EPA-260-R-06-002
August 2006
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in
the Florida Keys
Final Report
Leska S. Fore
Statistical Design
136NW40thSt.
Seattle, WA 98107
William S. Fisher
U.S. Environmental Protection Agency
Office of Research and Development
National Health and Environmental Effects Research Laboratory
Gulf Ecology Division
1 Sabine Island Drive
Gulf Breeze, FL 32561
and
Wayne S. Davis
U.S. Environmental Protection Agency
Office of Environmental Information
Environmental Analysis Division
Environmental Science Center
701 Mapes Road
Ft. Meade, MD 20755-5350
Printed with Oil on 100%
Post-consumer Chlorine
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
ABSTRACT
Measurements of coral reef condition were collected from stations in the Florida Keys National
Marine Sanctuary and Dry Tortugas National Park during 2003-04. Four assessment endpoints of
reef condition were derived from transect censuses and measurements of stony corals: total surface
area of all corals in the reef transect (ISA), percent live coral averaged across all colonies in the
reef transect (%LC), the sum of live surface areas for all colonies in a reef transect (LSA), and
percent live surface area within the reef transect (%LSA). TSA and LSA were highly correlated;
both measures were dominated by a few very large species of coral. Repeat samples were collected
within stations, within reefs at different stations, and at the same stations during different years. For
all measures, much of the variance was associated with different stations within reefs. Some stations
near Key West had much lower values for %LSA. Some of the lowest values for %LC were
associated with back reef habitats. For all measures, back reef habitat was most variable. Principal
components analysis identified unique species assemblages associated with back and transitional
reef habitats.
Variance estimates were derived from repeat samples to determine the minimum detectable
differences (MDD) in mean values of coral metrics that would represent a statistically significant
change in reef condition. MDD values were derived for a sampling design with repeat visits to the
same stations and a second design with repeat visits to different stations. For TSA and LSA, 20 or
more of the same stations should be sampled each year to detect a reasonable level of change. For
%LC and %LSA, the same or different stations could be sampled each year to provide the same
level of sensitivity to change and 10-20 stations are probably adequate. Strong agreement in coral
metrics for opposite halves of the belt transect suggested that a smaller area could be sampled at
each station.
ACKNOWLEDGMENTS
Field data were collected during ongoing collaboration with the National Oceanic and
Atmospheric Administration (NOAA) and the Florida Keys National Marine Sanctuary (FKNMS).
NOAA National Ocean Service provided the RV NANCY FOSTER as a research platform and the
National Park Service provided dive boat support. Data were collected by EPA dive teams (Gulf
Ecology Division) led by Jed Campbell and Bob Quarles. Leah Oliver helped with calculation of
candidate coral metrics and changes in the sampling methods. We appreciated the support from
EPA Chief Statistician, Dr. Barry Nussbaum, Environmental Analysis Division of the Office of
Environmental Information. Contract support was provided by Leska S. Fore, Statistical Design,
136 NW 40th St., Seattle, WA 98107 under contract with Perot Systems Government Services, Inc.,
8270 Willow Oaks Corporate Drive, Suite 300, Fairfax, Virginia 22031.
The report should be cited as:
Fore, L. S., W.S. Fisher, and W. S Davis. 2006. Bioassessment Tools for Stony Corals: Statistical
Evaluation of Candidate Metrics in the Florida Keys. EPA-260-R-06-002. USEPA Office of
Environmental Information, Washington. August 2006.
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
CONTENTS
Abstract ii
Acknowledgments ii
Introduction 1
Methods 3
Data collection 4
Reconciliation of sampling protocols 6
Data analysis 7
Results 10
Influence of geographic gradient and reef habitat type 11
Species assemblages associated with different reef habitat types 15
Sources of variance 16
Power analysis 18
Discussion 22
Natural sources of variability 23
Variance structure 24
Conclusions 25
Recommendations 26
References 28
Appendix A - Candidate Coral Metrics by Section (2003) 29
Appendix B - Candidate Coral Metrics for Stations (2003) 30
Appendix C - Candidate Coral Metrics for Stations (2004) 31
in
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
TABLES
Table 1. Steps in the design of an effective monitoring program 2
Table 2. Number of stations surveyed in back, fore and transitional
reef habitat for 2003 and 2004 4
Table 3. Coral species codes, genera, and species 6
Table 4. Variance estimates for ISA, %LC, LSA and %LSA 17
Table 5. %LC, LSA and %LSA for stations visited in 2003 and 2004 20
Table 6. MOD for two-sample t test and paired t test for 10,20, 30 and 50 stations 20
FIGURES
Figure 1. Reef locations in the Florida Keys National Marine Sanctuary 4
Figure!. LSA plotted against ISA by station 10
Figure 3. LSA plotted against TSAby species 11
Figured %LC, LSA, and %LS A by reef for 2004 12
Figure 5. Depth for each habitat type 13
Figure 6. %LC, LSA, and %LSA by habitat type for 2004 14
Figure 7. PCA with duplicate samples 15
FigureS. PCA with back, fore and transitional habitat types 16
Figure 9. Components of variance for ISA, %LC, LSA, and %LSA 17
Figure 10. %LC, LSA, and %LSA for five stations sampled in 2003 and 2004 19
Figure 11. Power curves for candidate coral metrics 21
IV
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
INTRODUCTION
Protection and management of natural resources depend foremost on the development of a reliable
and meaningful method to assess resource condition (Olsen et al. 1999). The primary purpose for
monitoring natural resources is to determine whether their condition has improved, declined, or
remains unchanged. Ward et al. (1986) provide a general framework describing the steps needed to
design an effective monitoring program (Table 1). The five key components of a mature program
are: 1) clearly stated program goals, 2) statistical design criteria, 3) a statistical sampling plan, 4)
detailed written protocols for data collection and analysis, and 5) an information reporting
procedure.
The current analysis focuses primarily on their Step 2 by evaluating the sensitivity and potential of
four candidate coral metrics to detect change in reef condition. The four coral metrics described in
this study are considered "candidate" metrics because they have yet to be tested for a response to
human disturbance. "Assessment endpoints," or "attributes," or "measurements" of the coral
assemblage are typically not labeled as metrics in the accepted terminology of biocriteria until they
demonstrate a correlation with disturbance (Jameson et al., 2001). High variability associated with
candidate coral metrics translates into a lower probability of detecting changes through time
because observed differences must exceed the range of natural variability typically observed. Coral
metrics that are less variable are more precise, and relatively smaller changes can be detected for
these measures. Evaluation of the variance structure for different coral metrics provides a
foundation for selecting among different sampling scenarios that differ according to the number of
stations visited each year, the frequency of sampling, or the type of data collected.
Statistical design criteria include definition of the sampling unit, delineation of the sample
population, and selection of sampling protocols and measures. Because reefs are a continuous
resource, no obvious method exists to objectively divide the resource into discrete sampling units.
