EPA/620/R-94/007
March 1994
FOREST HEALTH MONITORING
1991 GEORGIA INDICATOR
EVALUATION AND FIELD STUDY
Technical Director
Samuel A. Alexander
U.S. Environmental Protection Agency
Environmental Monitoring and Assessment Program
EMAP Center
Research Triangle Park, NC 27709
and
Program Manager
Joseph E. Barnard
U.S.D.A. Forest Service
U.S. Forest Service Laboratory
Research Triangle Park, NC 27709
Project Officer
Roger Blair
U.S. EPA Environmental Research Laboratory
200 SW 35th Street
Corvallis, Oregon 97333
U.S. Environmental Protection Agency
Office of Research and Development
Washington, D.C. 20460
Printed on Recycled Paper
-------
NOTICE
The research described in this report has been funded by the U.S. Environmental Protection
Agency. This document has been prepared at the EPA Environmental Research Laboratory in
Corvallis, Oregon, through Contract Nos. 68-C8-0006to ManTech Environmental Technology, Inc., 68-
CO-0049 to Lockheed Engineering and Sciences Company, and 68-DO-0106 to Statistical Consulting
Service. It has been subjected to the Agency's peer and administrative review and approved for
publication. Mention of trade names or commercial products does not constitute endorsement or
recommendation for use.
This report represents data from one year of field operations of the Environmental Monitoring
and Assessment Program (EMAP). Because the probability-based scientific design used by the EMAP
necessitates multiple years of sampling, there is uncertainty associated with these data. This
uncertainty will decrease as the full power of the approach is realized. Similarly, temporal changes and
trends cannot be reported, as these require multiple years of observation. Please note that this report
contains data from demonstration studies in one geographic region. Appropriate precautions should
be exercised when using this information for policy, regulatory or legislative purposes.
This report should be cited as follows: Alexander, S.A., M. Baldwin, W.A. Bechtold, D.L.
Cassell, S. Cline, T. Droessler, J.W.. Hazard, J.G. Isebrands, V.J. Labau, K.H. Riitters, H. Schreuder,
S.J. Steele, and M.S. Williams. 1993. Forest Health Monitoring: 1991 Georgia Indicator Evaluation
and Field Study, EPA/620/R-94/007. U.S. EPA Environmental Research Laboratory, Corvallis, OR.
-------
FOREST HEALTH MONITORING
1991 GEORGIA INDICATOR
EVALUATION AND FIELD STUDY
Approved by
Joseph E. Barnard
National Program Manager
Forest Health Monitoring
Samuel A. Alexander
Technical Director
Forest Health Monitoring
-------
FOREWORD
The Forest Health Monitoring (FHM) 1991 Georgia Pilot study was undertaken to conduct field
research for advancing forest monitoring science within the Environmental Monitoring and Assessment
Program (EMAP). Indicator developmental and operational monitoring research was conducted
simultaneously on plots. Indicator development objectives and criteria are detailed in the project plan
entitled FY911ndicator Evaluation Field Study for Environmental Monitoring and Assessment Program -
Forests (EMAP-F).
The pilot study was designed to test methods for quantifying vegetation structure,
photosynthetically active radiation (PAR), dendrochronology, and selected root fungi. Testing the
methods included comparing different data collection procedures for individual indicators, estimating
sampling efficiency (both of the sampling design and the sampling unit design), and evaluating spatial
variability. In addition, the accuracy and precision of tree height instruments were determined as part
of the pilot study.
The field work was conducted on a systematic grid consisting of 63 plots across the state of
Georgia. Figure 1 shows a map of Georgia with county boundaries and demo plot locations. Although
there were 63 potential plot locations, the land use for many was something other than forested (e.g.,
agricultural, urban, marsh). Several of the indicator studies utilized only 20 of the forested plots in
western Georgia. Figure 1 shows the pilot plot locations.
Figure 1. Locations of the 1991 Georgia Pilot
plots. The 20 plots used by several
indicators are circled.
-------
EXECUTIVE SUMMARY
The following summary provides a succinct review of the objectives and results of the 1991
Georgia pilot by indicator. The lessons learned have been used in planning and improving the 1992
field activities.
VEGETATION STRUCTURE
The objective of the vegetation structure study was to compare the operational and informa-
tional characteristics of area-based and point-based methods for quantifying vertical and horizontal
vegetation structure and to recommend a measurement system for vegetation structure for 1992 and
beyond.
The general superiority of the quadrat method for sampling vascular plant species richness at
both plot and regional levels was the most significant of several methodological differences found
between the quadrat and pole methods. Comparisons of quadrat and pole diversity indices and
species accumulation curves supported this finding. This finding is important because estimates of
species richness are the most basic and sensitive measurements of the status of biotic diversity.
Although estimates of species richness are basic and straightforward, they are not simple. Field
personnel must have a working knowledge of the regional flora, the ability to identify vascular plants
under field conditions based on experience or using regional taxonomic keys, and the ability to collect
and press unknown plant specimens for later identification.
In constrast, even though superior to the pole method, the quadrat method implemented in this
study usually sampled only 70 percent to 80 percent (range 66 percent to 107 percent) of a crudely
estimated total plant species richness of the plot and regional level. Therefore, suggestions to reduce
sample numbers per plot must be thoroughly evaluated before implementation, since the reliability of
species richness and other diversity calculations increases with sample size. The quadrat method for
measuring vegetation structure was recommended for use in future Forest Health Monitoring field
seasons based on these findings.
PHOTOSYNTHETICALLY ACTIVE RADIATION (PAR)
The primary objective of the PAR study was to develop an efficient and reliable method of using
ceptometers and quantum sensors for measuring forest canopy light environments in various stand
conditions. Several equipment problems were worked out so that PAR data collection can be
considered reliable. The importance of measuring diffuse PAR in open areas in addition to ambient
-------
PAR became evident. Statistics indicated that 7 points gave as good an estimate of PAR as 19 points,
thereby reducing field work and time.
DENDROCHRONOLOGY
The objectives for dendrochronology were to determine if the sampling intensity and tree
selection protocols were adequate for quantifing diameter at breast height (dbh) growth rates and
trends on a regional basis. Based on the variance component analysis, the sample intensity is
adequate. A graphical analysis of growth patterns showed that cores, grouped by species and age,
showed similar patterns of growth within groups. Specific species should be sampled where possible
(loblolly pine, for example), to minimize between-species variability in growth.
In addition, recommendations were made for improving equipment, field sampling, core
handling, and preparation. The recommendations should improve core quality and expedite the
measurement and analysis process. A specific recommendation was that cores should be prepared,
measured, and analyzed by one laboratory.
ROOT DISEASE
One objective was to determine the presence and severity of root disease using two root
sampling techniques: the two-root method and the cubic foot root collection method. The results
showed that the two-root method was more effective than the cubic foot method in detecting root
disease pathogens.
Another objective was to evaluate the cubic foot method for quantifying ectomycorrhizal fungi.
The field sampling procedure was simple, but the laboratory work was cumbersome and labor intensive.
The core sample was considered too large and the variability between cores was high. A smaller
volume soil sample and/or a soil subsampling procedure were suggested to improve the procedure.
TREE HEIGHT
The objective was to determine if the accuracy and precision of tree height measuring devices
were adequate for providing a measure of tree height change over time. The authors concluded that
a 10 percent error in measuring tree height was common, especially for trees over 12 m in height. A
10 percent measurement error is unacceptable for accurately estimating height change over a 5-year
period, therefore tree height is not recommended as an indicator at this time.
-------
TABLE OF CONTENTS
Section
Page
Forward i v
Executive Summary v
List of Figures X
List of Tables xi
Abbreviations and Acronyms . .. xi i
Acknowledgements xi i i
Contributors . xiv
Abstract • • • • xv
1. STATISTICAL METHODOLOGY 1-1
1.1 Introduction 1-1
1.2 Sampling Efficiency 1-1
1.3 Spatial Analysis : 1-4
2. VEGETATION STRUCTURE 2-1
2.1 Introduction 2-1
2.2 Objectives 2-1
2.3 Vegetation Structure Measurement Methods i 2-3
2.3.1 Sampling Design 2-3
2.3.1.1 Plot Selection 2-3
2.3.1.2 On-Plot Sampling Scheme , .2-4
2.3.2 Logistics ....... 2-4
2.3.2.1 Field Personnel Requirements 2-4
2.3.2.2 Training 2-4
2.3.3 Quality Assurance 2-4
2.3.4 Measurement Procedures 2-6
2.3.4.1 Point Quadrats .2-6
2.3.4.2 Area Quadrats 2-7
.. 2.4 Analytical Methods 2-8
2.4.1 Index Selection and Calculation ..... . .. .,. .:,...... .2-8
2.4.2 Statistical Analysis of Comparability of Pole and Quadrat Methods -...2-11
2.4.2.1 Operational Characteristics •,.-... .-......- ...:,-... 2-11
2.4.2.2 Informational Characteristics .. 2-11
2.5 Results and Discussion . .... . . . ..... .2-13
2.5.1 Operational Characteristics 2-13
2.5.1.1 Spatial Variation 2-13
2.5.1.2 Sampling Efficiency 2-13
2.5.1.3 Time and Labor Requirements ...... .. 2-17
2.5.1.4 Measurement Error ..... 2-17
2.5.2 Informational Characteristics 2-17
2.5.2.1 Species Diversity Indices 2-17
2.5.2.2 Species Accumulation Curves 2-18
-------
2.6 Summary of Results and Recommendations 2-22
2.6.1 Operational Results 2-22
2.6.2 Informational Results 2-23
2.6.3 Recommendations 2-23
3. PHOTOSYNTHETICALLY ACTIVE RADIATION (PAR): GEORGIA PILOT
STUDY 1991 3-1
3.1 Introduction 3-1
3.2 Objectives 3-1
3.3 Related 1991 PAR Studies 3-2
3.4 Methods 3-4
3.5 Results 3-4
4. DENDROCHRONOLOGY 4-1
4.1 Introduction 4-1
4.2 Objective and Scope 4-1
4.3 Methods 4-1
4.3.1 Sampling 4-1
4.3.2 Preparation and Measurement 4-2
4.3.3 Quality Assurance Procedures 4-2
4.3.4 Analyses 4-3
4.4 Results and Discussion 4-3
4.4.1 Quality Assurance 4-3
4.4.2 Regional Growth Patterns 4-3
4.4.3 Statistical Considerations 4-5
4.5 Conclusions and Recommendations 4-6
4.5.1 Equipment 4-6
4.5.2 Field Sampling 4-7
4.5.3 Handling and Preparation 4-7
5. ROOT DISEASE 5-1
5.1 Introduction 5-1
5.2 Evaluation of Root Disease Indicator 5-1
5.2.1 Objective 5-1
5.2.2 Materials and Methods 5-1
5.2.3 Results 5-2
5.3 Root Sampling for Evaluation of Root Diseases and Mycorrhizae 5-2
5.3.1 Objective 5-2
5.3.2 Materials and Methods 5-2
5.3.2.1 Root Disease Quantification 5-2
5.3.2.2 Mycorrhizal Quantification 5-3
5.3.3 Results and Discussion 5-3
-------
6. EVALUATION OF DIFFERENT INSTRUMENTS FOR MEASURING TREE HEIGHT 6-1
6.1 Introduction • 6-1
6.2 Methods 6-1
6.2.1 Study Planning and Preparation 6-1
6.2.2 Data Set Descriptions 6-2
6.2.3 Evaluation Techniques 6-3
6.3 Results and Discussion 6-15
6.4 Conclusions • • • 6-17
7. REFERENCES
7-1
-------
Figure
1.
2-1
2-2
2-3
2-4
3-1
3-2
3-3
3-4
3-5
3-6
4-1
4-2
6-1
6-2
6-3
6-4
6-5
6-6
6-7
6-8
6-9
6-10
Page
LIST OF FIGURES
Location of the 1991 Georgia Pilot plots xii
Forest Health Monitoring Program ecological assessment model for biotic diversity ... 2-2
On-plot sampling scheme for (a) point quadrats and (b) area quadrats
in relation to subplot 2-5
Typical examples of species accumulation curves based on the pole and quadrat
methods: (a) stratum 1, (b) stratum 2, (c) stratum 3, and (d) stratum 4 (quadrat only) 2-19
Regional species accumulation curves based on the pole and quadrat methods:
(a) stratum 1, (b) stratum 2, (c) stratum 3, and (d) stratum 4 (quadrat only) 2-21
Locations of PAR research sites and pilot sites in 1991 Forest Health Monitoring 3-2
Locations of the 1991 Georgia Pilot PAR plots 3-5
PAR sampling scheme for Georgia pilot study in 1991 3-6
Under-canopy and ambient PAR on a plot in the 1991 Georgia pilot showing the
amount of variability within plots 3-7
Percent transmitted PAR from data in Figure 3-4 3-9
Integrated PAR/houron the 1991 Georgia pilot PAR plots 3-10
Example 40-year growth trend 4-4
Example 90-year growth trend 4-5
True height versus estimated height using the Clinometer (all species) 6-9
True height versus estimated height using the Relaskop (all species) 6-9
True height versus estimated height using the Tele-relaskop (all species) 6-10
True height versus estimated height using the Enbeeco (all species) 6-10
True height versus estimated height using the Laser (all species) 6-11
True height versus estimated height using the Clinometer (softwoods only) 6-11
True height versus estimated height using the Relaskop (softwoods only) 6-12
True height versus estimated height using the Tele-relaskop (softwoods only) 6-12
True height versus estimated height using the Enbeeco (softwoods only) 6-13
True height versus estimated height using the Laser (softwoods only) 6-13
-------
LIST OF TABLES
Table
1-1
2-1
2-2
2-3
2-4
2-5
3-1
3-2
4-1
4-2
5-1
6-1
6-2
6-3
6-4
6-5
6-6
6-7
Page
Estimated Cost Components for the 1991 Georgia Pilot 1-2
Candidate Indicators of Biotic Diversity 2-3
Definition and Calculation of Plot-Level Values for Each Element of Heterogeneity
and Complexity of Vegetation 2-10
Mean Square and Error Estimates for Vertical Structure 2-15
Estimated Optimal Subsample Sizes 2-16
Differences in Hill Diversity Numbers from the Quadrat (Q) and Pole (P) Methods
by Vegetation Stratum 2-18
EMAP-FHM PAR Indicator: 1991 Georgia Pilot Plot Summary 3-3
1991 EMAP-FHM Georgia Pilot Means of %TPAR .3-8
Cost, Mean Square, and Error Estimates for Dendrochronology 4-6
Estimated Optimal Subsample Sizes 4-6
Ectomycorrhizal Tips per Liter of Soil 5-4
Regression Coefficients, Standard Errors, 95% Confidence Intervals, and R2 Values
for Untransformed Data and All Trees 6-4
Regression Coefficients, Standard Errors, 95% Confidence Intervals, and R2 Values
for Log Transformed Data and All Trees 6-5
Regression Coefficients, Standard Errors, 95% Confidence Intervals, and R2 Values
for Untransformed Data with All Hardwood Trees Removed 6-6
Regression Coefficients, Standard Errors, 95% Confidence Intervals, and R2 Values
for Log Transformed Data for Softwood Trees Only 6-7
Average Absolute Model Deviation and Number of Trees in Each 10-foot Height
Class for All Species, Model [1], Fit to the Data 6-14
Average Absolute Model Deviation and Number of Trees in Each 10-foot Height
Class for Softwood Trees Only, Model [1] Fit to the Data 6-14
Paired Comparison Test Results, Average and Median Absolute Errors Between
Crew Members, and Number of Observation 6-15
-------
ABBREVIATIONS AND ACRONYMS
ASCS
ASP
dbh
dm
EM
EMAP
EPA
FAA
FHM
GIS
GPS
MLRA
NLIN
MQO
PAR
PDR
QA
SAS
SCS
TPAR
USDA
VVS
Agricultural Stabilization and Conservation Service
annosus sampling procedure
diameter at breast height
decimeter
ectomycorrhizal
Environmental Monitoring and Assessment Program
U.S. Environmental Protection Agency
formalin, acetic acid, and alcohol
Forest Health Monitoring
Geographic Information System
Global Positioning System
major land resource area
(SAS) nonlinear regression procedure
measurement quality objective
photosynthetically active radiation
portable data recorder
quality assurance
Statistical Analysis System
U.S. Soil Conservation Service
transmitted photosynthetically active radiation
United States Deparment of Agriculture
vertical vegetation structure
-------
ACKNOWLEDGEMENTS
The authors thank the following peer reviewers for their constructive criticisms: Mark Hansen, U.S.
