This item is only available as the following downloads:
THE HEURISTIC ANALYSIS OF TEMPORAL AND SPATIAL VARIATION
WITHIN THE ZOOPLANKTON COMMUNITY STRUCTURE
AT THE CRYSTAL RIVER ESTUARY:
A MULTIVARIATE APPROACH
William Ingram III
A DISSERTATION PRESENTED TO THE GRADUATE
COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
I can not possibly acknowledge all the faculty, staff,
fellow students and friends who have contributed materially
to the fruition of this effort. However, I would like to
specially acknowledge a few individuals who were instru-
mental in the realization of this dissertation.
First, without my wife Janice's steady encouragement,
support, aid in preparation of the manuscript, and unending
patience, it would have been impossible to complete the
To Dr. Frank J. S. Maturo, Jr. goes my sincere appreci-
ation for his support in this project. His suggestions dur-
ing the planning of the project and the critical review he
provided during the writing, improved the clarity and
organization of the dissertation.
Through a long association with Dr. James T. McClave,
on this project and others, I have gained an appreciation
for the art and science of data analysis. This skill was
central to the accomplishment of this research project.
This study was supported by a contract with the Florida
Power Corporation, through the University of Florida Marine
Laboratory. The Computing facilities were provided by the
North East Regional Data Center (NERDC) and the Center for
Instructional and Research Computing Acti-vities.
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ....................... ............ ii
LIST OF TABLES .................................... vi
LIST OF FIGURES .................................. vii
ABSTRACT ........................................ ix
I. INTRODUCTION ................................... 1
II. MATERIALS AND METHODS ........................ 4
Study Site .................................. 4
Physical features ...................... 4
Hydrography ............. ............... 11
Hydrographic areas .................... 11
Circulation patterns .................. 13
Field Sampling ............................. 15
Station placement ...................... 15
Sample collection technique ........... 19
Sample size determination ............. 21
Biweekly Sampling Program .............. 23
Diurnal Sampling Program ............... 24
Laboratory Methods ........................... 26
Sample Processing ....................... 26
Zooplankton Categories ................. 27
Biomass Determination .................... 28
Data Analysis ................................. 29
General Considerations ................... 29
Approaches to community structure
analysis ............................ 31
Heuristic analysis ................. 35
Analysis models ..................... 38
Underlying assumptions .............. 39
Selecting the analysis model ........ 40
Displaying the results .............. 42
Biweekly Analysis ........................ 43
Determination of seasonality ........ 43
Community structure analysis ........ 47
Structure plot analysis ............. 48
Canonical discriminant analysis ..... 48
Diurnal Analysis ......................... 54
Diurnal response surface analysis ... 56
Dependent Variable Transformations ....... 60
III. Results ...................................... 62
Biweekly Analyses ............................. 62
Determination of Seasonal Boundaries ..... 62
Biweekly Canonical Discriminant Analysis 70
Canonical discriminant function I ... 78
Canonical discriminant function II .. 80
Canonical discriminant function III 82
'ooplankton Community Structure Analysis 84
Factor 1 ............................ 93
Factor 2 ........................... 96
Factor 3 ........................... 97
Structure Plot (SPLOTS) Analysis.......... 97
Diurnal Series ............................... 100
Factor Analysis ......................... 100
Stepwise Regression ..................... 102
Factor Loadings ......................... 116
Factor 1............................ 118
Factor 2 ............................ 119
Factor 4............................ 119
Factor 5............................. 121
Factor 7 ............................ 121
IV. Discussion..................................... 122
Biweekly Data ................................. 122
Canonical Discriminant Analysis.......... 122
Community Structure Factor Analysis....... 129
Diurnal Data................................... 135
Associations of Community Structure
with the Environment..................... 135
V. Summary and Conclusion ......................... 145
Appendix 1: Environmental Parameters......... 158
Appendix 2 : Biweekly Statistics.............. 173
Appendix 3 : Seasonal Distribution Maps....... 374
Appendix 4 : Diurnal Stepwise Regression...... 529
Appendix 5 : Diurnal Response Surfaces......... 556
LIST OF TABLES
1. Correlation coefficients between each
canonical variable and
the dependent variables .......................... 65
2. Determination of the number of
CDFA axes to retain for biweekly data ........... 71
3. Correlation coefficients between each
canonical variate and the dependent
variables for the biweekly data analysis ........ 85
4. Portion of zooplankton community structure
variation explained by the factor analysis ...... 86
5. ANOVA for biweekly community structure
factor analysis ................................. 94
6. Rotated factor pattern loadings of the
dependent variables for the biweekly
data ............................................ 95
7. Rotated factor pattern loadings of
the dependent variables for the
diurnal data analysis ........................... 101
8. Portion of the zooplankton diurnal
variation explained by
the factor analysis ............................. 103
9. Parameters selected by stepwise
regression for each factor ...................... 104
10. Response surface parameter coefficients
for diurnal analysis ............................ 105
LIST OF FIGURES
1. Hydrographic zones and location
of the study region ............................... 6
2. Physiographic features and station placement ..... 18
3. Cluster dendrogram of biweekly environmental
variable mean vectors ( a= 1st biweekly
in month, b= 2nd biweekly in month) ............. 63
4. Season means for CDF axis I and CDF axis II ...... 67
5. Plot of the geographic distribution of
canonical variate I. (a) Fall, (b) Winter,
(c) Spring, (d) Summer ............................ 73
6. Plot of the geographic distribution of
canonical variate II. (a) Fall, (b) Winter,
(c) Spring, (d) Summer ............................ 75
7. Plot of the geographic distribution of
canonical variate III. (a) Fall, (b) Winter,
(c) Spring, (d) Summer ............................ 77
8. Plot of the geographic distribution of
factor scores for factor 1. (a) Fall,
(b) Winter, (c) Spring, d) Summer ................ 88
9. Plot of the geographic distribution of
factor scores for factor 2. (a) Fall,
(b) Winter, (c) Spring, d) Summer ................ 90
10. Plot of the geographic distribution of
factor scores for factor 3. (a) Fall,
(b) Winter, (c) Spring, d) Summer ................ 92
11. Structure plots (SPLOTS) for the factor
scores for the three factors. (a) Factor 1
(b) Factor 2, (c) Factor 3 ....................... 99
12. Response surface plot for diurnal factor
scores for Factor 1 ............................ .. 107
13. Response surface plot for diurnal factor
scores for Factor 2 .............................. 109
14. Response surface plot for diurnal factor
scores for Factor 4 .............................. 111
15. Response surface plot for diurnal factor
scores for Factor 5 .............................. 113
16. Response surface plot for diurnal factor
scores for Factor 7 .............................. 115
Abstract of Dissertation Presented to the Graduate
Council of the University of Florida in Partial
Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
HEURISTIC ANALYSIS OF TEMPORAL AND SPATIAL VARIATION
WITHIN THE ZOOPLANKTON COMMUNITY STRUCTURE
AT THE CRYSTAL RIVER ESTUARY:
A MULTIVARIATE APPROACH
William Ingram III
Chairman: Frank J. S. Maturo, Jr.
Major Department: Zoology
A study was made of the spatial and temporal variations
of the zooplankton community at the Crystal River estuary.
Included in the study were considerations of the effects of
power plant operations on the structure of the zooplankton
community. The power plant was seen to have a discernible
effect on the community structure and its patterns of
The estuary was divided into 6 regions based on its
physical features. The environmental data collected during
the study were used to determine the seasons.
The interactive effects of season and region on the
zooplankton community structure were investigated through
nultivariate pattern recognition techniques. These effects
were most apparent during the warm seasons of the study.
The zooplankton community structure differences separated
the estuary into 2 regions: an inshore region consisting of
the inner estuary and the discharge side portion of the
mixing zone, and an outer region.
The zooplankton community differences were related to
shifts in the standing crop of thermally sensitive or
thermally tolerant organisms.
The thermally sensitive organisms included: Paracalanus
quasimoto, Lucifer sp., and Labidocera sp. These are
neritic organisms that are not adapted to the high ambient
water temperatures found in a subtropical estuary during the
spring and summer months.
The thermally tolerant organisms, which had higher
standing crops in the inshore areas, included: bivalve
larvae, Temora turbinata, barnacle larvae, and Tortanus
setacaudatus. These represent coastal zooplankters adapted
to higher ambient water temperatures usually found in the
inshore regions of subtropical estuary systems.
The differences between the season-region combinations
represent a differential response of the zooplankters to the
selection pressure of thermal stress and the influence of
the power plant operation. These differences are subtle and
do not include the dominant members of the zooplankton
The zooplankton community structure was decomposed by
factor analysis into 3 basic components: the seasonally
abundant meroplankters, the neritic holoplankters and the
The zooplankton community structure was decomposed by
factor analysis into 3 basic components: the seasonally
abundant meroplankters, the neritic holoplankters and the
coastal or estuarine holoplankters.
The response surface analysis of the diurnal data
estimates from field data the fitness surface for a group of
zooplankton categories. The fitness set mappings showed
temperature and salinity as the most important environmental
factors affecting the distribution of the zooplankton.
This indicates that the summertime temperature-salinity
combinations observed in the discharge canal and thermal
plume areas represent sub-optimal witnesses for all
components of the zooplankton community.
The results of these analyses validate earlier
laboratory and field studies. This validation is necessary,
especially for laboratory studies, to assure that the
conclusions drawn by the earlier studies are supported by
data collected in a field survey.
The analysis procedures followed provided an orderly
process for the successful investigation of a set of
"noisy," ecological field survey data. The approach
employed in this study can serve as a model for analysis
procedures in similar studies.
The importance of the zooplankton community to the ma-
rine, and particularly to the estuarine, environments has
long been recognized. In the estuarine regions, the zoo-
plankton serve as the primary food sources for many larger
consumers. As a result of their short generation times and
high metabolic rates, they are important in energy and nu-
trient cycling within the estuarine environment (Gunter,
1967; de Sylva, 1973; Williams et al., 1968).