For population censuses, for example, the household or individual taxpayer is used as the sampling
unit. For coral, the size of the sampling unit must be defined somewhat arbitrarily but in a way that
reliably characterizes the sampling location. The sample population was defined as stations within
reefs; however, different strata could be defined as subpopulations within the larger sampling frame,
e.g., fore, back or transitional habitats. The sampling protocols measured the total surface area of
coral in a reef transect, the percent live coral on each colony, the live surface area in the reef
transect, and the percentage of live surface area (see Fisher et al. 2006 for details).
The goal of this analysis was to evaluate the variability associated with the sampling protocols and
candidate coral metrics and apply the results to two sampling design options for assessing coral reef
condition through time. The two sampling designs were based on repeat visits to the same stations
each year and visits to new stations each year. The two sampling designs require different statistical
tests to determine whether change has occurred through time. More complex survey designs than
these exist, but the two statistical models presented here provide a starting point for judging the
relative merits of different sampling scenarios for the selected coral metrics (Skalski 1990, Larsen et
al. 2001).
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Table 1. Steps in the design of an effective natural resource monitoring program.
Relevant examples for coral reef monitoring are show for each step along with specific items within
each step (modified from Ward et al., 1986).
Step and description
Step 1: Evaluate information expectations
Identify resource condition targets (healthy living coral)
Define management goals (e.g., identify stressors)
Explicitly define monitoring goals (i.e., as statistical hypotheses)
Step 2: Define statistical design criteria
Define sampling unit (e.g., radial belt transect on reef)
Define the 'population' to be sampled (e.g., stations within the Florida Keys)
Evaluate variability of sampling protocols and measures (e.g., %LC, ISA, %LSA.)
Select appropriate statistical test (e.g., two-sample ttest)
Step 3: Design monitoring plan
Identify what to measure (e.g., stony corals)
Identify where to sample (e.g., randomly selected stations)
Define how frequently to sample (e.g., every year, every five years)
Step 4: Develop operating plans and procedures
Field sampling and analysis procedures
Data analysis procedures
Quality control
Data management
Step 5: Develop information reporting procedures
Identify target audience
Define report structure, format and content
Determine frequency and distribution of reports
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
METHODS
Data collection
Coral reef sampling stations were located in the Florida Keys National Marine Sanctuary and Dry
Tortugas National Park (Figure 1). During 2003, data were collected from 14 station visits to five
reefs (Table 2). During 2004, 35 station visits were made to 10 reefs. Some sampling stations were
also visited in the Dry Tortugas during 2004, but those data were unavailable for this analysis.
Back, fore and transitional reef habitat types were visited on most reefs during 2004. During both
years, some stations were sampled more than once (duplicate sampling). During 2003, some belt
transects were divided into two parts (sections) and data from each were kept separate.
At each station, corals were surveyed using radial belt transects which were delineated by the area
between two circles with radii of 8 m and 10m measured from permanent stakes (for additional
details on sampling or coral measures see Fisher et al. 2006). The radial belt encompassed 113.1 m2.
Within the radial belt transect, each stony coral colony was identified to species, volume of the
colony was estimated, and the percent live coral noted. From these data, four candidate coral
metrics were calculated for each station: total reef surface area (TSA) measured in m2 average
percentage live coral for all colonies (%LC), live surface area (LSA) measured in m2, and the
percentage of the total reef surface area covered in live coral (%LSA; Appendices I, II and III).
Total surface area in the reef transect (TSA) was derived from estimates of coral volume. Size
classes were defined as cubes, and the size class assigned to a colony was based on the smallest size
cube that could contain the colony. Colonies are not cube-shaped, so this method represents an
approximation. Estimates of surface area were based on the surface area of the five sides of the cube
(excluding the sixth side that faced the substrate). In 2003, corals were graded into five volumetric
size classes (1, 10, 50, 100 or > 100 L), but this did not sufficiently separate larger corals and
differences between 50 L and 100 L size classes were difficult to discern. In the second year, six
size classes were used: 1, 10, 100, 500, 2500 and > 2500 L.
Percent live coral (%LC) was graded into seven categories in 2003 (0, 0-20, 20-40, 40-60, 60-80,
80-100, and 100%); the values used in calculations were the midpoint of the ranges. In 2004, %LC
was graded into six categories (0, 1-25, 26-50, 51-75, 76-99, and 100%). For each station-visit, the
colony values of %LC were averaged to provide a station value.
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
"ĞĞ
Ğtğlğ
$#ğ
.?Sr"ğ
!\
Ğ / Catvttoti-
^-J*-* IhC-y^rttw
Figure 1. Reef locations in the Florida Keys National Marine Sanctuary. Source: NOAA Technical
Memorandum NMFS-SEFSC-427.
Table 2. Number of stations surveyed in each habitat type.
Zone name, reef code, reef name, and the number of samples collected in back, fore and transitional reef
habitat for 2004 data (2003 samples noted in parentheses). Some repeat samples were duplicate
samples of the same station and other replicates were from different stations within the same reef.
Zone
Dry Tortugas
Key West
Lower Keys
Middle Keys
Upper Keys
Reef
BK
LR
ED
RK
SK
WS
ES
LK
AR
SR
CR
MR
Reef name
Bird/Bush Key
Loggerhead Reef
Eastern Dry Rocks
Rock Key
Sand Key
Western Sambo
Eastern Sambo
Looe Key
Alligator Reef
Sombrero Reef
Carysfort Reef
Molasses Reef
Back
1
1
1
1
1
1
1
Fore
1 (1)
1
2(2)
4(1)
1
1
1
1
3
1
Trans Unknown type
(4)
(5)
1
2(1)
1
3
1
1
1
1
1
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
The live surface area (LSA) for an individual colony was calculated as the product of total surface
area (TSA) and percent living tissue (%LC) divided by 100 for each individual colony. LSA for
each station-visit equaled the sum of the colony values and measured in m2. The percent living
surface area (%LSA) was calculated as LSA divided by TSA multiplied by 100%.
To summarize, TSA was the total surface area of coral in the reef transect. %LC was the average
percent live coral for all colonies in the reef transect such that small and large colonies contributed
equally to the station average. LSA was the sum of the live surface area of all the individual
colonies found within the reef transect. % LSA was the percentage live surface area in the radial
belt. For the last two metrics, large colonies contributed more to the final percentage than did
smaller colonies.
Reconciliation of sampling protocols
Year 2003 was a pilot year for data collection and methods were better defined in 2004. A subset of
coral species were measured in 2003 when the methods were being tested. In 2004, all the stony
corals were measured with the exception of two species. Millepora alcicornis was excluded because
it is predominantly an encrusting rather than reef-forming species (Table 3). Siderastrea radians
was excluded because colonies were small and occurred in clusters, which made it difficult to
discern individual colonies. In addition, S. radians occur flush to the sea floor and provide no
structural relief for habitat. In addition, from 2003 to 2004, size classes and coverage classes
changed. Nonetheless, because estimates of section differences and annual differences could only
be obtained using the 2003 data, an effort was made to make the data sets as similar as possible. A
section was defined as half the belt transect. Variability was estimated for neighboring sections
using the 2003 data because data from the two halves of the transect were only identified separately
during that year. Replicate samples collected at different stations during 2003 and 2004 were used
to evaluate differences associated with measurement error. More stations were visited during 2004
and sufficient data were collected to evaluate the influence of reef habitat type and reef location on
candidate coral metrics. Repeat visits to five stations in 2003 and 2004 were using to estimate
annual variance.