Forest Service, North Central Forest Experiment Station, St. Paul, MN. Greg Reams, U.S. Forest
Service, Institute of Quantitative Studies, New Orleans, LA. Tim Strickland, U.S. Department of
Agriculture, Competitive Grants, Washington, D.C.
Appreciation goes to Terry Droessler and Susan Christie (ManTech Environmental Technology, Inc.)
for their contributions as editors.
In addition, editorial comments from Barbara Conkling helped improve the document.
xi 1
-------
Contributors
(listed alphabetically by organization, location, and name)
Bureau of Land Management, Corvallis, OR
D.L. Cassell (formerly of ManTech Environmental Technologies, Corvallis, OR)
Bureau of Land Management, Las Vegas, NV
S.CIine (formerly of ManTech Environmental Technologies, Corvallis, OR)
ManTech Environmental Technologies, Inc., Corvallis, OR
T. Droessler
Statistical Consulting Service, Bend OR
J.W. Hazard
Tennessee Valley Authority, Norris, TN
K.H. Riitters (formerly of ManTech Environmental Technologies, Research Triangle Park, NC)
USDA Forest Service, Anchorage Forest Sciences Lab, Anchorage, AK
VJ. LaBau
USDA Forest Service, Forest Inventory and Analysis, Southeastern Experiment Station,
Asheville, NC
W.A. Bechtold
USDA Forest Service, Rocky Mountain Station, Ft. Collins, CO
H. Schreuder
M. Williams
USDA Forest Service Research Lab, Rhinelander, Wl
J.G. Isebrands
S.J. Steele
U.S. EPA, U.S. Forest Service Lab, Research Triangle Park, NC
S.A. Alexander
Virginia Polytechnic Institute and State University, Blacksburg, VA
M. Baldwin
xiv
-------
ABSTRACT
The Forest Health Monitoring (FHM) 1991 Georgia Pilot study was undertaken to conduct field
research for advancing forest monitoring science within the Environmental Monitoring and Assessment
Program (EMAP). Indicator developmental and operational monitoring research was conducted
simultaneously on plots. The pilot study was designed to test methods for quantifying vegetation
structure, photosynthetically active radiation (PAR), dendrochronology, and selected root fungi. The
objective of the vegetation structure study was to compare the operational and informational
characteristics of area-based and point-based methods for quantifying vertical and horizontal vegetation
structure and to recommend a measurement system for vegetation structure for 1992 and beyond. The
primary objective of the PAR study was to develop an efficient and reliable method of using
ceptometers and quantum sensors for measuring forest canopy light environments in various stand
conditions. The objectives for dendrochronology were to determine if the sampling intensity and tree
selection protocols were adequate for quantifying diameter at breast height (dbh) growth rates and
trends on a regional basis. The objectives for the root disease study were to determine the presence
and severity of root disease using two root sampling methods (the two-root method and the cubic foot
root collection method) and compare the methods, and also to evaluate the cubic foot method for
quantifying ectomycorrhizal fungi. The objective of the tree height study was to determine if the
accuracy and precision of tree height measuring devices were adequate for providing a measure of tree
height change over time. The report presents the results and recommendations based on those
results.
xv
-------
-------
SECTION 1
STATISTICAL METHODOLOGY
D.L. Cassell and J.W. Hazard
1.1 Introduction
One of the objectives of these field activities was to assess statistically the reliability of
measurements for pilot indicators. The developmental stage of the various pilot indicators dictated the
statistical methodology, since, in some cases; concerns for logistics or feasibility were paramount.
Root sampling focused on feasibility and comparison of methods, as did tree height. The pilot
studies for these measurements were thus not designed with variance component estimation in mind.
Research on photosynthetically active radiation (PAR) has already included variance component
estimation and semivariogram estimation in a previous pilot (Riitters et al., 1991), and focused on other
avenues of research. The main goal of the vegetation structure leader was to compare quadrat and
pole sampling with feasibility studies, thus the plot layouts are not optimal for semivariogram estimation.
Variance component analysis for estimation of sampling efficiency was performed on both vege-
tation structure methods, as well as on dendrochronology data. These results provide the ingredients
for evaluating the efficiency of the sampling design and the sampling unit design. In addition, the
spatial variability of both vegetation structure methods was examined. Spatial variability is important
for two reasons. If the data for a measurement set are spatially correlated, the observations on those
measurements cannot be assumed to be independent, which would affect the variance component
analysis. Spatial correlation would also affect the variance estimates for the data if the indicator passed
to demonstration or implementation phase.
1.2 Sampling Efficiency
Sampling efficiency was investigated using standard statistical equations to combine information
obtained about sample variances with estimates of sampling cost to estimate optimum sample sizes
for the different sampling stages. A hierarchical multi-stage sampling model was used in each case.
Both methods of measuring vegetation structure used a three-stage sampling scheme incorporating
plots, subplots, and measurement stations (see Section 2). For the vertical pole method, the
measurement station was the pole point; for the quadrat method, the measurement station was the
entire quadrat. For the dendrochronology method (see Section 4), a four-stage sampling scheme was
used: plots, subplots, trees within subplots, and cores within a tree.
The cost components (Table 1-1) used in the evaluation are based on the average times
required to set up and make measurements on a new plot location, a new subplot in a given plot, and
further subsampling units. The subsampling costs are the times required for locating and measuring
1-1
-------
vegetation structure pole points, siting quadrats and measuring flora within the quadrat, and locating
and coring sample trees. These times were recorded as part of the logistics evaluations of the pilot
measurements.
TABLE 1-1. ESTIMATED COST COMPONENTS FOR THE 1991 GEORGIA PILOT
Cost Component Cost Estimate
Plot
Subplot
Pole point
Quadrat
Subplot + Tree
Core
3.33
0.21
Vertical Pole Method
0.05
Quadrat Method
0.40
Dendrochronology
0.66
0.08
The variance estimates were made by computing mean squares for the variables in the various
stages of the nested model. These mean squares were computed using the formulas in Sukhatme
(1954) and Cochran (1977), and were then substituted into the optimum sample size equations
provided in those texts. The sample size equations for a three-stage sampling design are:
Second stage: rru = (c1/c2)1/a * [(MS2 - MSg/kytMS, -
1/2
Third stage:
^opt -
* [MS3/(MS2 - MSg/k)]
[1/2
where cr and MS, are the sampling cost and the estimated mean square, respectively, for the ith stage
of sampling; m and k are the actual sample sizes used in the second and third stages of the design
respectively; and m^, and k^ are the respective estimated optimum sample sizes for these stages of
the design.
If adequate quality assurance data are available to provide an external estimate of the
measurement error at the final stage of sampling, then these sample size equations can be improved.
1-2
-------
If the estimate of the measurement error variance for the final sampling stage is se2, then the sample
size equations for the three-stage sampling design (Cassell, 1992) become:
Second stage: mopt = (C1/c2)1/2 * [(MS2 - (MS3 - se2)/k)/(MS1 -
Third stage:' kopt =
[(MS3 - se2)/(MS2 - (MS3 - se2)/k)f 2
where q, MS;, mopt and kopt are as on the previous page.
The optimum first-stage sample size (i.e., the total number of plots) is not provided by these
equations. This estimate can be made given a desired precision for a parameter over the population
of plots under study, or given a total fixed cost for surveying the population. But such a solution should
utilize alternate optimization techniques that are appropriate for the Forest Health Monitoring (FHM)
Program first-stage sampling design. The resource allocation formulas here are appropriate for simple
random sampling; the FHM design uses systematic sampling with post-stratification, which requires
different formulas.
This emphasizes that the objective of this study—determining an optimum plot design once a
plot location has been selected—gives little insight into the number of plots needed to characterize
regional forest health at some specified level of precision. On the other hand, an optimum single-plot
design is valid no matter how many plots are ultimately selected for measurement.
This estimation procedure does not assume that there are known bounds on the plot-level vari-
ability, but seeks to optimize the sampling distribution across the stages of measurement for the
estimated cost components. If particular indicators develop guidelines for plot-level variability limits,
then such limits will also be used to assess the arrangement of measurements within a plot.
Another important consideration is that the optimal plot design should be evaluated for the
objective of estimating a population parameter over many plot locations, and not for the objective of
estimating that parameter for any particular plot contained in that population. Optimal plot design for
a regional survey does not imply that every site-specific estimate will meet other precision requirements.
This tradeoff between obtaining precise answers for each site or for all sites in a population at once
makes large-scale surveys practical and simultaneously limits the inferences that can be drawn about
any one location.
A final consideration is that statistical optimality is not the only criterion for plot design. It may
be necessary to ensure that some measurements "cover the area" so that they may be related to other
measurements made on the plot. Or the design needed for an indicator may include special features
to ensure accurate characterization of relevant subsets of the plot. Indicator development is an
essential part of FHM, and the objectives of the indicator determine how the data are to be used and
what other considerations must be incorporated into sample selection.
1-3
-------
1 .3 Spatial Analysis
The vegetation structure measurements (Section 2) were taken on an hexagonal grid for pole
measurements, and in contiguous quadrats for the quadrat measurements. Semivariograms were
calculated for species abundance measurements taken at these measurement stations, using standard
formulas (e.g., Ripley, 1981). A semivariogram allows us to visualize the variability between
observations as a function of the distance between the observations. In essence, the variance of two
observations is computed as half the square of their distance. Then the variance for all pairs of data
points a given distance apart is calculated as the average of these pairwise calculations.
This is a reasonable way to calculate the variance for a pair of numbers. If we choose two
numbers, x and y, their mean z must be halfway between the two numbers. Thus the variance of the
pair {x,y} is given by:
= [(x - y)/2f + [(x - y)/2f
= 1/4 (x - yf + 1/4 (x - yf
The semivariograms permit us to visualize at what distance the spatial correlation levels off
(i.e., how far away the measurements can be when the correlation is roughly the same as for two
points on opposite sides of the plot). This provides useful information for deciding whether two data
points can be treated as statistically independent. When the data are independent, they appear to be
scattered at random. Spatial structure may appear in many forms, including clustering of the data
points, clustering of values of the data, or nonrandom order across the area.
1-4
-------
SECTION 2
VEGETATION STRUCTURE
S. Cline and D.L. Cassell
2.1 Introduction
Gathering information about the status of forest biotic diversity and how it is changing is one
objective of the Forest Health Monitoring (FHM) program. A vegetation structure indicator was
proposed originally to provide better assessment of non-tree, understory vegetation. This understory
vegetation comprises most of the plant species diversity in forests, is more sensitive to environmental
gradients, and has higher turnover rates and thus a potentially faster reaction time to stress than trees
(Daubenmire, 1968). Furthermore, vegetation structure is an important aspect of wildlife habitat
structure, which was appealing because monitoring animal habitat in EMAP may be a cost-effective
alternative to directly monitoring animal populations.
2.2 Objectives
Numerous candidate indicators representing compositional, structural, and functional aspects of
biotic diversity could be measured, depending upon the objectives of the monitoring program (Noss,
1990). Given the importance of the effects of habitat alteration upon biotic diversity (U.S. EPA, 1990),
structural features of landscapes and forest habitats are leading candidate indicators (Figure 2-1; Table
2-1).
The objectives for the vegetation structure indicator were:
To compare the operational characteristics of a point quadrat (pole) method and an area
quadrat method of measuring vertical and horizontal vegetation structure, including spatial
variation, sampling efficiency, time and labor requirements, and measurement error.
To compare the informational characteristics of a point quadrat (pole) method and an area
quadrat method of measuring vertical and horizontal vegetation structure, including species
accumulation curves, and species and structural diversity estimates.
To recommend a refined and streamlined measurement system for vegetation structure for
1992.
2-1
-------
.
DUJ
c
I "3 to
1 -1
« >.°
U.
CO
2
CD
73
.0
2
la
CD
•a
o
CD
CO
CD
CO
CO
CO
"5
_o
a)
o
o
CD
CO
0)
O
_c
o
O
ca
CD
X
1o
CD
£
CJ
CD
O)
U.
2-2
-------
TABLE 2-1. CANDIDATE INDICATORS OF BIOTIC DIVERSITY
B.
Coarse patch delineation: Landscape/Region level
Remote-based variables from small-scale (1:45,000) photos
forest area by class (conifer/deciduous/mixed)
land area by use type
landscape pattern (area, shape, juxtaposition, and connectivity of patches)
Fine patch delineation based on external features: Community/Ecosystem level
Remote-based variables from large-scale (1:6,000 or 12,000) photos
subdivision of patches by forest type
subdivision of patches by tree density and height
overstory cover, roughness, and patchiness
location and area of ecotones
number of vertical strata
Fine patch characterization based on internal features: Population/Species level
Ground-based variables from fixed area plots
tree species
tree diameter and basal area
tree density
Ground-based variables from pole and quadrat methods
profile of understory vegetation cover
patchiness of understory vegetation cover
canopy cover
species and growth-form composition
species richness
2.3 Vegetation Structure Measurement Methods
2.3.1 Sampling Design
2.3.1.1 Plot Selection
We selected 20 forested hexagons from the EMAP one-quarter interpenetrating national sampling
grid. The hexagons were distributed from the northern to the southern borders of western Georgia and
included a variety of forest types and elevations in the mountain, Piedmont, and coastal plain provinces
(Figure 1 in the Foreword). These sites presented realistic conditions to test the operational capabilities
and analytical difficulties that will typically be encountered in pilots and regional demonstration studies
in other mountainous regions of the eastern United States. The field sampling plot in each hexagon
2-3
-------
is located by selecting the FIA photo grid point closest to the hexagon centerpoint that is associated
with forest (Palmer et al., 1991, Chapter 5).
2.3.1.2 On-Plot Sampling Scheme
The standard FHM sampling plot was laid out in association with each hexagon; it is a set of four
fixed-area plots (each 1/24 acre or 168 m2) spread over about 1 ha (Figure 3-3; Palmer et al., 1991,
Chapter 5). The point and area quadrats employed different sampling schemes over this four-subplot
area. Pole point quadrats were set at 7 of the 19 points per subplot used to sample photosynthetically
active ration (PAR) (Figures 2-2a and 3-3). Meanwhile, at least three 1-m2 quadrats per subplot were
laid out side by side on randomly preselected azimuths (Figure 2-2b; subsection 2.3.4.2 [quadrat
layout]).
2.3.2 Logistics
2.3.2.1 Field Personnel Requirements
A botanist was responsible for vegetation structure and photosynthetically active radiation (PAR)
measurements in the 20-plot landscape pilot substudy. A primary qualification of this person was a
working knowledge of the flora in Georgia, including the ability to identify vascular plants under field
conditions based on experience or using regional taxonomic keys, and to collect and press unknown
plant specimens for later identification. One crew member assisted the botanist during pole point
measurements.
2.3.2.2 Training
The botanist and a field auditor were trained in the vegetation structure methods in a formal
session conducted the week before the start of the field season. Training consisted of an overview of
the point and area quadrat methods in a classroom session and a half-day of field instruction and
practice on plots selected to represent different forest structural and compositional features. The next
day, the botanist measured vegetation structure as part of the pilot crew to test the crew interactions
and work loads, to gain additional experience with the methods, and to estimate remeasurement errors
under more realistic conditions. Finally, a debriefing session was held at the end of the training session
to discuss results of "plot day" and remeasurement evaluations, and to make necessary adjustments
in the vegetation structure procedures before field data collection.
2.3.3 Quality Assurance
The auditor conducted a field audit at the beginning of the second week of the field season. The
auditor checked to see that the vegetation structure procedures were being followed, remeasured two
plots, and checked plant identifications. In addition, the botanist remeasured one preselected point and
area quadrat per plot to track data quality throughout the field season. The auditor and botanist
remeasurements were used to quantify measurement error (subsections 2.4.2.1 and 2.5.1.4).
2-4
-------
(a)
Subplot
Boundary
PAR
Sample
Point
and
Number
Area
Quadrat
Transects
(1-12)
12 Ft.
Between
Points
on Line
Pole Point
Quadrats
Azimuth From
Subplot Center
(b)
Outer subplot
boundary
1
t
1nr
2
3
4
5
6
7
8
9
10
11
12 13
14J15
To
Towards subplot
center
PAR Azimuth
Figure 2-2. On-plot sampling scheme for (a) point quadrats and (b) area quadrats in relation to
subplot.