The fact that many larger, often commercially impor-
tant, marine species spend an appreciable portion of their
immature life as temporary members of the planktonic commu-
nity adds to the importance of the zooplankton. Coastal
regions and estuarine environments often function as
spawning and nursery grounds for non-resident species, many
of which are commercially important (Gunter, 1967). Thus,
the estuarine areas of the world are well noted for their
contributions to the productivity and stability of marine
ecosystems. The productivity of these regions is important
to man not only as a food source, but also for recreational
functions (Reeve, 1973).
In addition, the estuaries and other coastal regions
are important to man as a sink for his waste products
(Cronin, 1967; Biglove and LaFleur, 1967; DeFalco, 1967).
Thermal waste is one of these products that has recently be-
come of interest to researchers (Odum and Kroodsma, 1976).
The consideration of the impact of thermal wastes on the en-
vironment is of increased importance and concern in a tropi-
cal or sub-tropical environment, such as is found in Flori-
da. Although estuarine organisms have typically evolved be-
havioral and physiological mechanisms that enable them to a-
dapt to large fluctuations in temperature and salinity
(Hutchinson, 1976; Kinne, 1964; Kinne, 1967), man-induced
perturbations to the magnitude and timing of these changes
may not be tolerated by the estuarine species.
With the recognition of the importance of the zooplank-
ton community to the estuarine environment and the potential
impact of thermal pollution in a sub-tropical environment,
this study was designed to investigate variation in the na-
tural and thermally stressed environments of a sub-tropical
The Crystal River estuary located in the vicinity of a
Florida Power Corporation steam generating plant provided
the ideal study site. This area represented the typical
shallow water, middle salinity estuary found in the sub-
tropical Gulf coast region of Florida.
The study consisted of two general components for data
collection: a diurnal series and a biweekly series. Each
component was designed to investigate a different portion of
the community variation problem.
The diurnal series collected data on the short-
term diel variation of the zooplankton community at differ-
ent localities and seasons. These data were used to analyze
the effect of environmental parameter variation on the
community. The parameters considered fell into five general
areas: localized micro-environmental conditions, spatial
positioning within the estuary (and to some extent within
the water column), thermal addition, temporal and tidal
The biweekly series collected data on the long-term, or
macro, variations experienced by the zooplankton community.
Environmental conditions were generally integrated over
time, thermal addition operating characteristics, seasonal
factors and spatial factors were examined for their effect
on the zooplankton community.
MATERIALS AND METHODS
The plant site is in Citrus County, Florida, approxi-
mately 12 km north of the town of Crystal River (fig. 1).
For the duration of this study the power plant consisted of
two oil fired steam generating units with a combined capaci-
ty of 897 megawatts (Mw). The cooling system was a "once-
through" type with a flow of 2410 m3/min. The maximum tem-
perature rise from intake to discharge was designed to be
6.1 C. Unit 1 began operation in July 1966; Unit 2 began
operation in November 1969. Unit 3, under construction dur-
ing the study period, began operation in July 1977. Unit 3
requires an additional 2580 m3/min. of cooling water flow.
The combined system has a net temperature rise of approx-
imately 8.1 C. The power plant is on the landward edge
of a tidal salt marsh. This marsh, dominated by Juncus
roemarianus, and others similar to it cover large areas
along the Gulf coast of Florida, often extending one or two
kilometers landward. Narrow bands of Spartina occur along
the marsh Gulf interface.
The coastal shelf adjacent to the plant site is part of
the drowned limestone karst topography characteristic of
this portion of west central Florida. It has a shallow
sloping bottom (45 km to the 9 m contour) and extends far
into the Gulf of Mexico (230 km to the 100 m contour) (Jones
et al., 1973). This area is part of what Tanner(1960) and
Walton(1973) have classified as the low wave-energy section
of the Florida Gulf Coast.
The immediate coastal area contiguous to the plant site
is comprised of a series of shallow basins separated by oys-
ter reefs. These oyster reefs have developed parallel to
the coastline and extend seaward 3 to 4 km. Among these
reefs are the estuarine bays characteristic of this area.
These bays recieve freshwater input from terrestrial run-off
and from two rivers. The salinities range from 17 to 30 ppt
and the normal water temperatures range from 14 to 30 C.
(McNulty et al., 1972; Maturo et al., 1974). Two major
community types occur within these estuarine bays.
On the landward edge of the reef systems are the inner
bays, characterized by shallow flats which are often exposed
at low tides or by strong winds. The average depth of these
bays is approximately one to two m. The average tidal
range in this region is approximately one m. This system
tends to be benthic dominated with dense summer growths of
sea grasses (Halodule, Thalassia, etc.).
Seaward of the shallow inner bays, the depth increases
to an average of two to three meters. These outer bays are
characterized by deeper flats where species of shade adapted
red and green algae (Gracilaria, Spyridia, and Caulerpa)
dominate the benthic flora (Van Tine, 1974; and McKellar,
1976). In these deeper bays the phytoplankton gain an im-
portant role in the daily production of organic matter.
Patches of attached Sargassum are also common. Bay ecosys-
tems similar to these at Crystal River have been described
from the Cedar Key region (Reid, 1954).
The two major freshwater sources to the area are the
Crystal River 4.8 km to the south, with an average flow of
1500 m3/minute, and the Withlacoochee-Cross Florida Barge
Canal complex, with an average flow of 2150 m3/minute. The
average combined flow of these freshwater sources is about
1.5 times the circulating water flow of the power plant.
The Withlacoochee River flows into the Gulf of Mexico end of
the unfinished Cross-Florida Barge Canal. Spoil islands
from the Barge Canal continue approximately 6.5 km into the
Gulf. These as well as the cooling water canal dikes for
the power plant, influence the local hydrographic circu-
lation patterns described below.
Pritchard (1967) defined an estuary as .a semi-en-
closed coastal body of water which has a free connection
with the open sea and within which sea water is measurably
diluted with fresh water derived from land drainage.
He further identified four classes of estuaries:
(1) drowned river valleys, (2) fjord-type estuaries, (3)
bar-built estuaries, and (4) estuaries produced by tectonic
processes. The Crystal River area fits the definition of a
bar-built estuary (Pritchard, 1967) as modified by Caspers
(1967). Caspers points out that many definitions of estuar-
ies are remarkably similar to and often include lagoons.
Estuaries are distinguished from lagoons by examining the
stability of their salinity. When the inflow of fresh water
into a separated basin develops a stable body of brackish
water, which is relatively uniform throughout the area, this
is termed a lagoon. However, if the mixing of fresh and
marine waters is not stable, but shows periodic changes, the
basin may be considered an estuary. This latter definition
applies to the oyster reef bay system at Crystal River.
Thus, although the region does not possess the physical
features of the classical estuary, as demonstrated by the
Chesapeake Bay estuary, it does exhibit the salinity gradi-
ents and instabilities observed in classical estuaries. In
addition, the oyster bar estuary system found at Crystal
River is characteristic of near shore systems all along the
Gulf coast (Odum et al., 1974).
The intake and discharge canals have been cut through
the salt marsh and coastal bay systems west of the plant
site, displacing approximately 1.1 km2 of marsh and 1.9km2
of bay (fig. 1). The intake channel extends from the plant
approximately 12.5 km into the Gulf of Mexico. It is
laterally confined with double bulkheading for the initial 5
km. The mean depth of the intake canal is approximately
6.5m, which is in contrast to the 1 to 3 m depth of the
adjacent bays. The width at mean low water (MLW) ranges
from 90 to 110 m. The intake canal was designed to
accommodate the movement of fuel barges to and from the
plant. The intake current velocity averages 9 cm/sec
(Carder, et al., 1974).
The discharge canal is significantly shorter than the
intake canal. The double bulkhead portion of the discharge
canal is approximately 2 km in length. The entire length of
the canal is 3.8 km. The discharge canal was designed with
a smaller cross-sectional area (4.5 m deep, 60 m wide)
in order to maintain a higher velocity (20 cm/sec.) and in-
sure adequate lateral flow entrainment and mixing upon dis-
charge to the shallower bay receiving waters (Carder et al.,
Detailed descriptions of the hydrography have been pre-
sented in technical reports to an environmental impact study
group (Florida Power Corporation, 1974). The methods em-
ployed during the hydrographic studies included: salinity-
temperature-time series, dye and drogue movement time
series, and circulation modelling and simulation. The
pertinent features of the reports will be discussed here
(Carder et al., 1974; Klausewics, 1974; Kemp, 1977).
From a physiographic and circulation point of view the
Crystal River estuary may be divided into three parallel
areas that run north-south (see fig. 2):
All water within the area bounded on the west by the
last Gulfward string of oyster bars and on the east
by the shore.
All water within the area bounded on the east by the
first area and on the west by an imaginary line
running north-south through the end of the intake
canal spoil banks.
All water west of the second area.
These general areas exist in bands along the Gulf coast
wherever the bar built estuary system is found. Area 1
corresponds to the area considered to be the estuary. Area
2 represents a mixing zone between the estuarine bay region
and the open Gulf water. In area 2 there are some oyster
bars and other physiographic relief features that present a
restriction to complete mixing with either the open Gulf or
the bay waters. In area 3 we have the open Gulf waters.
These three areas are further subdivided by the follow-
ing man-made or natural features found in the area:
The Withlacoochee River-Florida Barge Canal Complex,
as discussed above, provides several important
features. First by virtue of its fresh water input,
it is an important sustaining force of the estuarine
environment. Secondly, the spoil islands located off
shore from the mouth of the Withlacoochee River serve
to provide a northern boundary to the study area.
These spoil islands, like the discharge and intake
canal dikes, form an effective barrier to long shore
currents. These barriers tend to drive the long
shore currents out offshore into the Gulf of Mexico.
This has the effect of limiting the source of Gulf
water to that brought in by the tidal cycle, thus,
in this region of low wave energy beaches, increasing
the residency time of the water mass.
The cooling water circulation system of intake and
discharge canals serves to draw primarily from region
3. Thus providing a source for introducing Gulf
origin water into the estuarine system. However,
because of the barrier affect of the intake dike,
this mixing only has an effect on the discharge side
of the system.