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Table 3. Coral species code, genera and species name.
Species code, genus, species and whether the taxon was excluded from sampling in 2003 or 2004.
Species Code
ACER
ARAL
AGAR
AFRA
ALAM
CNAT
DSTO
DCLI
DLAB
DSTR
MDEC
MFOR
MMIR
MAD
MMEA
MALC
MCOM
MANN
MCAV
MFAV
MFRA
MALI
MDAN
MFER
MLAM
MYCET
PAST
PPOR
SRAD
SSID
SIDsp
SBOU
SMIC
Genus
Acropora
Acropora
Agaricia
Agaricia
Agaricia
Colpophyllia
Dichocoenia
Diploria
Diploria
Diploria
Madracis
Madracis
Madracis
Madracis
Meandrina
Millepora
Millepora
Montastrea
Montastrea
Montastrea
Montastrea
Mycetophellia
Mycetophellia
Mycetophellia
Mycetophellia
Mycetophellia
Porites
Porites
Siderastrea
Siderastrea
Siderastrea
Solenastrea
Stephanocoenia
Species
cervicornis
palmata
agaricites
fragilis
larmarckii
natans
stokesii
clivosa
labyrinthyformis
strigosa
decactis
formosa
mirabilis
meandrina
alcicornis
complanata
annularis
cavernosa
faveolata
franksii
aliciae
danaana
ferox
larmarckiana
astreoides
porites
radians
siderea
bournoni
michelini
Excluded?
2003
2003
2003
2003
2003
2003
2004
2004
For all analyses of 2003 data, changes were made to match the 2004 protocols where possible. Two
species excluded in 2004 were excluded from the 2003 data (Millepora alcicornis and Siderastrea
radians). Size classes also changed from 2003 to 2004. In 2003, five size classes were used: 1, 10,
50, 100, and >100 L. In 2004, the SOL size was not used and three new large size classes were
added which yielded size classes of 1, 10, 100, 500, 2500, and >2500 L. For the size class >100 L a
value of 200 L was used; for the size class >2500 L a value of 5000 L. To make the 2003 data more
similar to the 2004 data, the 50 L size class was combined with the 100 L size class. From 2003 to
2004, %LC cover categories were reduced from seven to six categories. For the 2003 data, %LC
categories were modified to match the method used in 2004 (0, 1-25, 26-50, 51-75, 76-99, and
100%).
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
For most of the analyses involving the 2004 data, no changes were made to the data or calculations
of the coral metrics. An exception was the comparison of the 2003 and 2004 data to evaluate
changes through time. To make the two data sets as similar as possible, coral taxa excluded from the
sampling in 2003 were also eliminated from 2004 data for the annual comparison only. In addition,
the three largest size classes were collapsed to the largest size class recorded in 2003 (>100 L) to
make the candidate coral metrics more comparable for the annual comparison only.
Data analysis
A variety of replicate and duplicate samples were available to answer questions about patterns of
variance for the candidate coral metrics. Unfortunately, individual estimates of variance were
derived from different sets of stations because a limited number of repeat samples were available.
For the purposes of this analysis, "replicate" samples refer to repeat samples taken within a reef
while "duplicate" samples are repeat censuses of the same station within a reef.
Sampling designs also differed from 2003 to 2004. Half belt transects (sections) were sampled in
2003 only and these data were used to estimate variance associated with neighboring sections from
the opposite side of the belt transect. Duplicate samples were collected in both years and these data
were used to estimate variance associated with same-day revisits. Data from 2004 were used to
evaluate geographic differences in coral metrics associated with location along the archipelago.
Data from 2004 were used to compare coral metrics for different reef habitat types (back, fore, and
transitional). One station was eliminated from the analysis of variance for 2004; station WS04C had
an artificially high value for %LC and %LSA because dead A. palmata andM complanata were
accidentally excluded from the survey.
Data were used to address three questions:
What is the optimal size for a coral sampling unit?
What are the sources of variance for the candidate coral metrics and what are their
relative contributions to the total variance?
How much change in coral condition could we detect through time for a given sampling
effort?
Optimal sample size. To answer the first question, data from 2003 were used to compare %LC,
LSA and %LSA values from opposite sides of the radial belt. All 10 stations from 2003 had coral
survey data kept separate for each side of the belt (section) or two replicate samples for each station.
In some cases, the same section was surveyed more than once by different divers. In these cases,
%LC, LSA and %LSA were calculated for each diver and averaged for the section.
Sources of variance. Sources of variance for %LC, LSA, and %LSA that could be evaluated from
the available data included differences within reefs (stations), differences associated with habitat
type (fore, back, or transitional), microhabitat differences (neighboring sections), and measurement
error (duplicate samples). Back reef was defined as the area between the reef crest and the shore,
excluding sandy lagoons. Fore reef included the area from the reef crest that begins to slope down
7
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
into deeper water toward the bank or shelf. Transitional reef was the area sloping away from the
fore reef into deeper water; transitional reef is also referred to as the deep fore-reef.
An ANOVA model was used to estimate variance associated with opposite sections of the radial
transect where the 10 stations represented a single factor and the two replicate sections at each
station were used to estimate variance. A second ANOVA model was used to estimate variability
associated with 2003 duplicate samples using two stations with three duplicate samples each. For
2004, a third ANOVA model was used to estimate variability associated with 29 stations and three
duplicates at three of those stations for 2004. One duplicate sample for 2004 was eliminated
(WS04C) because dead A palmata andM complanata were inadvertently left out of the survey. A
fourth ANOVA model was used to estimate variance associate with annual duplicate sampling at
five stations sampled in 2003 and 2004. Each of these four ANOVA models was applied separately
to the four candidate coral metrics. Because the estimates of variance were derived from small
sample sizes as well as different study areas sampled at different times, the estimates should only be
considered approximate at best.
Because replicate samples within reefs were associated with different reef habitat types, that source
of variability could not be quantified for these data. Graphical analysis also revealed that although
coral metrics differed by reef habitat types, the differences were not consistent across reefs.
Additional analysis of habitat type was pursued using an exploratory multivariate technique,
principal components analysis (PCA). The total surface area of each species at each station was
used in PCA to evaluate patterns associated with reef habitat type.
Power analysis. Statistical power is defined as the probability of detecting a change when a change
truly occurs (Peterman 1990). Statistical power is a function of the number of samples collected, the
variance of the measure of interest, and the level of acceptable uncertainty (i.e., probability of a
Type I or Type II error).