2-5
-------
2.3.4 Measurement Procedures
2.3.4.1 Point Quadrats
Summary of Method
The crew raised a 10-m telescoping pole vertically through the forest vegetation at preselected
grid points, and recorded "hits" of live leaves, branches, and stems by species and height (to the
nearest decimeter interval). The "pole" method provides objective, quantitative estimates of the vertical
and horizontal distribution of vegetation cover and its species composition.
The pole method was implemented on each plot as follows: (1) set up the grid of measurement
points on each subplot; (2) determine the division of crew member responsibilities; and (3) at each grid
point, record ground surface substrate and any species of live vegetation that touch the fully extended
pole.
Grid Layout
The crew collected data on all subplots at points 1, 2, 4, 6, 8, 10, and 12 of the 19-point,
hexagonal sampling grid established for PAR measurements (Figure 2-2a). Where a point was
obstructed by a tree bole, a very tall stump, or very large rocks, the point was moved further along the
direction line until it was physically possible to place the pole on the ground. The distance moved and
the height that the pole was elevated above the ground were then recorded. Number flags were used
to mark each grid point and left in the ground for PAR measurements.
Crew Member Responsibilities
Crews consisted of two members: observers 1 (botanist) and 2 (assistant). At each point,
observer 2 held the pole vertically while observer 1 estimated and recorded, on a standard form, the
vegetation intersecting the pole.
Recording
Plot and subplot numbers were recorded, along with starting time and (later) time finished. Next,
the substrate under the pole was recorded at the 000 height. The choices were as follows: mineral
soil/sediment, rock, standing water/flooded, stream, dead wood, litter/duff, live roots, moss (by species
or type, if known), and lichen (by species or type, if known).
Next, starting at the ground line and moving upward, live vegetation intersected by the pole was
recorded to the nearest decimeter (dm) height interval by species. Thus, the completed data for each
"hit" consisted of entries in a height and species column, and the data for a completed point consisted
of a series of height-species entries for all hits between 000 and 100 dm.
2-6
-------
Measurement Quality Objectives (MQOs)
The main sources of procedural error in the pole method were pole movements off vertical and
inherent bend, or "play" as the pole was extended. These errors were exacerbated by some uncon-
trollable environmental and site conditions, including wind that caused pole and vegetation movement,
steep slopes that made the pole hard to steady, and low and thick vegetation overhead that obscured
the view of the pole tip at upper positions. A bubble level was attached to the pole as a leveling guide.
The inherent bend in the pole was controlled by following strict protocols for placing, aligning, and
raising the pole; this did not eliminate the play, but made it consistent between measurements (i.e., the
same bend in the same direction).
For the training session, the MQOs were as follows: difference in total number of hits per point,
± 1; difference in 0.5-m intervals of recorded hits, ± 1; misidentification of plant genus, no differences;
and misidentification of plant species, 5 percent.
2.3.4.2 Area Quadrats
Summary of Method
The general strategy was to assess the number and abundance of vascular plant species present
on contiguous square meter quadrats (Mueller-Dombois and Ellenberg, 1974). Visual estimates of
cover were made for each plant species in each of four strata on each quadrat, in stratum 1, all
ground substrates and nonvascular plant species were recorded. Then, starting with stratum 2, the
following were measured: small herbs, shrubs, and tree seedlings (stratum 2); large shrubs and tree
saplings (stratum 3); small and large trees (stratum 4). The method provides quantitative estimates
of the vertical and horizontal distribution of vegetation cover and its species composition.
Quadrat Layout
The quadrats were laid out on 1 of 12 randomly preselected transect lines running parallel (offset
to either side by one meter) to one of the azimuths used to establish the point quadrats (Figure 2-2b).
Beginning at the outer edge of the subplot, a square meter quadrat frame was laid down and plant
measurements were made. The frame was then moved to the next contiguous square meter area
along the transect line and the measurements were repeated. The crew measured three to six qua-
drats in this manner on each subplot. The number of quadrats sampled per subplot varied because
the botanist was instructed to sample no more than one hour per subplot; thus, a variable number of
quadrats were sampled depending upon vegetation density and diversity.
Crew Member Responsibilities
The botanist made all measurements for this method.
2-7
-------
Recording
Plot and subplot numbers were initially recorded, along with starting time and (later) ending time.
An index of cover abundance was estimated visually for every species and substrate identified within
each stratum (Daubenmire, 1968, with Bailey and Poulton, 1968 modification). The cover classes we re:
(1) 0 to 1 percent (0.5 percent midpoint), (2) > 1 percent to 5 percent (3.0 percent midpoint), (3) > 5
percent to 25 percent (15.0 percent midpoint), (4) > 25 percent to 50 percent (37.5 percent midpoint),
(5) > 50 percent to 75 percent (62.5 percent midpoint), and (6) > 75 percent (87.5 percent midpoint).
Next, the cover of substrates within each quadrat was estimated by cover class. The substrates were
the same as those used for the pole method. The strata were: (1) 0 m (ground surface), (2) > 0 to
1.5 m, (3) 1.5 to 10 m, and (4) > 10 m. Thus, the completed data for each stratum consisted of entries
in substrate/species, strata, and cover class columns, and the data for a completed quadrat consisted
of a series of species-strata-cover entries for all strata present.
Measurement Quality Objectives (MQOs)
The main sources of procedural error in the quadrat method were inconsistencies in enumerating
all vascular plant species in the sampling areas and subsequently in identifying the plant species, and
to a lesser extent, estimating vegetation cover classes, especially in stratum 4. These errors were
exacerbated by several uncontrollable environmental and site conditions, including dense understory
vegetation, phenological and maturity state of each plant species, and low and thick vegetation
overhead that obscured the view of the taller vegetation. Errors in enumerating and identifying species
were limited by hiring a botanist familiar with the flora in Georgia. Reference marks were placed on
the sampling frame as guides during cover estimation.
For the training session, the MQOs were as follows: difference in total number of species per
subplot, ±10 percent or ± 2 species, whichever was less; difference in estimated cover of species, ±
1 cover class; misidentification of plant genus, no differences; and misidentification of plant species,
1 in 20 (5 percent).
2.4 Analytical Methods
2.4.1 Index Selection and Calculation
The structure of a biological community refers to its species composition and the pattern of
species abundances, where abundance is expressed as frequency, biomass, productivity, or any similar
importance value (Pielou, 1975). Community structure is known to reflect site environment and change
predictably with environmental pollution or disturbance (e.g., Brenchley, 1958; Patrick, 1968; Kempton
and Taylor, 1976). Thus, there is obvious appeal to a summary statistic or index that captures
community structure and is sensitive to environmental changes such as pollution and disturbance.
Advantages of such a summary include quantification, ease of interpretation and communication, and
2-8
-------
ability to evaluate environmental impacts upon communities in different regions, since species names
are discarded (Kempton and Taylor, 1974).
For this study, we selected Hill's (1973) series of diversity measures as summary statistics for
the community structure of vascular plants sampled with the pole and quadrat methods on each pilot
plot. We chose these indices because they are among the most widely used in the ecological literature
and provide diversity numbers that are among the easiest to interpret ecologically (Ludwig and
Reynolds, 1988). Hill's family of diversity is calculated as follows:
where Pi is the relative abundance of the ith species. Three members of this series are in common use
(i.e., A = 0,1, and 2. When A = 0, N0 equals S, or species richness, the total number of species in the
sample regardless of abundance. When A = 1, N, equals the exponent of the Shannon index, eH/
(Shannon and Wiener, 1949). When A = 2, N2 equals the reciprocal of the Simpson index, 1/D
(Simpson, 1949) (Table 2-2). Each of these diversity measures expresses the "effective" number of
species in a sample, a measure of the number of species weighted by their abundance. As A
increases, less weight is placed on the rare species, so typically N0 > N, > N2.
Plot-level Hill diversity indices were calculated by strata using pole and quadrat data (Table 2-2).
Frequency data from all points, or cover data (class midpoints) from all quadrats, were accumulated
across each plot to construct a species list and to determine the relative abundance (p,) of each species
for use in the diversity indices. Diversity calculations were made separately for each stratum. With this
analytical approach, diversity measures indicate the community structure of organisms similar in habitat
or microhabitat, size, life history traits, and resource utilization (Hurlbert, 1971).
The status or trend in the values of Hill's diversity measures is one way to assess ecological
condition or environmental quality. For example, the studies of Brenchley (1958), Patrick (1968), and
Kempton and Taylor (1976) show that communities in a polluted or recently disturbed environment are
typically characterized by a few species having very uneven distributions; these sites would have
relatively low Hill diversity values. In contrast, in more stable environments, communities tend to have
a larger number of species with a much more even species abundance distribution; these communities
have larger Hill diversity numbers. Interpretation of patterns or trends in diversity indices must be done
with caution, however, because numerous environmental factors influence diversity and any strong
relationship between diversity and some environmental impact is best described as correlative, rather
than causative (Magurran, 1988).
2-9
-------
Q
uu
CC
uu
ts
X
LJL
O
UJ
01
LU
I
CC
2
HI
a.
O
§1
OUJ
zo.
ES
UJ O
00
CM
UJ
CO
"eo
!3
1
"55
3
o
CL
o
is
c
•— •
CO
CO
0)
£.
D>
EJ>
<
Formula
Definition
w
c
ui
CD
Q.
j=
CO
CO
2
1
CD
'o
CD
Q.
CO
O CD
ir 5*
o co
r"^ CD
r~ ^
3 CO
1^
O.B
CO
.32
o
CD
Q.
CO
*+—
CD £
.0 co
E *-*
3 '£
C 3
CD ir
ffS.
II
CO Q.
II E
ICO to
.£
TJ
CD
Q.
CO
CO
CO
CD
'o
CD
s-
"o
CD
.C»
•••t CO
?=• 03
CD «
£§
CO
CO
CD
XT
CC
CD
;>
*"
^-^
CM
£•,
to
1
CO
CD
"fX
\J
CD
Q.
CO
.£?c5
1_ ^
—. O
o w
CO.Q
o
s
u
§
T3
Jk
.ic.
*— • —
J5 co
CD CD
w 'o
II CD
CL CO
^*
.t;
™
2
The distribution of abundance
among the species in a comrr
CO
CO
1
1
UJ
>»i CD
"ro £
o-g
CD g
S o^^
^r Q3 O. Q.
s i^i
^ Zl TJ w
T3 = £ CD
CO -^ o"
*~ "*^ di ^
g n c: «
D) — ^ *C
co X2 'g «
•iv rrj f^i fi\
•^ CO (}
> ^ogi
f ®ls
CD <-•£:£
co o o co
ZD Z o co
CD
CD
5 sr-
c •*
0 C53
co c"
SZ 0
~
CD CD
1 1
E | •
CO COO
to to
Ǥ CO S
>;c <° c
= ' - 3? - •
ro -2 S -2 •§.
'to .52 d ~ .SS E
^ CO 3 CO CO CO
" ^ ^ M >^ -g
.2 .« " co •£ CD
g .g g_ n .g B
« 8 8 §88
Variation due to the spatial
arrangement of plant cover
11
0_ Q
2-10
-------
2.4.2 Statistical Analysis of Comparability of Pole and Quadrat Methods
2.4.2.1 Operational Characteristics
Spatial Variation
Semivariograms were computed for the quadrat and vertical pole data, using the method dis-
cussed in Section 1.3. For both data sets, the variables examined were the diversity indices,
mentioned above, calculated for pole or quadrat within each stratum. The vertical structure
measurements were taken on an hexagonal grid within each subplot, and in contiguous quadrats.
Sampling Efficiency
The sampling efficiency for the vertical structure and quadrat protocols was investigated using
the statistical methodology discussed in Section 1.3. This efficiency analysis compares the variability
accrued at each stage of sampling with the added costs of a sample at that stage. The analysis then
produces estimates of the number of samples at each stage required to provide a given overall regional
variance estimate at the lowest overall cost.
Time and Labor Requirements
The elapsed time to complete pole and quadrat measurements was calculated for each subplot
and plot. Paired Student's t-tests (normal distribution) and the Wilcoxon signed ranks test (distribution
free) were used to test the null hypothesis H0: time needed to complete measurements did not differ
by method.
Measurement Error
We analyzed two data sets: (1) data collected by the botanist and remeasured by the auditor,
and (2) data collected and remeasured by the botanist. For each data set, the three diversity indices
described in Section 2.4.1 were calculated using data from the original measurement and the remeas-
urement. Paired Student's t-tests (normal distribution) and the Wilcoxon signed ranks test (distribution
free) were used to test the null hypothesis H0: mean difference between remeasurements = 0.
2.4.2.2 Informational Characteristics
Species Diversity
Plot-level diversity values were calculated using data generated by the pole and quadrat meth-
ods. The measurement data collected with the pole were grouped in the same height strata as the
quadrat data, so that the resulting diversity calculations were comparable. The pole and quadrat
diversity calculations were compared for strata 1 (ground layer, 0 m), 2 (> 0 to 1.5 m), and 3 (> 1.5 to
10 m), but not for stratum 4 (> 10 m). Paired Student's t-tests (normal distribution) and the Wilcoxon
2-11
-------
signed ranks test (distribution free) were used to test the null hypothesis H0: mean difference in
diversity values based on pole and quadrat methods = 0.
Species Accumulation Curves
Plot Level. To evaluate how well the vegetation structure methods sampled the species richness
of each plot, species accumulation curves were constructed by stratum for each plot using point (pole)
and area quadrats (stratum 4 quadrat only). Each plot consisted of the mean number of plant species
accumulated in relationship to sample area or size. The mean number of plant species was calculated
from 100 sequences of samples drawn at random. For example, for a given plot and stratum, 12
quadrats were drawn at random without replacement (e.g., quadrats 2, 9, 4, 8, 10, etc., and the
cumulative number of plant species was recorded for each quadrat. This process was repeated 100
times. Then the cumulative number of species in the 100 quadrats drawn first, second, third, etc., were
averaged and plotted in relationship to quadrat sequence.
Then the SAS nonlinear regression procedure (NLIN) was used to fit several different models to
the species accumulation curves to determine which model form fit the data best. The models used
were: (1) mean species richness (MSR) = a + b * 1n quad (or point) no. (logarithmic), (2) MSR = a
* EXP b/quad (or point) no., and (3) MSR = a * {1-EXP[-b*quad (or point) no.]} negative exponential).
The latter two equations have defined asymptotes, but the first one does not. The goodness of fit of
each model was evaluated by comparing the models' residual mean squares by stratum by plot, and
ranking them in ascending order. Finally, once the best fitting equation had been identified for each
curve, it was used to estimate the number of plant species sampled if 40 point or area quadrats were
measured. The percentage of species sampled with current methods in relation to expanded sampling
(n » 40) was used as a general guideline for estimating how well current methods sample total plant
species richness.
Regional Level. The procedure used to generate plot-level curves was then repeated for the
western Georgia region. First, the total number of substrates and plant species in each stratum was
determined by plot. Then, for a given stratum, the 20 plots were drawn at random without replacement
(e.g., plot nos. 3208571, 3308563, 3108551) and the number of plant species accumulated in
relationship to the number of sample plots. This process was repeated 200 times and the cumulative
numbers of species drawn first, second, third, etc., were averaged and plotted in relationship to plot
sequence. As before, the SAS NLIN procedure was used to fit the same models to the species accu-
mulation curves. Similarly, the percentage of species sampled with current methods in relation to
expanded sampling was used as a general guideline for estimating how well current methods sample
total plant species richness in the region.
2-12
-------
2.5 Results and Discussion
2.5.1 Operational Characteristics
2.5.1.1 Spatial Variation
Semivariograms permit us to visualize the spatial correlation structure, that is, how far away the
measurements can be when the correlation is roughly the same as for two points on opposite sides of
the plot. This provides information useful in deciding whether two data points can be treated as
statistically independent.
The semivariograms looked similar for all strata except stratum 2. In all the strata except stratum
2, there was no sign of significant spatial structure at the resolution of the sampling, due in part to the
irregularity of occurrences for strata 3 and 4, which led to large numbers of missing values in the data.
The lack of apparent spatial structure for stratum 1 in the quadrat measurements may indicate that the
correlation structure is not particularly evident at distances of a meter or more. This does not indicate
that there cannot be spatial structure at distances shorter than one meter.
The semivariograms for stratum 2 consistently showed more correlation between the closest data
points than between the most distant ones for both measurement methods and for all three diversity
measures. This correlation structure indicates that for the data from stratum 2, we cannot consider the
diversity observations for adjacent measurement stations to be independent.