The Crystal River provides another freshwater input
to the estuary. This input enters the system from
the south. The southern, or intake, portion of the
study area is open to influence from long shore
Thus, physical features also play an important role in the
makeup of the environment to which the zooplankton community
Timed collection of tidal data, including current speed
and direction, salinity and temperature data, over three
seasons and all tidal regimes were reported by Carder
(1973,1974) to the Crystal River Environmental Impact Study
Group. Those data are summarized below.
On the average the source waters come primarily from
area 3 (46.9%), and about equally from areas two (26.9%)
and 1 (26.3%). Thus the majority of the water entrained by
the intake canal is of Gulf origin. This water is then
heated by traveling through the plant's condensers and re-
turned to the environment via the discharge canal. An exam-
ination of what occurs on each of the tidal cycles gives a
more detailed picture of the dynamics of the interaction of
the tidal cycle and the pumping action of the power plant.
During ebb tide, a profile of the canal shows surface
waters flowing Gulfward, while the deeper water, of Gulf
origin, is flowing toward the plant. The current set up by
the circulating pumps is strong enough to counter balance
the tidal effect along most of the canal. This establishes
a point of no return. Brackish, warm water enters the sur-
face water of the canal on the latter stages of the ebb
tide, which has delivered near-shore water far enough west
to be available for entrainment.
During flood tide, the brackish water that has moved
in front of the canal mouth is pushed shoreward and replaced
by cooler, more saline water of Gulf origin. Some of the
brackish water is entrained by the power plant. The entire
cycle of entering the intake canal and being entrained by
the power plant takes about twelve hours.
Cross sectional analysis of the intake canal waters
over a full regime of tidal cycles shows an interesting
stratification of the water layers. The water column in the
intake canal is divided into a deeper, cooler, more saline
layer of Gulf origin and an upper layer of estuarine or mix-
ing zone origin. This is significant as most of the power
plant's draw is from the deeper portion of the intake canal.
Based on the physical features described above, the
study area was divided into six regions (see fig. 1). The
selection of the boundaries for these regions was a result
of considering the topography of the estuary floor, studies
performed on the physical oceanography, water mass resident
times, benthic organism distribution, sediment type distri-
bution patterns, and a degree of arbitrary boundary assign-
ment. The area, average depth, and volume varies greatly
between the six regions depending, to a large extent on the
physiography of the bottom.
Region 1 represents the shallow, limited tidal exchange
estuarine region on the intake side of the power plant. Re-
gion 2, beyond the major oyster bar barrier on the intake
side of the power plant, represents the mixing zone of estu-
arine waters and the more saline Gulf of Mexico waters.
Region 3, which is the largest and deepest region, covers
the portion of the study region that is beyond the dikes of
the intake/discharge canals and represents the open waters
of the Gulf of Mexico. Region 4 represents the special
environment of the intake canal. Region 5 is the area
containing the plume of the thermal discharge. Region 5
bounded on the Gulfward side by oysterbars, as is Region 1,
represents an estuarine region. Region 6, is basically
comparable to Region 2, except it is subjected to the
thermal discharge that comes from Region 5 across the oyster
bar barrier. Region 6 represents a mixing zone of the Gulf
waters and the thermally impacted estuarine waters.
The stations to be sampled were arranged within the re-
gions as follows (fig. 2):
Region 1 contains two stations. Station 1 is located
adjacent to Negro Island, in shallow water (usually less
than one meter). This station could be reached only during
high tide periods.Station 2 is situated between the inner
and outer parallel lines of oyster bars.
Region 2 has one station, station 3, located centrally
to the its boundaries.
Region 3 contains two stations. Station 4 is placed
south of the end of the solid intake dike at the beginning
of the intake canal spoil islands. Station 5 was located
slightly Gulfward to the end of the intake canal spoil is-
Region 4, the intake canal, also has two stations.
Station 6 was located at the mouth of the enclosed intake
canal, and station 7 was situated at the intake screens.
Region 5 had three stations. Station 8 was located at
the point of thermal discharge from the power plant into the
discharge canal. Station 9 was placed at the point of dis-
charge from the enclosed portion of the discharge canal into
the receiving estuary. Station 10 was situated at the cen-
ter of the thermal plume area.
Region 6 had one sampling point, station 11, which was
located centrally to its boundaries.
Samples were collected at each of the eleven stations
biweekly from November 1973 until September 1974. All samp-
les up to and including July 1974 were processed. Those
collected after July 1974 were not processed but were ar-
Sample collection technique
The samples consisted of a complement of net tows.
This complement of subsamples was designed to sample as
thoroughly and as completely as possible the water column
environment found at the station.
Minimally, the sample consisted of a surface tow of a
half-meter 202 micron standard plankton net equipped with
a General Oceanics flow meter. Where tidal and physiograph-
ic features made it possible, a simultaneous depth 202 mi-
cron sample was also collected. Depth samples were taken by
employing a subsurface depressor, which was designed to
create a drag on the towing line.
Each sample consisted of two separate tows, which were
combined and fixed in the field in a jar of buffered 10%
formalin. These sample jars were then transported to the
laboratory for detailed examination. The length and speed
of towing the sample collection nets, one minute at a speed
of two knots, was designed to provide a 202 micron sample
which filtered an average volume of fifteen cubic meters.
For each set of samples obtained a complete complement
of environmental parameters was measured at each station.
The parameters measured included: salinity, water tempera-
ture, air temperature, wind speed, wind direction, surface
current speed, current direction, dissolved oxygen, secchi
disk, tidal condition, precipitation, and sea conditions.
In addition a concurrent study, examining the phyto-
plankton at the same sampling stations, collected a comple-
ment of water chemistry parameters. The parameters measured
levels of dissolved: organic carbon, phosphate, nitrate,
nitrite, ammonia and silicate (Hopkins and Gibson, 1974).
Sample size determination
Prior to the beginning of this project, it was necessary
to determine the desired sample size for the field collec-
tion effort. As a result of the industry funding for this
project, the upper bound for the collection effort was not
necessarily limited to that which a single investigator
could perform. Rather the requirements of the study for
accuracy and precision and the variablity of the data could
shape the sampling effort. Thus an analysis of previously
collected preliminary data was made to determine the sample
To do this an estimate of the variance of the zooplank-
ton community population vector is required. This problem
is somewhat complex, since the ideal situation would be to
predict the sample size required at all stations at all sea-
sons for all categories of the zooplankton community. As
each category would have different estimates of required
sample size, the variances associated with the observed
abundances of calanoid copepods was selected. This category
was selected because of its overwhelming importance to the
zooplankton community, both in numbers and biomass.
An iterative formula was used to determine the sample
size (Sokal and Rohlf, 1969). The formula is:
n hl r) 6% + Z(,-ptjI
Where n= number of replications,
0' = the true standard deviation (estimated by
( = the smallest true difference that it is
desired to detect,
~ = degrees of freedom of the sample standard
0( = significance level for the type I error,
p = significance level for the type II error,
and -t = values from a two tailed t-table with
degrees of freedom.
An alpha=.01 and an alpha=.05 with a p=.50 were
selected for investigation. The error degrees of freedom
were 420, thus Z-tables were used in place of t-statistics.
The coefficient of variation for the Calanoid Copepod data
was 28%. The acceptable level of difference for detection
was set at 25%. This resulted in a required n of 9 or 17
samples for testing at the .05 or .01 level, respectively.
Therefore, the sampling program should gather at least 9 or
17 samples for each station/temporal period. The time
period selected for examination was the season.
An additional constraint placed upon the sampling
design was the requirement for sampling at regular inter-
vals. Since it was impossible to predict the timing of the
seasons, sampling across the transect of time was to occur
at constant intervals.
The solution to this problem was to sample biweekly,
collecting two samples at each station. This would provide
sampling on regular intervals. If the seasons averaged
three months in length, the temporal period would have
twelve samples, or if the seasons were four months long the
temporal block would have sixteen samples. These data were
utilized in the final design of the sampling protocol.
Biweekly Sampling Program
For logistical reasons, the stations sampled biweekly
were divided into three groups and assigned to sampling
teams: the discharge area (stations 8,9,10, and 11), the
inner bay of the intake area (stations 1,2,6, and 7), and
the outer bay of the intake area (stations 3,4, and 5).
All stations within a sampling area were sampled twice
within a tidal cycle, in a randomized order. Thus producing
at an ideal station two surface 202 micron samples
comprised of four separate tows) and two depth 202 micron
samples. Depth samples were not collected at stations 1,
2 and 9 because the water was too shallow, nor at station 8
since local mixing of the water column was assured.
Diurnal Sampling Program
During the period of time that biweekly samples were
being collected on a regular basis, a quarterly diurnal sam-
pling schedule was also being conducted. The purpose of
this sampling protocol was to collect sufficient samples
from the intake and discharge sides of the estuary and exam-
ine the samples for short term variations.
The method of collecting the samples was basically the
same as for the biweekly sampling, except that no tows were
pooled during the diurnal collections. The same environmen-
tal parameters as were collected during the biweekly series
were collected during the diurnal series. Each quarter the
detailed protocol for accomplishing the diurnal sampling was
slightly different. Therefore, each quarterly diurnal samp-
ling period will be described separately.
The fall quarterly diurnal sampling program was per-
formed during November 12-15, 1973. Stations 7,3,11, and 9
were sampled on successive days. Each station was sampled
over a 24 hour period, beginning at 0700. Two surface tows
with a 202 micron net were taken each hour, usually at the
first and last quarters.
The winter quarter diurnal series was collected during
January 21-25, 1974. The winter quarterly diurnal followed
the same sampling regime as outlined for the fall quarterly
diurnal sampling program.
The spring quarterly sampling program was performed on
April 29, 1974. The sampling schema for the spring diurnal
changed somewhat from that followed in the fall and winter.