Variance estimates were used to calculate the minimum detectable difference (MDD) for various
sampling scenarios for TSA %LC, LSA, and %LSA. For each candidate coral metric, the statistical
power was calculated to detect a change based on visits to 10, 20, 30, and 50 reef stations each year.
The statistical power of each of the four coral metrics was evaluated separately. Two statistical
models were used for each measure: a two-sample t test based on two sets of stations sampled in
different years and a paired t test based on repeat samples from the same stations in different years.
Two sample t test. "Two sample" refers to the fact that different reefs are visited on each occasion,
and two (different) samples are collected. Two variance estimates were available to use for this
model. During 2003, 10 stations were surveyed and during 2004, 29 stations were surveyed.
Variance estimates were selcted from 2004 data because more stations were sampled and because
the sampling protocol used was more robust.
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
MDD was calculated using the following equation (Zar, 1984):
MDD>
Where s2 = station variance for 2004,
n = the number of stations surveyed,
ta(i), v = the t value for an a of 0. 1 for a 1 -sided test,
fy(i), v = the t value for |3 of 0. 1 for a 1 -sided test, and
v = 2n-2.
a and |3 were set to 0. 1 because, in the context of resource monitoring, we are just as concerned that
we might miss a change (Type II error, |3) as we are that we might incorrectly conclude a change
has occurred (Type I error, a; Dayton 1998, Di Stefano 2001, Yoccoz et al. 2001). A 1-sided test
was selected because we are primarily concerned with loss of coral through time.
Paired t test. A "paired" test refers to the fact that each station is paired with itself for the second
sampling event and the same stations are visited during each visit. This model is based on a one-
sample t test in which the one sample is composed of the differences between each station during
the first and second visit. If the average differences are significantly greater (or less) than zero, a
change in condition would be indicated. Variance was calculated for the differences between each
coral metric observed at five stations during two different years.
The equation for MDD was very similar to the equation above:
MDD"
Where Sd2 = the variance of the differences between stations for repeat visits,
n = the number of stations surveyed,
^a(i), v = the t value for an a of 0. 1 for a 1 -sided test,
^p(i), v = the t value for |3 of 0. 1 for a 1 -sided test, and
v = n - 1.
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
RESULTS
During 2003, 23 coral taxa were identified to species and two were identified to genus
(Mycetophellia and Siderastred). A total of 28 coral species were identified and recorded in 2004.
Species within the genus Madracis were sometimes identified only to genus.
For the 2004 data, LSA was highly correlated with TSA (r = 0.92; Figure 2). One exception to this
relationship was SK01, which had several large colonies ofMontastrea cavernosa andM
faveolata, and Acroporapalmata with a high percentage of dead coral. In 2004, LSA ranged from
2.3-121.4 m2 and TSA ranged from 5.3-206.1 m2. When LSA was plotted against TSA for each
coral species, the correlation was again high (r = 0.95; Figure 3) with the exception of one species,
A. palmata which had a higher percentage of dead coral than other species. Because TSA and LSA
were so highly correlated, subsequent analysis focused on LSA alone.
140
120
100
80
60
40
20
0
FLKSK01
A
0 20 40 60 80 100 120 140 160 180 200 220
TSA (m2)
Figure 2. LSA was highly correlated with the TSA for 2004 stations measured using
a 113 m2 radial belt transect (N = 29; r2 = 0.84; Pearson's r = 0.92, p< 0.01).
10
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
240
200
160
< 120
80
40
MFAV
ARAL
PAST
SSID
A
MCA\,
100 200 300 400 500 600 700
ISA (m2)
Figure 3. LSA was highly correlated with the ISA for species summed across all
stations (r2 = 0.90; Pearson's r = 0.95, p < 0.1, N = 29 taxa).
Influence of geographic gradient and reef habitat type
%LC, LSA, and %LSA showed different patterns across stations when stations were ordered along
a geographic gradient from Key West to the ENE toward Carysfort Key (Figure 4). %LC did not
show an association with the geographic gradient. Station values for %LC ranged from 62-94%
with the exception of two stations, both of which were located in back reef habitat. RK03 and ES03
had much lower values of 28 and 26%. Duplicate values at station WS04 showed a much higher
variability for %LC than did duplicates at ES01 or CR02 stations.
LSA also failed to show an association with geographic gradient (see Figure 4), nor were the
outliers consistently associated with a particular type of reef habitat. In contrast, %LSA indicated
that a higher percentage of the areal coverage of the reef tended to be alive for stations located
further to the ENE. Some stations closer to Key West had a lower %LSA. In addition, some of the
lowest values for %LSA were associated with back reef habitat. For both LSA and %LSA, the
variance associated with duplicate samples was higher at station WS04 than at either ES01 or CR02
stations.
Reef habitat types differed in average depth with transitional reefs deepest, followed by fore reefs
and back reefs which were the most shallow (Figure 5). Coral metrics also differed according to reef
type (Figure 6). For some stations, %LC was much lower on back reefs. For several back reefs,
%LSA was much lower; in addition, %LSA was slightly lower in fore reef habitat than in
transitional habitat.
11
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
1 UU
90
80
70
60
50
40
30
on
+
jfi
. * *
" g o . ' +
4- n -|~ "r
i- ''r
+ +
+
ğ Back
ğ Ğ Trans
* + Fore
1 ^U
120
100
80
t
< 60
en
40
20
0
+
*
B
ğ
+
+
T + f i
+
* m +
* Ğ
SK RK ED WS ES LK SR AR MR CR
SK RK ED WS ES LK SR AR MR CR
< 50
en
SK RK ED WS ES LK SR AR MR CR
Figure 4. Shown for each reef are values for %LC, ISA, and %LSA observed in 2004. Reef habitat types
are noted. Reefs are arrayed from left to right in order from Key West toward the ENE to Carysfort Key.
Three duplicate samples were collected on WS (fore reef habitat), ES (transitional), and CR (fore), N = 35.
12
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Q_
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
100
90
80
70
60
50
40
30
20
Median
I I 25%-75%
T Non-Outlier Range
0 Outliers
* Extremes
Trans Fore Back
90
80
70
60
< 50
3
se 40
30
20
10
0
Trans Fore
Back
140
120
100
80
< 60
t/3
40
20
0
Trans Fore Back
Figure 6. Average %LC values were similar in different habitat types. %LSA was higher on transitional reef
habitats and %LSA was higher on fore reef than back reef habitat (upper panel). In contrast, ISA was lower in
transitional habitat and highest in some back reef habitats (lower panel). Back reef habitats were more variable
for all three coral metrics.
14
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Species assemblages associated with different reef habitat types
Principal components analysis (PCA) based on the total surface area (ISA) of every species at each
station identified the species assemblages associated with different reef habitat types. As a check
that PCA could successfully identify similar coral assemblages, duplicate samples from the same
station were compared. The three duplicates from three stations plotted very closely together when
station visits were plotted for the first two factors from the PCA (Figure 7). For the same plot, when
stations were identified by reef types, stations in back reef habitat plotted closely together (Figure
8). Some transitional habitat stations plotted close together. Fore reef habitat stations were also
grouped, but some stations were also spread among other stations that were characterized as back or
transitional habitats. Correlation between the factors and each coral species indicated thatAcropora
palamata was the primary species characterizing back reef habitat. The transitional habitat stations
in the upper right quadrant of Figure 8 were characterized by Diploria clivosa, D. strigosa,
Dichocoenia stokesii, Madracis decactis, Meandrina meandrina, and Solenastrea bournoni.