The evidence of interdependence among diversity observations for stratum 2 data from adjacent
measurement stations suggests that it might be beneficial to move the measurements farther apart
within each subplot, as has already been done for quadrat measurements in pilots conducted during
the 1992 field season (the three quadrats per subplot were 7.8 m apart). Furthermore, the evidence
indicates that we must be cautious about assuming statistical independence of observations within a
subplot for the diversity indices for stratum 2. The consequences for subsequent analyses are
discussed in the following paragraphs.
2.5.1.2 Sampling Efficiency
The cost components (Table 1-1) used in the evaluation are based on the average times required
to set up and make measurements on a new plot location, a new subplot in a given plot, and further
subsampling units. The subsampling costs are the times required for surveying the diversity at the
various layers at a measurement station, whether that measurement station is a quadrat or a vertical
pole point. These times were recorded as part of the logistics evaluations of the pilot measurements.
For both the vertical habitat structure and the quadrat methods, a three-stage sampling scheme
was used: plots, subplots within plots, and measurement stations within subplots. The variables
analyzed using this technique are the three diversity measures discussed previously, aggregated within
2-13
-------
strata for each measurement station. Diversity measures were not calculated for stratum 1 (the sub-
strate under pole points), so only strata 2 and 3 were examined for the vertical pole measurement.
Due to a large number of stations with no observations for strata 3 and 4 on the quadrats, it was not
feasible to calculate eH/ or 1/D for these stations; thus strata 3 and 4 were examined for the quadrats
using only the measure S. Remember that for both measurement methods, stratum 2 showed a spatial
correlation structure and hence the variance estimates and resulting computations for stratum 2 must
be viewed as approximate.
The measurement error estimates were calculated for the quadrats only, since quadrat-level
quality assurance data bases were available. Thus the vertical pole data were evaluated using the
equations presented in Section 1 that do not take measurement error into account and the quadrat data
were evaluated using those equations in Section 1 that incorporate measurement error. The
measurement error variances for the quadrats were estimated from two sources. The field crew
member remeasured one quadrat per plot, and the original and remeasured data were compared to
obtain a remeasurement variance.
Also, two complete plots were remeasured during an audit and 24 quadrats were measured by
both the auditor and field crew member. These data were examined for signs of spatial structure
before any variance estimates were computed. Although there were signs of spatial structure in the
data, the differences between the original and audit measurements at each quadrat did not show any
spatial correlation structure. This is to be expected, since the difference in two measurements at a
quadrat should be due only to measurement error, including crew-to-crew variability, and should not
be affected by the spatial structure of the vegetation at other quadrats on the plot.
The two measurement error estimates were pooled to generate a more reliable error estimate
with more degrees of freedom, using the formula:
n.
712-2
where s\ is the error variance used in the calculations, s? (i = 1,2) is the two variances calculated from
the audits and the remeasurements, and ns -1 is the respective degrees of freedom for those error esti-
mates. This formula is equivalent to using a Kalman filter (Chatfield, 1989) to combine the two error
estimates. It also has the form of the standard maximum likelihood estimates for a pooled variance
to be used in a one-way analysis of variance with two levels.
Tables 2-3 and 2-4 show the mean squares and measurement error estimates used in the anal-
yses for vertical structure and the quadrats. See Table 1-1 for the costs of the sampling stages.
2-14
-------
TABLE 2-3. MEAN SQUARE AND ERROR ESTIMATES FOR VERTICAL STRUCTURE
Stratum 2
1/D
eH'
S
Stratum 3
1/D
eH'
S
Stratum 1
1/D
eH'
S
Stratum 2
1/D
eH'
S
Stratum 3
S
Stratum 4
S
Plot Mean
Square
2.38
2.75
3.27
0.773
0.918
1.26
0.324
0.471
3.08
4.74
10.92
46.86
11.46
6.27
Subplot Mean
Square
Pole
0.612
0.690
0.846
0.701
0.908
1.30
Quadrat
0.0698
0.162
4.21
4.22
8.22
11.38
5.19
0.0440
Residual Mean
Square
0.753
0.808
0.880
0.297
0.376
0.552
0.0621
0.0946
1.04
2.36
3.50
8.94
1.37
0.0875
Error Variance
0.0288
0.0513
0.477
0.607
0.741
0.548
0.282
0.0663
Using the formulas in Section 1.2, the mean squares and variances (Table 2-3) were used to
compute the estimates shown in Table 2-4. These estimates are the optimal number of subplots and
the optimal number of measurement stations per subplot. A measurement station is defined as a ver-
tical pole point for the pole method or a quadrat-level diversity estimate for the quadrat method.
2-15
-------
TABLE 2-4. ESTIMATED OPTIMAL SUBSAMPLE SIZES
Plot
Stratum 2
1/D
eH>
S
Stratum 3
1/D
eH>
S
Quadrat
Stratum 1
1/D
eH>
S
Stratum 2
1/D
eH'
S
Stratum 3
S
Stratum 4
S
Number of
Subplots
1.90
1.88
1.93
4.18
4.43
4.54
1.74
2.33
5.61
3.96
3.61
1.76
2.75
0.31
Stations within Each Subplot
2.50
2.43
2.27
1.38
1.36
1.38
0.55
0.39
0.27
0.50
0.45
0.72
0.34
0.55
Rounding up the estimates, we see that, given an appropriate number of plots, four subplots per
plot and one or two quadrats per subplot are enough for regional estimation. Similarly, four or five
2-16
-------
subplots with three vertical pole measurements per subplot would provide enough information for
regional estimation. More data per subplot are currently being collected, with an appropriately smaller
plot-level variance estimate. So, based on these calculations, the plot design appears to be adequate
for this indicator.
2.5.1.3 Time and Labor Requirements
The quadrat method required one person, whereas the pole method required two people. Con-
sequently, the elapsed times were multiplied by the number of persons needed for each method, to
arrive at person-time, an estimate of the total effort required to complete measurements on each plot.
Analysis of paired subplot person-times (n = 70) showed that the pole method took significantly longer
(43 minutes) than the quadrat method (25 minutes; paired t-test, t = -7.17, prot>. > t 0.0001). The
mean elapsed time per subplot (n = 20) was 92 minutes for the quadrat method and 155 minutes for
the pole method.
2.5.1.4 Measurement Error
Botanist Measurements Remeasured by Auditor
The diversity indices S, eH', and 1/D calculated for each stratum from botanist and auditor
measurements were not significantly different (paired t-tests, prob. > t ranged from 0.162 to 0.795).
The S values for stratum 1 were the only measurements that approached significant difference (t =
1.926, prob. > 10.067). This difference was related to the identification of substrates and nonvascular
plants in stratum 1, not vascular plant species.
Botanist Measurements Remeasured by Botanist
The diversity indices S, eH', and 1/D calculated for each stratum from botanist measurements and
remeasurements were not significantly different (paired t-tests, prob. > t ranged from 0.141 to 1.000).
Again, the S values for stratum 1 were the only measurements that approached significant difference
(t = 2.092, prob. > t 0.055).
2.5.2 Informational Characteristics
2.5.2.1 Species Diversity Indices
The null hypothesis H0: mean difference in diversity values based on pole and quadrat methods
= 0 was rejected for S in all three strata and for eH' in stratum 3 (Table 2-4, paired t-tests, prob. > t
ranged from 0.0001 to 0.023). In strata 1 and 2, S values based on the quadrat method were greater
than those based on the pole method; the higher diversity numbers with the quadrat method suggests
better performance than the pole method in capturing the species richness of stratum 2. in contrast,
2-17
-------
in stratum 3, both S and eH> values based on the quadrat method were less than those based on the
pole method, suggesting better performance with the pole than the quadrat method. Meanwhile, 1/D
values did not differ significantly between methods, although the differences in strata 2 and 3
approached significance at the 0.05 level (Table 2-5, paired t-tests, prob. > t ranged from 0.084 to
0.096).
TABLE 2-5. DIFFERENCES IN HILL DIVERSITY NUMBERS FROM THE QUADRAT (Q) AND
POLE (P) METHODS BY VEGETATION STRATUM3
HILL DIVERSITY NUMBER
Stratum
1
2
3
A = 0 (S)
Q>P
P = 0.0001
Q>P
P = 0.0001
P>Q
P = 0.0132
A = 1 (e"')
Q = P
P = 0.2105
P = Q
P = 0.8007
P>Q
P = 0.0234
4 = 2 (1/D)
Q = P
P = 0.341 7
P = Q
P = 0.0835
P = Q
P = 0.0958
1 Paired West.
Two patterns emerge from these results. First, as A changes from 0 to 2, (i.e., as the
weights given to rare species decline), the significance of the differences between methods decreases.
Thus Hill diversity numbers based on lower A values are more likely to distinguish smaller differences
between methods. Second, as the stratum changes from 1 to 3, superior performance in capturing
species richness shifts from the quadrat method to the pole method. Thus, if species diversity is
relatively high, as in stratum 2, the quadrat method should be employed; but if species diversity is
relatively low, as in stratum 3, the pole method may be employed.
2.5.2.2 Species Accumulation Curves
Plot Level
Methods Comparability. We compared species accumulation curves from the quadrat and pole
methods to determine which method was better at capturing the vascular plant species richness of each
stratum. At each plot, there is some unknown plant species richness S. Typically this total species
richness is not fully captured by sampling; however, the method that captures relatively more plant
species may be judged better or superior because its sample S is relatively closer to a total S for the
plot. By this criterion, we judged the quadrat method to be superior to the pole method in strata 1 and
2, but within stratum 3, the sample Swas similar for both methods [Figure 2-3(a-d)].
2-18
-------
O 3
"a
§1
C/}
a
?£g~
o £
CO
G
'o
a.
ssiosdg
0
II
5
ce '
CO
CO <
CO
o
CO <
it
o .
2
£-
0 . <
o,
'
>
^ O (O O IO o lA'A
o
o
ca
o
CO
. o
Z~- !-,
1
O 3
<° 2:
OJ
"o.
'g 1
o -^
"* H
**-4
o
o ^
CO
o
M
0
0
CO CO CM CU ~H — «
Q
CM"
E
a
~ £5
£! «
II . — .
3 S,
a "
•O p
3 -3
or ~
! ^
s
^ . .
S "o
II ^3
t!T 03
3 E
O _,--.
CU CC
if
cr
E
co
CD
a
CU
jo
o
^J
a.
O
O
co
punoao jo
satoadg
jo
2-19
g>
-------
Goodness of Fit of Model. We further analyzed the species accumulation curves of the quadrat
method by testing for goodness of fit to several different empirical models. In subsection 2.4.2.2, we
explain that the known values for comparison were the species numbers predicted at n = 40 quadrats,
using three nonlinear equations. The asymptotic, negative exponential function described the species
accumulation curves best in strata with high species richness, such as strata 2 and 3 (mean plant
species richness of 33 and 9, respectively). In contrast, the non-asymptotic, logarithmic function was
slightly better than the negative exponential function in strata with low species richness, such as strata
1 and 4 (mean plant species richness of 6 and 3, respectively). The best fitting, parameterized models
served two functions: (1) a convenient summary description of each species accumulation curve and
(2) a mechanism for estimating species richness in relation to changing sample size (see next para-
graph).
Efficiency of Sampling Vascular Plant Species Richness. We evaluated the efficiency of the
quadrat method in sampling plant species richness in different strata by dividing the species richness
observed using the current sampling strategy (mean quadrats per plot = 11.5) by the predicted species
richness with 40 quadrats per plot, expressed as a percent. The best fitting nonlinear models,
described in the previous paragraph, were used to predict species richness values. The mean effici-
encies of the current quadrat sampling strategy in relation to the logarithmic function were 77 percent
for stratum 1, 72 percent for stratum 2, 72 percent for stratum 3, and 66 percent for stratum 4.
Meanwhile, the mean efficiencies of the current quadrat sampling strategy in relation to the negative
exponential function were 96 percent for stratum 1,87 percent for stratum 2,94 percent for stratum 3,
and 99 percent for stratum 4. Lower mean sampling efficiencies were indicated for the logarithmic
function than for the negative exponential function, because the negative exponential function was
asymptotic, but the logarithmic function was not. Thus, estimating conservatively, at least 70 percent
of the vascular plant species richness was sampled with the current quadrat sampling strategy.
Regional Level
Methods Comparability. Regional species accumulation curves from the quadrat and pole
methods supported the plot-level results by showing the clear superiority of the quadrat method for
sampling regional plant species richness in stratum 2, and the similarity of the methods for sampling
within strata 1 and 3 [Figure 2-4(a-d)].
Goodness of Fit of Models. The regional species accumulation curves of the quadrat method
were also tested for goodness of fit with the same empirical models used for plot-level data. The
negative exponential function described the species accumulation curves much better than either the
logarithmic or the alternative asymptotic function in all but stratum 1, for which the logarithmic function
was slightly better. The mean regional species richness was lowest in stratum 1 (15), followed by
strata 4 (30), 3 (77), and 2 (305). These results followed the similar patterns found for the plot-level
curves.
2-20
-------
. 2 •S.
O o.
1
= t
o o
cy o
CO CT
OOOOOOOOO'
-------
Efficiency of Sampling Vascular Plant Species Richness. We evaluated the efficiency of the
current quadrat method in sampling regional plant species richness in different strata by the same
procedure used for the plot-level results. Using predicted values from the negative exponential
function, the mean efficiencies were 107 percent for stratum 1, 79 percent for stratum 2, 85 percent
for stratum 3, and 70 percent for stratum 4. Using predicted values from the logarithmic function, which
had much poorer fits except for stratum 1, the mean efficiencies were 79 percent for stratum 1, 76
percent for stratum 2, 75 percent for stratum 3, and 75 percent for stratum 4. Thus, all nonlinear
models suggested similar mean sampling efficiencies, and overall results indicate that about 75 percent
of the regional vascular plant species richness was sampled with the current sampling strategy.
2.6 Summary of Results and Recommendations
2.6.1 Operational Results
Measurements based on four subplots per plot and one or two quadrats per subplot were
deemed optimal for regional estimation of the S, eH', and 1/D diversity indices. Similarly, four or five
subplots with three vertical pole measurements per subplot were optimal for similar regional diversity
estimations based on the pole method. Thus, the current plot design and the current number of
samples per subplot were adequate, although perhaps fewer measurements could be taken on each
subplot for the vegetation structure indicator. Although any of these diversity indices can be calculated
based on any number of samples, the reliability of such diversity calculation increases with sample size
(Magurran, 1988). Species accumulation curves at both the plot and regional levels indicated that even
with current sample sizes, plant species richness was undersampled, varying from 66 percent to 98
percent of total plant species richness, depending upon stratum. Consequently, the suggestion of
reducing samples per subplot must be thoroughly evaluated before implementation.
The quadrat method actually required less total effort (person-time) than the pole method, since
the quadrat method required only one person and the pole method required two people. Therefore,
the quadrat method has a distinct practical advantage since the field work can be completed more
quickly and it is easier for groups to collect compatible data.
Quadrat diversity calculations were both reproducible (observer versus auditor) and repeatable
(observer versus observer). Even though pole remeasurements were not available from 1991, 1990
analyses showed that trained crews could reproduce one another's measurements (Riitters et al.,
1991). So, although both field methods have known procedural errors, the errors do not significantly
affect the reproducibility of the diversity indices derived from their respective data.
2-22
-------
2.6.2 Informational Results
Species richness (S) varied significantly between pole and quadrat methods. For strata 1 and
2, S estimates were greater with the quadrat method than with the pole method; opposite results were
found for stratum 3. Meanwhile, the eH/ and 1/D indices were methodologically comparable.
Species accumulation cuives indicated the quadrat method sampled plant species richness better
than the pole method for strata 1 and 2 at both the plot and regional levels. In contrast, the pole
method was similar or slightly superior to the quadrat method in sampling species richness within
stratum 3.
Species accumulation curves based on quadrat sampling were best described overall by the
negative exponential function, especially in strata with higher species richness, such as strata 2 and
3. Meanwhile, the logarithmic function was similar or slightly better than the negative exponential
function in strata with low species richness, such as strata 1 and 4.
The quadrat method typically sampled an estimated 70 percent to 80 percent (range 66 percent
to 107 percent) of total plant species richness at the plot and regional levels, depending upon stratum
and predictive equation. The effect of undersampling was greater for strata 2 and 3, where species
richness was higher, compared to strata 1 and 4.
2.6.3 Recommendations
We recommend the quadrat method for measuring vegetation structure in future Forest Health
Monitoring field seasons. The overall plant species richness and diversity in forests might be more
efficiently estimated by implementing one or more of the following sampling refinements:
Sampling strata 1 and 4 less often (e.g., every second or third quadrat) because of lower
species richness.