Stations 3 and 11 were dropped and simultaneous surface and
depth samples were added. Rather than sampling each station
on successive days as was done in the fall and winter, sta-
tions 7 and 9 were sampled concurrently. The length of time
to complete the diurnal was shortened from four days to one
day. The total number of samples collected remained the
The summer diurnal sampling program was performed on
July 25, 1974. The sampling procedures were identical to
the spring diurnal sampling program.
Samples were returned to the laboratory at the Univer-
sity of Florida Marine Laboratory (UFML), Gainesville,
Florida. Each sample was subjected to the following
(1) The samples were split using a Folsom plankton
splitter. The first split resulted in a sample re-
serve which was archived at the UFML, and the sub-
sample which was to be counted.
(2) The subsample which was used for counting was flush-
ed with water through a series of graduated sieves
to separate the organisms into different size class-
es. The sieve sizes used were: 2000, 850, 600, 300,
and 202 microns. The greater than 4000 micron and
less than 202 micron portions were discarded. This
would reduce some of the bias introduced through
gear selectivity that might result from net clog-
(3) Each size class was split using the folsom plankton
splitter until a countable number was reached. In
general, the plankton counter split the sample until
a portion containing 250 to 2000 organisms/size
class was obtained.
(4) The samples were counted to determine the standing
crop values for the zooplankton categories.
(5) The biomass determination was made.
(6) Upon the completion of the sample, the data were
keypunched, verified and added to the data base.
All standing crops were expressed in numbers of or-
The following zooplankton categories were identified in
the processing of the samples:
Penaeid Shrimp Larvae
Other Shrimp Larvae
Initially biomass was to be determined by drying the
entire size class subsample and weighing it. However,the
amount of non zooplankton debris in the samples precluded
this approach. Also, the volume of samples processed made
this approach unfeasible. Therefore, an alternative method
of biomass determination was selected. The approach chosen
was motivated by the assumption that the within sizeclass
variation in organism biomass was not detectable given the
precision and accuracy of the balances being used. This
assumption was tested and supported by an analysis of
variance. For each organism category-season-sizeclass
combination, 5,000 to 10,000 organisms were selected from a
random assortment of samples. These were dried in tared
weighing containers and then their biomass was determined.
These data were analyzed for significant effects due to
season, size class and category. The results of this
analysis were used to calculate the biomass values to be
added to the data base.
As we are made aware of the need for effective resource
management, the necessity to develop the tools to under-
stand the patterns of natural fluctuations in communities,
and to examine the effect of man's activities on these
patterns comes into focus (Cronin, 1967). However, in spite
of its importance, there have been few studies of the
zooplankton community dynamics that have applied modern
statistical methods to the analysis of the problem (Holt,
Until recently, the investigation of community spatial
and temporal dynamics was restricted to the examination of a
few variables in a one-at-a-time mode, i.e. a univariate ap-
proach (Crovello, 1970; Gould and Johnston, 1972). In order
to draw conclusions about the community as a whole, the in-
vestigator had to combine the results of many single analy-
ses into a comprehensible and consistent picture.
Several major problems are encountered in this approach
to the analysis of data collected during field ecology stud-
ies. Except in the most carefully controlled experiments,
the number of variables that may potentially enter into con-
sideration often reaches an unmanageable number. The inde-
pendent variables are often interrelated, complicating the
structure of the analysis model, and precluding the use of
conventional univariate statistical analysis methods.
Employing the univariate approach in a multivariate
situation may lead the researcher to ignore important rela-
tionships that exist between the variables. The interac-
tion, or synergistic effect, is the type of relationship
most likely to be ignored (Harris, 1975; Cooley and Lohnes,
1971). It is this interaction effect that is often of most
interest to the biologist (Alden et al., 1976).
These problems in the analysis and interpretation of
the data often cause confusing and contradictory results
which will lead to the conclusion that the data collected by
the field study are too variable for meaningful interpreta-
tion. This belief often frustrates the investigator
and results in the selection of a simplistic or cursory
analysis as the only approach for dealing with these "messy"
However, the results of Cassie's investigations suggest
that zooplankters are not randomly distributed throughout
the water column (Cassie, 1959, 1960, 1961, 1962, 1963a,
1963b, 1967a, 1967b, 1969a, 1969b). Rather, they seem to
show distributions that are understandable and highly cor-
related with the current conditions and historical aspects
of their immediate environment.
Cassie demonstrated that an estuary is characterized by
a unique profile of physical and chemical parameters that
affect reproduction, survival, and the movement of estuarine
organisms. These physical factors interact to define physi-
ologically and ecologically distinct environments within the
estuary which will shape the structure of its biological
community. There are several approaches to the analysis of
community variation data to discover the underlying struc-
ture of the biological community.
Approaches to community structure analysis
One approach is to remove the survey atmosphere from
the study and perform a tightly controlled experiment. The
salinity-temperature studies reviewed by Alden et al. (1976)
present good examples of this approach. Alden (1979) has
improved on this method by bringing the laboratory into the
field. The reduction in the "noise" in the data is accom-
plished by reducing the number of environmental effects
studied and usually limiting the range of these effects to a
small number of discrete levels. This is usually achieved
by designing a laboratory experiment to test the hypothesis
and then extrapolating to the field environment.
The selection of this approach insures that the data
will be well structured and can be analyzed with classic
univariate statistical procedures. The major disadvantage
is that the number of environmental effects that can be con-
trolled in an experiment is usually three or less. In
addition, the assumption that the performance of an organism
in a laboratory environment is comparable to its action in
the field is not always well founded. This is not to say
that laboratory studies are not well advised for ecological
research. However, there are definite limitations to their
A second approach involves retaining the field survey
method of obtaining the data and employing statistical anal-
ysis techniques for explaining the noise in the data.
The most commonly employed procedure for analyzing
field survey data is the "indicator" variable approach.
Pielou (1969) provides an excellent review of these methods.
The indicator variable approach may take two directions.
The first is to select a subset of the potential dependent
variables and only consider them in the analysis. For
example, many zooplankton field studies have dealt almost
exclusively with the most abundant copepods. The advantage
of this method is that it reduces the number of dependent
variables that must be considered to a manageable number.
Another analysis technique employs derived variables that
have the property of integrating into a single variable the
information about the biological community that is stored in
each of the separate original variables. A typical example
is to employ biomass or species diversity as the variable
for analysis. When such summary statistics are employed,
valuable information concerning the patterns of interaction
between the variables may be lost, although simplicity is
gained through a reduction in the analytic dimensionality
(Bary, 1964; Pielou, 1966; Holt, 1976).
The key to the analysis of ecological data generated
from field survey programs is in reducing the dimensionality
of the problem to one that is manageable by a human. The
dimensionality of a problem is defined by the number of sep-
arate variables that must be in active consideration simul-
taneously in order to accomplish the solution and interpre-
station. Psychological studies in computer program design
have shown that man is capable of keeping track of a maximum
of approximately seven separate factors simultaneously
(Martin, 1973). This concept applies to the analysis of
data as well. If the dimensionality of the problem exceeds
seven variables, the ability of the ecological researcher to
deal with the problem is extremely limited.
Another problem in analyzing data serves to limit the
acceptable dimensionality even further. The ability to for-
mulate meaningful interpretations of the biological or eco-
logical implications of the data analysis results is greatly
enhanced by the researcher's ability to visualize the re-
sults. The connection between numbers and meaning is often
achieved by a graphical presentation of the results.
(Lindgren, 1968) Thus, the practical dimensionality of the
analysis is in the range of three to five.
The accomplishing of the reduction in dimensionality
has been the goal, although often an unwritten or a subcon-
cious one, of the myriad of data analysis methodologies that
have been proposed for analyzing ecological field data
(Crovello, 1970). The approaches listed above all have the
structure and dimensionality of the solution predetermined
by the analysis technique. For example, many studies employ
the Shannon-Weiner species diversity index, or some compar-
able derivative, to examine community variation. The dimen-
sionality of the solution is one, at least from the depen-
dent variable point of view. Regardless of the magnitude or
pattern of the observed variation, all information about
that variation is constrained to be mapped into the single
variable: species diversity.
The approach selected in this study was quite differ-
ent. The philosophy was to select data analysis methods
that were responsive to the underlying structure of the sys-
tem under investigation. In other words, these analysis
procedures allowed the data to dictate the dimensionality of
the problem, rather than to enforce one that is
predetermined. Thus, these methods would investigate and
describe the patterns of variation found in the data.
This approach to data analysis is closely aligned to
Tatsouka and Tiedeman's description of the structure of
science (1954) as a hypothetico-deductive-observational pro-
cedure. They recognize within the scientific method a step
that leads from the initial observational data to the set of
theoretical constructs, or hypotheses. This step is termed
"creative invention" and includes assistance from statisti-
cal methods and analogies from principles in related scien-
In the creative invention stage at which theory is in-
itiated, primarily by the use of analogies, statistics has a
heuristic role in the discovery and refinement of cons-
tructs. This is a different role than is usually assigned
to statistics, which comes toward the end of the scientific
process in the testing of hypotheses deduced from theory
against empirical results.
Statistical procedures that fulfill this heuristic-
construct seeking role are termed statistical pattern
recognition techniques (Lachenbruch, 1975; Cooley and
Lohnes, 1971). They include discriminant analysis, factor
analysis, cluster analysis, etc. and fall within the mul-
tivariate general linear hypothesis (Finn, 1974). Through
pattern recognition techniques, one can reduce the dimen-
sionality of the data by uncovering the underlying patterns
of variation, while simultaneously accounting for the
maximum amount of the variation observed in the data.
These methods depend heavily on linear functions fitted
to complexes of variables by multivariate procedures as the
method for specifying the details of the constructs and the
relationship between constructs. In the beginning stages of
a scientific endeavor, which is the position enjoyed by all
survey studies, the heuristic uses of these procedures are
far more important than the hypothesis testing. As Cattell
(1966) argued in his support of the observational aspects of
the survey approach over the manipulative experimental ap-
proach, the potency of this approach is high because
it takes life's own manipulations and by statistical finesse
teases out the causal connections among data that could not
These heuristic procedures do not allow for testing of
inferential hypotheses in the traditional statistical sense.