Stations in fore habitat and some stations in transitional habitat were characterized by a mix of
several species with Montastreafaveolata and M. franksii being the most highly negatively
correlated with PCA factor 1.
6
5
4
3
2 2
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
6
5
4
3
& 2
ro 1
0
-1
-2
-3
++
+
-7
-5 -4
-3 -2
Factor 1
-1
Figure 8. Factors 1 and 2 from PCA for data collected in 2004. Each point represents a
station visit. Back, fore, and transitional reef habitat types are indicated.
Sources of variance
Patterns of variance differed according to the candidate coral metric being considered (Table 4). As
for previous comparisons, TSA and LSA showed similar patterns and only LSA is reported here.
For LSA, the greatest percentage of the total variability was associated with differences in reefs.
Relatively smaller percentages of the total variance were due to year, duplicates and sections
(Figure 9).
For %LC, the total overall variance was dominated by differences associated with duplicate samples
and year differences. These results were somewhat difficult to interpret because duplicate variance
estimates were derived from only three stations and two of the stations had nearly identical values
for all three duplicates while one station (WS04) had very different values for duplicate samples.
The latter sample greatly influenced the estimate of variance due to duplicate sampling. For %LSA,
variance associated with duplicate sampling represented a large proportion of the total variance,
again due primarily to the very different values observed at WS04.
16
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Table 4. Variance estimates for TSA, %LC, LSA and %LSA.
Estimates of variance for different sources of variability and four coral candidate metrics. Variance estimates
were derived from different data sets and ANOVA models. Shown are year of data collection, variance
estimates for TSA, %LC, LSA and %LSA, number of stations from which the estimates were derived, number
of replicate sections, and number of duplicate samples (repeat measure of the same station). WS04C was
left out of the calculations for 2004 because dead A. palmata and M. complanata were left out of the survey.
Source
Year
TSA
%LC LSA %LSA Stations Reps Dupes
Sections
2003
94.73
16.79 27.73 50.06
10
2 each
Same day reps 2003
Same day reps 2004
Stations 2004
66.20 2.64 13.66 23.50 2
315.99 83.63 67.94 142.11 3
2361.28 98.23 932.99 198.28 29
Annual
2003-04 250.82 89.36 38.96 19.72
3 each
2-3 each
2 each
110
CD c
O) CO
co -sz
= 5
CD >
g
CD
CL
TSA %LC LSA %LSA
Section
Dupes (2004)
Year
Station
Figure 9. Relative contribution to the overall variance of candidate coral metrics due to section
differences (2003 data), same-day duplicate sampling (2004 data), year differences (2003 and 2004),
and station differences (2004 data).
17
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
No measurable variance could be associated with habitat type for any of the candidate coral metrics
because the differences were not consistent across reefs. For example, for LSA at stations SK, RK
ES, and LK much lower values were recorded in back reef than in other habitat types. In contrast, at
ED, WS and CR, the highest values for LSA were found in back reef habitat.
Patterns observed for the variability analysis represent a statistical approach to what can be seen in
Figure 4. Taking %LC as an example, reefs were fairly similar in values of %LC and that pattern is
reflected in the very low percentage of the total variance associated with reef. In contrast, the values
for duplicate samples on reef WS extended beyond the range of values for most other reefs and that
pattern is reflected in the high percentage of the variance associated with duplicate sampling.
Annual differences could only be assessed from visits to five stations during 2003 and 2004 (Figure
10). %LC was higher at all five stations during 2003. Observed differences in %LC may have been
due to differences in the sampling protocol. The inclusion of small colonies in 2003 may have
results in higher values for %LC particularly if smaller coral colonies are younger and tend to be
healthier. Agreement between stations for both LSA and %LSA was very good.
Power analysis
A key point to consider when interpreting the results of the power analysis, is that the ranges of
values for TSA, %LC, LSA, and %LSA were much smaller for these five stations than for the larger
set of 29 reefs sampled in 2004. For example, LSA in 2004 ranged from 23-121A m2, but the
highest value observed for the five repeat sites was 45.4 (see Figure 5; Table 5).
As expected, smaller differences in coral metrics would be statistically significant for a paired
model than for the two-sample model (Table 6). This difference is expected because pairing a
station with itself eliminates the variance associated with different stations. Differences in MDD for
the two sampling designs were more extreme for TSA and LSA than for %LC or %LSA (Figure
11). Both TSA and LSA were too variable to detect a change for a design based on 10 stations using
a two-sample test. The MDD subtracted from the mean of the observed values went below zero. For
both %LC and %LSA, the MDD was small enough that the difference between the observed mean
and the change in the mean that would indicate a significant change was also relatively small. For
these candidate coral metrics, the power to detect change was high and several significant
increments of change could be detected before the mean values declined to zero.
18
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
60
60
65
70
75 80
%LC 2003
85
90
95
100
90
80
70
60
50
40
30
20
10
10 20 30 40
50 60 70
LSA 2003
90 100
10
10 20 30 40 50
%LSA 2003
60
70
80
Figure 10. Comparison of %LC, LSA, and %LSA for five stations sampled in both 2003
and 2004. Dotted lines indicate the line of perfect agreement. The ranges of the axes
match the range of values observed from the larger 2004 data set of 29 stations to provide
perspective on differences observed in 2003 vs. 2004.
19
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
For TSA and LSA, the paired test was much more likely to detect a change through time. For %LC
and %LSA, the paired test was also more sensitive, but the difference in statistical power was not
that much greater than for the two-sample test. The primary advantage associated with a two-sample
test is that twice as many sites can be visited during the same time period.
Table 5. %LC, LSA, and %LSA for both 2003 and 2004.
Values observed for %LC, LSA and %LSA during each year, the difference between years, and below,
the mean and variance of the differences for each coral metric.
Station 2004
ED01 65.3
SK01 69.1
SK02 77.9
SK03 69.1
WS03 61 .6
Average (d)
Variance (d)
%LC
2003 d
90.0 -24.7
77.9 -8.8
82.6 -4.7
76.1 -7.1
73.1 -11.5
-11.4
61.9
LSA
2004 2003 d
18.0 14.6 3.3
37.8 39.5 -1.7
27.2 45.4 -18.2
17.3 13.0 4.3
18.6 13.5 5.2
-1.4
94.9
%LSA
2004 2003 d
52.6 56.8 -4.3
53.5 56.4 -2.9
61.3 49.3 12.0
55.2 59.9 -4.7
54.7 52.5 2.2
0.5
49.0
Table 6. MOD for two-sample t test and paired t test.