Sampling stratum 3 more intensively (e.g., use pole to determine presence/absence and as
reference during cover estimates), but less often (e.g., alternating quadrats), because of
moderate species richness.
Sampling stratum 2 more often (e.g., four or five quadrats per subplot), to more fully capture (>
80%) species richness.
1.
2-23
-------
-------
SECTION 3
PHOTOSYNTHETICALLY ACTIVE RADIATION (PAR)
J. G. Isebrands, S. J. Steele, and K. H. Riitters
3.1 Introduction
Measurements of solar radiation intercepted by the canopy are fundamental to the interpretation
of the productivity and function of plant communities (Norman and Campbell, 1989). Photosynthetically
active radiation (PAR) is the quantity of light between the 400-700 nm wavebands of the spectrum, and
is the part of the spectrum used by plants for photosynthesis. We can estimate the percentage of PAR
transmitted by a plant canopy by calculating the ratio of PAR under the canopy to ambient incoming
PAR. This ratio can be related to canopy condition as well as leaf area index and can be combined
with growth measurements to estimate growth efficiency, an important indicator of forest health (Waring
and Schlesinger, 1985). We can also combine PAR with companion measurements of vegetation
structure and/or remote sensing to assess canopy condition with a multivariate indicator approach.
Reliable measurements of transmitted solar radiation are difficult to achieve on the ground, and
are typically characterized by significant temporal and spatial variability. Ambient PAR measurements
vary, depending upon cloud conditions, time of day, and solar angle (i.e., location and time of season).
Thus, obtaining a reliable estimate of %TPAR over a range of temporal and spatial conditions is a
challenge. The Georgia pilot study was designed to test the feasibility of using a portable integrating
radiometer called a ceptometer (Decagon Devices, Inc., Pullman, Washington) for estimating PAR and
transmitted PAR across the various forest types and stand conditions in Georgia.
3.2 Objectives
The general objective of this and related research was to develop and evaluate PAR as an
indicator of canopy condition that can be used in monitoring the health and ecological condition of U.S.
forests. This objective includes evaluating alternate methods of making PAR measurements and
developing knowledge that will enable the Forest Health Monitoring (FHM) program to interpret the
data.
The regional pilot and demonstration tests focus on developing a suite of concurrently measured
indicators, including PAR, in an operational setting. This rationale leads to the following specific
objectives of the PAR Georgia Pilot study:
1. Develop an efficient and reliable method of using a ceptometer and quantum sensors for
measuring forest canopy light (PAR) environments under different stand conditions.
2. Develop and test procedures for linking PAR measurements to vegetation structure
measurements.
3-1
-------
3. Develop and test procedures for linking ground measurements of PAR to photointerpreted
measures of stand and canopy attributes.
The Georgia Pilot field study helped accomplish the first specific objective by providing the data
needed to (1) evaluate and recommend new or modified field sampling procedures, instrument
modifications, and field data handling procedures for measuring PAR with a ceptometer based on
experiences gained from the 1990 20/20 pilot study, and (2) recommend efficient sampling procedures
to achieve specified precision for various forest types and stand conditions. The Georgia Pilot met the
second objective by measuring PAR and vegetation structure on common sample points on the
subplots on the same day and relating the measurements quantitatively. The third specific objective
of linking photo- and ground-based measures is a longer term objective. The spatially-referenced PAR
measurements will be correlated with forest canopy attributes derived from the 1:12000- and 1:6000-
scale photography when those photos are available.
3.3 Related 1991 PAR Studies
Photosynthetically active radiation was initially measured as part of the 1990 FHM 20/20 study
in the Northeast and Southeast. Results of that study indicated that PAR could be measured reliably
with a ceptometer on days with uniform sky conditions (e.g., clear or cloudy), but that variable
conditions were more problematic. This experience prompted us to modify the methods employed in
the Georgia Pilot.
FHM Pilot sites
>{< Off-Frame research sites
Rgure 3-1. Locations of PAR research sites and pilot sites in 1991 Forest Health Monitoring.
3-2
-------
In 1991 a smaller scale PAR companion pilot study was conducted in California and Colorado
to test the applicability of PAR methods to western forest stands and conditions. Other PAR activities
were conducted by a PAR indicator team consisting of interested scientists from throughout the country.
Their goal was to evaluate PAR as a potential indicator for the FHM Program. In 1991, research PAR
studies were conducted in Wisconsin, Michigan, Tennessee, North Carolina, Colorado, and Idaho
(Figure 3-1). In total, the PAR team measured PAR on a wide variety of forest types, including
hardwoods and conifers, at many different geographical locations in the East (Table 3-1) and the West
TABLE 3-1. EMAP-FHM PAR INDICATOR: 1991 GEORGIA PILOT PLOT SUMMARY
Plot
#
1
2
3
4
5
6
7
8
9
10
11
12
13
Hex#
3408521
3408435
3308385
3308481
3308476
3308563
3308318
3208365
3208571
3108368
3108551
3108431
3008467
Date
Sampled
6-17
6-19
6-25
6-30
7-1
7-2
7-8
7-9
7-13
7-19
7-22
7-23
7-24
Forest Type
Oak-Gum-Cypress
Oak-Pine, Oak-Hickory
Oak-Pine, Oak-Hickory
Loblolly
Oak-Pine, Oak-Hickory
Loblolly
Oak-Gum-Cypress
Loblolly
Loblolly
Loblolly
Loblolly
Oak-Pine, Oak-Hickory
Oak-Pine, Oak-Hickory
Cloudiness
Index
PAR/Hour
1200-1400
Hrs
102585
69524
44172
57274*
103845*
98646
57513
112296
73166
114347
82615*
89370*
71676
Ambient station on less than one hour; value extrapolated to one hour
3-3
-------
3.4 Methods
The PAR measurements were to be made on 20 locations selected in the western half of
Georgia. The plot selection rules were dependent on the needs of all participating indicators, as well
as on logistical constraints. To meet the objectives of the PAR portion of the pilot project, the 20
selected stands were to be representative of available locations (to provide estimates of expected
regional variability of terrain, forest type, and stand conditions). The PAR measurements were made
during a six-week timef rame beginning about June 15, after full canopy development and before canopy
senescence. Measurements were taken at only 13 locations, due to logistics, weather, and equipment
problems (Figure 3-2; Table 3-1).
The on-plot sampling scheme for PAR measurements was done on an ambitious sampling grid
superimposed on the standard FHM four-point fixed area subplot clusters. Under-canopy PAR was
measured at 19 points on each subplot for a total of 76 sample points per plot (Figure 3-3). Moreover,
to achieve uniformity, the PAR team adopted this standard PAR sampling grid for all PAR studies
across the country. Under-canopy PAR measurements were made with a ceptometer. Specific
operating details can be found in the FHM Field Methods Guide. Synchronized ambient PAR
measurements were also made in the open at each site with quantum sensors. Transmitted PAR was
then calculated as the ratio of PAR under canopy to PAR in the open, expressed as a percentage.
The PAR measurements were made during a standard sampling window from 1100 hrs to 1300
hrs standard zone time (i.e., 1200 and 1400 hrs daylight savings time). This window was necessary
to ensure accurate measurement of the percentage of transmitted PAR at each site.
3.5 Results
Figure 3-4 shows a typical PAR data set from a plot in Georgia; under-canopy PAR and ambient
PAR are plotted for each of the 76 sample points. The data are from hexagon number 3408521 which
contains an oak-gum-cypress forest type. Figure 3-5 shows transmitted PAR (calculated from data in
Figure 3-4) for the same plot. This graph illustrates the kind of variability in canopy condition that the
PAR indicator can detect within a FHM plot.
Table 3-2 summarizes all the transmitted PAR data by plot location on a subplot and plot basis.
The data can be cross referenced to Table 3-1, which summarizes all locations according to hexagon
number, date, location, forest type, and cloudiness index on the dates sampled.
In Georgia, we experienced some early problems with our instrument reliability, but most were
corrected immediately. For example, the ambient PAR station did not function well when it was moved
often or when it experienced intense heat. Inexpensive dataloggers and fabricated quantum sensors
were used at the beginning of the field season rather than standard factory equipment. The quantum
sensors had to be replaced by factory equipment in the first week; the datalogger was not replaced
3-4
-------
and continued to cause problems throughout the year. The Decagon ceptometer performed well for
all plots except one. After minor repairs, that problem was corrected and no further problems occurred.
The PAR equipment was inoperable only during heavy rain.
Figure 3-2. Locations of the 1991 Georgia Pilot PAR Plots.
3-5
-------
SAMPLE #/LOCA TION
1-19 SUBPLOT 1
20-38 SUBPLOT2
39-57 SUBPLOTS
58-76 SUBPLOT4
suaftor
North
Figure 3-3. PAR sampling scheme for Georgia pilot study in 1991. Subplot example on left shows
sampling point layout according to azimuth. Diagram on right shows 19-point grid for
each subplot of the 4 fixed area subplot cluster. A total of 76 points were sampled per
location.
As mentioned earlier, 13 of the 20 proposed locations in the Georgia pilot have complete PAR
data. The seven missing plots resulted from:
Four ambient station datalogger malfunctions.
One rainy day.
One ceptometer malfunction.
One absence of a suitable ambient station.
3-6
-------
w
•D
O
jc.
h-
Sample Point Number
Figure 3-4. Under-canopy and ambient PAR on a plot (3408521) in the 1991
Georgia pilot showing the amount of variability within plots.
In the pilot studies, as well as at the other research sites, we investigated the optimal number
ot PAR sampling points needed per subplot. Although 19 sample points were measured throughout
all 1991 studies, we wanted to determine how well we could estimate PAR with a smaller sample size,
based on only the center and 6 corner points of each subplot hexagon (e.g., points 1, 2, 4, 6, 8, 10,
and 12 of subplot 1 in Figure 3-3). The advantages of using seven sample points were that the
sampling time would be considerably less and the environmental impact would be lower during
sampling, due to less trampling of understory vegetation. Based on the means, standard errors, and
variances shown in Table 3-2, our 1991 studies clearly showed that seven points gave as good an
estimate of transmitted PAR as 19 points. This result meant that in 1992 we could reduce our sample
points by two-thirds and reduce the PAR indicator timeframe from 2 hours to 1 hour or less, and, at the
same time, do minimal understory vegetation damage to the site.
3-7
-------
CL
h-
tfl
ui
CL
OJ
v-^
D_
55
U.
O
'MEANS
n
o!
EC
S
o
u.
cL
•£
01
c\l
CO
11,1
_J
S S. »"
£-!
• .
'a
11
5|
|
i
i
n
t
i
o
i
*
CM
f
1
3
Ifi
i-
i
CM
*r c"
ss
*SLfe>
o o>
•e co
ss
11
"*?"S*
*^w>
si
rC o
§.=.
V O
at r^
en r-.
cTCf
OJ CM
gs
=•=
is
2 !«•
-
11
CO
II
in"* en*
CO CM
CD h»
S§
n ui
co in
11
CM *-
cn to
52
•«• "C
II
52. 2.
co tri
5— "**
CM
r^ ri
to
II
II
SS
11
2.2*.
o co.
O> CO
~ ~
2.—
0.2.
CO CM
0.2.
0.0.
S5
o. o.
o o
CO O
in CD
£ i**
co
1
CM
sl
2.2.
— o
in «r
s^
uTc"?
£.£.
0,53.
CM -^
5.S.
CM ^;
CD in
CM, CM.
0 G,
tN. CO
*& -
to*"
STifn1
S.S.
V CO
*- o
2-
CM
SI
*-
T** ff
ElS.
01 *^
ss
ST eo*
«g- to
gj S
cq cq
CO O
CO N
s g
S2.£.
CO T-
s?
__
5 2
S" Jq"
CM in
si
5S"
5 5
in co
2 i^
CO
•B.-B."Q.£
« ic to o
« (/> tf> -Q
"o *o *H 3
ft *t * Z
S » • '
£, St B B
1 '§ 'I 'I
| £[£ £
°* £. £ £
S co co w
3 *o "o "o
3-8
-------
20 40 60 76
Sample Point Number
Figure 3-5. Percent Transmitted PAR from data in Figure 3-4;
%TPAR = under-canopy PAR/ambient PAR x 100.
Note the variability over the entire plot.
Our off-frame PAR studies in 1991 showed that transmitted PAR can be dramatically affected
by cloudiness. It was already well known that PAR is affected by sun angle, which varies by
geographical location and time of year. These factors influence important indicator criteria such as
index period stability and signal-to-noise ratio. Plots of a cloudiness index (e.g., integrated PAR over
measurement period) for the Georgia Pilot are given in Figure 3-6. Values over 100,000 indicate clear
days; as expected in Georgia, most days encountered by the crews were cloudy. Cloudiness index
is a continuous variable that was measured on every site. Therefore, cloudiness could be used as a
covariate or a regression parameter to adjust the transmitted PAR values from the Georgia Pilot,
reducing noise in the data.
3-9
-------
130
Plot Number
Figure 3-6. Integrated PAR/hour on the 1991 Georgia pilot PAR plots.
Units are micromoies/meter2/hour.
Data from the pilots and the other PAR research indicated that diffuse PAR should be measured
in the open areas in addition to ambient PAR. Our off-frame preliminary results suggest that diffuse
PAR can be used to estimate canopy condition and leaf area index without the variability experienced
on cloudy days with ambient PAR only. Diffuse PAR allows us to calculate beam fraction, which is
equal to 1 - (diffuse PAR/ambient PAR). Diffuse PAR is measured quite simply by shading a
companion quantum sensor co-located in the open areas with the quantum sensor for measuring
ambient PAR, with a small disk. Beam fraction was very stable over a wide range of cloudiness
conditions in the off-frame PAR studies and is being measured in all 1992 PAR studies.
Additional PAR team activities in 1991 included:
Comparing of different algorithms used in defining ambient PAR and mean-value estimators
of transmitted PAR.
3-10
-------
Comparing of different methods and models for estimating LAI. The transmitted PAR data
was incorporated into different models and the LAI estimates were compared to results
received from a plant canopy analyzer on the same sites (LI-COR, Inc.; Lincoln, Nebraska).
Examining shoot-level measurements and how they related to stand-level PAR transmission
measurements to see if these measurements can assist in detecting significant change in
PAR transmission over time.
In summary, the information gained from the Georgia and Western pilots and the other PAR
research findings have advanced the PAR indicator methodology significantly. In 1992, we have
incorporated better, more reliable equipment and included a shading device to collect diffuse PAR at
the ambient PAR station. Moreover, we are taking the measurements on only 7 points per subplot
rather than the 19 previously used. We hope that these advances will enable PAR to become more
repeatable, reliable, and responsive to changes over time in a variety of forest ecosystems throughout
the country.
3-11
-------
-------
SECTION 4
DENDROCHRONOLOGY
T. Droessler
4.1 Introduction
The tree cores extracted in Georgia in 1991 were used for determining annual diameter at
breast height (dbh) growth. Dendrochronology, the systematic study of annual increment obtained from
growth rings in trees, was being considered as an indicator in 1991.
4.2 Objective and Scope
The objectives for dendrochronology were to determine if the sampling intensity and tree
selection protocols were adequate for quantifying dbh growth rates and trends on a regional basis (in
this case, for the state of Georgia or regions within).
4.3 Methods
4.3.1 Sampling
Briefly, a plot consists of a cluster of four subplots, with a center subplot (subplot 1) and three
outlying subplots at 0, 120, and 240 degrees (subplots 2, 3, and 4, respectively). The subplot radius
was 7.3 m (24 ft) and the distance from the center of subplot 1 to the centers of the outlying subplots
was 36.6 m (120 ft). Figure 3-3 shows a schematic of the plot layout. See the FHM Field Methods
Guide for a more detailed description of plot structure, layout, and sampling methods.
Tree cores were collected on subplots 2 and 4 according to the following protocol. A random
compass bearing was obtained from a portable data recorder and followed to the intersection of the
subplot boundary at 7.3 m (24 It), forming an inner sampling boundary. An outer sampling boundary
was positioned at 11.0 m (36 ft). Proceeding in a clockwise direction between the sampling boun-
daries, the first dominant or co-dominant live tree 12.7 cm or greater (5 in) was identified as the sample
tree. The azimuth and distance of the sample tree from the subplot center were recorded.
Tree cores were extracted at a fixed height, usually 1.37 m (4.5 ft). If the sample tree was on
a slope, the first core was extracted parallel with the slope. If the tree leaned, the core was extracted
90° from the lean. A second core was extracted 90° from the first core. The cores were to contain the
pith and be free of knots, pitch pockets, and compression wood. Additional cores were extracted if
4-1
-------
necessary. A second dominant or co-dominant sample tree was located within the described sampling
boundaries by proceeding in a clockwise direction from the first sample tree. A maximum of eight cores
were extracted at a plot (two cores per tree from two trees on two subplots). The cores were stored
in labelled straws and the straws stored in a tube. The tubes were mailed to the U.S. EPA laboratory
in Las Vegas, Nevada, for tracking, drying, and storage.
4.3.2 Preparation and Measurement
The preparation included drying, stabilizing the core by mounting it in a wood stick, and
machine and hand sanding the core so the growth rings were clearly visible on an even, flat surface.
Stokes and Smiley (1968) present a formal description of tree core preparation. Cook and Kairiukstis
(1990) provide several sections dealing with core handling.
Tree cores were glued into wooden mounts, machine and hand sanded, and dated by counting
from the bark to the pith. The year adjacent to the bark was known as 1991.
Ring widths were measured to the nearest 2 urn (0.0001 in), with an incremental measuring
machine equipped with a linear glass encoder. The encoder was interfaced with a microcomputer
containing software that recorded ring widths by year for each core. As each ring width was measured,
it was simultaneously plotted on the screen against a composite of other cores at the same location.
A stereomicroscope in combination with a monitor was used to measure ring widths. The
magnified core image was transmitted to a monitor via a video camera. An electronic cross-hair
projected onto the monitor screen was the reference point for measuring a ring width. No attempt was
made to determine the number of rings not sampled or the distance to an estimated pith position.
4.3.3 Quality Assurance Procedures
All cores were first hand-dated from the bark end to the pith end, with decade years marked
directly on the core surface. As the cores were measured, the measurement year was compared to
the marked decade for agreement and visually compared to a composite plot of all previously measured
cores on the same plot. Any obvious discrepancies were examined and, if necessary, corrected at the
time of measurement. A minimum of five percent of the cores were randomly selected and measured
by two technicians. The correlation between the independent measurement data for each core was
calculated. In addition, the cores were shipped to the U.S.D.A. Forest Service Institute for Quantitative
Studies in New Orleans for an independent measurement.
4-2
-------
4.3.4 Analyses
The data used for analysis consisted of 138 tree cores from 84 trees from 33 plots. Total tree
age was not estimated from the cores because the pith was usually not included. The length of record
ranged from 2 to 99 years.
Van Deusen (1992) used two grouping levels to enhance a graphical display and analysis of
growth trends in natural loblolly pine (Pinus taeda L) in the southeastern United States. The first
grouping was based on similar median stand age estimated from the cores. The second grouping was
by 10-year age classes. The graphical results indicated that several trends could be distinguished from
the cores.
For the 1991 Georgia tree cores, the raw increment by year was plotted for each core so that
the consistency within and between trees on a subplot and plot could be visually compared. The
within-tree cores exhibited similar growth magnitude and trends, although with some variability. The
average increment by year, calculated from two cores per tree (if two cores were taken), was plotted
by tree. Next, the data were subset to pine species for which the length of record was 19 years or
greater. The average increment by year plots were compared between pine trees of approximately the
same age on a subplot and were found to exhibit similar growth patterns. The plot average increment
by year was then calculated and plotted.
4.4 Results and Discussion
4.4.1 Quality Assurance
Nine cores were randomly selected for remeasurement by a different technician. The
correlation coefficient between the two measurements was 0.99 for eight of the nine cores and 0.78
for the remaining core.
The independent laboratory measurement by the U.S.D.A. Forest Service laboratory in New
Orleans encountered a few potential false rings. Since a potential false ring does not obscure regional
trends in growth, the data were considered adequate for the purpose of this report.
4.4.2 Regional Growth Patterns
Figures 4-1 and 4-2 are representative of two common pine growth trends for trees approxi-
mately 40 years and 90 years old, respectively. Figure 4-1 depicts a tree with high early growth rates
which then sharply decline; the growth trend would be expected to level off in the next few years. Such
a trend would be expected for a tree facing increasing competition from stand dynamics typical of
4-3
-------
plantations. Figure 4-2 shows a tree with a steady or level growth trend with periods of increasing and
decreasing growth rates about the trend; such a trend would be typical of natural pine stands.
12
10-
8-
E
CD
E 6-
E
o
JZ
Ct
2-
Hexid=3108431 Sub=4Tree=1
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990
Year
Figure 4-1. Example 40-year growth trend.
A formal analysis of growth trends is beyond the scope of this report. The graphical approach
was sufficient for showing the utility of tree cores for depicting regional growth trends from which
expectations of future growth rates could be determined.
4-4
-------
E
E
Hexid=3408376 Sub=2 Tree=2
jc
Q
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990
Year
Figure 4-2. Example 90-year growth trend.
4.4.3 Statistical Considerations
The sampling efficiency for the dendrochronology protocols was investigated using the
statistical methodology discussed in Section 1, Statistical Methodology. This efficiency analysis
compares the variability accrued at each sampling stage with the added costs of a sample at that stage.
-The analysis then produces estimates of the number of samples at each stage to provide a given
overall regional variance estimate at the lowest overall cost.
The cost components (Table 4-1) used in the evaluation are based on the average times
required to set up and make measurements on a new plot location, a new subplot in a given plot, and
further subsampling units. The subsampling costs are the times required for locating and coring sample
trees. These times were recorded as part of the logistics evaluations of the pilot measurements.
For dendrochronology, a four-stage sampling scheme was originally proposed: plots, subplots
within plots, trees within subplots, and cores within trees. The variables analyzed using this technique
are the average core increments for the last five, ten, and fifteen years, with 1990 being the last full
year available.
Analysis of variance components for the four-stage model led to a design alteration. Essentially
all the variability between subplots was explained by tree-to-tree variability. Not only were the mean
squares not significantly different from zero, but the variance estimates were consistently negative.
This led to an alternative model: plots, trees within plots, and cores within trees. The cost component
of trees within plots (Table 4-1) encompasses the time required to establish the subplot and select an
appropriate sample tree. Over 80 percent of this time is due to the time required for sample tree
selection.
4-5
-------
Table 4-1 shows the costs, mean squares, and measurement error estimates used in this
analysis. The measurement error estimates were calculated from duplicate readings on tree core
Increments.
TABLE 4-1. COST, MEAN SQUARE, AND ERROR ESTIMATES FOR DENDROCHRONOLOGY
Effect
Plot
Subplot/Tree
Core
Meas. Error
Cost($)
3.33
0.66
0.08
-
5- Year Average
14.40
5.85
0.369
0.000634
10- Year Average
32.62
4.76
0.324
0.000138
15- Year
Average
4.94
1.58
0.121
0.000091
Using the formulas given in Section 1, these values gave the following estimates for the optimal
number of trees and cores per tree (Table 4-2).
TABLE 4-2. ESTIMATED OPTIMAL SUBSAMPLE SIZES
Trees per Plot
Cores per Tree
5- Year Average
1.49
0.73
10- Year Average
0.86
0.76
15- Year Average
1.30
0.81
Rounding up, we see that given an appropriate number of plots, two trees per plot and one
core per tree would provide an adequate estimate for regional estimation. Thus the current design is
obtaining enough material for dendrochronological purposes. There is currently not enough information
to determine if the variability at the plot level meets program needs, and so two trees with only one
core per tree may not be adequate for all program goals. No information is available to determine if
an adequate sample is being obtained for elemental analysis.
4.5 Conclusions and Recommendations
4.5.1 Equipment
Obtaining high-quality increment cores requires adherence to several guidelines. The
increment borer equipment must be sharp and free of sap and dirt. The borer must be carefully started
4-6
-------
in the tree with even pressure to prevent corkscrewing. Using an increment borer starter is highly
recommended for maintaining consistent directional pressure on the borer to prevent corkscrewing.
Corkscrewing was evident on many of the cores. A core usually breaks at each bend in the corkscrew
when being glued in the mounting stick.
4.5.2 Field Sampling
The objectives for dendrochronology were to determine if the sampling intensity and tree
selection protocols were adequate for quantifying dbh growth rates and trends for the state of Georgia.
The pine cores were much easier to extract, handle, prepare, and measure than cores from most other
tree species. Cores from oak were more difficult to extract and often had reaction wood that made
measuring increments difficult.
If all plots were completely forested, all sample trees cored, all cores completely labelled, and
all cores appropriate for analysis, a maximum of 264 cores would have been available from the 33
plots. The cores from hardwood trees were much more difficult to accurately date and measure than
the cores from softwoods. Hardwood cores had a greater frequency of potentially missing or false
rings. The ring boundaries were often obscured by reaction wood, making accurate measurement of
increment difficult.
The pith was frequently missed by one to many rings. Since the original intent for extracting
the cores was for tree core elemental analysis and not dbh increment, a core that included the pith was
not as critical. If the core misses the pith by many rings (cm), obtaining an accurate measure of
increment becomes increasingly difficult towards the pith. In addition, the increment obtained when
approaching the pith is not from the same radius from bark to pith. The increment is from an arc of
radii extending from the original radius to a radius 90° to the side of the pith from which the core was
actually taken. For example, if a core misses to the right of the pith, the measurements near the pith
represent increments from radii up to 90° to the right of the bark-to-pith radii.
Many of the mounted cores did not have complete label information. All codes from the field
label were to be transferred to the mounting stick. Evidently the field labels were not completed
properly. All information required on a label should be recorded before leaving the site. The hexagon,
subplot, tree, and core numbers, and species must always be completely filled in for proper sample
identification and tracking beginning in the field and continuing to analysis.
4.5.3 Handling and Preparation
Cores should be handled as little as possible between extraction and measurement. Each core
should be placed in a straw or some storage container in a consistent manner (pith end first), leaving
the bark attached. Many of the cores had the bark end removed, making it difficult to determine if just
the bark or the bark and last one to several years were removed. Bark ends may have broken off after
4-7
-------
the core had been shipped from the field. Each handling results in more broken cores and the potential
for broken segments to be lost, get out of proper sequence, or be mixed with portions of cores from
other trees. Two mounted cores consisted of portions of cores from different species.
The cores used in this study were shipped from the field to Las Vegas for tracking, drying, and
storage. Some cores were glued in crude holders. The cores were shipped to Virginia Polytechnic
Institute and State University for gluing into core holders. The core holders were made by making saw
cuts into a wood stick so that the core could be glued into the rectangular slot. The rectangular slot
provided less surface area contact than a rounded slot for the core to be completely stabilized for
surfacing. A good core holder has a curved slot routed out that matches the size of the core. The core
should snap into place when being glued in the holder. No amount of sanding, measurement, or
analysis will overcome problems with an improperly extracted, dried, and/or mounted core.
A tree core measurement laboratory should be identified to which cores can be sent for drying,
mounting, dating, and measurement. The quality control then becomes the responsibility of a clearly
identified laboratory.
4-8
-------
SECTION 5
ROOT DISEASE
£>.A. Alexander and M. Baldwin
5.1 Introduction
As part of the pilot research project entitled FY91 Indicator Evaluation Field Study for Environ-
mental Monitoring and Assessment Program - Forests (EMAP-F), we examined root samples in plots
in western Georgia for root disease. Sampling began June 16, 1991, and lasted through September
4, 1991. We used two root sampling techniques: the two-root method (Section 6 of the FHM Field
Methods Guide), in which two roots of the sample trees were directly sampled; and the cubic foot
method (Section 9 of the FHM Field Methods Guide), in which roots were taken from a cubic foot of
soil near a sample tree, but not specifically identified with a particular tree. A soil core was taken from
the center of the cubic foot sample for quantification of ectomycorrhizae. Root and soil core samples
were stored, refrigerated, and shipped on ice weekly to the Forest Pathology Laboratory at Virginia
Polytechnic Institute and State University (VPI&SU).
In 37 plots, a total of 120 trees were sampled by the two-root method; 21 of those plots were
also sampled by the cubic foot method. Root isolations were made specifically for Leptographium
procerum (Kendrick) Wing., Armillaria(Fr.:Fr.) Staude spp. and Heterobasidion annosum (Fr.:Fr.) Bref.
To maintain quality assurance, a training workshop was held for FHM field crews in Asheville, North
Carolina, June 10-14,1991. A field audit of the procedures was conducted on July 1,1991. The field
crew was debriefed on October 1 and 2,1991. Details of the training and debriefing can be found in
the FY91 Field Study Operations Report.
5.2 Evaluation of Root Disease Indicator
5.2.1 Objectives
The objectives were to: (1) determine the incidence of root disease; and (2) compare the two
detection methods.
5.2.2 Materials and Methods
The two-root samples received at the Forest Pathology Laboratory at VPI&SU were logged in
with the date they were received and assigned an index number. This number is part of a sequential
indexing system designed for sample tracking and data recording. A metal identification tag with the
index number accompanied each sample through processing. The samples were stored under
refrigeration until processed. There were four root chips per sample, two from each of two roots per
5-1
-------
sample tree. Chips were at least 2.5 cm3 to provide enough tissue for fungal isolations. Samples were
washed under running tap water for five minutes to remove dirt and debris, air dried on paper toweling
for 15 minutes, and then flamed briefly with 95% ethanol to sterilize the surface. The edges and bark
of each chip were removed using sterile pruners to expose clean wood. Blocks of wood approximately
1 cm3 were removed and aseptically placed on sterile agar growing media. This processing employed
one general medium (2% malt extract agar) and three media selective for L. procerum (McCall and
Merrill, 1980), Armillaria spp. (Russell, 1956) and H. annosum (Alexander and Skelly, 1973).
One block from each root chip was placed on each plate. All samples were plated within one
week of receipt. Plates were incubated at 20°C for three weeks and examined under a binocular
dissecting microscope (7x to 60x) for the presence of pathogenic fungi. Pure cultures were obtained
through conidial transfer and deposited in the VPI&SU Forest Pathology Laboratory culture collection.
The original plates were re-examined after 60 days for slow growing fungi before being autoclaved and
discarded.
5.2.3 Results
L procerum was isolated from three trees on three separate plots. H. annosum was isolated
from two trees on two separate plots. No Armillaria species were isolated.
5.3 Root Sampling for Evaluation of Root Diseases and Mycorrhizae
5.3.1 Objective
The objective was to determine whether the cubic foot root collection method for detection of
pathogens, as described by Alexander (1989), can be used to obtain samples appropriate for
quantifying ectomycorrhizal fungi. We also compared this method to the two-root technique for
pathogen detection.
5.3.2 Materials and Methods
5.3.2.1 Root Disease Quantification
The cubic foot method was adapted from the Annosus Sampling Procedure (ASP) (Alexander,
1989). Because the ASP was developed for detection of annosus root disease in pine plantations, we
used this method only when the sample tree was a pine species. We selected an area of 30.5 cm2
(12 in2) one to four meters from the sample tree, in the direction of other pine trees, if present, to
maximize the number of tree roots sampled. The duff layer was removed and the square excavated
to a depth of 30.5 cm. All tree roots 0.32 cm (0.12 in) and larger were collected and placed in a plastic
bag. Soil was returned to the hole and the duff layer replaced. The soil core and root samples were
5-2
-------
placed on ice for shipment to the Forest Pathology Laboratory at VPI&SU. In all, crews sampled 21
plots.
Samples were logged in when received and assigned the same index number as the
accompanying two-root sample described in Section 5.2. They were stored under refrigeration and
evaluated for root disease within seven days of receipt. Pine roots were examined for disease
symptoms of white stringy rot, resin soaking, and staining. All other types of roots were discarded.
The number of symptomatic roots and the total number of roots were recorded. After the roots were
washed and surface sterilized with a 10% bleach solution, a 1 -cm segment of each root was aseptically
plated onto each of the four media used for the two-root method. Sample plates were incubated and
examined as described in Section 5.2.
5.3.2.2 Mycorrhizal Quantification
Within the square described in 5.3.2.1, a soil core 76 mm (3 in) in diameter and 152 mm (6 in)
deep was taken through the litter layer and placed in a plastic bag. Mycorrhizal soil cores were washed
for 10 minutes in a North Carolina State University Semi-Automatic Soil Elutriatoronto a 0.7-mm screen
to remove the soil from the roots. The wet roots were placed in plastic bags and frozen until they could
be counted. At that time, the bags were removed from the freezer and thawed in warm water; the
contents were placed in water in a standard glass petri dish 100 mm in diameter. Roots were
examined under a binocular dissecting scope (7x to 60x) and the active ectomycorrhizal (EM) tips were
counted. Criteria used to determine active and inactive EM were taken from Harvey et al. (1976).