Instead, the statistical methods are viewed as quantitative
tools for exploring patterns to be revealed in data. Thus,
the "test statistics" produced by the statistical methods,
and their concomitant significance levels can not be inter-
preted as tests of significance. However, the test sta-
tistics may be used as a quantitative measure of the
strength with which the observed data support the proposed
pattern. In order to avoid lengthy, repetitive explanations
at each point in the analysis, the terms significant and
nonsignificant will often be employed when referring to the
use of these quantitative measures as aids in decision
The question besetting the ecological researcher in
deciding to employ multivariate analysis procedures is which
from among all the many possible methods should be used?
The answer to this question lies in how many sets of vari-
ables and how many populations are included in the design of
Those heuristic studies of the data which involve a
single population and a single set of variables are classed
as interdependence models by Kendall (1957). The methods
employed to analyze those data are principal components,
factor and cluster analysis (Morrison, 1967; Sneath and
Sokal, 1974). Principal components and factor analysis
methods are also employed to reduce a complex, large set of
dependent variables to a smaller set of independent factors
or components prior to further analyses (Atchley, 1974).
When one is dealing with more than one set of vari-
ables, or more than one population, Kendall (1957) terms
these interdependence models.
In the case of more than one set of variables, the pro-
cedure usually described is canonical correlation. However,
based on the experience obtained in this study, the results
obtained by combining factor analysis, stepwise multiple re-
gression, and response surface fitting, as described below,
are more satisfactory.
If the data call for the analysis of inter-populational
variation, one of two analysis approaches is recommended.
If the intent of the analysis is to discover the differences
between known groups, then the techniques of discriminant
analysis are required. On the other hand, if the goal is to
discover what, if any groups there may be, the methods of
principal components and factor analysis are suggested.
In these methods several underlying assumptions are in-
herent: first, the data are assumed, in some manner, to have
the properties of the multivariate normal distribution.
This may not always be the situation. In some instances a
transformation of the data is the solution. In other cases,
the admonition of Tukey (1962, 1969) may be the only answer.
Tukey suggested that data analysts must be "willing to seek
for scope and usefulness rather than security", and to be
"willing to err modestly often in order that inadequate evi-
dence shall more often suggest the right answer". Further,
one must use scientific judgement more than mathematical,
but not one to the exclusion of the other.
Second, the structures uncovered are linear in nature.
This might raise criticisms that structures in nature are
seldom linear. This is answered in two ways. If an under-
lying structure is hypothesized to have a particular form,
as was the case in this study, then appropriate transforma-
tion may be employed (Cassie, 1960, 1962, 1963a). Also, the
purpose of these methods is exploratory analysis; thus
searching for linear functions is probably appropriate, as a
Third, the multivariate techniques that are subsumed by
the general linear hypothesis attack the problem of reducing
the original variable space to the minimum number of dimen-
sions needed to describe as much of the relevant information
contained in the original observations as is possible, i.e.,
the dimensionality problem. Different multivariate tech-
niques, i.e., different models within the general linear
model, differ in the types of information that they
Selecting the analysis model
Factor analysis and principal components analysis "dis-
cover" the underlying pattern of variability such that the
number of variables may be reduced to a smaller set of fac-
tors or components which account for the observed variabil-
ity in the data.
Canonical correlation is a procedure for factoring two
variable sets simultaneously, with the goal of extracting
factors which are uncorrelated within their respective vari-
able sets but provide maximum correlation of factor pairs
across variable sets. Canonical correlation discovers the
factors which express the maximum redundancy, or overlap,
between the two set of variables.
The approach to the variable set interdependency prob-
lem which was used in this study has a different orientation
than canonical correlation. It finds the set of factors
that best describe the observed variation in one set of var-
iables, the dependent variables. It then finds the set of
variables, from amongst the predictor variables, that have
the greatest R-square with the dependent variable factors.
This approach finds the important patterns of variation of
the dependent variables and then finds the patterns of vari-
ation in the predictor variables that best explain the vari-
ation patterns of the dependent variables.
Canonical discriminant analysis addresses the problem
of discriminating between known groups. It is a procedure
for forming a set of variables into a group of factors, the
canonical discriminant functions, that are viewed as axes
spanning the variable space. These axes are constructed in
such a manner that they maximally separate the groups
(Lachenbruch, 1975). The factors are formed so that they
describe the observed variation between groups.
The multivariate analysis of variance, MANOVA, is a
generalization of the univariate ANOVA. The MANOVA extends
the researcher's ability to investigate the differences be-
tween groups. Just as the t-test and linear contrasts are
employed in the univariate ANOVA to examine intergroup dif-
ferences, canonical discriminant analysis can be used in
conjunction with the MANOVA to partiton a specific component
of the variation for discrimination analyses.
Each of these methods is best used for a particular
type of problem. The first step in selecting the appropri-
ate multivariate pattern recognition procedure is to identi-
fy what pattern of variability needs to be explored.
As was stated above, probably the single most important
feature of this analysis approach is that the data are used
to define the dimensionality of the analytic procedure.
Each of these procedures provide quantitative, objective mea
sures for selecting the cutoff point in the dimension reduc-
Displaying the results
The final aspect of the heuristic analysis process is a
consideration of the data presentation. The best technique
for presenting and assimilating the results of pattern rec-
ognition analysis is visual. The use of graphics is essen-
tial to the analysis process. Of course, the use of graph-
ics places some limitation on the dimensionality that is ar-
bitrary. It is not yet possible to represent easily more
than three dimensions. However, that can be circumvented by
plotting three dimensions and controlling the values of the
higher dimensions to a low, intermediate, and high value.
This produces a series of "frames" that can be viewed side-
by-side, or in sequence.
The detailed techniques of the multivariate analyses
will not be described here. The interested reader is
referred to the texts referenced in this work.
The analysis of the biweekly data had two main objec-
tives: first, the discovery of the structure of the major
zooplankton communities and the patterns of spatial and
temporal variation they exhibit; second, what components of
the zooplankton communities could be used to discriminate
between the areas defined for the estuary. This second ob-
jective may be likened to a search for indicator organisms.
The area of particular interest is the thermally affected
regions of the discharge canal and thermal plume. The ques-
tion is: can one distinguish the thermally affected regions
from the other regions within the study area?
Both of these objectives call for multivariate analysis
techniques. However, as each asks a slightly different
question of the data, two different sets of analyses are
Determination of seasonality
As discussed above, the factors of primary interest
were the seasonal and spatial patterns of variation. The
regions were determined, as described above, by examinations
of the hydrographic features of the estuary. The
determination of the season boundaries was not as easy. The
seasons along the west coast of central Florida may be di-
vided into wet or dry and cool or hot. The influences of
offshore Gulf of Mexico water ameliorate the changes obser-
ved in the atmospheric conditions. The interaction of these
factors indicate that the traditional number and kind of
seasons and their boundary points may have little meaning
within the subtropical estuary found at Crystal River.
Thus, the data collected on environmental parameters were
used to define the number, kind and boundary points for the
seasons during the study period.
The determination of the seasonality for the study
period involved two main steps. First, perform a cluster
analysis to use the observed data to define the seasonal
breaks and then employ MANOVA and CDFA as tools to investi-
gate the strength of the separation of the cluster analysis-
defined seasonal groups and to identify the environmental
variables responsible for the group separations.
The following constraints were imposed on the analysis
of the weather data:
1) The variables were selected to characterize the en-
vironment of the zooplankton community. They can be divided
into three groups:
Those variables that directly measured a facet
of the estuarine environment to which the
zooplankton community was exposed (water
temperature, salinity, water chemistry values,
water clarity measures, etc).
Those variables that directly measured an aspect
of the water transport mechanisms which affect the
zooplankton community ( tidal variables and wind
Those variables that measured an external effect
which was thought to exert or respond to a force
shaping the environment and its seasonality
(rainfall, daily sunshine, and organic carbon).
2) Biweekly averages of environmental conditions for
the intake and open Gulf areas would be used as data points.
This would attempt to remove from the seasonal factor any
effect of the power plant's thermal addition (see Appendix
3) The seasons would have to be contiguous temporally.
Thus, the seasonal analysis would be employed to indicate
the general pattern of the seasons, but not to dictate the
The cluster analysis method employed was based on the
techniques discussed by Johnson (1967). It is an algorithm
for partitioning objects into optimally homogeneous groups
using a single linkage joining technique. It forms
nonoverlapping hierarchical clusters using the euclidean
distance as the metric. The alogorithm was structured such
that it would have the following features:
1. The input should consist of the n(n-l)/2
similiarity measures among the n objects.
2. There should be a clear, explicit, and intuitive
description of the clustering; i.e., the clusters should
3. The clustering procedure should be essentially in-
variant under monotone transformations of the similarity
The next step was to examine the data in order to as-
certain the reasons for the clusters forming as they did.
If the distribution of environmental data that caused the
observed clustering pattern could be used to form a reason-
able explanation of the seasonality, then the cluster defin-
ed seasons would be used for the remainder of the biweekly
Two techniques were used to examine the data for sea-
1. A MANOVA was performed on the vector of environ-
2. The MANOVA was supported by a graphic display which
plotted each of the variables against time (see appendix 1).
The graphic presentation was used primarily to support
and aid in the interpretation of the MANOVA results.
The model used for the MANOVA was a oneway analysis
with season as the main effect. The seasons used were the
ones determined by reconciling the clusters formed during
the cluster analysis and the need for temporally contiguous
seasons. Thus the observed separation between seasons might
not be expected to be as distinct as that between the orig-
Following the seasonal analysis all of the samples were
assigned to their correct season, in preparation of the bi-
Community structure analysis
The discovery of the zooplankton community structure
involved a factor analysis of the vector of zooplankton cat-
egories. This resulted in series of factors that represent
those zooplankton categories that demonstrated the same pat-
tern of variation, or in other words, each factor represent-
ed a zooplankton sub-community. Following the factor analy-
sis, the factor scores were analyzed with an ANOVA to deter-
mine if the zooplankton sub-communities displayed patterns
of variation that could be explained by spatial or temporal
Structure plot (SPLOTS) analysis of biweekly data
The ANOVA of the community structure factor scores pro-
vides an analysis of the temporal progression of the
spatial patterns of the zooplankton community structure by
considering the study region as a unit. An equally impor-
tant question involves the temporal variation of the commun-
ity structure at a particular region. Consideration of this
aspect allows the investigation of which geographic areas
are more variable with regards to their community structure.