MOD for four candidate coral metrics for 10, 20, 30 and 50 stations sampled during each time period. As
an example, for %LC and 10 different stations sampled per visit, a decline of 11.8% would likely represent
a statistically significant change; if the same 10 stations are sampled on both occasions, a decline of
6.9% would likely represent a significant change (a = p = 0.1).
Number of
stations
Mean
10
20
30
50
TSA
2 sample
45.0
57.8
40.1
32.5
25.1
Paired
45.0
21.5
14.6
11.8
9.0
%LC
2 sample
74.3
11.8
8.2
6.6
5.1
Paired
74.3
6.9
4.7
3.8
2.9
LSA
2 sample
24.5
36.3
25.2
20.4
15.8
Paired
24.5
8.5
5.8
4.7
3.6
%LSA
2 sample Paired
55.2
16.8
11.6
9.4
7.3
55.2
6.1
4.2
3.4
2.6
20
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
50
40
30
20
10
10
20 30
Number of stations
50
30
25
20
< 15
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
DISCUSSION
Stations sampled for this study represented a very large geographic area from Carysfort Key at the
northeastern end of the Florida Keys, south and west past Key West to the Dry Tortugas. Although
the survey design used to collect data for this study was not ideal for evaluating patterns of
variance, the data were not collected with this specific purpose in mind. The primary purpose of the
original project was to refine and test sample survey protocols to determine how coral should be
censused and how best to summarize information for each colony and the reef transect. As part of
that study, some replicate and duplicate samples were collected. From these pilot data, the relative
contribution of the different sources of variance to the overall estimates of reef condition could be
evaluated; however, the limited number of replicate samples and the fact that variance estimates for
different sources were made from different stations should be kept well in mind. In other words, the
results from this analysis are likely reliable for relative comparisons, but the exact values for
variance may not be accurate. Because coral reef data are expensive to collect, many data sets must
serve more than one purpose. Although more replicate samples would yield more robust
conclusions, the data available for the Florida Keys can, nonetheless, provide valuable insight into
the best approaches for coral reef monitoring and protection.
The four candidate coral metrics summarized different aspects of reef condition and also behaved
quite differently in terms of their variance structure and their association with natural features. Coral
species contributed unequally to the coral metrics due to large differences in size. Five species
contributed disproportionately to LSA calculations and swamped the contributions of other smaller
colonies. These species were Acroporapalmata, Montastrea cavernosa and M. faveolata,
Siderastrea siderea, and Porites astreoides. Similarly, %LC and %LSA summarized two very
different aspects of reef condition. %LC treated all colonies equally, averaging across colonies
irrespective of size. In contrast, %LSA measured live tissue on the reef surface as a function of area
while ignoring the identity of individual coral species.
TSA and LSA tracked each other closely for all analysis and a reasonable conclusion seems to be
that the living surface area depends heavily on the amount of coral surface area available. Patterns
of variance for %LC and %LSA were also somewhat similar to each other. These results indicate
that the most efficient sampling designs for these two sets of variables may be different.
TSA and LSA varied locally throughout the Florida Keys; in contrast, %LC was much less variable
from reef to reef. %LSA showed marked differences from reef to reef, with the lowest values
observed around Key West on fore and back reef habitat. If the observed differences are associated
with human land use, %LSA may be a strong indicator of reef condition. Many of the highest
observed values for TSA and LSA were associated with stations near Key West, though high values
were also observed on reefs near the northeast end of the sampled area.
Because the candidate coral metrics summarize different aspects of reef condition, they may also be
sensitive to different aspects of human disturbance. LSA may be a better indicator of physical
damage while %LC may be a better indicator of global stressors such as increasing ocean
temperatures. %LC and %LSA may also be sensitive to local stressors such as nutrient enrichment
or sediment associated with terrestrial land use practices.
22
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
Natural sources of variability
Three natural sources of variability could be assessed with these data: geographic gradient, reef
habitat type, and depth. %LC showed very little association with geographic gradient with most
stations showing similar values. An exception was two stations with much lower %LC, both of
which were located on back reef. LSA and TSA showed a broader range of values from reef to reef
with the largest differences observed around Key West. %LSA showed the strongest association
with geographic gradient in that there was a trend toward greater %LSA in the more northeastern
reefs.
For all four candidate coral metrics, stations on the same reef were very different. It was not
generally true that stations within reefs were more similar than stations on different reefs. Lack of
replicates within reefs meant that the effects of different locations within a reef could not be
separated from different habitat types within a reef.
Although stations within reefs varied according to habitat type, the differences were not consistent
across reefs, e.g., back reefs were not consistently lower for %LSA. The only consistent pattern was
that back reefs were the most variable and fore reef habitat was slightly more variable than
transitional habitat for all candidate metrics.
Depth and habitat type were strongly associated with each other: transitional habitat was deepest,
followed by fore reef habitat and then back reef habitat. Whether the reef habitat type or the depth
was the more important factor underlying observed differences could not be determined from the
data. Depth may be an inherent feature of the type of reef habitat. Given coral dependence on light,
depth may also be a factor in determining which coral were associated with different types of
habitat.
To further explore patterns in the coral assemblage associated with reef habitat type, an exploratory
technique (PC A) was used to identify which coral species were more typical of the different habitat
types. Back and transitional habitats had a small subset of coral species that characterized the
assemblage. These results further support the idea that reefs differ by habitat type and any survey
design developed to assess reef condition should consider either stratifying on habitat type or only
assessing one reef habitat. Creating strata for different subpopulations can greatly reduce the
variance associated with the final estimates of reef condition when the strata differ for whatever
reason, either due to underlying natural differences or the influence of human disturbance.
Variance structure
ANOVA was used in two different ways for this study, both to compare the relative contribution of
different sources of variance to the total variance of each candidate coral metric as well as to
estimate the actual variance values which were then used to calculate the MDD. Unfortunately, the
estimates of variance associated with different reef sections, duplicate samples, annual differences
and station differences were all derived from different stations. A better design would estimate each
source of variance using the same locations. The difficulty associated with estimating variance from
different reefs is that the individual stations may be more variable in general (regardless of the
reason) and, therefore, the particular aspect of variance being estimated from those locations may be
inflated. Although the actual numbers derived for variance from this study may be approximate,
23
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
several general conclusions could be drawn from the data that can guide the design of future
surveys.
During 2003, data from half-sections of the each belt transect were kept separate. Calculations from
the two halves of the belt transect showed high agreement for all four coral metrics and the variance
estimates associated with differences from one side of the transect to the other were relatively small.
The close agreement suggest that the sampling unit (i.e., the belt transect) could be smaller. Smaller
sampling units translate into less field time at each station and provide the opportunity to visit more
sites. When surveying large areas, a small amount of information collected from more locations is
often preferred to highly precise information about fewer locations because larger sample sizes
yield more accurate estimates of resource condition.
Variance analysis formalized the observations first summarized in simple graphs of the candidate
coral metrics. For TSA and LSA, the greatest differences were associated with different stations.
Neighboring sections and duplicates varied little and, furthermore, stations varied little from year to
year. In contrast, %LC varied little from station to station, most of its variability was associated
with duplicate sampling and annual differences. However, although these sources of variability
represented a large percentage, their actual value was still quite low. This translated into a high
precision and a good ability for %LC to potentially detect even very small changes in reef
condition. %LS A had the greatest proportion of its variability associated with duplicate sampling;
unfortunately, this result was driven almost entirely by the high variability in duplicates found at
one station (WS). %LSA for different years at five other stations showed high agreement which
translated into a good ability of %LSA to detect a change through time for the MDD analysis.
Duplicate samples at stations represented the measurement error associated with the sampling
protocol. Duplicate same-day samples ideally should yield the same values because coral condition
does not change within such a short period of time. For coral surveys, measurement error could be
due to differences in crews, difficulty assigning size category classes or coverage classes of live
coral, or missed coral. The small number of reefs with duplicate samples made it difficult to draw
robust conclusions regarding the magnitude of the measurement error, particularly because the three
stations with duplicates in 2004 showed such markedly different patterns. Two stations showed very
consistent values for all coral metrics, but a third, WS, showed very divergent values for duplicates.
With a small data set, it's difficult to know which stations would be more typical. To better answer
this question and estimate the measurement error more accurately, duplicate samples from 10-20
stations would be needed.
The data set used to estimate year to year variability was quite small (five stations) and included
only two years. A more reliable data set would have repeat samples from 20-30 stations sampled for
three to five years. As a consequence, actual variance estimates are most certainly approximate;
however, MDD values derived from the variance estimates provide information about the relative
sensitivity of the different measures. For example, TSA and LSA need a paired sampling approach
to detect change through time because the observed values varied greatly from station to station. In
contrast, %LC could be successfully monitored using different stations each year. For this analysis
only two time periods were considered, but the approach could be expanded to a trend model based
on multiple years of sampling when better estimates of annual variability become available
(Urquhart et al. 1998).
24
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
The two statistical models used in this analysis demonstrated that a much smaller change can be
detected when a site is paired with itself through time than when new sites are sampled each year.
This difference in sampling design parallels a larger issue which is the trade-off between the
assessment of status and trends. To assess the general, overall status of a large regional resource, we
are more confident when the assessment is based on visits to many locations. In contrast, the most
sensitive design for detecting trend through time visits the same locations each year. Allocation of
resources depends on which aspect of monitoring is more important for resource protection.
CONCLUSIONS
The field sampling protocol used to estimate candidate coral metrics provided reliable, repeatable
estimates of coral condition. In addition, all coral metrics had good statistical precision for detecting
change through time. The physical area sampled at each station may actually be larger than needed
to assess a station. Association between geographic location and some coral metrics may indicate a
natural gradient or a difference in coral condition which should be considered when designing a
sample survey for a large geographic area. Reef habitat type was associated with different species
assemblages as well as much of the variability within a reef. Differences within reefs should be
considered when designing a monitoring plan by either stratifying on homogeneous habitat types or
by sampling only in habitats that are easy to identify. All four candidate coral metrics had adequate
statistical precision to detect change through time based on a reasonable number of stations
sampled. The candidate metrics quantified very different aspects of coral reef condition; because of
this difference, their variance structure also differed. Consequently, a single sampling design may
not be optimal for all measures. Development of a long-term monitoring plan for a resource as
extensive and complex as coral reefs is an iterative process. Results from this study support the
implementation of these field collection protocols and coral metrics to monitor and protect coral
reefs.
25
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
RECOMMENDATIONS
Connect measures of reef condition to stressors or measures of site condition.
A good indicator is characterized by three features: 1) it represents an aspect of the biota that
is biologically meaningful, 2) it correlates with independently derived measures of site
condition or stressors, and 3) it has adequate statistical precision to detect a change should a
change occur (Fore 2003). The first and third features have been demonstrated for these
candidate coral metrics, but the second test remains. A strong relationship between
indicators and stressors also must be established before any assessment based on coral
condition can be interpreted. TSA and LSA were more variable across stations than %LC
and %LSA. Whether these differences reflect underlying natural variability or human
disturbance would influence the choice of sampling design.
Reconsider the size of the sampling unit
Because coral reefs are continuous resources, dividing them into discrete sampling units is
somewhat arbitrary (Stevens and Olsen 1999). On the one hand, sampling units should be as
small as possible to minimize time spent in the field collecting data. On the other hand,
enough data should be collected to adequately characterize the sampling location and
minimize measurement error associated with microhabitat differences. The close agreement
for these candidate coral metrics from opposite sides of the radial belt transect suggests that
a smaller sampling area may be adequate. A simple way to decide would be to continue with
the radial belt transect method as is, but divide the radial belt into four parts in the field and
note the quadrant for each coral surveyed. The variance from four quarters, two halves, and
duplicates could be compared for each coral metric and the optimal sampling unit size
determined.
Recognize habitat types in future survey designs.
Much of the variance in these four coral metrics was associated with different habitat types
although differences could also be more simply explained by differences in locations on the
reef; the two potential sources of variance could not be distinguished for these data. It was
clear, however, that back reef habitat was much more variable than other habitat types for all
candidate coral metrics. Could back reef habitat be further divided into subcategories? The
key point for managing this source of variance is to continue to distinguish between habitat
types when collecting data, but also to consider only sampling the habitat type that is the
least variable or, alternatively, the best indicator of reef condition.
Revisit at least 10 of the stations.
Variability through time is one of the most difficult aspects of variance to assess but critical
for designing long-term monitoring plans (Larsen et al. 2001). If possible, at least 10 stations
should be revisited soon and in subsequent years.
Design experiments with a balanced design.
Any future survey design is likely to have replicate sampling included or imposed by the
funding agency. A balanced design provides much more reliable variance estimates and is
characterized by equal sampling of stations for each aspect of variance being assessed. For
26
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
example, if variance of habitat type and variance of stations within reefs are to be assessed,
the design should include the same number of replicates at each station and the same
locations should be used to evaluate both. A specific example might select 10 reefs and
sample two stations each in back, fore and transitional reef areas. Selecting site locations
that are nearby eliminates underlying natural sources of variability that the sampler may be
unaware of. In general, a better design would have two replicates from 10 stations rather
than three replicates from six stations.
27
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
REFERENCES
Dayton, P.K. 1998. Reversal of the burden of proof in fisheries management. Science 279:821-822.
Di Stefano, J. 2003. How much power is enough? Against the development of an arbitrary
convention for statistical power calculations. Functional Ecology 17:707-709.
Fisher, W. S., W. P. Davis, R. L. Quarles, J. Patrick, J. G. Campbell, P. S. Harris, B. L. Hemmer
and M. Parsons 2006. Characterizing coral condition using estimates of three-dimensional
colony surface area. Environmental Monitoring and Assessment, (online; print version not yet
available)
Fore, L.S. 2003. Developing Biological Indicators: Lessons Learned from Mid-Atlantic Streams.
EPA 903/R-003/003. U.S. Environmental Protections Agency, Office of Environmental
Information and Mid-Atlantic Integrated Assessment Program, Region 3, Ft. Meade, MD.
www.epa.gov/bioindicators/
Jameson, S.C., M.V. Erdmann, J.R. Karr, K.W. Potts. 2001. Charting a course toward diagnostic
monitoring: A continuing review of coral reef attributes and a research strategy for creating
coral reef indexes of biotic integrity. Bulletin of Marine Science 69(2):701-744.
Larsen, D.P., T.M. Kincaid, S.E. Jacobs, and N.S. Urquhart. 2001. Designs for evaluating local and
regional scale trends. BioScience 12:1069-1078.
Olsen, A.R., J. Sedransk, D. Edwards, C.A. Gotway, W. Liggett, S. Rathbun, K.H. Reckhow and
L.J. Young. 1999. Statistical issues for monitoring ecological and natural resources in the
United States. Environmental Monitoring and Assessment 54: 1-45
Peterman, R.M. 1990. Statistical power analysis can improve fisheries research and management.
Canadian Journal of Fisheries and Aquatic Sciences 47: 2-15.
Skalski, J.R. 1990. A design for long-term status and trends monitoring. Journal of Environmental
Management 3 0:13 9-144.
Stevens, D.L. and A.R. Olsen. 1999. Spatially restricted surveys overtime for aquatic resources.
Journal of Agricultural, Biological and Environmental Statistics 4:415-428.
Urquhart, N.S., S.G. Paulsen and D.P. Larsen. 1998. Monitoring for policy-relevant regional trends
overtime. Ecological Applications 8:246-257.
Ward, R.C., J.C. Loftis and G.B. McBride. 1986. The "data rich but information poor" syndrome in
water quality monitoring. Environmental Management 10:291-297.
Yoccoz, N.G., J.D. Nichols and T. Boulinier. 2001. Monitoring of biological diversity in space and
time. Trends In Ecology & Evolution 16: 446-453.
Zar, J.H. 1984. Biostatistical Analysis, 2nd ed. Prentice-Hall, Inc., Englewood Cliffs, NJ.
28
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
APPENDIX A
CANDIDATE CORAL METRICS BY SECTION (2003)
Study area, station name, section, initial of diver, and values for coral reef measures collected in 2003.
For sections with duplicate samples, average values are shown.
Study Area
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Station
BK06
BK06
BK06
BK06
BK06
BK06
LR05
LR05
LR05
LR05
LR05
LR05
Section
NS
SN
NS
SN
NS
SN
NS
SN
NS
SN
NS
SN
Diver
BQ
BQ
JC
JC
MP
MP
BQ
BQ
JC
JC
MP
MP
ISA
23.2
20.7
21.6
19.4
34.1
28.1
15.3
12.9
20.1
7.7
19.1
9.6
%LC
87.7
89.1
84.0
86.9
81.6
87.3
90.9
87.0
88.1
93.3
87.8
92.5
LSA
15.6
16.0
14.1
16.2
19.1
20.7
13.5
8.7
16.3
5.6
14.5
5.9
%LSA
67.4
77.6
65.1
83.5
56.1
73.6
88.3
67.1
81.2
72.4
75.7
61.9
29
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
APPENDIX B
CANDIDATE CORAL METRICS FOR STATIONS (2003)
Study area, station name, and values for coral reef measures collected in 2003.
Study Area
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Dry Tortugas
Key West
Key West
Key West
Key West
Key West
Station
BK06
BK06
BK06
BK07
LR05
LR05
LR05
LR06
LR07
ED01
SK01
SK02
SK03
WS03
ISA
43.9
41.0
62.2
39.6
28.2
27.7
28.7
37.6
28.4
25.8
70.0
92.1
21.7
25.6
%LC
88.4
85.4
84.2
84.5
88.9
90.3
90.0
87.0
89.5
90.0
77.9
82.6
76.1
73.1
LSA
31.7
30.3
39.8
30.0
22.2
21.8
20.4
30.1
21.4
14.6
39.5
45.4
13.0
13.5
%LSA
72.2
73.8
64.0
75.7
78.6
78.8
71.1
80.3
75.2
56.8
56.4
49.3
59.9
52.5
30
-------
Bioassessment Tools for Stony Corals:
Statistical Evaluation of Candidate Metrics in the Florida Keys
APPENDIX C
CANDIDATE CORAL METRICS FOR STATIONS (2004)
Study area, station name, duplicate sample identification, and values for coral reef measures collected
in 2004.
Study area
Middle Keys
Middle Keys
Upper Keys
Upper Keys
Upper Keys
Upper Keys
Upper Keys
Key West
Key West
Key West
Lower Keys
Lower Keys
Lower Keys
Lower Keys
Lower Keys
Lower Keys
Lower Keys
Lower Keys
Upper Keys
Upper Keys
Key West
Key West
Key West
Key West
Key West
Key West
Key West
Middle Keys
Middle Keys
Key West
Key West
Key West
Key West
Key West
Key West
Station Dupe
AR01
AR02
CR01
CR02 A
CR02 B
CR02 C
CR03
ED01
ED03
ED04
ES01 A
ES01 B
ES01 C
ES02
ES03
LK01
LK02
LK03
MR01
MR02
RK02
RK03
SK01
SK02
SK03
SK04
SK05
SR01
SR02
WS02
WS03
WS04 A
WS04 B
WS04 C
WS05
ISA
5.6
5.8
35.4
31.6
53.4
49.3
110.1
60.7
123.0
38.5
41.8
45.9
41.5
43.2
33.4
157.8
162.3
5.3
85.7
15.2
85.2
45.9
206.1
62.0
59.9
28.4
42.4
8.9
26.5
140.6
54.8
110.4
161.3
79.7
56.6
%LC
79.3
75.6
70.7
83.3
73.3
77.5
83.2
65.3
73.5
81.5
75.8
72.5
78.2
69.7
26.0
76.7
83.8
82.6
77.8
72.4
72.2
28.0
69.1
77.9
68.4
80.9
80.0
76.4
74.3
87.2
61.6
93.7
67.2
86.2
80.0
LSA
4.1
4.4
23.8
24.2
31.9
31.3
75.3
24.1
83.7
29.9
25.5
25.3
24.2
19.3
4.1
100.9
121.4
2.3
50.2
8.8
54.9
3.6
73.9
27.4
30.4
19.8
10.8
6.8
13.7
107.1
25.4
61.2
36.6
55.1
28.1
%LSA
72.8
75.7
67.2
76.5
59.6
63.5
68.4
39.7
68.0
77.6
60.9
55.1
58.3
44.7
12.2
63.9
74.8
42.4
58.5
57.5
64.4
7.9
35.8
44.3
50.7
69.5
25.5
76.7
51.6
76.2
46.3
55.5
22.7
69.2
49.6
31
------- |