from
the same study area measured in fully stocked aspen stands. Identical field methods were

used in data collection. Figure 2.75 shows samples grading from xeric to mesic along the

163

first principal coordinate, whereas the second dimension effectively separates samples
from these two different forest types, suggesting that differences in species composition
or relative abundance reflect successional gradients expressed along the second principal
coordinate.

The implications of Figures 2.74 and 2.75 are that species composition and
relative abundance varies along environmental and temporal gradients. Van Groenewoud
(1992) recommended that the only way to get reasonable results using CA or DCA is to
restrict the sampling mainly to one gradient, or pre-stratify the samples before analysis to
represent mainly one gradient at a time. These preliminary analyses reported for

discussion purposes support this recommendation.

 

 

 

 

2.5
2
1.5 1'
a“,
N I g H
V I" ”I
D a
0.5 a .4
A in a
A r I r . a
A Am A
0 44““ .4 H u
M H
‘4‘ A 4‘ ”a H
‘ ‘1'!
-0.5 A I
A A H
-1
-1 -0.5 0 0.5 I 1.5 2 2.5
CA 1

Figure 2.75. 88 samples in CA dimensions 1 and 2 of commonly
occurring species space labeled by forest type: H denotes
hardwoods and A denotes aspen.

164

The are several management implications of these findings. First, ecological land
units, ecological species groups, or habitat types should be nested within and defined for
different climatic zones or regions. Second, assemblages of species used as indicators of
environmental conditions may need to be developed for early as well as late successional
forest types. Third, extrapolation of ecological land classifications outside of the regions
in which they were developed is likely unsound without verification and adjustment to

local conditions.

CHAPTER III
SUMMARY AND CONCLUSIONS

The use of a series of complementary ordination, clustering, graphical, and tabular
analyses of floristic and soils data was effective in defining seven multifactor ecological
land units for northeastern lower Michigan. Any single analysis of any single factor
would not have allowed the development of an ecologically meaningful classification of
samples or species. Iterating between ordination and clustering results aided in objective
exploratory partitioning of samples incommonly occurring and high quality species
space. Ordination of samples in soil variable space corroborated ordination space
partitioning of samples in species space, and identified important soil variables
subsequently used in the interpretation of ordination and clustering of samples based on
ground-flora. Reanalysis of ground-flora data composed of species with high correlation
to the first five dimensions of a CA further refined the classification. Graphical and
tabular methods and comparisons facilitated the interpretation of mutual changes in
ground-flora and soil conditions, and these methods were essential in the development of
the final classification.

Seven Ecological Land Units were classified. ELU 1 consists of the sugar maple-
basswood-blue cohosh plant association on moderately well-drained sandy loams over
sandy clay loam substratums on moraines and ground moraines; some pedons have loamy
sand caps. The ground-flora of sites sampled for ELU 1 consists of Caulophyllum
thalictroides, Dryopteris spinulosa, Allium tricoccum, Carex deweyana, and Carex

annectens. These species form the Caulophyllum species group. The overstories of sites

165

166

sampled for ELU 1 are dominantly composed of sugar maple, basswood, white ash, and
beech.

ELU 2 consists of the red oak-sugar maple-viola plant association on well-drained
sandy loams over sandy clay loam substratums on moraines and ground moraines; some
pedons have loamy sand caps. The ground-flora of sites sampled for ELU 2 consists of
Viola species including Viola pubescens and Viola canadensis, Prenanthes alba, Ribes
cynobasti, Solidago hispida, and a diverse assemblage of tree seedlings including Acer
saccarum, T ilia americana, F agus grandifolia, and Ostrya virginiana. These species
form the Viola species group. The overstories of sites sampled for ELU 2 are composed
of red oak, sugar maple, basswood, white ash, red maple, beech, paper birch, and
ironwood.

ELU 3 consists of the red oak-basswood- miterwort plant association on
well-drained sandy loams over silty clay loam substratums on moraines and ground
moraines; some pedons have loamy sand caps. The ground-flora of sites sampled for
ELU 3 consists of Mitella diphylla, Osmorhiza chilensis, Athyriumfelix-femina, Hepatica
acutiloba, Carex plantaginea, Actaea rubra, Galium triflorum, and T ilia americana .
These species are form the Mitella species group. The overstories of sites sampled for
ELU 3 are composed of red oak, basswood, sugar maple, paper birch, white ash, red
maple, ironwood and beech.

ELU 4 consists of the red oak-red maple-bellwort plant association on well
drained sands with sandy B8 to coarse to fine loamy B2t horizons overlying fine loamy

substratums on ice-contact topography. The ground-flora of sites sampled for ELU 4

167

consists of Uvularia perfoliata, Trillium grandiflorum, Viola Spp., Fraxinus americana,
and Prenanthes alba. These species form the Uvularia species group. The ground-flora
of ELU 4 contains elements of both northern hardwood and oak ecosystems, including
species within the Uvularia species group, and Viburnum acerifolium, Hamamelis
virginiana, and Vaccinium angustifolium. The overstories of sites sampled for ELU 4 are
composed of red oak, red maple, white oak, paper birch, and white ash.

ELU 5 consists of the red oak-red maple-vibumum plant association on well
drained sands with sandy Bs overlying coarse to fine loamy substratums on ice-contact
topography. The ground-flora of sites sampled for ELU 5 consists of Viburnum
acerifolium, Hamamelis virginiana, Sanicula trifoliata, Cretaegus spp., Amelanchier
Spp.,and Oryzopsis asperifolia. These species form the Viburnum species group. The
overstories of sites sampled for ELU 5 are composed of red oak, white oak, black oak,
and red maple.

ELU 6 consists of the mixed oak-red maple-big leaf aster plant association on
excessively well-drained sands of ice contact topography. The ground-flora of sites
sampled for ELU 6 consists of Aster macrophyllum, Quercus rubra, Lonicera
canadense, and Vaccinium angustifolium. These species form the Aster macrophyllum
species group. The overstories of reference sites sampled for ELU 6 are composed of red,
black, and white oak, and red maples.

ELU 7 consists of the northern pin oak-white oak-bearberry plant association on
excessively well-drained sands of outwash plains. The ground-flora of sites sampled for

ELU 7 consists of Arctostaphylos uva-ursi, C ladina spp., Comptonia peregrina,

168

Apocynum androsaemifoliumspr, Melampyrum lineare, Dicranum polysetum,
Gaylussacia baccata, and Quercus velutina. These species form the Arctostaphylos
species group. The overstories of reference sites sampled for ELU 7 are composed of
upland pin, black, and white oak.

Two secondary goals of this research met while classifying multifactor ecological
land units were: (i) the documentation of a series of complementary multivariate, tabular,
and graphical analyses of ground-flora, soils, and overstory data in the classification
process, and (ii) the determination if the classification of samples and species into
multifactor ecological land units and ecological species groups could be improved by
eliminating ubiquitously distributed or superfluous ground-flora species prior to
application of several ordination and clustering methods.

The display of results at each step of the classification process and in subsequent
discussions provided documentation of the sequence of analyses of floristic and soil data
sets. Objective numerical analyses used in conjunction with a minimum of subjective
interpretation of graphical and tabular results were effective in developing the
classification. Clustering results used in initial exploratory ordination space partitioning
of samples in species and soil variable space provided a means of objectively subdividing
continuous distributions of samples in ordination space. Interpretation of subsequent
analyses by comparing initial results, diagrams, and tables were useful in deriving the
final classification. A final set of synthesis tables was used to derive an integrated
classification of samples and species. Results obtained could thus be repeated by other

investigators with a minimum of subjective decisions affecting the final classification.

169

Summary of Vegetative Analyses in Northeast Lower Michigan - Ordination

The first dimension of a correspondence analysis of ground-flora ordinated
samples along a moisture and nutrient gradient. ELU’s graded from mesic to xeric along
the first axis as sample weights graded from negative to positive. ELU 1 consisted of
moderately well drained sandy loam solas to depths of 100 cm. overlying sandy clay loam
substrates. ELU 2 consisted of well drained sandy loam solas to depths of 30 cm.
overlying sandy clay loam substrates. ELU 3 consisted of sandy loam solas to depths of
30 cm. overlying silty clay loam substrates. ELU 4 consisted of shallow sandy caps
overlying fine loamy substrates. ELU 5 consisted of deep sandy caps overlying fine
loamy substrates at depths greater than 100 cm. ELU 6 consisted of deep sandy soils with
moderate spodic horizon development. ELU 7 consisted of deep sandy soils lacking
spodic horizon development.

Interpretation of the second principal coordinate of a CA of ground-flora was less
straightforward than interpretation of the first principal coordinate. It appears that the
second dimension separated samples based on a successional stability gradient. The idea
of a successional gradient along this axis was corroborated by interpretation of the
squared cosines of samples in species space. Samples classified as ecological land units
1, 6, and 7 were most strongly correlated with the first axis, whereas both ecological land
units 4 and 7 were more closely associated with axis 2. Ecological land unit 4 consists of
ground-flora typifying both mesic northern hardwood and dry-mesic and xeric oak flora.
Based on ground-flora composition and soil textures, this system has the potential to

succeed to mesophilic sugar maple-white ash ecosystems, or dry-mesic oak-red maple

170

ecosystems, depending on natural or anthropogenic disturbance regimes.

CA and DCA were both effective in distinguishing ecological land units in
commonly occurring and high quality species space for ecological units classified in
northeastern lower Michigan. Both CA and DCA were robust to the inclusion of
ubiquitously distributed ground-flora species, and the use of high quality ground flora in
ordination procedures had little if any advantage over use of commonly occurring species.

For all data sets analyzed, detrending removed the curved configuration of the
first and second dimensions of CA, but CA revealed the same patterns in the first and
third dimensions as DCA revealed in the first and second. The implication is that either
method may be used to discern underlying structure of vegetative data sets, but
investigators need to interpret the first and third as opposed to the first and second
dimensions of CA to arrive at the same results as DCA. Apparently information on
secondary gradients expressed in the second dimension of a DCA is deferred to higher
axes in CA.

A benefit to using CA instead of DCA lies in the geometric interpretation of a CA.
The fidelity of a species to reduced subspaces can be quantified and interpreted in
ordination diagrams, or used to condition data for subsequent clustering procedures by
eliminating ubiquitously or spuriously distributed individuals based on species sums of

squared cosines in respective subspaces.

Summary of Vegetative Analyses in Northeast Lower Michigan - Clustering

Agglomerative clustering of commonly occurring ground-flora using three linkage

171

methods resulted in relatively consistent classification of sample groups, ordering of
groups, and assignment of samples to respective groups for each of the three
agglomerative clustering procedures. Five of seven ecological units were distinguished
by clustering of commonly occurring species.

Agglomerative clustering of samples based on high quality produced more
consistent results than clustering of commonly occurring species in terms of clusters
defined and sample assignment to clusters among all clustering procedures. Six of seven
ELU’s defined in the final classification were distinguished by these clusters. None of
the agglomerative procedures for either data set were effective in distinguishing between
ecological units 5 and 6. The consistent recovery of the same cluster structure by three
different agglomerative linkage methods and divisive clustering indicated that the derived
classifications were valid. The interpretation of these groupings and final classification
of samples and species was made by tabularly examining patterns in soil characteristics
and vegetation.

Classification of species using agglomerative clustering procedures produced
inconsistent results. Agglomerative classification of a reduced data set for which species
were eliminated based on interpretation of the TWINSPAN synthesis table also failed to
identify consistent clusters. Consequently agglomerative clustering was not used to
identify ecological species groups.

Classification of species by TWINSPAN identified eight species groups and two
outliers. Groups consisted of species assemblages grading from highly mesophilic

species typifying northern hardwood ecosystems to pyrophilic xeric species typifying oak

172

and pine-oak ecosystems. The final seven ecological species were defined by interpreting
TWINSPAN clustering, ordination, and tabular results of ground-flora in conjunction

with interpretations of analyses of soils data.

Summary of Analysis of Soils Data in Northeast Lower Michigan - Ordination

Principal component analysis of twenty-eight soil variables identified important
soil characteristics that differentiated multifactor ecological land units and also produced
ordination diagrams that grouped samples into meaningful classes of soil, ground-flora,
and overstory conditions. Ordinations of samples in soil variable space in the first two
dimensions of a PCA distinguished all ecological land units, with only a few outliers not
grouping within respective classes. With the exceptions of ELU’s 6 and 7, ordinations of
samples in variable space in the first and third dimensions failed to group ELU’s,
however. The soils data set was not summarized well in low dimensional subspaces, and
patterns in the third and higher dimensions did not assist in the interpretation of samples
in variable space or variables in sample space.

Important variables based on high positive first dimensional weights included soil
textures in substrata, then textures in sola. Important variables based on high negative
first dimensional weights included variables describing depths to textural discontinuities,
thickness of sandy Bs horizons, and depths to mottles and watertables. Important
variables based on high positive second dimensional weights included depths to mottles,
dominant texture of sola and substrates, soil textures in substrata, and depth to

watertables. Important variables based on high negative second dimensional weights

173

included drainage class codes, textures of the top 30 cm., a coded variable describing
depths to textural discontinuities, and textures of B horizons.

The most important soil variables based on high positive first dimensional factor
loadings, or variable correlations with variates derived in the first principal component,
included a coded variable describing substrate textures, textures between 100 and 150
cm., and sola textures. The most important soil variables based on high negative first
dimensional factor loadings included depth to heavy textured horizons, and depths to
mottles and water tables. The most important soil variables based on high positive
second dimensional factor loadings included depths to mottles, dominant texture of sola
and substrates, textures between 100 and 250 cm., and depth to watertables. The most
important soil variables based on high negative second dimensional factor loadings
included drainage class code, texture of the top 30 cm., and a coded variable describing
substrata textures.

The first PCA axis ordinated samples along a textural and moisture gradient.
Textures of the upper 30 cm. and 100 cm. graded from sands to sandy loams with
increasing sample weights along the first axis. Textures of the lower 150 cm. graded
from sands to sandy clay loam with increasing sample weights along the first axis. Soil
drainage graded from excessively well drained to well drained with increasing sample
weights along the first axis.

The second PCA axis ordinated samples along a substrate texture and drainage
class gradient. Silt content of the soil horizons at depths of 100 to 250 cm. increased with

increasing sample weights along the second axis . Soil drainage graded from well drained

174

to moderately well drained with decreasing sample weights along the second axis.

Summary of Analysis of Soils Data in Northeast Lower Michigan - Clustering
Agglomerative clustering of samples based on twenty-eight soil variables using
Ward’s and average linkage methods failed to identify the same groups as ordination
space partitioning of samples in soil variable space, or as the ordination and clustering of
samples based on ground-flora. Interpretation of clustering results indicate that soil
groupings were highly variable, and simply included all possible combinations of soil
textures in sola and substrata. Results were not corroborated by ordination, graphical,
and tabular analyses of soils or ground-flora data, or clustering of ground-flora. Results

were therefore discarded from firrther consideration in formulating ELU’s.

Summary of Final Classifications of Ecological Land Units in Northeast Lower Michigan

Seven ecological land classification units (ELU’s) and ecological species groups
were defined based combinations of ground-flora, soils, and overstory conditions.
Comparisons of soils data to clusters of samples defined by high quality species revealed
two distinctly different soil-substrata conditions within one cluster that included samples
consisting of deep sands and sands underlain by fine loamy substrates. This cluster was
subdivided into two ecological land units as a consequence. The remaining groupings of
samples based on high quality ground-flora species showed consistent patterns in soil
characteristics.

The clustering of samples based on twenty-eight soil variables failed to group

175

samples according to ecological units. Nonetheless, data on soil characteristics was
essential in formulating ecological units in several regards. First, tabular inspection of
soil variables revealed textural and drainage patterns responsible for patterns in ground-
flora assemblages. Second, soil-plant relationships were used as a basis for formulating
ecological species groups. Third, differences in soil characteristics were used as a basis
for partitioning continuously distributed samples in species space. Fourth, differences in
soil characteristics were used as a basis for subdividing one large cluster of samples
objectively grouped by agglomerative and divisive clustering techniques. Fifth, the
ordination space partitioning of seven groups of samples in soil variable space that
matched the ordination space partitioning of samples in species space corroborated the
classification of samples based on ground-flora.

Overstory composition was not a basis of classification in this research, but
nonetheless units had relatively consistent composition of overstory species.
Examination of overstory composition within combined soil-ground-flora units
corroborated the meaningful formulation of seven ecological land units based on
congruent changes in ground-flora and soil variables, and served as a validation of the

ecological significance of these units.

Summary of Use of High Quality Species in the Classification of Ecological Land Units
in Northwestern Lower Michigan
The utility of eliminating ground-flora species poorly correlated with the first five

dimensions of a CA in cluster analysis was corroborated in the analysis of a second

176

independently derived classification by Host (1987). CA and DCA were both effective in
distinguishing ecological land units in commonly occurring and high quality species
space. CA was robust to the inclusion of ubiquitously distributed ground-flora species,
and the use of high quality ground flora in ordination procedures had little if any
advantage over use of commonly occurring species. Use of high quality species in DCA
obscured results in the first and second dimensions. DCA was as effective as CA in
distinguishing units in the first and third dimensions of high quality species space.

TWINSPAN clustering of samples based on high quality ground-flora species
improved the consistency of groups derived and assignment of samples to groups over
clustering based on commonly occurring species. TWINSPAN clustering of 76 samples
based on high quality species effectively grouped samples into six of nine units defined
by Host (1987).

Agglomerative and divisive clustering of 76 samples separated samples from two
different climatic regions (Albert et al. 1986). Elimination of these samples and
reanalysis of 56 samples within a homogeneous climatic zone enabled divisive
(TWINSPAN) clustering to effectively identify all nine ecological units previously
classified by Host (1097). This was significant because soils data that was critical to the
original classification was not used in the reanalysis, indicating that high quality species
were effective phytometers of key soil variables.

Agglomerative clustering of 76 samples based on commonly occurring and high
quality species was only effective in grouping samples into four of nine units defined by

Host (1987). Agglomerative clustering of 56 samples within a climatically homogeneous

177

area based on high quality ground-flora species was only effective in grouping four of
nine units defined by Host (1987).

These results support the use of a series of complementary methods to effectively
classify multifactor ecological land units. In contrast to clustering analyses of data sets
from northeastern lower Michigan, agglomerative clustering failed to produce satisfactory
results in northwestern lower Michigan. A full analysis of ground-flora and soils data
using a series of ordination, clustering, and tabular methods iteratively for northwestern
lower Michigan would have revealed these inconsistencies, enabling selection and
interpretation of the most appropriate ordination and clustering procedures. Furthermore,
selection of key differentiating species strongly associated with environmental gradients
would have been improved if species sums of squared cosines in low dimensional
subspaces had been determined and species with low fidelity eliminated from

consideration for both clustering analyses and inclusion within ecological species groups.

LIST OF REFERENCES

178

LIST OF REFERENCES

Abrams, M.D., D.G. Sprugel, and D.I.Dickman. 1985. Multiple successional pathways
on recently disturbed jack pine sites in Michigan. For. Ecol. And Mgrnt. 10:31-48.

Albert, D.A., S.R. Denton, and B.V. Barnes. 1986. Regional landscape ecosystems of
Michigan. School of Natural Resources, U. Michigan, Ann Arbor. 32 pp.

Allen, RB, and R.K. Peet. 1990. Gradient Analysis of forests of the Sangre de Cristo
Range, Colorado. Can. J. Bot. 68:193-201.

Allen, R.B., R.K. Peet, and W. L. Baker. 1991. Gradient analysis of latitudinal variation
in Southern Rocky Mountain Forests. J. Biogeogr. 18:123-39.

Archambault, L., B.V. Barnes, and J .A Witter. 1989. Ecological species groups of oak
ecosystems of southeastern Michigan. For. Sci., 35:1058-1074.

Arno, S.F., D.G Simmerman, and RE. Keane. 1986. Characterizing succession within a
forest habitat type - an approach designed for resource managers. Res. Note INT-357.
Ogden, UT: USDA. Forest Service, Intermountain Forest Experiment Station. 8 p.

Avers, RE, and BF. Schlatterer. 1991. Ecosystem Classification and Management on
National Forests. In Proceedings of the 1991 Symposium on Systems Analysis in Forest
Resources.

Backus, I. 1993. Ecotone versus ecocline: vegetation zonation and dynamics around a
small reservoir in Tanzania. J. Biogeogr. 20:209-18.

Bailey, R.G. 1987. Suggested hierarchy of criteria for multi-scale ecosystem mapping.
Landscape and Urban Planning 14:313-319.

Bailey, R.G. 1988. Problems with using overlay mapping for planning and their
implications for geographic information systems. Env. Manage. 12:11-17.

Bailey, R.G. 1991. Design of ecological networks for monitoring global change. Envim.
Conserv. 18: 173-176.

179

Barnes, B.V. 1984. Forest ecosystem classification and mapping in Baden -
Wurttemberg, West Germany. In: Forest Land Classification: Experience, Problems,
Perspectives. Proceedings of the Symposium, Madison, WI, March 18-20. pp. 49-65.

Barnes, B.V., K.S. Pregitzer, T.A. Spies, and V.H. Spooner. 1982. Ecological forest site
classification. J. For. 80:493-498.

Barnes, B.V., K.S. Pregitzer, T.A. Spies, and V.H. Spooner. 1982. Ecological forest site
classification. J. For. 80:493-498.

Barton, A.W. 1994. Gradient analysis of relationships among fire, environment, and
vegetation in a southwestern USA mountain range. Bull. Torr. Bot. Club 121:251-65.

Bekele, T. 1994. Phytosociology and ecology of a humid Afromontane forest on the
Central Plateau of Ethiopia. J. Veg. Sci. 5:87-98.

Birks, H.J.B., S.M. Peglar, and HA. Austin. 1994. An annotated bibliography of
canonical correspondence analysis and related constrained ordination methods
1986-1993. Botanical Institute, University of Bergen, Bergen, Norway.

Boeye, D., and RF. Verheyen. 1994. The relation between vegetation and soil chemistry
gradients in a ground water discharge fen. J. Veg. Sci. 5:553-60.

Braun-Blanquet, J. 1932. Plant Sociology: the study of plant communities (authorized
English translation of Pflanzensoziologie, translated, revised, and edited by G. D. Fuller
and H. S. Conrad). McGraw-Hill, New York, 439 pp.

Bray, J .R. and J .T. Curtis. 1957. An ordination of the upland forest communities of
southern Wisconsin. Ecol. Mono., 50:131-151.

Brenner, RN. and J .K. Jordan. 1991. The Role of an Ecological Classification System in
Forest Plan Development and Implementation. In: Proceedings of the 1991 Symposium
on Systems Analysis in Forest Resources, 1991, Charleston, SC. March 3-6.

Bruggink, J .L. 1988. Oak stump sprouting potential on outwash ecosystems of northern
lower Michigan. MS. Thesis. Dept of Forestry, Michigan State Univ. 104 pp.

Cajander, AK. 1926. The theory of forest types. Society for the Finnish Literary. 108
PP-

Carmean, W.H. 1979. Site index comparisons among northern hardwoods in northern
Wisconsin and Upper Michigan. Res. Pap. NC-169. St. Paul, MN: USDA. Forest
Service, North Central Forest Experiment Station. 17 p.

180

Causton, DR. 1988. An introduction to vegetation analysis: principles, practice, and
interpretation. Unwin Hyman, London. 342 pp.

Cleland, D.T, P.E. Avers, H.W. McNab, W.E. Russell, T.King, C.C. Jensen. 1996. The
National Hierarchy of Ecological Units. In: Ecosystem Management In Natural
Resources. Ed: Haney and Boyce. Yale Press. 12 pp.

Cleland, D.T., T.R.Crow, J .B. Hart, E.A. Padley. 1994. Resource management
perspective: remote sensing and GIS support for defining, mapping, and managing forest
ecosystems: 243-264. In: V.A. Sample (ed.) Remote Sensing and GIS in Ecosystem
Management. American Forests, Forest Policy Center. Island Press. Washington, DC.
369 p.

Cleland, D.T., T.R. Crow, P.E. Avers, and J .R. Probst. 1992. Principles of Land
Stratification for Delineating Ecosystems. In: Proceedings of Taking an Ecological
Approach to Management National Workshop.

Cleland, D.T., J .B. Hart, K.S. Pregitzer, and CW. Ramm. 1985. Classifying oak
ecosystems for management. In: (J .E. Johnson, ed.) Proceedings of Challenges for Oak

Management and Utilization, March 28-29, University of Wisconsin, Madison. pp.
120-134.

Coffman, M. S., E. Alyanak, J. Kotar, and J .E. Penis, 1983. Habitat classification field
guide. 2nd printing. Cooperative for Research on Forest Soils, Michigan Technological

University, Houghton, MI, 62 pp.

Chatfield, C., and A.J. Collins. 1980. Introduction to multivariate analysis. Chapman
and Hall, Ltd., London. 246 pp.

Coffey, G.N. 1912. US. Dept. Agr. Bur. Soils Bull. No. 85, p. 7-40.

Cook, EC. 1996. Implications of modern successional theory for habitat typing: a
review. For. Sci 42(1). pp. 75-75.

Cooper, S. V., K.E. Neiman, R. Steele, and D.W. Roberts. 1987. Forest habitat types of
northern Idaho: A second approximation. USDA For. Serv. Gen. Tech. Rep. INT-236,
135 pp

Corliss, J .C. 1974. ECOCLASS--A method for classifying ecosystems. P. 264-271 in
Foresters in Land-Use Planning. Proc. 1973 Nat. Convention Soc. Am. For., Washington,
DC.

181

Curtis, J .T. 1959. The vegetation of Wisconsin. Madison, WI: University of Wisconsin
Press.

Daubenmire, R. and J .B. Daubenmire. 1968. Forest vegetation of eastern Washington
and Northern Idaho. Washington State University Technical Bulletin 60, Washington
Agricultural Experiment Station, 104 pp.

Delcourt, H.R., T.A. Delcourt, and T. Webb. 1983. Dynamic plant ecology: the spectrum
of vegetative change in space and time. Quant. Sci. Rev. 1:153-75.

Denton, DR. and B.V. Barnes. 1988. An ecological climatic classification of Michigan:
a quantitative approach. For. Sci. 34:119-138.

Digby, P.G.N., and RA. Kempton. 1987. Multivariate Analysis of Ecological
Communities. Chapman and Hall, Ltd. London.

Dokuchaev, V.V. 1893. About the question of overvaluation of lands of Europe and
Asiatic Russia. Collected Works, vol. Vi, Acad. Sci., U.S.S.R., 1953.

Driscoll, R.S.; and others. 1984. An ecological land classification framework for the
United States. USDA Forest Service. Miscel. Pub. 1439. Washington, DC. 56 p.

Drude, O. 1890. Handbuch der Pflanzengeogrphie. Engelmann, Stuttgart.

Dzwonko, Z. 1993. Relations between the floristic composition of isolated young woods
and their proximity to ancient woodlands. J. Veg. Sci. 4:693-8.

Eyre, RH. 1980. Forest cover types of the United States and Canada. Society of American
Foresters. Washington, DC. 148 p.

F enneman, NM. 1938. Physiography of eastern United States. McGraw-Hill Book
Company. 714 p.

F em ndez-Palacios, J. M., and J. P. de Nicol s. 1995. Altitudinal pattern of vegetation
variation on Tenerife. J. Veg. Sci. 6:183-90.

Formann, R.T.t. and M. Godron. 1986. Landscape Ecology. Wiley, New York.

Foster, D. R. 1992. Land-use history (1730-1990) and vegetation dynamics in central
New England, USA. J. Ecol. 80:753-72.

Fowlkes, E.B, R. Gnanadesikan, and J .R. Kettenring. 1988. Variable Selection in
Clustering. J .of Classification 5:205-228.

182

Franklin, S. B., P. A. Robertson, J. S. Fralish, and S. M. Ketler. 1993. Overstory
vegetation and successional trends of Land Between the Lakes, USA. J. Veg. Sci.
4:509-20.

Franklin, J ., and M. Merlin. 1992. Species-environment patterns of forest vegetation on
the uplifted reef limestone of Atiu, Mangaia, Ma'uke and Miti'aro, Cook Islands. J. Veg.
Sci. 3:3-14.

Gallant, A.L.; T.R. Whittier, D.P. Larsen, J .M. Omernik, R.M. Hughes. 1989.
Regionalization as a tool for managing environmental resources. US. Environmental
Protection Agency. EPA/600/3-89/060. Corvallis, OR. 152 p.

Gauch, HG. 1982. Multivariate analysis in community ecology. Cambridge University
Press, London and New York. 574 p.

Gilliam, F.S., N.L. Turrill, M.B. Adams. Herbaceous-layer and overstory species in
clear-cut and mature central Appalachian hardwood forests. Ecological Applications:
5(4). pp 947-955.

Greaney, P. J ., 1987. Patterns of tree height growth in upland forests of northern Lower
Michigan. MS. Thesis, Department of Forestry, Michigan State University, East
Lansing, MI, 104 pp.

Greenacre, M.J. 1993. Correspondence analysis in practice. Academic Press, London.

Greenacre, M.J. and E.S.Vrba. 1984. Graphical display and interpretation of antelope
data in African wildlife areas using correspondence analysis.

Greenacre, M.J. 1984. Theory and Applications of Correspondence Analysis. Academic
Press, London.

Grigal, DP. and HF. Ameman. 1970. Quantitative relationships among vegetation and
soil classifications from northeastern Minnesota. Can. Jor. Of Bot. 48:555-566.

Grigal, DP. 1984. Shortcomings of soil surveys for forest management. IN: Forest Land
Classification: Experience, Problems, Perspectives. Proceedings of the Symposium,
Madison, WI, March 18-20. pp. 148-166.

Gysel, Leslie W., and F. Stearns. 1968. Deer browse production of oak stands in central
lower Michigan. USDA Forest Service Res. Note NC-48.

Haase, R. 1990. Community composition and soil properties in northern Bolivian savanna
vegetation. J. Veg. Sci. 1:345-52.

183

Hahn, J .T. 1984. Tree volume and biomass equations for the Lake States. Res. Pap.
NC-250. St. Paul, MN: USDA. Forest Service, North Central Forest Experiment
Station. 10 p.

Hahn, J .T., and W.H. Carmean. 1982. Lake States Site Index Curves Formulated. Gen.
Tech. Rep. NC-88. St. Paul, MN: USDA. Forest Service North Central Forest
Experiment Station. 5 p.

Hannah, P. R. and R. Zahner. 1970. Nonpedogenetic texture bands in outwash sands of
Michigan: their origin, and influence on tree growth. Soil Sci. Soc. Amer. J.
34:134-136.

Hart, J .B. 1988. Ecological classification system: Basic concepts, ecosystems vs
ecological land units, and statistical comparison of different classification systems. m:
Proceedings of USDA FS Eastern Region "Ecological Classifiication System Workshop",
May 11-15, 1987. Traverse City, MI.

Hauser, M., and L. Mucina. 1991. Spatial interpolation methods for interpretation of
ordination diagrams. Coenoses 6: l 5 1-68.

Hill, M.O., and HG. Gauch. 1980. Detrended correspondence analysis: an improved
ordination technique. Vegetatio 42:47-58.

Hill, MO. 1979. TWINSPAN - a FORTRAN program for arranging multivariate data
on an ordered two-way table by classification of the individuals and attributes. Section
of Ecology and Systematics, Cornell University, Ithaca, New York.

Hills, GA. 1952. The classification and evaluation of site for forestry. Ontario Dep.
Lands and Forests. Res. Div. Rep. 24.

Hix, D.M. 1988. Multifactor classification of upland hardwood forest ecosystems of the
Kickapoo River watershed, southwestern Wisconsin. Can. J. For. Res. 18: 1405-1415.

Host, G.E.; P.L. Polzer, D.J. Mladenoff, M.A. White, T.R. Crow. 1996. A quantitative
approach to developing regional ecosystem classifications. Ecological Applications.
6(2):608-618.

Host, G.E., C.W. Ramm, E.A. Padley, K.S. Pregitzer, J .B. Hart, and D.T. Cleland. 1993.
Field sampling and data analysis methods for development of ecological land
classifications: an application on the Manistee National Forest. USDA Forest Service,
North Central Forest Experiment Station General Technical Report NC-l62. 47 p.

184

Host, GE. and K.S. Pregitzer. 1992. Geomorphic influences on ground-flora and
overstory composition in upland forests of northwestern lower Michigan. 1992. Can. J.
For. Res. 22:1547-1555.

Host, G. E. and K.S. Pregitzer. 1991. Ecological species groups for upland forest
ecosystems of northwestern Lower Michigan. Forest Ecology and Management
43:87-102.

Host, G.E., K.S. Pregitzer, C.W. Ramm, D.T. Lusch, and D.T. Cleland 1988. Variations
in overstory biomass among glacial landforms and ecological land units in northwestern
Lower Michigan. Can. J. For. Res. 18:659-668.

Host, GE. 1987. Spatial patterns of forest composition, successional pathways and
biomass production among landscape ecosystems of northwestern Lower Michigan.
Ph.D. Dissertation, Michigan State University, East Lansing, MI. 202 pp.

Host, G.E., K.S. Pregitzer, C.W. Ramm, J .B. Hart, and D.T. Cleland. 1987. Landforrn
mediated differences in successional pathways among upland forest ecosystems in
northwestern Lower Michigan. For. Sci. 33:445-457.

Jensen, M.C., C. McMicoll, and M. Prather. 1991. Application of ecological
classification to environmnental effects analysis. J. Env. Qual. 20:24-30.

John, E., and M. R. T. Dale. 1990. Environmental correlates of species distributions in
saxicolous lichen community. J. Veg. Sci. 1:385-92.

Johnson, RS. 1992. Oak overstory/reproduction relations in two xeric ecosystems in
Michigan. For. Eco. and Mgmt. 48 (1992) 233-240.

Jones, R.K. and others. 1983. Field guide to forest ecosystem classification for the clay
belt, site region 3e. Ministry of Natural Resources. 123p. Ontario, Canada.

Jones, R.K. 1994. Site classification: its role in predicting forestland responses to
management practices. pp. 1 87-218 In: Impacts of Forest Harvesting on Long-terrn Site
Productivity. Edited by W.J. Dyck, D.W. Cole, and NB. Comerford. Chapman and Hall,
London, England.

Jongman, R. H. G., C. J. F. ter Braak, and O. F. R. van Tongeren, editors. 1987. Data
Analysis in Community and Landscape Ecology. Pudoc, Wageningen, The Netherlands.

Jordan, J .K. 1982. Application of an integrated land classification. In: Proceedings,
Artificial Regeneration of Conifers in the Upper Lakes Region. October 26-28, 1982.
Green Bay, WI. pp. 65-82.

185

Kent, M., and P. Coker. 1992. Vegetation description and analysis: a practical approach.
Belhaven Press, London.

Kemer, A. Von Marilaun. 1895. The natural history of plants, their formed, growth,
reproduction, and distribution. Blackie, London.

Kotar, J. J .A. Kovach, and CT. Locey. 1988. Field guide to forest habitat types of
northern Wisconsin. UW-Madison and Wise. Dep. Nat. Resour.

Koppen, J .M. 1931. Grundriss der Klirnakunde. Walter de Grayter; Berlin, 388 pp.

Krzanowski, W.J. 1988. Principles of multivariate analysis: a user's perspective.
Clarendon Press, Oxford.

Kuchler, A.W. 1964. Potential natural vegetation of the conterrninous United States.
Am. Geog. Soc. Spec. Pub. 36. ll6pp.

Liebart, L., A. Morineau and K. M. Warwick. 1984. Multivariate descriptive statistical
analysis: Correspondence analysis and related techniques for large matrices. John Wiley
and Sons, New York. 231 pp.

Major, J. 1969. Historical development of the ecosystem concept. In: (G.M. Van Dyne,
ed.) The Ecosystem Concept in Natural Resource Management. Academic Press, New
York. pp. 9-22.

Marquis, DA. and R. Brenneman. 1981. The impact of deer on forest vegetation in
Pennsylvania. USDA F S Gen. Tech. Rep. NE-65. pp. 1-7.

McIntosh, RP. 1976. Ecology since 1900. pp. 353-372. In B.J. Taylor (ed.) Issues and
Ideas in America. University of Oklahoma Press, Norman. 380 p.

McIntosh, RP. 1978. (Ed.) Phytosociology. Benchmark papers in ecology, Volume 6.
Academic Press, Stroudsburg, PA. 388 p.

McIntyre, S., and S. Lavorel. 1994. How environmental and disturbance factors influence
species composition in temperate Australian grasslands. J. Veg. Sci. 5:373-84.

McNab, W.H. 1987. Rationale for a multifactor forest site classification system for the

southern Appalachians. In: Proceedings of 6th Central Hardwood Forest Conference. Feb
24-26. Knoxville, TN. p. 283-294.

186

Meier, A.J., S.P. Bratton, and DC. Duffy. 1995. Possible ecological mechanisms for
loss of vemal-herb diversity in logged eastern deciduous forests. Ecological
Applications: 5(4) pp 93 5-946.

Milligan, G.W. and MC. Cooper. 1988. A Study of Standardization of Variables in
Cluster Analysis. Jor. of Classification. 52181-204.

Minchin, PR. 1987. An evaluation of the relative robustness of techniques for
ecological ordination. Vegetatio 69:89-107.

Moon, DE. 1984. Forest land resources inventory in British Columbia. IN: Forest Land
Classification: Experience, Problems, Perspectives. Proceedings of the Symposium,
Madison, WI, March 18-20. pp. 66-81.

Morrison, DP. 1976. Multivariate statistical methods. McGraw-Hill, USA. 415 pp.

Morzuch, B.J. and GA. Ruark. 1991. Principal components regression to mitigate the
effects of multicollinearity. For. Sci. 37:191-199.

Mueller-Dombois, D., Ellenberg, H. 1974. Aims and methods of vegetation ecology.
John Wiley and Sons, New York. 547 p.

Newcomer, J .A. 1984. Principal components analysis of total tree form in seven central
Pennsylvania hardwood species. Forest Sci. 30: 64-70.

Omernik, J .M. 1987. Ecoregions of the conterrninous United States. Annals of the
Association of American Geographers 77:118-125.

O'Neill, R.V., D.L. DeAngelis, J .B. Waide, and T.F.H. Allen. 1986. A hierarchical
concept of ecosystems. Princeton University Press, Princeton, NJ.

Odland, A., H. J. B. Birks, and J. M. Line. 1990. Quantitative vegetation-environment
relationships in west norwegian tall-fem vegetation. Nord. J. Bot. 10:511-33.

Odum, ER 1969. The strategy of ecosystem development. Science 154:262-270.

Oksanen, J. 1988. A note on the occasional instability of detrending in correspondence
analysis. Vegetatio 82:29-32

Oksanen, J ., and P. Huttunen. 1989. Finding a common ordination for several data sets by
individual differences scaling. Vegetatio 83:137-45.

187

Padley, EA. 1989. Associations among glacial landforms, soils, and vegetation in
northeastern lower Michigan. Ph.D. Dissertation, Michigan State University, East
Lansing, MI. 202 pp.

Palmer, M. W. 1990. Spatial scale and patterns of species-environment relationships in
hardwood forests of the North Carolina piedmont. Coenoses 5:79-87.

Palmer, M. W. 1993. Putting things in even better order: the advantages of canonical
correspondence analysis. Ecology 74:2215-30.

Parker, A. J. 1994. Latitudinal gradients of coniferous tree species, vegetation, and
climate in the Sierran-Cascade axis of Northern California. Vegetatio 115:145-55.

Peet, R.K., R.G. Knox, J .S. Case, and RB. Allen. 1987. Putting things in order: the
advantages of detrended correspondence analysis. Am. Nat. 131 (6):924—934.

Peet, R.K., and O.L. Loucks. 1977. A gradient analysis of southern Wisconsin forest.
Ecology 58: 485-499.

Pielou, EC. 1984. The interpretation of ecological data: A primer on classification and
ordination. John Wiley and Sons, New York. 263 pp.

Pregitzer, K.S. 1981. Relationships among physiography, soils, and vegetation of the
McCormick experimental Forest, Upper Michigan. Padley, EA. 1989. Ph.D.
Dissertation, University of Michigan, Ann Arbor, MI. 205 pp.

Pregitzer, K.S. and B.V. Barnes. 1982. The use of ground flora to indicate edaphic
factors in upland ecosystems of the McCormick Experimental Forest, Upper Michigan.
Can. J. For. Res. 12:661-672.

Pregitzer, K.S. and B.V. Barnes. 1984. Classification and comparison of upland
hardwood and conifer ecosystems of the Cyrus H. McCormick Experimental Forest,
upper Michigan. Can. J. For. Res. 14:362-375.

Prentice, H. C., and W. C. Cramer. 1990. The plant community as a niche bioassay:
environmental correlates of local variation in Gypsophila fastigiata. J. Ecol. 78:313-25.

Reed, R. A., M. W. Palmer, R. K. Peet, and P. S. White. 1993. Scale dependence of
vegetation-environment correlations: a case study of a North Carolina piedrnont
woodland. J. Veg. Sci. 42329-40.

Retuerto, R., and A. Carballeira. 1991. Defining phytoclirnatic units in Galicia, Spain, by
means of multivariate methods. J. Veg. Sci. 2:699-710.

188

Rowe, J .S. 1980. The common denominator in land classification in Canada: an
ecological approach to mapping. For. Chron. 56:19-20.

Rowe, J .S. 1984. Forestland classification: limitations on the use of vegetation. In:
Forest Land Classification: Experience, Problems, Perspectives. Proceedings of the
Symposium, Madison, WI, March 18-20. pp. 132-147.

Roberts, MR. and N.L. Christensen. Vegetation variation among mesic successional
forest stands in northern lower Michigan. Can. Jor. Bot. Vol 66. pp 262-272.

Russell, W.E., and Jordan, J .K. 1991. Ecological Classification System for Classifying
Land Capability in Midwestern and Northeastern US. National Forests. In: Proceedings
of the Symposium, Ecological Land Classification: Applications to Identify the
Productive Potential of Southern Forests, Charlotte, NC, January 7-9. USDA-Forest
Service, General Technical Report SE-68.

Rydgren, K. 1994. Low-alpine vegetation in Gutilia National Park, Engerdal, Hedmark,
Norway, and its relation to the environment. Sommerfeltia 21 :1-47.

Salt, SW. 1979. A comment on the use of the term emergent properties. Am. Nat. 113.
Pp. 145-148.

Shimper, A.F.W. 1903. Plant geography on a physiological basis. Clarendon Press,
Oxford.

Schouw, J .F . Grundtraek til en almindelig plantegeographie. Glydendal, Copenhagen.

Smalley, G.W. 1986. Site classification and evaluation for the Interior Uplands. USDA
Forest Service. Technical Publication R8-TP9. Southern Region, Atlanta, Georgia.

Soil Survey Staff. 1975. Soil taxonomy. USDA Soil Conserv. Serv. Publ. No. AH-436.

Spies, T.A., and B.V. Barnes. 1985a. A multifactor ecological classification of the

northern hardwood and conifer ecosystems of Sylvania Recreation Area, Upper
Peninsula, Michigan. Can. J. For. Res. 15:949-960.

Spies, T. A. and B.V. Barnes. 1985b. Ecological species groups of upland northern
hardwood-hemlock forest ecosystems of the Sylvania Recreation Area, Upper Peninsula,
Michigan. Can. J. of For. Res., 15:961-972.

Stoddard, AM. 1979. Standardization of Measures Prior to Cluster Analysis. Biometrics
35:765-773.

189

Swanson, F .J ., T.K. Kratz, N. Caine, and R.G. Woodmansee. 1988. Landforrn effects on
ecosystem patterns and processes. BioScience 38:92-98.

ter Braak, C. J. F. 1988. CANOCO - an extension of DECORANA to anayze
species-environment relationships. Vegetatio 75: 159-60.

ter Braak, C. J. F. 1986. Canonical correspondence analysis: a new eigenvector technique
for multivariate direct gradient analysis. Ecology 67:1167-79.

van Groenewoud, H. 1992. The robustness of Correspondence, Detrended
Correspondence, and TWINSPAN Analysis. J. Veg Sci. 3:239-246.

van Humbolt, A. 1807. Ideen zu einer Geographic der Pflanzen nebst einem
Naturgemalde der Tropenlander. Cotta. Tubingen. 182 p.

Vel zquez, A. 1994. Multivariate analysis of the vegetation of the volcanoes Tl loc and
Pelado, Mexico. J. Veg. Sci. 5:263-70.

Warming, E. 1895. Plantesamfund, grund traek af den okologiske plantegeografi.
Philipsen, Copenhagen.

Warming, E. 1909. Oecology of plants: an introduction to the study of plant
communities. Oxford University Press, Oxford. 422 p.

Wartenberg, D., S. Person, and J .E. Rohlf. 1987. Putting things in order: a critique of
detrended correspondence analysis. Am. Nat. 129: 434-448.

Wertz, W.A., and J .A. Arnold. 1972. Land Systems Inventory. USDA Forest Service;
Intermountain Region, Ogden, UT.

Whitney, G.G. 1991. Relation of plant species to substrate, landscape position, and
aspect in north central Massachusetts. Can. J. For. Res. 21 :1245-1252.

Whittaker, RH. 1960. The vegetation of the Siskiyou Mountains of Oregon and
California. Ecol. Monogr.30:279-338.

Zak, DR. and K.S. Pregitzer. 1990. Spatial and temporal variability of nitrogen cycling
in northern lower Michigan. For. Sci. 36:367-3 80.

Zak, D.R., G.E. Host and K.S. Pregitzer. 1989. Regional variability in nitrogen
mineralization, nitrification, and overstory biomass in northern lower Michigan. Can. J.
For. Res. 19:1521-1526.

190

Zak, D.R., K.S. Pregitzer, and GE. Host. 1986. Landscape variation in nitrogen
mineralization and nitrification. Can. J. For. Res. 16:1258-1263.

Zhang, J .-T., and E. R. B. Oxley. 1994. A comparison of three methods of multivariate
analysis of upland grasslands in North Wales. J. Veg. Sci. 5:71-6

APPENDICES

191

APPENDIX A

Normality Test (Kolmogorov-Smirnov) for soil variables

02TH K—S Distance = 0.104 P = 0.231 Passed
AlTH K-S Distance = 0.189 P = <0.001 Failed
AlV K-S Distance = 0.227 P = <0.001 Failed
ETH K-S Distance = 0.145 P = 0.016 Failed
T3OSICL K—S Distance = 0.301 P = <0.001 Failed
BDEP K-S Distance = 0.177 P = <0.001 Failed
BIN K-S Distance = 0.337 P = <0.001 Failed
BVA K-S Distance = 0.170 P = 0.002 Failed
BSTH K-S Distance = 0.105 P = 0.222 Passed
BSICL K-S Distance = 0.204 P = <0.001 Failed
CDEP: K-S Distance = 0.114 P = 0.138 Passed
GWT K-S Distance = 0.415 P = <0.001 Failed
MOTDEP K-S Distance = 0.301 P = <0.001 Failed
DCL K-S Distance = 0.325 P = <0.001 Failed
UTSIL K-S Distance = 0.207 P = <0.001 Failed
LTSI K-S Distance = 0.274 P = <0.001 Failed
TXSIL K-S Distance = 0.279 P = <0.001 Failed
MAXSICL K-S Distance = 0.155 P = 0.007 Failed
DOMSICL K-S Distance = 0.296 P = <0.001 Failed
ACV K-S Distance = 0.487 P = <0.001 Failed
ACLS K-S Distance = 0.219 P = <0.001 Failed
ACSL K—S Distance = 0.223 P = <0.001 Failed
ACSCL K—S Distance = 0.252 P = <0.001 Failed
LD150 K—S Distance = 0.158 P = 0.005 Failed
TILLD K-S Distance = 0.277 P = <0.001 Failed
EFFDE K-S Distance = 0.116 P = 0.121 Passed
SLOP K-S Distance = 0.139 P = 0.026 Failed
ASP K-S Distance = 0.231 P = <0.001 Failed
ELEV K-S Distance = 0.146 P = 0.015 Failed
BTH K-S Distance = 0.151 P = 0.010 Failed
EDC K-S Distance = 0.076 P = 0.669 Passed
BIC8 K-S Distance = 0.225 P = <0.001 Failed
PF K-S Distance = 0.198 P = <0.001 Failed
DEPSCL K-S Distance = 0.242 P = <0.001 Failed
SICL K—S Distance = 0.301 P = <0.001 Failed

A test that fails indicates that the data varies significantly
from the pattern expected if the data was drawn from a
population with a normal distribution.

A test that passes indicates that the data matches the pattern
expected if the data was drawn from a population with a normal
distribution.