While the data to be investigated are the same as for
the previous biweekly analyses, the vantage point from which
it will be viewed is somewhat different. The analytical
tools to be employed are also different. In this situation,
the view of the data will be limited to a single area con-
sidered for all seasons. Thus, three dimensional geographic
plots will not provide the insight necessary. Instead, the
structure, or box and whisker, plots described by Tukey
(1975) will be used.
Canonical discriminant analysis
The second analysis problem calls for the class of mul-
tivariate analysis known as canonical discriminant function
analysis (CDFA) coupled with the multivariate analysis of
variance (MANOVA). The rationale employed was to propose a
conceptual model for the observed pattern of community vari-
ation that could be explored using these statistical tech-
niques. This was accomplished using a two step procedure:
1) Propose a model that grouped the myriad of in-
dependent variables into broad classes of independent fac-
tors. Form a MANOVA and subject the main effect or inter-
action Sum of Squares and Cross Products (SSCP) matrices to
2) The CDF variates and their correlations with
the original dependent variates along with the mapping of
the factor means in the CDF space would provide information
to be used in interpretation or in formulating further des-
The model employed for the biweekly sampling program
for step one for the MANOVA and the ANOVA was:
Yijk = Region + Stat + Seas + Biwk +
Region*Seas + error.
Y = The dependent response variables, the zooplankton
category counts (see Appendix 2);
Region = The six regions of the study area;
Stat = The eleven stations situated throughout the
Seas = The seasons of the sampling period;
Biwk = The biweekly period in which the sample was
If the test statistic resulting from the MANOVA or
ANOVA for the interaction term surpasses the cutoff point,
the main effects will not be interpreted. In this case, the
presence of the main effects in the model will serve to par-
titon, and thus reduce, the error sum of squares. However,
if the test statistic for the interaction term does not sur-
pass the cutoff point, then the main effect terms will be
assumed to affect the response independently and analyzed
A CDFA constructed to discriminate between the twenty-
four possibly distinct season/area means will result in fif-
teen separate orthogonal axes. It is definitely possible
that not all of these new variates, linear combinations of
the original variables, should be considered as valid. The
CDF's, or axes, considered to be invalid, or not represent-
ing valid pattern discriminators, may be discarded.
Bartlett (1941, 1947) has proposed a test of the signi-
ficance of canonical correlations which may be applied to
the test of significance of CDF's.
The test statistic lambda is defined as The
null hypothesis that the groups can not be discriminated by
the CDFI's is tested by a function that is distributed
approximately as a chi-square with pl*P2 degrees of freedom:
9-( I- ) .
nd = (p-
If the null hypothesis can be rejected, the contribution of
the first CDFI to A' can be removed and the significance
of the remaining p2-1 CDFI's can be tested by the chi-square
statistic with (pl-l)(p2-1) degrees of freedom.
In general, with r CDF's removed from / the signif-
icance of the remainder can be tested in the same fashion by
the chi-square statistic.
It should be noted that initial calculation of A yields a
test statistic that is equivalent to the Wilks' Criterion
(Cooley and Lohnes, 1971).
The calculation of this test statistic was performed
for the season/region interaction as an aid in deciding
which of the CDF's, if in fact any, could be discarded in
further analyses and interpretations. Another quantitative
aid used in deciding how many of the CDF's to consider for
further analyses was the percent of the total between groups
variation for which a particular CDF accounted.
The next step in the analysis of the biweekly data in-
volves the interpretation of the CDF's. The basis for this
interpretation is founded in the following points:
The Canonical Discriminant Analysis method forms new
variates that are linear combinations of the original vari-
ates. This is accomplished by solving for the eigenstruc-
ture of a determinental equation. The determinental equa-
tion that is solved to generate the CDF's is formed so as to
emphasize, or discriminate between, the differences of the
treatment mean vectors, in this case the season/area combi-
The CDF's are orthogonal to each other. Each function,
given the linearity and orthogonality constraints expressed
above, "explains" the greatest amount of between groups var-
iation possible, consistent with the functions that have
been extracted before it.
Thus, the CDF's represent independent axes that span a
space of reduced dimensionality. The orientation of the
axes within this space is such that the projection of group
separation along any single axis is maximized.
The axes, which are linear functions of the original
variates, can be analyzed to determine which of the many
original dependent variates are responsible for the observed
separation of the treatment mean vectors. This is accom-
plished by computing the correlation coefficient of the
The correlation coefficients are used in a subjective
interpretation to determine which of the original variates
contribute significantly to the variation of the axis. The
larger the absolute value of the coefficient the greater the
effect a variation in the single dependent variate will have
on the CDF score. The sign of the coefficient tells us
whether an increase in the value of the original variate
will tend to raise or lower the CDF score.
The greatest operational problem encountered in per-
forming a CDFA involves the interpretation of the original
variable-CDFA variate correlation coefficients. There has
not been an objective quantitative method proposed for
determining which of the coefficients are the important
ones. In this area the CDFA is subjective. The selection
criterion employed in this study is analogous to determining
the descending runs of the coefficients and partitioning the
group of coefficients into homogeneous subsets. The list of
coefficients is sorted into descending order with regard to
coefficient magnitude but not sign. This sorted list is
then examined for breaks in the descending run of coeffi-
cients. The first "significant" break is selected as the
cutoff point. In most instances this procedure provides a
clearcut selection point, but occasionally the dividing line
is not clear and an arbitrary decision must be made.
When analyzing an interaction effect, it is useful to
look at the area to area variation within a single season
and of course, the reverse, the season to season variation
within a single area. The overall impact of the area to
area viewpoint is best seen through using geographically
oriented plots, or maps, of the CDF variate scores. The
primary technique employed here is to examine the relief of
the geographic plot (or map it is a map of the geographic
distribution of the CDF score). The greater the relief, the
greater the discrimination of the CDF for the particular
season being investigated. Any of the areas that are at the
same level of relief would appear to be a plateau and can
not be distinguished by the CDF. In addition, to support
the validation of the CDF coefficient analysis, the same
mapping treatment is applied to the original variates singly
(see appendix 3).
The analysis and interpretation then involves the
integration and synthesis of all these quantitative indices
and data display techniques to relate these data into a bio-
The Diurnal Analysis
The diurnal analysis procedure, while conceptually the
same as the biweekly analysis, was operationally quite
different. Although the analysis of both sampling programs
employed multivariate pattern recognition techniques, there
the similiarity ends. The biweekly program was designed to
examine macrovariation in the zooplankton community. It
collected data to look at large scale variation in both time
and space at the study region. As a result of this, the
region and season variables, were used in the independent
portion of the multivariate linear model. These factors
were used instead of the continuous environmental variables
that actually drive the seasonal and spatial variation. The
effect these continuous variables have on the zooplankton
community is an integrative one and probably not measurable
in a meaningful fashion by single point sampling on a
biweekly schedule. In the light of this situation, the
approach was to model with the large scale "known" factors
and then to work backwards utilizing external knowledge to
synthesize an interpretation. This interpretation is then
presented both as an explanation of the functioning of the
zooplankton community and also as a hypothesis to be tested
by future workers.
In the case of the diurnal data, the samples were col-
lected frequently enough to warrant using the continuous en-
vironmental variables on the independent side of the linear
model. Therefore, a regression approach was selected. How-
ever, in a regression approach several problems can arise,
and probably will arise in most survey-type data collection
The first problem centers around performing a multivar-
iate multiple regression analysis. Although, the techniques
are well developed for the multivariate extension of multi-
ple regression, generally termed Canonical Correlation, it
is the author's opinion is that Canonical Correlation poses
severe problems in the interpretation of the results.
An alternative to the Canonical Correlation approach
involves the use of Factor Analysis, or Principle Components
Analysis, and multiple regression. A form of the approach
used in this study was employed by Atchley(1974). This ap-
proach provides that the maximum amount of the dependent
variable set variation will be used in the regression. The
analysis approach is outlined below. The discussion below
will identify each step in the four step procedure.
Diurnal Response Surface Analysis
The first step is to identify the dependent variable
set. In this study, it was the twenty-seven element zoo-
plankton community standing crop vector. Next is the iden-
tification of the independent variables: water temperature,
salinity, dissolved oxygen, pyroheliometer measure, tidal
height and tide direction.
The initial step in analyzing the data involves
performing a Factor Analysis on the dependent variable set.
A discussion of the details of Factor Analysis and which op-
tions to specify is beyond the scope of this section (the
interested reader is referred to a multivariate analysis
text). The purpose for performing the factor analysis is to
orthogonalize and reduce the dimensionality of the dependent
variate set. Thus, by starting with n-variates, the Factor
Analysis will derive a new set of variates, usually much
less than n in number. These new variates are formed under
the constraints that they: are linear combinations of the
original variates, are orthogonal, or independent, to each
other, and explain the maximum amount of the original set's
variance, subject to the first two constraints.
Once the dependent variate set has been made tractable,
albeit, at the expense of some loss of information, the next
step is to begin work on the independent side of the model.
One of the major problems with multiple regression, espe-
cially in field survey data, is the multicolinearity of the
independent variables. This means that the independent var-
iates are in fact not independent from one another. This
can cause problems and result in misleading results from the
linear model. The approach selected for this study was a
screening of the independent variables employing stepwise
regression. By utilizing stepwise regression, one is
reasonably confident that the variables that passed the
screening procedure have a low level of multicolinearity.
The stepwise method was a combination of the forward
and backward model building techniques. The technique be-
gins by finding the one-variable model that produces the
highest R-square. For each of the remaining independent
variables F-statistics are then calculated. These F-statis-
tics reflect the variable's contribution to the model if it
were to be included.