Individual tips were counted whether they were single or part of a complex structure. For each sample,
tips representative of each morphological type were preserved in formalin, acetic acid, and alcohol
(FAA) for later staining and microscopic confirmation of Hartig net. No attempt was made to quantify
the different morphological types. The logging and washing process took approximately one hour per
sample; it took an average of four to five hours per sample (range: 1-20 hours) to count EM tips. We
sampled 20 plots.
5.3.3 Results and Discussion
Of the 20 plots sampled, eight contained roots showing disease symptoms. The mean number
of roots in cubic foot samples from these eight plots was 20 ± 11, of which 2 ± 1 were symptomatic.
The only symptom observed was resin soaking. None of the root pathogens of concern were isolated
using the cubic foot method. A qualitative comparison of the root sampling methods indicates that the
two-root system may be as effective as the cubic foot method in detecting root disease pathogens. No
pathogenic fungi were isolated directly from symptomatic roots in cubic foot samples from eight of the
plots, even though the two-root method confirmed root disease in two sample trees from two of those
plots.
5-3
-------
TABLE 5-1. ECTOMYCORRHIZAL TIPS PER LITER OF SOIL
Plot Number
3408435
3308476
3308481
3308563
3208365
3208446
3208417
3108551
3108431
3108368
3008386
3108311
3008175
3208273
3208148
3208571
3308323
3408148
3408382
3408218
n
2
1
2
3
4
3
2
2
4
4
4
1
2
1
2
1
3
3
2
2
EM Tips per Liter
991 ±101
2,964
248 + 152
606 ± 566
1,136 + 614
1,021 ±713
841 ± 741
1,697 + 42
2,403 + 1 ,753
1,792 ±2,292
430 ± 366
5,144
1,386 ±31 9
2,518
1,673 ±202
2802
1,375 ±620
233 ±175
967 ± 503
574 + 303
The mycorrhizal study was intended to perform an overall evaluation of the procedure. The
field sampling method was easily performed over the soil types and conditions found in the study.
Laboratory processing of the samples was cumbersome and time consuming. One sample alone took
20 hours to count. The mean number of EM tips in the 20 plots sampled was 1,540 ± 1,150 per liter
of soil (Table 5-1). These numbers were within ranges that could be expected over a variety of soil
types and stand ages. Harvey et al. (1979) found a range of 200 to 1,700 EM tips per liter in three
forest habitat types in western Montana. The variability of the mycorrhizal counts was too high to make
comparisons among the plots. This was to be expected, since ectomycorrhizal populations are known
to be highly variable in distribution within an area (Marks et al., 1967; Alexander, 1985), and few
5-4
-------
samples were taken. In addition, within-sample variability has been shown to be very high. Alexander
(1985) estimated that, based on core diameter, 44 to 431 cores would need to be taken in a sitka
spruce plantation to minimize within- and between-sample variability. Core size (0.6895 L) was very
large, and the distribution of EM within the samples was irregular, preventing subsampling. Work is
needed to reduce the core size and/or develop a suitable subsampling procedure. The number of
samples taken per plot will have to be increased to compensate for the high degree of variability.
5-5
-------
-------
SECTION 6
EVALUATION OF DIFFERENT INSTRUMENTS FOR MEASURING
TREE HEIGHT
W.A. Bechtold, V.J. LaEJau, H. Schreuder, and M.S. Williams
6.1 Introduction
Tree height growth is a potential indicator of forest health, since height growth may be restricted
by anthropogenic or natural factors that impose stress on forests. The existing instruments for
measuring standing tree height are adequate for most applications, if the tree base and top are well
defined and clearly visible. However, the variance and bias of the height estimates could be large for
tall trees in dense stands or for trees that do not have well-defined tops. These errors in measurement
could be as large or larger than tree growth between two successive measurements.
An opportunity to evaluate some of the currently available height measurement instruments
(including a laser-driven instrument) for forest health monitoring presented itself in the summer of 1991.
Tree climbers, who were employed to collect foliage samples for chemical analysis, measured the true
height of standing trees. Readings from several different height measuring instruments were compared
to the values obtained by the tree climbers. We used these data to determine if any of the instruments
were accurate enough to make using tree growth as an indicator of forest health practical.
6.2 Methods
6.2.1 Study Planning and Preparation
All data collected were from Forest Health Monitoring (FHM) plots in Georgia. True heights for
all trees were taken while crews collected foliage samples from the tree crowns. One crew member
climbed as far as possible up each tree. From that point, the climber used a pole to measure the
remaining distance to the top of the tree. To determine when the pole was at the top of the tree,
sightings were taken from the ground and by the crew member in the tree. When all crew members
were in agreement, they calculated the total height from the ground to the top of the pole. Although
some measurement error existed in this method, no alternative method could have been implemented
that would have met the time and cost constraints, and still represented realistic measurement
situations, such as varying terrain, canopy, tree height distributions, and species mix.
Readings from a Laser Height Finder (Jasumback, 1991), a Suunto Clinometer (Husch et al.,
1982), a Spiegal Relaskop, an EEnbeeco Clinometer, and a Spiegal Tele-relaskop (Bitterlich, 1978;
Husch et al., 1982) were recorded at the same locations using a tripod to steady each instrument. This
6-1
-------
procedure required adapting a tripod mount for the Suunto Clinometer, which is designed to be hand-
held. All other instruments had thumb-screw systems to accommodate tripods.
A mix of scales on the instruments was used so that no two instruments successively utilized the
same scale. The Suunto Clinometer and the Enbeeco used the percent scale, the Relaskop used the
topo scale, and the Tele-relaskop and the Laser Height Finder used the degree scales. This mix of
scales was intended to interrupt the tendency to drive the current readings to be the same as those
obtained with the previous instrument. The following formulas give the scale conversion for each of
the instruments:
Clinometer [Percent Scale]
Height = (Top reading + Base reading) x (Level Distance/100)
Relaskop [Topographic Scale]
Height = (Top reading + Base reading) x (Level Distance/66)
Tele-relaskop [Degree Scale]
Height = [Level distance x tan(Top reading)] + [Level distance x tan(Base reading)]
Enbeeco [Percent Scale]
Height = (Top reading + Base reading) x (Level distance/100)
Laser [Degree Scale]
Height = [Level distance x (tan(Top reading)] + [Level distance x tan(Base reading)]
Before these computations could be made, all level distances had to be computed, since the distances
collected in the data set were slope distances. The formula for converting slope distance to level
distance is:
Level Distance = Slope Distance x [cos(slope)], where slope = cotan(percent slope/100).
As an example, given a slope distance of 73 feet on a 4% slope, Level Distance = 73 x
{cos[cotan(4/100)]} = 73 x (cos 2.29) = 73 x .9992 = 72.94 feet (22m).
6.2.2 Data Set Descriptions
The original data set contained instrument readings from six different observers, but only one
observer collected enough data to permit reasonable comparisons among all five instruments.
Therefore, most of the analyses that follow are based on the data collected by the single observer. To
determine if differences between observers impacted the results, all analyses were repeated with a
pooled data set that included all observers. Detailed output from the pooled data sets are not
presented in this report, but the few differences are noted in the results. In addition, only 15
observations were made with the Laser Height Finder, which was an preproduction model and not
6-2
-------
available until late in the field season. Due to the small number of observations collected, it was
difficult to make any definite conclusions about the Laser Height Finder.
6.2.3 Evaluation Techniques
Four linear models were fit to the data to test the bias and efficiency of the height measurements,
and to determine if there is an upper limit to the reliability of the instruments. The models were first
fit using all species of trees together, then softwood trees only. Hardwoods were removed to examine
the effect of extracting additional variability caused by poorly defined central stems typical of many
hardwood species. Tables 6-1 through 6-4 show the results of the regression analysis.
Ideally, the correlation between true height and measured height is 1 , thus the first model was
specified as
ht = Phm + e, [1]
where h, is the true height and hm is the measured height. The error terms, e, are assumed to be
normally distributed with mean 0 and variance o2.
The assumption of equal variance in the error term is highly suspect. A reasonable assumption
is that the error in measurement increases with tree height. Therefore the model
eht _ .Jihm _
- B + £,
[2]
was fit to remove the effect of the heteroscedasticity.
For completeness, intercept terms were added to [1] and [2], yielding
ht = a + phm + e
and
« _
fc? —
, c
T C-j.
[3]
[4]
Analysis of the coefficients from values given by models [1H41 was used to test if a bias in
estimation existed for any of the instruments. Confidence intervals for models [1] and [2] allowed the
hypothesis
H0 : P = 1
to be tested. For models [3] and [4], the hypothesis
H0 : a = 0, J3 = 1
was tested.
6-3
-------
TABLE 6-1. REGRESSION COEFFICIENTS, STANDARD ERRORS, 95% CONFIDENCE
INTERVALS, AND R2 VALUES FOR UNTRANSFORMED DATA AND ALL TREES
Instalment
Clinometer
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Relaskop
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Tele-relaskop
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Enbeeco
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Laser
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Parameter
0.9888
3.6659
0.9329
0.9835
3.6330
0.9283
0.9950
4.4237
0.9266
0.9746
3.3680
0.8835
1.0008
4.4341
0.9387
Standard
Error
0.0084
1 .3941
0.0228
0.0090
1 .5225
0.0248
0.0160
2.5880
0.0430
0.0088
1 .5062
0.0372
0.01858
4.9531
0.0718
Confidence
Interval
(0.9721, 1.0056)
(0.8954, 6.4364)
(0.8876,0.9781)
(0.9655, 1.0016)
(0.6074, 6.6586)
(0.8791 , 0.9776)
(0.9628, 1.0272)
(-0.7771,9.6245)
(0.8403, 1.0130)
(0.9570,0.9921)
(0.3666, 6.3693)
(0.8093, 0.9576)
(0.9609, 1.0406)
(-6.2665, 15.1346)
(0.7836, 1.0938)
R2
0.9462
0.9501
0.9372
0.9410
0.8990
0.9047
0.9505
0.9537
0.9250
0.9293
n
n=90
n=90
n=51
n=76
n=15
6-4
-------
TABLE 6-2. REGRESSION COEFFICIENTS, STANDARD ERRORS, 95% CONFIDENCE
INTERVALS, AND R2 VALUES FOR LOG TRANSFORMED DATA AND ALL TREES
Instrument
Clinometer
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Relaskop
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Tele-relaskop
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Enbeeco
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Laser
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Parameter
1 .0003
0.2074
0.9485
0.9984
0.1228
0.9678
1.0024
0.3822
0.9065
0.9963
0.1375
0.9623
1.0017
0.2665
0.9380
Standard
Error
0.0021
0.0669
0.0168
0.0021
0.0727
0.0182
0.0048
0.1312
0.0349
0.0023
0.0735
0.0183
0.0041
0.2086
0.0501
Confidence
Interval
(0.9961 , 1 .0045)
(0.0745, 0.3403)
(0.9151,0.9819)
(0.9942, 1.0027)
(-0.0216, 0.2673)
(0.9316, 1.0041)
(0.9926, 1.0120)
(0.1046,0.6599)
(0.8362, 0.9767)
(0.9918, 1.0008)
(-0.0090, 0.2839)
(0.9257, 0.9988)
(0.993, 1.010)
(-0.1842, 0.7172)
(0.8297, 1.0462)
R2
0.9702
0.9731
0.9687
0.9697
0.9215
0.9321
0.9726
0.9735
0.9597
0.9642
n
n=90
n=90
n=51
n=76
n=15
6-5
-------
TABLE 6-3. REGRESSION COEFFICIENTS, STANDARD ERRORS, 95% CONFIDENCE
INTERVALS, AND R2 VALUES FOR UNTRANSFORMED DATA WITH ALL
HARDWOOD TREES REMOVED
Instrument
Clinometer
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Relaskop
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Tele-relaskop
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Enbeeco
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Laser
Without Intercept (model [1])
P
With Intercept (model [3])
a
P
Parameter
0.9846
3.9622
0.9262
0.9837
4.1243
0.9229
0.9983
4.3524
0.9335
0.9707
3.6061
0.9197
0.9899
5.1553
0.9185
Standard
Error
0.0096
1.6137
0.0255
0.0106
1.8004
0.0284
0.0195
3.1092
0.0501
0.0098
1.7017
0.0259
0.0171
4.3162
0.0621
Confidence
Interval
(0.9655, 1.0037)
(0.7403,7.1841)
(0.8753,0.9771)
(0.9625, 1.0048)
(0.5297, 7.7189)
(0.8661 , 0.9797)
(0.9589,1.0377)
(-1.9418, 10.6467)
(0.8320, 1.0350)
(0.9511,0.9903)
(0.1984, 7.0138)
(0.8679, 0.9715)
(0.9517, 1.0284)
(-4.6086, 14.9192)
(0.7780, 1.0589)
R2
0.9480
0.9523
0.9363
0.9410
0.8961
0.9012
0.9535
0.9569
0.9542
0.9605
n
n=68
n=68
n=41
n=59
n=11
6-6
-------
TABLE 6-4. REGRESSION COEFFICIENTS, STANDARD ERRORS, 95% CONFIDENCE
INTERVALS, AND R2 VALUES FOR LOG TRANSFORMED DATA FOR SOFTWOOD
TREES ONLY
Instrument
Clinometer
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Relaskop
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Tele-relaskop
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Enbeeco
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Laser
Without Intercept (model [2])
P
With Intercept (model [4])
a
P
Parameter
0.9996
0.2321
0.9421
0.9989
0.1242
0.9681
1.0031
0.3595
0.9133
0.9956
0.1370
0.9620
1.0000
0.3349
0.9199
Standard
Error
0.0025
0.0761
0.0190
0.0025
0.0834
0.0208
0.0060
0.1596
0.0403
0.0026
0.0815
0.0202
0.0042
0.1800
0.0432
Confidence
Interval
(0.9946, 1.0047)
(0.0802, 0.3842)
(0.9042, 0.9801)
(0.9938, 1.0040)
(-0.0422, 0.2907)
(0.9266, 1.0097)
(0.9911, 1.0152)
(0.0364, 0.6827)
(0.8318, 0.9949)
(0.9904, 1.0009)
(-0.0262, 0.3002)
(0.9220, 1.0024)
(0.9907, 1.0092)
(-0.0723, 0.7421)
(0.8221,1.0177)
R2
0.9702
0.9739
0.9694
0.9704
0.9220
0.9312
0.9743
0.9755
0.9730
0.9805
n
n=68
n=68
n=40
n=59
n=11
6-7
-------
R* values were used as an indicator of goodness of fit. For this study, R2 is defined as
SSE
R2 = 1 -
SST
where
i-l
SST = £(/*,. - h)2
i=l
and n^ is the estimated height generated from the models described above, and h is the mean of the
true heights.
To determine if there is a specific tree height at which ocular estimates are no longer accurate,
a visual comparison of model errors was performed. The visual comparison entailed graphing the
measured height versus the true height to determine if a significant error pattern could be identified.
Figures 6-1 through 6-5 show the fit of model [1] to the data for all trees. Figures 6-6 through 6-10
show the fit of model [1] to the data with softwoods only.
After the bias and efficiency of each instrument had been evaluated, the next goal was to
determine if any instrument was accurate enough to measure the growth of trees between successive
growth cycles. Time constraints would not allow us to collect data at two time periods for comparison,
so we evaluated the magnitude of the error for a single measurement. We determined that the growth
of a tree could not be measured accurately enough if the amount of error associated with a single
measurement was as large as or larger than the expected growth between successive measurements.
A comparison of model errors was performed by computing the average absolute error between
the model estimates and the true heights by 10-foot (3.0 meter) height classes. These values were
computed using the formula
where nh is the number of trees in a given height class. Ten-foot (3.0 meter) height classes were used.
Average errors in each height class provided a good indicator of how well tree growth could be
measured with each of the instruments. If the average error in a given height class was large relative
to the expected growth between successive measurement periods, no useful information about tree
growth could be collected. Table 6-5 gives the error analysis results for all trees and Table 6-6 gives
the analysis when data for hardwood trees were removed.
6-8
-------
CD
-
CO Q.
-i =
w co
05 ~—
« -
CO
CD CO
05
C\I
CD
£
I
(ui) }i)&
D)
"
TJ
'
0> Q.
re _"
~ 're
D> (B
'
LL
(ai) wBiej-i ami
6-9
-------
0>
•I
•o
1
g>
CD
£ "ST
•O .£
CD O
1 8.
•- —
11.
CO O
Z3 O
CO CD
CD 2
>•§
£ UJ
.5> CD
CD -C
I.i
i= 3
c
0)
g>
a
(tu)
- i
2>
„
a s
(ui) }ii6|8H erui
^ s.
T3 CO
CD =
08 ^B,
w Q.
s^
W CO
co a
I?
0) CD
> "33
£1-
O)
f
CO
CD
6-10
-------
&
£
c
g> o
" > S
i EO
1» .5> CO
1 sf
cp
CD
£
I
(w)
1- 8
S
(m)
g>
'
<» "to"
to .2
E ••=
S w
•J --i_r—
OS w
D)
X
•n CD
OJ >
to «*
I "§)'
if
,<- o
I— 3
U)
CD
I
D)
6-11
-------
.
ID
OJ "O
• a}
s «g»s
£!§
9
CO
I
O)
IE
I f
C3) C
'5 O
•S= CO
Si
^ o
CO CO
-3
>
(ui) }i|6|8H erui
6-12
-------
.
g> .
« T3
18
th co —
x > I
•§ 2=3
CO
g>
a.
(LU) wSi
g
£
O)
'5
I
•o
.£
UJ
E s;
.^"c
O O
•^ CD
® O.t
m 9
E|'.
11
CO O
3 O
CO CD
11
£LU
D) 0)
-O.C
0 g>
2-i
1- 3
0)
CD
£
O)
S
(LU)
6-13
-------
TABLE 6-5. AVERAGE ABSOLUTE MODEL DEVIATION AND NUMBER OF TREES IN EACH 10-
FOOT HEIGHT CLASS FOR ALL SPECIES, MODEL [1] FIT TO THE DATA3
Height Class
ft (m)
10-20 (3.0-6.1)
20-30 (6.1-9.1)
30-40 (9.1-12.2)
40-50 (12.2-15.2)
50-60 (15.2-18.3)
60-70 (18.3-21.3)
70-80 (21.3-24.4)
80-90 (24.4-27.4)
90-100(27.4-30.5)
Clinometer
0.80, 5
1.20,4
1.33,4
2.22, 12
5.09, 15
2.89, 11
2.69, 13
5.86, 13
3.84, 3
Relaskop
0.71,5
0.42, 4
1.15, 14
1.76, 12
4.18, 15
2.12, 11
3.23, 13
6.96, 13
7.98, 3
Tele-relaskop
2.05, 3
0.50, 3
3.94, 8
4.56, 7
6.03, 9
2.66, 6
3.37, 8
3.01,5
11.31,3
Enbeeco
0.61,5
1.35,3
1.51, 11
2.51, 10
5.34, 11
2.06, 9
1.97, 13
5.16, 12
5.44,3
Laser
1.37,2
0.03,1
0.64, 1
3.38, 3
3.32, 4
6.45, 4
* The model used is h, = phm.
TABLE 6-6. AVERAGE ABSOLUTE MODEL DEVIATION AND NUMBER OF TREES IN EACH 10-
FOOT HEIGHT CLASS FOR SOFTWOOD TREES ONLY, MODEL [1] FIT TO THE
DATA3
Height Class
ft (m)
10-20 (3.0-6.1)
20-30 (6.1-9.1)
30-40 (9.1-12.2)
40-50 (12.2-15.2)
50-60 (15.2-18.3)
60-70 (18.3-21.3)
70-80 (21.3-24.4)
80-90 (24.4-27.4)
90-100 (27.4-30.5)
Clinometer
0.97, 4
1.46,3
1.24, 11
2.74, 6
6.24, 10
3.07, 9
2.60, 12
5.02, 10
3.94, 3
Relaskop
0.70, 4
0.39, 3
0.93, 11
1.25, 6
4.98, 10
2.25, 9
3.04, 12
6.46, 10
8.00, 3
Tele-relaskop
2.02, 3
0.48,3
4.78, 6
6.78, 4
8.77, 5
2.40, 5
3.39, 7
3.08, 5
11.26,3
Enbeeco
0.65, 4
1.45,3
1.33,9
3.21, 4
7.48,6
2.29, 8
1.65, 12
4.49, 10
5.29, 3
Laser
1.71,2
3.56, 3
3.30, 4
4.44, 2
a The model used is h,
6-14
-------
The top of the tree is sometimes not well defined, which means that picking a point at which
to measure the top of the tree can be very subjective; two different people may choose different points
as the top of a tree. For some of the trees in the data set, more than one crew member took meas-
urements. Two crew members took a number of measurements on the same trees using all five
instruments. We used a paired comparison to determine if these two crew members made significantly
different estimates of tree height. The hypothesis tested was H0 : u. = 0, where u. was the average
difference between height measurements. The average absolute error and median absolute error
between the two observers was also compared. The formula used for the average absolute error was
where hti and h2i are the height measurements taken by the two different crew members for tree /. The
median absolute error is the median of the [h^ - ha| values. Results of the tests between crew
members are given in Table 6-7.
TABLE 6-7. PAIRED COMPARISON TEST RESULTS, AVERAGE AND MEDIAN ABSOLUTE
ERRORS BETWEEN CREW MEMBERS, AND NUMBER OF OBSERVATIONS
Instrument
Clinometer
Relaskop
Tele-relaskop
Enbeeco
Laser
H0 : u. = 0
Accept H0
Accept H0
Accept HQ
Accept H0
Accept H0
Average Abs. Error
2.97
2.56
8.98
2.59
1.71
Median Abs. Error
1.2
1.2
6.9
0.7
0.9
n
29
28
5
19
7
6.3 Results and Discussion
Tables 6-1 through 6-4 show results for fitting models [1] through [4] to the data set with all
species and to the data set with only softwood trees. The average absolute errors by diameter class
are given in Tables 6-5 and 6-6.
For the Clinometer, the models without intercept terms (models [1] and [3]) had confidence
intervals containing p = 1, regardless of whether all species or only softwood species were considered.
In contrast, for models with intercept terms (models [2] and [4]), no confidence interval contained either
a = 0 or p = 1. R2 values ranged from 0.9480 to 0.9739. When all species were considered, the R2
6-15
-------
values were consistently second largest. When only softwood trees were considered, the R2 values
were third largest, regardless of whether the data were transformed or which model was fit.
The results for the Relaskop were similar to those for the Clinometer. For the models without
intercept terms (models [1] and [3]), the confidence intervals for p contained 1. When the intercept
term was added (models [2] and [4]), the confidence intervals contained a =0 and p = 1 for both of the
log transformed data sets, which may imply that the errors in fit were due to heteroscedasticity in the
data. R2 values ranged from 0.9363 to 0.9704. For all species, the R2 values were always third
largest. When only softwoods were considered, R2 values were fourth largest.
For the Tele-relaskop, the confidence intervals contained p = 1 for the models without intercept
terms. When intercept terms were added, the confidence intervals contained a = 0 and p = 1 for the
untransformed data using both the data set for all species and the data set for softwoods only. Using
R2 values as an indication of fit, the Tele-relaskop produced the worst fit for all models and data sets
considered.
The Enbeeco was the only instrument that did not have confidence intervals containing p = 1 for
all models without intercept terms. When model [1] was fit to the softwood data, the assumption that
h, s hm was rejected. When intercept terms were added, model [4] had confidence intervals containing
a - 0 and p - 1. R2 values for the Enbeeco were consistently the largest when all tree species were
considered. When only softwood trees were considered, the R2 value for model [2] was the largest for
all instruments and second largest when model [4] was fit.:
The Laser has the distinction .of being the only instrument whose confidence intervals contained
P s 1 and a = 0 under every model and data set. The R2 values were fourth largest when all species
were considered. R2 values were the largest for the softwood data, except when model [2] was fit. In
that case, the R2 value was second largest.
The analysis of the absolute average error by diameter class is given in Tables 6-5 and 6-6. In
all height classes less than 40 feet (12.2 m), the average absolute errors were generally less than 1.5
feet (0.5 m) with the exception of the Tele-relaskop, which in the 40-foot (12.2 m)class had error values
of 3.94 for the data set for all species and 4.78 for the data set for softwoods only. For trees greater
than 40 feet (12.2 m) tall, the error values increase. The error values range from 0.03 to 11.31 feet
(0.01 to 3.45 m), with most values falling in the two- to five-foot (0.6-1.5 m)range. In most cases,
removing the softwood trees from the data reduced the average absolute error. No instrument differed
greatly from the others. The Tele-relaskop was the only instrument to produce average errors
exceeding 10 feet (3.0 m).
The paired comparison test showed no significant difference between observers for any of the
instruments. The average and median absolute errors showed similar results for the Clinometer,
Relaskop, Enbeeco, and Laser. The Tele-relaskop had very large average and median absolute error
6-16
-------
values, which could be due in part to the small number of observations available for two observers; it
is consistent with the single observer resufts, however.
6.4 Conclusions
Results were similar for the Clinometer, Relaskop, and Enbeeco. The Tele-relaskop appeared
to be slightly less accurate. Even though the results for the Laser indicate that it is the best instrument,
the effectiveness of the Laser is still difficult to determine because of the limited amount of data
available. In addition, the Laser height finder used in this study was an early preproduction model.
Numerous design improvements have been made on current models, which may improve the accuracy
of the Laser. The analysis of average and median errors indicates that errors of two to five feet (0.6-
1.5 m) are common, especially for trees larger than 40 feet (12.2 m). This may be acceptable when
the objective is to measure current height. However, when height growth is the objective, and the
measurement cycle is as little as four to five years, average errors could exceed the amount of tree
growth expected for many species and geographic areas. In addition, these errors could be
conservative estimates of the expected error from repeated measurements, because tree heights would
not necessarily be taken from the same point at each time period. Additional sources of variation could
include new estimates of slope, slope distance, and tree lean at each time period. All these factors
would increase the measurement error. With such large errors, height data obtained from standard
instruments is not reliable enough at this time to make height growth a reliable indicator of forest health.
6-17
-------
-------
SECTION 7
REFERENCES
Alexander, I.J. 1985. Measuring mycorrhizal biomass: principles and problems. Pages 136-139 in
R. Molina, ed. Proc. 6th North American Conference on Mycorrhizae. Bend, Oregon. 471 pp.
Alexander, S.A. 1989. Annosus root disease hazard rating, detection and management strategies in
the Southeastern United States. Pages 111-116. Symposium on Research and Management
of Annosus Root disease in Western North America. Gen. Tech. Rept. PSW-116. 177 pp.
Alexander, S.A., and J.M. Skelly. 1973. A comparison of isolation methods for determining the
incidence of Fomes annosus in living loblolly pine. Eur. J. For. Path. 4:33-38.
Bailey, A.W., and C.E. Poulton. 1968. Plant communities and environmental relationships in a portion
of the Tillamook burn, northwestern Oregon. Ecology 49:1-13.
Bitterlich, W. 1978. Single tree measurements by the Tele-relaskop—a highly efficient tool for forest
inventories. Eighth World Forestry Congress, Jakarta, Indonesia.
Brenchley, W.E. 1958. The Park Grass Plot at Rothamsted. Rothamsted Experimental Station.
Harpenden.
Cassell, D.L. 1992. Estimating optimal resource allocation for environmental monitoring networks.
Pages 225-229 in 1991 Proceedings of the Section on Statistics and the Environment.
American Statistical Association, 1429 Duke Street, Alexandria, Virginia.
Chatfield, C. 1989. The Analysis of Time Series, An Introduction. Fourth edition. Chapman and Hall,
New York.
Cochran, W.G. 1977. Sampling Techniques. Third edition. John Wiley and Sons, New York.
Cook, E.R., and L.A. Kairiukstis (eds). 1990. Methods of Dendrochronology. Kluwer Academic
Publishers, Boston.
Daubenmire, R. 1968. Plant Communities: A Textbook of Plant Synecology. Harper & Row, New
York.
Harvey, A.E, M.J. Larsen and M.F. Jurgensen. 1976. Distribution of ectomycorrhizae in a mature
Douglas-fir/larch forest soil in western Montana. For. Sci. 22:393-398.
Harvey, A.E, M.J. Larsen and M.F. Jurgensen. 1979. Comparative distribution of ectomycorrhizae in
soils of three western Montana forest habitat types. For. Sci. 25:350-358.
Hill, M.O. 1973. Diversity and evenness: a unifying notation and its consequences. Ecology
54:427-432.
Hurlbert, S.H. 1971. The nonconcept of species diversity: a critique and alternative parameters.
Ecology 52:577-586.
Husch, B., C.I. Miller, and T.W. Beers. 1982. Forest Mensuration. Third edition. John Wiley &
Sons,, New York. 402 pp.
7-1
-------
Jasumback, T. 1991. Laser Height Measurement. Timber Tech Tips. Gov. Pub. 9124 2336-MTDC.
USDA Forest Service Technical and Development Program.
Kempton, R.A., and L.R. Taylor. 1974. Log-series and log-normal parameters as diversity discrimin-
ants for the Lepidoptera. J. Anim. Ecol. 43:381-399.
Kempton, R.A., and L.R. Taylor. 1976. Models and statistics for species diversity. Nature
262:818-820.
Ludwig, J.A., and J.F. Reynolds. 1988. Statistical Ecology. A Primer on Methods and Computing.
John Wiley & Sons, New York.
Magurran, A.E. 1988. Ecological Diversity and Its Measurement. Princeton University Press,
Princeton, New Jersey. 179 pp.
Marks, G.C., N. Ditchburne, B.C. Foster. 1967. A technique for making quantitative estimates of
mycorrhiza populations in Pinus radiata forests and its applications in Victoria. Pages 67-83
Proc. Int. Union Forest Res. Org. XIV 1967, Munich.
McCall, K.A., and W. Merrill. 1980. Selective medium for Verticicladiella procera. Plant Dis.
64:277-278.
Mueller-Dombois, D., and H. Ellenberg. 1974. Aims and Methods of Vegetation Ecology. Wiley &
Sons, New York. 547 pp.
Norman, J., and G. Campbell. 1989. Canopy structure. Pages 301-325 in R. Pearcy, J. Ehleringer,
H. Mooney, and P. Rundel, eds. Plant Physiological Ecology. Chapman and Hall, London.
Noss, R.F. 1990. Indicators for monitoring biodiversity: a hierarchical approach. Cons. Biol.
4:355-364.
Palmer, C.J., K.H. Riitters, T. Strickland, D.L. Cassell, G.E. Byers, M.L. Papp, and C.I. Liff. 1991.
Monitoring and Research Strategy for Forests—Environmental Monitoring and Assessment
Program (EMAP). EPA/600/4-91/012. U.S. Environmental Protection Agency, Washington,
D.C.
Patrick, R. 1968. The structure of diatom communities in similar ecological conditions. Amer. Nat.
102:173-183.
Pielou, E.C. 1975. Ecological Diversity. John Wiley & Sons, New York.
Riitters, K., M. Papp, D. Cassell, and J. Hazard (eds). 1991. Forest Health Monitoring Plot Design and
Logistics Study. EPA/600/S3-91/051. U.S. Environmental Protection Agency, Office of
Research and Development, Research Triangle Park, North Carolina.
Ripley, B.D. 1981. Spatial Statistics. John Wiley and Sons, New York.
Russell, P. 1956. A selective medium for the isolation of basidiomycetes. Nature 177:1038-1039.
Shannon, C.E., and W. Wiener. 1949. The Mathematical Theory of Communication. University of
Illinois Press, Urbana. 117pp.
7-2
-------
Simpson, E.H. 1949. Measurement of diversity. Nature 163:688.
Stokes, M.A., and T.L. Smiley. 1968. An Introduction to Tree-Ring Dating. University of Chicago
Press, Chicago.
Sukhatme, P.V. 1954. Sampling Theory of Surveys, with Applications. Iowa State College Press,
Ames.
U.S. EPA. 1990. Threats to Biological Diversity in the United States. PM-223X. Office of Policy,
Planning, and Evaluation, U.S. Environmental Protection Agency, Washington, D.C. 57 pp.
Van Deusen, P.C. 1992. Growth trends and stand dynamics in natural loblolly pine in the south-
eastern United States. Can. J. For. Res. 22:660-666.
Waring, R.H., and W.H. Schlesinger. 1985. Forest Ecosystems: Concepts and Management. Aca-
demic Press, Orlando, FL.
*U.S. GOVERNMENT PRINTING OFFICE: 1994-550-001/00174 Region 5.
7-3
-------
------- |