The second variable to be entered into the model is
selected by comparing the F-statistics to the minimum entry
level. If none exceed the minimum entry level, the proce-
dure stops. Otherwise, the variable with the largest F-sta-
tistic is entered into the model. The F-statistic for the
remaining independent variables is then recalculated, and
the evaluation process is repeated.
After a variable is added, the procedure examines all
the variables already included in the model, and deletes any
variable that does not produce an F-statistic greater than
the minimum stay criterion. Only after this check can
another variable be added to the model.
Variables are thus added singly to the model until no
remaining variables produce F-statistics greater then the
minimum entry, or the variable to be added was just deleted.
The next step involves the statistical modelling
procedures of response surface techniques, which is a
special set of the ANOVA/Multiple Regression class of the
General Linear Model. This process entails selecting each
factor, or component, as a dependent variable then selecting
each of the independent variates that passed the stepwise
regression screening. Next, a linear model is formed with
terms for all linear effects, quadratic effects, and cross-
product effects. The resulting model is manipulated using
the full model/reduced model approach so that the appropri-
ate tests may be performed for the significance of the class
of linear, quadratic, or crossproduct effects. Then the
model is reformulated including only those effects that
proved to be significant. The only exception to this invol-
ves the linear terms. If any of the higher order terms are
significant, then include the linear terms also. With the
new model use the partial regression coefficients to test
for the significance of a specific term. Now the model is
formulated once more and the coefficients for the included
terms are determined.
The final step creates a plot of the predicted values
for the response surface. This plot is used to interpret
the relationships of the dependent and independent varia-
bles. The dependent variables are factors, linear combina-
tions of the original variables, which represent a portion
of the zooplankton community that displays a common pattern
of variation. The factor loadings are original variable-
factor score correlation coefficients and are used in a
manner similar to the CDF analysis.
Dependent Variable Transformations
Prior to the analysis of the data by any of the pattern
seeking models, it was thought necessary to determine if the
data needed to be transformed in any manner. This is
desirable since the analysis procedures assume that the
error terms follow a normal distribution, either multivari-
ate or univariate, and that effects specified in the various
models are additive.
The transformation that seemed to be appropriate was
the log transformation, X = logl0(X+l). From a theoretical
point of view, Cassie (1957) found that zooplankton in
oceanic situations displayed an over dispersed pattern of
the log-normal type (see Pielou, 1969). The data from an
earlier study were tested and found to conform to the log-
normal assumptions (Maturo et al., 1974).
Another justification for using the log transformation
is discussed by Cassie (1960). He suggests that a log-nor-
mal distribution of plankton may result from the relative
abundances of plankton species being geometrically related
to the variations in the physical properties of their envir-
Determination of Seasonal Boundaries
The results of the cluster analysis of environmental
variables displayed two well separated groups (see fig.
3). The dividing point occurred between March and April. A
less well defined separation occurred within each of the two
major groups. This indicated that the study could be
divided into four seasons. The individual sample points
were not temporally contiguous. However, the general
grouping was apparent. The fall and summer seasons (seasons
were named based on their temporal proximity to the "true"
seasons) were cohesive, well separated from the other
samples and reasonably temporally contiguous.
For the purpose of the MANOVA/CDF analysis the biweekly
sampling periods were divided into seasons such that: the
fall season consisted of the November and December biweekly
samples; winter included the January and February biweekly
samples; spring contained the March and April biweekly
samples and the first biweekly sample in May;
BIWEEKLY SAMPLING PERIODS
N N D J F F D J M
O O E A E E E A A
V V C N B B C N R
7 7 7 7 7 7 7 7 7
3 3 3 4 4 4 3 4 4
M A M J A
A PA U P
R R Y N R
7 7 7 7 7
4 4 4 4 4
Fig. 3. Cluster dendrogram of biweekly environmental
variable mean vectors ( a= 1st biweekly in month
b=2nd biweekly in month).
J J J
U U U
N L L
7 7 7
4 4 4
and summer was assigned as the season for the remaining
The MANOVA test statistics showed conflicting results
ranging from significant (Pillai's Trace, p=.0305) to
nonsignificant (Wilk's Criterion, p=.2527). The data do not
present a clearly detectable seasonal difference. However,
there is an indication that the difference does exist. This
result may be a consequence of the requirement for temporal
An examination of the characteristic roots showed that
approximately 95% of the between seasons variation could be
explained by the first two Canonical Descriminant Functions
(see table 1).
The correlations between the original variables and the
two CDF's provided a mechanism for interpreting the possible
meaning of the CDF's (see table 1). The first CDF,
explaining approximately 78% of the observed between season
variation,had its highest correlations occurring with water
temperature (-0.121) and pyroheliometer (-0.115).
Correlations of secondary importance were found with organic
carbon (0.072), dissolved oxygen (0.068), and salinity
(0.063). The second CDF, explaining approximately 17% of
the observed between seasons variation, had its highest
correlation with salinity (0.197), with correlations of
slightly lesser importance with wind speed (-0.162) and
organic carbon (-0.157).
I H m ofCmn r- r-N 4
a) > C D 0 C D00 D0 0 0 COD0 Q0 m
a 00000000000000 NN
C C C) I II I NNNN
VD CO D-
m to i
0 -M C0nZ '7 Z
>) 0. r o- 4 E
4J 000< < E
" W U 1 1 u >
J U 1 1 1 1W1
w0 0 lm Oc 0M EQ 2 2 WZW
4- Z a.ZWO w O E. 0 E
C C Z(A-C .OO O< m U U) AZ E
C) 0 uEmuz0 D '4'3 IO
C)iif f k r' ~ ''t ^ -^' i o s r
L. < o o o o oo o o^w a
A plot of the means of the CDF variates was examined to
obtain a visual image of the relationship of the four
seasons to each other when mapped onto the CDF-space (see
Considering each axis separately the seasons fall into
groups. The first CDF served to place the seasons into
three groups: 1) fall; 2) winter and spring; and 3) summer.
The second CDF formed two groups: 1) fall, summer and
winter; and 2) spring. However, when the two axes are
considered together, the four seasons separate.
An examination of the univariate ANOVA's shows that
most of the variables included for analysis display no
detectable differences when grouped into the proposed
seasons. However, the variables emphasized by the
CDF-original variable correlations do display significant
F-values for the ANOVA. In addition, the variable S103 also
shows at least marginal significance (p=0.034). With this
information a re-examination of the CDF-original variable
correlations showed that SI03 could be considered of
The next phase in the interpretation of the seasonal
analysis employed the plots of the original variables.
Having used the MANOVA and CDFA to provide an understandable
pattern from the data, the next step is to compare the
CRYSTAL RIVER ENVIRONMENTAL DATA
SEASONAL MEANS FOR CDFI AND CDFII
45.0 47.5 50.0 52.5
CDF I AXIS
Fig. 4. Season means for CDF axis I and CDF axis II.
proposed pattern to the original data and evaluate its
goodness of fit. This evaluation will be done qualitatively
through visual inspection.
The comparisons of the original data plots for the
variables emphasized by the first CDF show the following
1. The variables with negative correlations, water
temperature, pyroheliometer, and organic carbon, have
basically the same pattern. The values remain constant, or
decrease slightly, from fall through winter. At the
beginning of spring an increase is seen which continues
through the summer. In the late spring, there is a decrease
in temperature and pyroheliometer values, but this is more
likely an anomalous condition, such as a short cold snap.
The pattern shown by the organic carbon variable follows
that observed for the water temperature and pyroheliometer
variables except that the summer values level off and then
decrease. Also, all of the variables, except possibly water
temperature, display one or more rather dramatic outliers.
These outliers tend to cloud the picture and probably have a
strong effect on the significance levels of the test
statistics from the original MANOVA.
2. The variables with positive correlations, salinity
and dissolved oxygen, each have their own pattern. The
salinity starts in the fall with high values (28-30 ppt.),
falls steadily to a low in the early spring (20-22 ppt.),
and then begins to climb again in the summer. The dissolved
oxygen begins in the fall with low values (7.0), climbs in
to a high in the winter (8.5) and then drops steadily
through the spring and the summer (6.0).
The comparisons of the original data plots for the
variables emphasized by the second CDF involved many of the
same variables as examined for the first CDF. This suggests
that the environmental variables are not responding to the
seasons in a strictly linear fashion. Thus the CDFA was
able to decompose these original variables into two linear
components that each serve to separate the seasons in a
different manner. The original variable that was emphasized
by the second CDF that was not emphasized by the first CDF
was wind speed. The pattern for wind speed across the
seasons shows relatively low values for fall and summer (6
mph) and higher average values for winter and spring (8-10
mph). Spring shows the highest average values.
The combination of these various steps in the analysis
allows a characterization of each of the seasons determined:
1. Fall: The highest salinity, with the rest of the
parameters that are considered important measuring as
considerably lower than those experienced in the spring or
2. Winter: Similar to the Fall, except that it
exhibits one of the lowest average salinities.
3. Spring: Has the highest Organic Carbon, wind
speed, and dissolved oxygen. The rest of the selected
variables are markedly higher than those displayed by the
Winter or Fall.
4. Summer: Shows an increase from the Spring except
for Organic Carbon, Dissolved Oxygen, and wind speed. The
Organic Carbon and Dissolved Oxygen mean values are the
lowest observed for the study. The water temperature and
pyroheliometer values are the highest mean values observed
during the study.
Biweekly Canonical Discriminant Function Analysis
The results of the MANOVA showed that the test
statistic for the interaction effect surpassed the cutoff
point (Pillai's Trace, p=0.0001; Wilks' Criterion,
p=0.0001). This indicates that the regions respond
differently for different seasons. Thus, the next step was
to proceed to the CDFA.
The application of Bartlett's test to the Eigenvalue
data for the season/region interaction showed that three
CDF's, or axes, are to be examined (see table 2). These
three axes explain a total of 61.28% of the variation that
was the result of the season/region interaction (see table
3). Thus, by applying the technique of Canonical
Discriminant Analysis, the dimensionality of the
discriminant space has been reduced from twenty-seven to
U 0 00 0 0 '10000000000
O0 0 r, V H
-4L 0m CbUU, 4 1NNHHH
V4 C M CDr(Nn0 mmUmomnIT(nN
m m c o emma vuort rCOONcOMO
E0 cC ^ -m (N-0 Na0mrc
& M oH mrm
o0 33 )
three, with a reduction in "explained" variation from 100%
The examination of the geographic plots gives a
striking visual explanation of why the interaction was
significant (see fig. 5-7). The pattern observed is
different across the areas for each season. It is important
to perform the MANOVA before proceeding to the graphic
pattern analysis. The MANOVA provides a quantitative level
on which to judge whether the observed pattern is
significant or random error.
Canonical discriminant function I
The first CDF explains 34.36% of the between
season/area variation. The zooplankton categories
emphasized by this axis provide the greatest discrimination
between season/region combinations. Examination of the
seasonal geographic plots shows that the major relief is
found upon mapping the Spring and Summer seasons (fig.
5a-d). The Fall season displays little relief, in
comparison to the Spring and Summer. The major feature of
the Fall-CDFI map is the depression in the vicinity of
region 6 and a lesser depression in Region 2. The Winter
season has even less relief than does the Fall season. The
only discernible relief feature in the Winter season is a
depression located in Region 1.
The Spring and Summer seasons show very similar relief
patterns for their CDF1 maps. The highest portion of the
map is located in the open Gulf of Mexico waters of Region
3. From there the raised portion of the map descends
gradually to Region 2 forming what appears as a plateau
covering the non-thermally stressed portion of the study
area. The plateau continues as a lower ridge along the
intake canal and into the barge turning basin. The plateau
descends sharply to Region 1, the next highest region. The
discharge canal and thermal plume area, Region 5, and Region
6 have the lowest relief observed on the map.
The first Canonical Discriminant Axis separates the
regions during the hot portion of the year, the Spring and
Summer seasons. Further the CDF axis divides the study
region into two basic portions: the first consisting of
Regions 2 and 3, and the second containing Regions 1, 5 and
The differences in the community structure for the high
offshore plateau and the nearshore valley can be
investigated by an examination of the individual zooplankton
category/CDF correlation coefficients. The nearshore areas
possess lower standing crops for those categories which have
large positive coefficients and higher standing crops for
those categories that have large negative coefficients.
These coefficients point out those categories that provide
discernible differences between the nearshore and offshore
areas. They are not necessarily, and probably will not be,
the most numerous members of the community. For example,
Acartia tonsa is not emphasized by any of the first three
Canonical Discriminant Functions. Rather, the emphasized
categories are those which display the differences in the
The cutoff point for the first CDF was selected at
0.25. This resulted in three positive categories and four
The negative categories include bivalve larvae,
Temora turbinata, barnacle larvae, and Tortanus setacauda.
The positive categories include Paracalanus quasimoto,
Lucifer sp., and Labidocera sp. (table 3).
Canonical discriminant function II
The CDFII explains 15.85% of the between season/region
variation. Examination of the seasonal geographic plots
show that, with the exception of Fall, each seasonal map
displays a major relief feature (see fig. 6a-d). As in the
first Canonical Discriminant Function the major difference
is between the near shore and off shore areas. However, the
maps generated by this Canonical Discriminant Function show
temporally progressing patterns when viewed chronologically.
The Winter season shows a plateau centered around
Region 2 and extending partially into Region 3. This
plateau drops off sharply, resulting in a broad plain
covering Regions 1, 5, 6, and the discharge portion of 3.
In the Spring season, the relief of the map has changed.
The plateau extends shoreward encompassing Regions 2, 3 and
6. Only the immediate nearshore areas remain in the plain,
and the elevation of the plain has risen dramatically. The
trend observed in the transition from the Winter season to
the Spring season is continued as the Summer season CDF map
is considered. The map for the Summer season shows the
plateau has continued shoreward on the intake side to
include Region 1. This leaves only Region 5, the area
containing the discharge canal and the central portion of
the thermal plume, at a lower elevation. Although the
chronological sequence is disrupted, the Fall season, which
actually precedes Winter temporally, shows the plateau
encompassing the entire study region.
Thus the CDFII is useful in tracking season/region
differences that occur in a component of the zooplankton
community across time. It also serves to separate the
offshore areas from the nearshore areas. The separation of
community types exposed by this axis is most prominent in
the Winter season.
The cutoff point for the CDFII was more difficult to
select. Two possible breaks were present for selection.
The cutoff could be placed at 0.29 or at 0.40. In either
situation, only positive coefficients possess values with
sufficient magnitude to be selected (table 3).
If the first cutoff point is selected, there are five
categories identified as important to the determination of
the CDF score. Those categories are: other shrimp larvae,
chaetognaths, gastropod veligers, Euterpina acutifrons, and
Temora turbinata. If the cutoff point is extended to
include the lower value, then the following zooplankton
categories are added to the list: polychaete larvae, crab
larvae, Paracalanus crassirostris, Oithona sp., and bivalve
One distinctive difference between CDFI and II concerns
the signs of the important coefficients. The second
Canonical Discriminant Function contains only positive
correlations that have magnitudes greater than either cutoff
point. This indicates that the differences seen in the map
are the result of presence or absence of the important
Canonical Discriminant Function III
The CDFIII explains 11.07% of the between season/area
variation. Examination of the CDF seasonal maps present a
different analytical situation than experienced with the
first two CDF's. They presented a reasonably simple,
consistent series of patterns across the seasons. Both axes
appeared to be discriminating between the nearshore,
thermally affected regions and the more offshore regions.
This does not seem to be the case with the third
CDF (see fig. 7a-d). In this situation the patterns
displayed show differences between the intake and discharge
portions of the study region. However, these patterns also
seem to influenced by the nearshore/offshore effect on the
The best approach to take in this situation is to
describe the observed variation pattern on a season by
The Fall season map shows a high plateau in Region 6.
This plateau drops to a lower one in Regions 1 and 2.
Region 5 is a low plain as is Region 3. The lowest
elevation is in Region 3.
The Winter season map shows its highest plateau
connecting Regions 2 and 3. Regions 1 and 5 are lower than
the previously mentioned plateau. The lowest elevation on
the map is found in Region 6. Visually, the Winter season
map appears to be opposite of the Fall map.
The Spring map shows one high plateau covering Regions
1, 5 and 6. This is opposed by a low plain in Region 3,
which descends into a valley covering Region 2. This
seasonal map displays a nearshore/offshore contrast similar
to ones seen earlier.
The Summer map shows almost no relief and thus may be
This CDF has two potential cutoff points: either 0.19
or 0.17. Selecting the first cutoff point results in ten
coefficients being selected, three positive and seven
negative. Extending the cutoff to the second point results
in an addition of one more positive coefficient (table 3).
The positive categories include: Tortanus setacauda,
medusae, other shrimp, and Euterpina acutifrons. The
negative categories include: Metis holothuriae, Temora
turbinata, tunicates, Oithona sp., penaeid shrimp,
polychaete larvae, and Longipaedia helogolandica.
Zooplankton Community Structure Factor Analysis
The vectors of log transformed zooplankton standing
crop scores from the biweekly data were factor analyzed.
The principal axis method of initial solution coupled with a
varimax rotation was employed. A minimum eigenvalue of 1.0
was used to provide the cutoff for the number of factors to
be selected for rotation.
Three factors were retained (see table 4). They
accounted for 86.0% of the original biweekly zooplankton
H VOOW'4Nmm aN m0 cNmo mrrmm v a, c 0
ew i mm- m Or-'O N 0%or-
[ a0 OO N N -4-4N 0(O -,T O4 MO Cr ol OO N, O 1
a) m CCfl i .0
CUD 0 0 0 00000 00 094
fI I1 I0 II
c Um om
c ZU0ooooooo 000o0 ooooooo o00oo0 oooon
0 0 I I <
V-4 v m w- r-v D mrmkwv 0)
S r-IL 0 r- NNNCNqNr C-- n 00 C4mCO O L C
< O rCOaN rCnr-OO O ONmvm NO -4(OOOiOMr 0
(( U 0 0 000 o ooooooooo00
0*4*' I 1 111 1 1I l I I I 4
0 >1 0 U3 0 4O
U. 4l 02 3 a (a0 0
0 4 (a (m 0 u > Q
# m CD O l Mr O U *
(- 0 0- 1 D 0 :3 0 > 0)
30 M s- 0 o1(-'0 a ) O O 0 2a -
C < m 0 00 0 U 4-. -4() 0 2 i > 20 Z0)
00 0 00 m02 u w M 0 4000 0 00 >O.
00 .ci0 0.10 a u-4 40U > a 0> ae
4 -1 0, 3 w (a :3 li :3 ao B o1 q $ -40 4 a) d) 2i E> 10
S 44 ~02 00 O> 02 0 -40 02
01 4 42l U r-O O0 *)U14. 00 >14 0-4 .0 (V > M > (
44 Z 0 4-J 1 >0 0020 0 .0141
w w 0i 0 wlO CO C U > m q 0 Vc B
MI 0 > l r M 4l W W C UMm 1 o 0wC
02 0 w400 C020000 0 a, f 0> = 04 w (00) 201 0
00 > 0 0202 mji)0t 4J M m r_ w w w c 0 r 0
(a U2 -'0- a C4 4m 'aSn4J0 20 > ( 0C 44 a4B) C0 .
04 U0 ( 000 0 > ul a 00 U to (a Z C
uM ),a- W (aa a- ).0 )4( :3
CL I.'jc =cUCmA
Portion of Zooplankton Community Structure Variation
Explained by the Factor Analysis
FACTOR PORTION EXPLAINED CUMULATIVE PORTION
1 0.454 0.454
2 0.275 0.729
3 0.131 0.860
^ u &
.....~ ,, m+
4J o '
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EIJ8XP0O7_JKTU27 INGEST_TIME 2012-09-24T13:49:01Z PACKAGE AA00011858_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC