Comparisons with Tidal Forests of Florida's Gulf Coast and the Roanoke River, ...... The Forested Wetlands of the Southern United States held in Orlando, FL, July 12- .... After graduating high school, he attended Nicolet Area Technical College ...
FRESHWATER TIDAL FOREST COMMUNITIES SAMPLED IN THE LOWER SAVANNAH RIVER FLOODPLAIN
By JAMIE DUBERSTEIN
A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2004
Copyright 2004 by Jamie Duberstein
ACKNOWLEDGMENTS I thank my parents for their continuous support and encouragement, and for instilling in me the belief that people can make positive contributions to the world in any way they choose, so long as they set their mind to it and make the required effort. That mindset is what gives me the freedom to pursue my dreams, while at the same time driving me forward in my occupational development as a natural resources ecologist. I also thank my sister for all of her support and assistance. I thank my advisor, Dr. Wiley Kitchens. Wiley graciously gave me many opportunities to use my experience, imagination, and background knowledge to propose alternative solutions to situations. His timely reminders of the pertinent ecological principles and statistical approaches were always helpful, and their blatancy sometimes humbling. I am particularly grateful to him for passing on some of his knowledge of wetland systems. I thank my committee members, Dr. William Conner and Dr. Michael Binford, for their editorial contributions to my thesis, as well as comments given during my defense. I also thank Dr. Conner for his recommendations during the early stages of my research. I sincerely thank Mark Parry, Janell Brush, Scott Berryman, Zach Welch, Adam Cross, AnnMarie Muench, and Joey Largay for their help and dedication in the field. I also thank the entire staff at the Savannah National Wildlife Refuge for all of their logistic and moral support, particularly William “Russ” Webb, Robert Rahn, and John Robinette. iii
TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................................................................................. iii LIST OF TABLES............................................................................................................ vii LIST OF FIGURES ......................................................................................................... viii ABSTRACT....................................................................................................................... ix CHAPTER 1
INTRODUCTION AND STUDY AREA ....................................................................1 Introduction...................................................................................................................1 Location of Study Area.................................................................................................3 Hydrology .....................................................................................................................6 Soil and Underlying Bedrock .......................................................................................9 Tree Species................................................................................................................10 Lower Floodplain History...........................................................................................11
2
METHODS .................................................................................................................12 Vegetative Sampling...................................................................................................12 Species-Area Curve .............................................................................................12 Sampling Design .................................................................................................13 Soil Analysis...............................................................................................................14 Chemical Constituents.........................................................................................15 Organic Matter Content and Bulk Density..........................................................16 Statistical Analyses.....................................................................................................17 Species Importance Values..................................................................................17 Insignificant Data Removal.................................................................................18 Rare species..................................................................................................18 Outlying plots...............................................................................................19 Insignificant environmental variables ..........................................................20 A priori Landscape Grouping..............................................................................20 Exploratory Data Analyses..................................................................................21 Cluster analysis ............................................................................................21 Indicator species analysis .............................................................................22
iv
Multi-response permutation procedures.......................................................24 Nonmetric multidimensional scaling ordinations ........................................25 Classification and Regression Tree .....................................................................26 3
RESULTS ...................................................................................................................28 Introduction.................................................................................................................28 Exploratory Data Analyses .........................................................................................29 Cluster Analysis...................................................................................................29 Indicator Species Analysis ..................................................................................30 Multi-response Permutation Procedures..............................................................34 NMS Ordinations.................................................................................................35 Autopilot.......................................................................................................35 Subsequent ordinations ................................................................................37 Classification and Regression Tree Analysis .............................................................46 Descriptions of Communities .....................................................................................48 Shrub Community ...............................................................................................48 Water Tupelo Community...................................................................................50 Swamp Tupelo – Tag Alder Community ............................................................51 Water Oak – Swamp Bay Community ................................................................52
4
DISCUSSION.............................................................................................................55 Tidal Forest Communities in Sampled Areas of the Savannah River Floodplain......56 Comparisons with Tidal Forests of the Lower Chesapeake Bay................................57 Community Description ......................................................................................57 Environmental Factors.........................................................................................58 Comparisons with Tidal Forests of Florida’s Gulf Coast and the Roanoke River, NC ..........................................................................................................................59 Comparisons with Bottomland Hardwood Soils ........................................................61 Community Description ......................................................................................61 Soil Properties .....................................................................................................61 Future Research Needs ...............................................................................................62
APPENDIX A
TIDAL FOREST COMPUTATIONS BASED ON NATIONAL WETLAND INVENTORY .............................................................................................................64
B
SPECIES NAMES AND ABBREVIATIONS...........................................................65
C
SPECIES X PLOT DATA MATRIX .........................................................................66
D
SOIL PROPERTY X PLOT DATA MATRIX ..........................................................70
E
CORRELATION OF SPECIES AND SOIL CONSTITUENTS TO AXES FOR RUNS SUBSEQUENT TO AUTOPILOT MODE ....................................................76
v
LIST OF REFERENCES...................................................................................................77 BIOGRAPHICAL SKETCH .............................................................................................82
vi
LIST OF TABLES Table
page
2-1
Species removed from analyses. ..............................................................................19
2-2
Environmental variables collected. ..........................................................................20
3-1
Monte Carlo results of species indicator value. .......................................................32
3-2
Significant indicators species in 4 clusters...............................................................33
3-3
MRPP results for groups of plots. ............................................................................35
3-4
Pearson’s coefficients of determination (r2) and Kendal’s tau values of environmental variables to axes for autopilot mode of NMS ordination. ................36
3-5
Proportion of variance represented by axes in NMS ordination. .............................37
4-1
Published nutrient values (mg/kg) of forested wetland soils. ..................................62
vii
LIST OF FIGURES Figure
page
1-1
Locations of tidal forests documented in the United States. ......................................1
1-2
Location of study areas ..............................................................................................4
1-3
Projected 0.1 ppt salinity zones during and after operation of tide gate ....................5
1-4
Mean annual discharge at USGS monitoring station near Clyo, Georgia..................9
2-1
Species-area curve....................................................................................................13
2-2
Locations of plots and the a priori group they were placed in .................................22
3-1
Cluster dendrogram ..................................................................................................31
3-2
Summary of the 7 indicator species analyses...........................................................32
3-3
NMS ordination: dahoon holly and Virginia willow................................................39
3-4
NMS ordination: fetterbush and wax myrtle............................................................40
3-5
NMS ordination: water tupelo..................................................................................41
3-6
NMS ordination: swamp tupelo and tag alder..........................................................42
3-7
NMS ordination: water oak and swamp bay ............................................................43
3-8
NMS ordination: biplot of axis 2 vs 1 ......................................................................45
3-9
NMS ordination: biplot of axis 3 vs. 2 .....................................................................46
3-10 Classification and regression tree.............................................................................47 3-11 Community locations within the sample areas.........................................................49 3-12 Average stems per acre for communities and a-prior group ....................................53
viii
Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Master of Science FRESHWATER TIDAL FOREST COMMUNITIES SAMPLED IN THE LOWER SAVANNAH RIVER FLOODPLAIN By Jamie Duberstein August 2004 Chair: Wiley M. Kitchens Major Department: Natural Resources and Environment Two freshwater tidal forest stands were sampled in the lower Savannah River floodplain. Multivariate statistics were used to help describe community composition. Plots were agglomerated using cluster analysis, indicator species characteristic of each community were identified, and multiple response permutation procedures were used to test significance differences between the groups. Trends were examined using nonmetric multidimensional scaling ordinations of plots in species space with vector overlays of edaphic factors. Finally, a classification and regression tree analysis was used both in a confirmative fashion, to compare varying results based on community size, and in a predictive fashion, characterizing communities based solely upon soil properties. Four communities were found: 1) shrub 2) Nyssa aquatica 3) Nyssa biflora – Alnus serrulata and 4) Quercus nigra – Persea palustris. The shrub community has the most homogeneous mix of species and the highest stem density per hectare of small diameter trees. This community also contains the rarest
ix
species documented. It exists on substrate that has a very high organic matter content (>78%), with high levels of Ca2+, Mg2+, Na+, and electrical conductivity. In general, this community is found in areas relatively far removed from tidal creeks and drainages. The Nyssa aquatica community has the highest density and greatest basal area of Nyssa aquatica canopy trees among the communities in this study. Decreased development of the shrub layer is a general maxim. The soil has, on average, the highest concentration of PO43-. This community is found near tidal creeks and drainages in the western study site. The Nyssa biflora – Alnus serrulata community has a well developed canopy in terms of tree heights and abundances. Nyssa biflora dominates the canopy, along with the highest amount of Taxodium distichum found in any of the communities. The shrub layer of this community is relatively well developed and dominated by Alnus serrulata, Cornus foemina var. foemina, and Cephalanthus occidentalis. Soils of this community have high electrical conductivity and Na+ concentration, though not nearly as high as the shrub community. This community is generally found associated with tidal creeks and drainages in both the eastern and western stands. The Quercus nigra – Persea palustris community has a canopy layer with uniform distribution of Nyssa biflora, Nyssa aquatica, ash, and Taxodium distichum. The shrub layer of this community is dominated by smaller “tree” species rather than “shrub” species. On average, the soils have the lowest values of organic matter, Na+ concentration, Ca2+ concentration, and electrical conductivity found in this study. This community is found in areas of greatest tide-water flux.
x
CHAPTER 1 INTRODUCTION AND STUDY AREA Introduction Along the coastal plain of the southeastern Unites States exists a unique mosaic of forest habitats that are expressive of the junction of several water sources: an alluvial river system, groundwater seepage, and a tidally driven hydrologic backflow of fresh water. These freshwater tidal forests are little studied. The few accounts of tidal forests include those found along the Pamunkey River in the lower Chesapeake Bay (Doumlele et al. 1985, Rheinhardt 1991, 1992, Rheinhardt and Hershner 1992), the Roanoke River in North Carolina (Wharton et al. 1982), the Altamaha (Wharton et al. 1982) and Savannah rivers of Georgia, as well as the Apalachicola, Suwannee, St. Marks, and Yellow rivers in the panhandle of Florida (Wharton et al. 1982) (Figure 1-1).
Tidal forests documented
Figure 1-1. Locations of tidal forests documented in the United States. All occur within the southeastern region of the country.
1
2 These unique ecosystems occur where large river systems meet a tidally forced backflow, situated just upstream of freshwater tidal marshes. Geographic extents of most tidal forests have not been calculated, but they seem to be proportionally sized to the tidal range for each river system, whether those relationships be directly related to depth of overland flow or indirectly related to mean water-table height as a result of tidal backpressure (Rheinhardt and Hershner 1992). The extents of tidal forests situated along the Gulf of Mexico are likely relatively small while in the Virginia part of the lower Chesapeake Bay there are a total of approximately 3500 ha on three different rivers (Rheinhardt 1992). The Savannah River floodplain has a tidal range of up to 3 m, resulting in comparatively large areas of tidal forest. There are approximately 3900 ha of truly tidal forest in the Savannah River floodplain, with an additional 500 ha of seasonally tidally flooded forest, and 150 ha of temporarily tidally flooded forest, calculated from national wetlands inventory geographical information systems coverages (Appendix A). The National Wetland Inventory, which follows the classification system developed by Cowardin et al. (1979), has classified the northern portion of the western study stand as being non-tidal, when it clearly is (personal observation). While the National Wetland Inventory is useful for general habitat quantification such as this, comparisons made using this information should be broad and take into consideration the potential for minor misclassifications. The objectives of this study were to identify the suite of tree species that occupy tidal forests of the Savannah River basin, and explore factors that can help to explain general tree communities. The working hypothesis is that species composition and densities, together with substrate characteristics will be reflective of the topographical
3 and hydrologic history of the area. Further hypotheses regarding topographical factors (elevation, ridge and swale location, etc.) and hydrological factors (hydroperiod, groundwater vs. overland water source) may be formed once the biological and edaphic factors measured in this study have been analyzed. To accomplish this I will look at: how the plots naturally group into communities based on their species compositions, specific species abundances that may be indicative of each community, general landscape position of each plot, and soil characteristics that may be used to typify each community. Location of Study Area The Savannah River Bird Refuge was originally established on April 6, 1927, by Executive Order Number 4626. That decree set aside a total of 953 ha as a preserve and breeding ground for native birds. In 1940, Presidential Proclamation 2416 renamed the refuge the Savannah National Wildlife Refuge (SNWR). Throughout the years, a variety of parcels were added to the SNWR through several executive orders, acquisitions using both duck stamp funds and Land and Water Conservation funds, exchange of spoilage rights, exchanges for power line right-of-ways, and several purchases in title fee. The current total acreage of the SNWR is now 11,239 ha (Graves 2001) situated along the borders of Georgia and South Carolina. The U.S. Army Corp of Engineers and the U.S. Geologic Survey (USGS), the organizations from which all location designations and water data were obtained for purposes of this study, use the river mile (RM) as the unit of measure for distances along a river, and cubic feet per second (cfs) as a measure for discharge. Therefore, the International System of Units (SI) convention for these measurements will be broken and the current U.S. convention followed. The study area lies within SNWR boundaries (Figure 1-2). Two forest stands were chosen based on projected salinities from a
4 hydrologic model used to predict interstitial salinities (Pearlstine et al. 1990). The eastern stand is located 26 RM from the Atlantic Ocean adjacent to the Little Back River, a distributary of the main Savannah River. The western stand is located 27 RM from the mouth of the Atlantic adjacent to the main channel of the Savannah River. Salinities in the eastern stand area (Figures 1-2 and 1-3) during the 14 years of operation of the tide gate (described in later sections) were projected to be in excess of 0.1ppt, whereas salinities in the western stand were below 0.1ppt (Pearlstine et al. 1990). # #
West
#
#
#
#
#
East
#
# #
#
# #
#
#
#
#
# #
#
#
# #
#
#
# #
#
# #
# #
#
1:10,000
#
1:10,000
# #
#
#
#
#
West
South Carolina
East
Little Back River
Savannah National Wildlife Refuge r ive hR na van Sa
Savannah
Atlantic Ocean
Georgia
Figure 1-2. Location of study areas. Study plots are indicated by green dots. The Savannah National Wildlife Refuge is indicated by the cross-hatched area.
5
West
East
During Tide Gate
New Cut
After Tide Gate
3.2 km (2 mi)
Savannah Figure 1-3. Projected 0.1 ppt salinity zones during and after operation of tide gate. Modified from Pearlstine et al (1990). Higher salinity occurs to areas south of 0.1 ppt zone during each time period. Inset picture shows tide gate in operation. Even though salinity at the 0.1ppt level is quite low, it was believed that the community compositions, particularly the subcanopy structure, in the two areas may vary as a function of differing salinity stress during the tide gate era. Sampling points (plots) and rationale will be described in Chapter 2.
6 Savannah, Georgia has an average annual temperature of 19ºC, with the highest monthly average of 27.8ºC in July and lowest monthly average of 9.6ºC in January (NOAA 2002). The average frost free season is 226 days long (90% confidence), occurring between March 30th – October 31st (NOAA 1988). Average annual precipitation is 126 cm with an average high of 14 cm in June (NOAA 2002). Hydrology Hydrology has been widely recognized as the major factor in the determining the community distributions of wetlands plants (Conner et al. 1981, Parsons and Ware 1982, Wharton et al. 1982, Mitch and Gosselink 2000), as well as bottomland hardwood community development and succession (Larson et al. 1981). The community composition of freshwater tidal forests is also likely to be greatly affected by both existing and past hydrologic conditions. Changes imposed upon the Savannah River have been documented as the cause of vegetation shifts in marsh macrophytes (Latham 1990). Although tree species in swamps are unlikely to respond as quickly to changing hydrologic conditions as compared to marshes, especially if annuals are an important component of the marsh plant community (Rheinhardt and Hershner 1992), the long-term effects of dam construction, tide gate installation and decommission, and rising sea level are unknown. Bottomland hardwood forests can be broken down into two main types based on their primary source of water and subsequent nutrient load: blackwater swamps and redwater swamps. Blackwater swamps arising in the coastal plain receive water inputs principally through precipitation and are typically nutrient poor. Alluvial floodplain forests, also known as redwater swamps, receive floodwater from rivers draining
7 Piedmont watersheds and are relatively nutrient rich due to the physical and chemical breakdown of rock. The lower Savannah River undergoes a regular, semidiurnal flooding regime and is a salt-wedge type estuary (Hansen and Rattray 1966). Tidal range of the Savannah River marshes is in excess of 3 m with flow reversals 28 RM upstream of the river mouth. Tidal ranges at the study sites of the Savannah tidal forests, however, are only 1.5-2 m on average, which are approximately comparable to the 1m mean tidal range in the tidal freshwater swamp along the Pamunkey River, Virginia (Doumlele et al. 1985). Positioned upstream of tidal freshwater marshes and downstream of bottomland hardwood forests, the tidally influenced forest of the Savannah River basin is classified by U.S. Fish and Wildlife Service convention (i.e. Cowardin et al. 1979) as being a palustrine system, forested wetland class, broad-leaved deciduous subclass with a permanently flooded-tidal modifier (PFO1/2T). Hydrologic conditions resulting from the range and consistency of the semidiurnal tides keep soils saturated for the entire year in most areas of the tidal forest, even during drought conditions (personal observation). In 1977 a one-way tidal flap gate was installed at RM 14 as a mechanism for minimizing the amount of maintenance dredging in the shipping channel (Front River) of the Savannah River. In-flowing water was allowed to pass upstream through the gate during the tidal flood stages. The one-way flap gate was shut at slack tide, and the entire volume of entrained water was forced to flow through a diversion channel (New Cut, Figure 1-3) and out the main channel during ebb tide, thereby increasing the velocity and scour through the harbor area. However, the blockage caused salt water intrusion into the Little Back River and Middle River portions of the Savannah River. With each tidal
8 cycle, the salt wedge was pushed further upstream, resulting in a dramatic shift in vegetation from freshwater species to those that are more tolerant of oligohaline conditions (Georgia Ports Authority 1998). Salinity projections by Pearlstine et al. (1990) (Figure 1-3) indicate portions of the tidal forest of the Savannah River floodplain area were impacted from operation of the tide gate. Although not likely as dramatic as the diversion of the Santee River into the Cooper River that caused a reduced growth rate in water tupelo (Nyssa aquatica) (R.A. Klawitter, personal communication, in Wharton et al. 1982), the increased salinity may have been a factor affecting the community makeup in some areas of the Savannah River tidal forest. In 1991 the tide gate was taken out of operation, with the subsequent closure of New Cut in 1992. In 1993-94 the shipping channel was further deepened by 1.2 m. To date, salinity levels in the tidal forest stretches of the lower Savannah River floodplain have returned to below 0.5 ppt (personal observation), the level used to define a “freshwater” system (Cowardin et al. 1979). The Savannah River, arising in the southern Appalachian Mountains, is an alluvial river and has the 5th largest discharge in the southeastern coast next to the Mississippi, Alabama, Apalachicola, and Altamaha rivers. Freshwater inputs to the basin are from inland runoff from the (approximately) 25,500 km2 drainage area. A comparably sized watershed of an alluvial floodplain may be expected to flood from 18% to 40% of the year (Bedinger 1981). Mean discharge of the Savannah River is 16,060 cfs at USGS station #02198500 (Fig 1-4) located at RM 61 near Clyo, Georgia. Aside from the natural seasonal and lunar fluctuations, the discharge of the Savannah River is governed by a series of three dams: The J. Strom Thurmond Dam and
9 lake was constructed at RM 237.7 in 1954; The Richard B. Russell Dam and lake, located at RM 275.1, was constructed in 1963; The Hartwell Dam and lake, at RM 304.7, was constructed in 1983. Meade (1976) found that the reservoirs trap 85% to 90% of incoming sediment. Sediment inputs to the continental shelf have been decreased by 50% since 1910 as a result of the reservoirs and dams. Therefore, the recharge of the sediment load to the river south of the reservoirs must come from the river bed, banks, and floodplain. This affects the tidal forest drastically by reducing sediment inputs to the floodplain, as well as increasing erosion on the floodplain and tidal creeks.
40000
1200
1000
800 20000 600 15000
400 10000
200
5000
2000
1990
1980
1970
1960
1950
1940
0
1930
0
Year
Figure 1-4. Mean annual discharge at USGS monitoring station #02198500 near Clyo, Georgia. Soil and Underlying Bedrock The underlying bedrock is geologically recent coastal plain sedimentary rocks composed of marsh and lagoon deposits from the Pleistocene and Holocene epochs
Discharge (cubic meters per second)
25000
Tide Gate Decomissioned
Discharge (cubic feet per second)
30000
Russell Dam
Tide Gate Initiated
35000
Hartwell Dam
Thurmond Dam
1400
10 (Quaternary Period). Technically referred to as the Pamlico Shoreline Complex, the underlying bedrock is composed predominantly of sand and sandy clay with marsh and lagoonal facies which were deposited at former high sea levels (GA DNR 1976, 1977). General soil descriptions from the U.S. Department of Agriculture, Soil Conservation Service, indicate the soils in the eastern stand are Levy Soils, which are very poorly drained, nearly level soils on the lower coastal plain. The surface layer is very dark gray silty clay loam 20 cm thick. The underlying material, to a depth of 152 cm, is gray silty clay over silty clay loam (USDA 1980). General descriptions of soils in the western site indicate that they are composed of Angelina and Bibb soils, also frequently flooded and poorly to very poorly drained. These two soil series occur together, in approximately a 4:2:4 ratio of Angelina:Bibb:other (“other” being Chipley, Kershaw, and Ocilla soils). They have been formed in recent deposits of sediments washed from soils on the coastal plain. Surface layers are very dark gray loam about 8 cm thick (Angelina) or light brownish gray loamy sand about 46 cm thick (Bibb). The underlying areas are black to light-gray sand to silty clay loam (Angelina) or mottled light-gray to greenish-gray coarse sand to sandy loam (Bibb). The clay content between depths of 25 and 102 cm within the Bibb series is less than 18 percent (USDA 1974). Tree Species Individual tree species of the Savannah River tidal forest (Appendix B) have been documented as being part of many other forested wetland communities. The communities include those described in the bottomland hardwood forest community profile (Wharton et al. 1982): six zones of seasonally flooded bottomland forests, and descriptions of freshwater tidal forests of the Suwannee and St. Mark’s rivers in Florida.
11 Descriptions of freshwater tidal forests of the lower Chesapeake Bay (Doumlele et al. 1985, Rheinhardt 1991, Rheinhardt 1992, Rheinhardt and Hershner 1992) also contain identical species, though community structure differs greatly from those in the Savannah River floodplain. Lower Floodplain History The lower basin has been severely altered to facilitate a variety of anthropocentric benefits. In the mid 1700’s much of the tidal portions of marsh and forest along the Savannah River were transformed to rice cultivation. Through this process the trees were cut down and moved out of the way or burned, and the stumps largely removed (Doar 1936). The presence of some remaining large stumps adjacent to Rifle Cut, a man-made tidal creek, suggests that the tidal forest may have extended at least 5 RM further downstream. After the 1863 issuing of the Emancipation Proclamation by President Abraham Lincoln, and the subsequent ending of the Civil War in 1865, rice cultivation in the tidal marshlands failed and much of the land was abandoned (McKenzie et al. 1980). Aerial photography shows signs of rice field drainage creeks that had been constructed without the associated land clearing that lie just south of the eastern site along the Little Back River. This is likely where rice development in the tidal floodplain ended and it is probable that the currently existing forest area wasn’t logged for any reason. Cypress stumps endure for many years, and their presence may indicate what the original forest on a given site was like (Wharton et al. 1982). It follows that the absence of obviously logged stumps throughout the area the tidal forest currently occupies is a good indicator that this area was not logged in recent history.
CHAPTER 2 METHODS Vegetative Sampling Species-Area Curve A pilot study was undertaken to determine the tree species richness and diversity within the study areas. It was quickly evident that the eastern site had greater diversity, so efforts were focused within its boundaries. In all, ten nested quadrats were cataloged for information regarding species and size. The smallest reasonable area to be quantified was assumed to be 25m2 (5X5). By lengthening each quadrat by 5m on two ends, each nested quadrat, then, consisted of one of each quadrat size: 25, 100, 225, and 400m2. With this information, a species-area curve (Cain 1938; Kent and Coker 1992) was developed to determine the minimum quadrat area (equivalent to minimal area for the community) for obtaining the data for this study. Unlike the traditional species-area curves that use a progressive doubling of the quadrat size (Kent and Coker 1992), our species-area curve necessitated an algorithm that could incorporate several samples of the same quadrat size. To accomplish this, the computer program Sigma Plot 8.02 (SPSS Inc. 2001) was used to perform a nonlinear regression, resulting in an optimal quadrat size of 100m2, or 10x10m (Figure 2-1).
12
13
Number of Unique Species
20
15
10
5
0 25
100
225
400
Size of Quadrat (square meters)
Figure 2-1. Species-area curve computed with non-linear regression (r2=0.58) based on 10 points with 4 nested quadrat sizes. Sampling Design Plots were positioned in a stratified random manner in each of the two stands in order to represent each stand in a way that would reflect the heterogeneity of the study area. However, since differences between the two stands were evident by mere observation, the stratification (see below) was unique to each stand. One obvious difference was the greater structural diversity in tree communities within the eastern stand. The gradient between understory and overstory is continuous for most of the eastern stand whereas the majority of the western stand has a notable gap between the understory and overstory. Soil conditions, principally stability of the substrate layer, were also obviously more variable in the eastern stand. For these reasons, the stratification of the eastern stand divided it into 4 equal sized quarters. Four points were randomly placed in each of the quarters using a random number generator and a geographical information systems (GIS) coverage of the area. The western stand was divided into 2 approximately equal sized sections; one north of HWY I-95 and one south of the highway. Eight points were randomly placed in each section using the same tools
14 as used in the other stand. That made for a total of thirty-two 10x10m plots in two stands. A quadrat was flagged off at each point by measuring due north from the point 10m, then due east 10m, south 10m, and west 10m. Within each quadrat (plot) all tree species ≥ 1.38m (4.5 ft, or breast height) were identified and measured for diameter at breast height (DBH). The canopy position of each tree was recorded as being in one of three groups: understory, sub-canopy, or canopy. This was determined using relative, rather than absolute heights since the overall community structure differed markedly between quadrats. Soil Analysis Sample nutrient concentrations, organic matter content, pH, and bulk density of the general site area were obtained by taking samples from the substrate floor (i.e., not on microtopographical highs or “hummocks”) at each plot. Although Rheinhardt (1992) found that there were no statistical differences between organic matter content of hummocks and hollows (i.e., the substrate floor), I felt as though the soil properties in hummocks would be more representative of the specific species living on them, rather than the proximal tree community as a whole. It was also thought that the hummocks may be more variable in regard to their nutrient concentrations. Further, microtopographical highs contain a dense root structure, making sampling further biased by both the higher values of organic matter, as well as increased effect of soil compaction from excessive pressure on the soil corer. Two samples were taken at each sampling location to a depth of 12.6 cm using a 6.9 cm diameter aluminum soil corer with holes drilled in the sides to allow for water drainage and accurate measurement. All samples were placed in a freezer upon return
15 and frozen until processing. When processed, all samples were thawed and then oven dried at roughly 50ºC (120°F) for at least 2 weeks to adequately remove all moisture. From one of the two samples, material was passed through a 2mm sieve, homogenized, and sent to the lab for analysis. Material passing through the sieve included mineral matter as well as organic matter. The second soil sample was homogenized, weighed to determine bulk density, and combusted to determine organic matter content by the loss on ignition method. Chemical Constituents All nutrient concentrations in the soil samples were determined by the Analytical Research Laboratory at the University of Florida. Concentrations of phosphorous (P), potassium (K), calcium (Ca), magnesium (Mg), zinc (Zn), manganese (Mn), sodium (Na), and iron (Fe) were determined by Mehlich extraction with 5g (4cm3) soil to 20mL 0.05 HCl + 0.0125 M H2SO4. Electrical conductivity and chloride ion (Cl-) concentrations were determined using a 2:1 water to soil ratio with 250 cm3 soil. Values were multiplied by the bulk density of the soil for each site to standardize the quantities of nutrients present rather than just the concentration. Macronutrients such as P, K, and N, are well established as being very important in plant nutrient needs. The availability of C, N, and P may prove critical in determining the health of a system (Salisbury and Ross 1992), and Fe and Mn concentration levels have been found to be elevated in hydric soils (Gambrell et al. 1989). Soil pH has also been found to be at least partially correlated to extractable Fe and Mn (Gambrell et al. 1989), therefore we recorded soil pH with an Oakton pH 6 Acorn series meter in the hole created by removal of the soil core.
16 Organic Matter Content and Bulk Density Soil organic matter (SOM) was assumed to be one of the most important soil properties for analyzing soils-to-tree relationships. Wharton et al. (1982) found that amount of SOM varies between the National Wetland Technical Council zones and is a useful variable to examine when comparing blackwater and alluvial floodplains (Wharton et al. 1977). Given the relationship to floodplain characteristics, and its assumed relationships to species assemblages, the method of obtaining values for SOM was carefully chosen. The loss on ignition (LOI) method (Klawitter 1962) was used to determine SOM. While earlier studies (eg. Wakeman and Stevens 1930, Robinson 1939) recommend the Walkley-Black method (a chromic acid oxidation, Walkley and Black 1934) for conventional soils, LOI is preferable for hydric or highly organic soils (Broadbent 1953, Storer 1984, Deutsch 1998) of the type encountered in this study. The LOI method involved combusting the samples in an ignition furnace at 500ºC for eight hours. The formula for calculating percent organic matter for each plot was: %OM = [(weight loss due to ignition)/(dry soil weight)]*100 Two runs were done: one with a 2 g sub-sample, and one with an entire column of soil. In the first run two 2 g sub-samples were averaged. The burnt sub-samples were then added to the remaining sample for that site and sent to the laboratory for analysis (see chemical constituents, above). In the second run the full amount of an additional sample, collected June 2003, was used. Since this sample was composed of entirely organic-free constituents after burning, and since chemical analyses were already done on the previous run, the burnt remains were of no further use and discarded.
17 Bulk density was computed by dividing the dry weight of each soil sample by the known volume of soil collected (470.9 cm3). Values of soil organic matter, bulk density, nutrient concentration, electrical conductivity, and amount of nutrients present in each plot are listed in Appendix D. Statistical Analyses Species Importance Values Accurately representing a particular tree species’ contribution to the community makeup of a given plot is perhaps the most important step in community analysis. When studying distribution of tree species, two main factors must be taken into consideration: how many, and how large. Due to inherent differences of plot structure, a method of representing the competitive interactions at each plot is imperative. For example, one plot may be comprised of many small, shrubby trees whose collective basal area is small. Conversely, a plot may be made up of relatively few big trees with a large cumulative basal area. The tidal forests along the Savannah River floodplain have structures described in both scenarios. Importance values are an optimal way of dealing with large differences in structural diversity, while still accurately representing the importance of a species in a plot. Originally developed by Curtis and McIntosh (1950, 1951), importance values have been used in many studies of eastern North American forests (McCune and Grace 2002), including studies of the tidal freshwater swamps of Virginia by Doumlele (1985), Rheinhardt (1991, 1992) and Rheinhardt and Hershner (1992). One rationale for their use is the fact that importance values are not overly sensitive to extremes of structural diversity, as are measures of relative dominance or relative frequency alone. The conversion of the species by plot data to importance values has yet another advantage. It
18 essentially is a standardization transformation of the data. Standardizations of this type are widely used in gradient analyses because it increases the strength of the relationship between species dissimilarity and ecological distance for moderate or long gradients (Faith et al. 1987). For this study the importance values were computed in a manner similar that of Curtis and McIntosh (1950, 1951), with the elimination of the relative frequency term (for more information see Kent and Coker 1992). The value is the average of two components: 1.
2.
Relative Density: Number of individuals of a particular species Total number of individuals of all species
* 100
Relative Dominance: Average basal area of a particular species * number of that species in that plot * 100 Total basal area of all species in that plot In this way, importance values summed over all species within a plot add up to 100.
Species importance values for each plot are listed in Appendix C. Insignificant Data Removal A full matrix of 28 species x 32 plots was modified by the removal of rare species and an outlying plot. The resulting matrix, which will be referred to as the primary matrix, contains 20 species and 31 plots. A second matrix containing all environmental variables was edited in a way that only meaningful data were retained; the resulting matrix will be referred to as the secondary matrix. Specifics of data scaling and deletion follow. Rare species Rare species were removed from the analyses in an effort to tighten patterns and enhance the detection of relationships between community composition and environmental factors. Using an approximate rule of thumb offered by McCune and
19 Grace (2002), those species that were present in fewer than 5% of the plots (i.e., 2 plots or fewer) were removed from the analyses. Although deletion of rare species is considered inappropriate when examining patterns in species diversity (Cao et al. 1999), it is often helpful for multivariate analysis of community structure (McCune and Grace 2002) such as nonmetric multidimensional scaling ordination. In total, 8 species were removed (Table 2-1). Table 2-1. Species that were removed from analyses. Species Inkberry (Ilex glabra) Highbush blueberry (Vaccinium corymbosum) Sweet bay (Magnolia virginiana) Groundsel tree (Baccharis halimifolia) Black alder (Ilex verticillata) Black willow (Salix nigra) Water elm (Planera aquatica) Laurel oak (Quercus laurifolia)
Plots found NE3 NE3, NW4 NW1 NW2, W14 SE1, W12 SE2 W13 W13
Outlying plots Following the removal of the 8 rare species, an outlier analysis was done to detect entire plots that were functioning as outliers. This was done by calculating the average distance, using the Sorensen distance measure, from each sample unit to every other sample unit. Those plots that were more than 2 standard deviations from the mean for average distance were considered outliers. A plot located in the western site (W13) was removed from the analyses. This plot was comprised of mostly canopy and sub-canopy trees, including (predominantly) swamp tupelo (Nyssa sylvatica var. biflora), with some bald cypress (Taxodium distichum) and ash (Fraxinus spp.) trees of similar canopy position. Relatively few shrubs were cataloged in this plot, likely resulting in the outlying nature. A single water elm (Planera aquatica)
20 sapling (DBH < 1cm) and a laurel oak (Quercus laurifolia) sapling (DBH 4.1 cm) were also found in this plot only. Edaphic properties were not dissimilar to other plots. Insignificant environmental variables Environmental variables were first scaled to reflect the same order of magnitude as the data in the primary matrix. To accomplish this task, values for particular variables were multiplied or divided by orders of 10 so that the resultant value was as near the range of 10-100 as possible. Following the relativization, NMS procedures were used to determine the correlations between environmental variables and the main dissimilarity matrix obtained from the primary matrix. In an effort to discern true relationships between tree communities and soil properties, quantitative soil variables that had a coefficient of determination (r2) less than 0.392 to any of the axes for any rotation were removed from the environmental matrix (Table 2-2). Table 2-2. Environmental variables collected in the tidal forests of the Savannah River floodplain. Only variables with a Pearson’s correlation (r2) of at least 0.392 were retained for further analyses. Variables retained Organic matter Ca concentration Mg concentration Electrical conductivity Na concentration Cu present Bulk density P present
Variables removed pH Zn present P concentration Mn concentration K concentration Mn present K present Cu concentration Ca present Fe concentration Mg present Fe present Zn concentration Na present
A priori Landscape Grouping Each plot was placed into one of three physiognomic categories based on their landscape position and assumed hydrogeomorphologic differences (Figure 2-2): 1) Plots that are proximal to either the main channel of the Savannah River or a large distributary. These plots are likely to be of higher elevation and have higher mineral content since
21 they are associated with the natural levee of the river. 2) Those plots associated with tidal creeks and drainages. Lower in elevation than the latter group, the proximity of these plots to tidal rivulets in the floodplain likely results in intermediate drainage conditions and soil mineralization as compared to the other two groups. 3) Plots relatively far removed from tidal creeks and drainages and, therefore, from the main channel of the Savannah River. These are essentially the backswamp sites furthest removed from the main rivers, experiencing decreased water flux with each tidal cycle. Relative isolation leads to very poor drainage, ponding, as well as increased residence time and accumulation of organic matter and nutrients in the soil. A categorical variable was added to the secondary matrix to reflect this grouping. Exploratory Data Analyses Unless otherwise stated, all exploratory analyses were done using the statistical software PC-ORD for Windows, version 4.27 (McCune and Mefford 1999). Similarly, unless otherwise noted, the distance measure used was Sorensen (Bray-Curtis) due to its non-parametric nature. Cluster analysis A hierarchical, polythetic (multiple species), agglomerative clustering was done on sample units based upon the importance value of each species in each plot. The clustering routine utilizes the Sorensen distance measure in combination with a flexible beta (β = -0.25) linkage method (McCune and Grace 2002). Group memberships from the cluster analysis were written to the secondary matrix and then used as categorical variables to assist with an indicator species analysis.
22 W6 6 W11 11
#
West
W88
# #
East er Middle Riv
W11
# #
#
#
13 W13 2W2
M
W12 12
ai
n
7W7
#
W9 14 9 W14 #
4W4 15 W15 5W5 16 3W3 W16 #
Little Back Rive r
Main S avanna hR
iver
#
W10 10
Sa va nn ah #
Ri ve
r
#
#
#
#
West
1:10,000
East
NW4
#
NW3
NE2
NE3 #
# #
# #
NW1
#
#
NW2 NE4
NE1
#
SW4 Legend
Associated with tidal creeks Proximal to main channel or large distributary
ererr Riviivve ckkRR B Baaackc ttelleeB Litiittl LL
Distant from drainages
#
SE3
#
#
SW2
# #
SW3
SW1
SE2
#
SE4
#
SE1
1:10,000
Figure 2-2. Locations of plots and the a priori group they were placed in. Indicator species analysis To assist with pruning of the cluster dendrogram, several indicator species analyses were performed. The general procedure is based on Dufrene and Legendre’s (1997) method. The groups to which each plot belonged, computed from the cluster analysis, were used as categorical variables in which to compute relative abundance and relative frequency for each indicator species analysis. A requisite of this analysis is that each group must be comprised of at least two or more plots, therefore the maximum number of
23 groups that could be analyzed with data from this study was eight. Logically, the minimum number of groups was two, since placing all plots in one group leaves nothing to compare and contrast. It follows that a total of seven separate indicator species analyses were performed, ranging from 2 to 8 groups. The analyses are based upon values for each species (j) as it pertains to that group of plots (k): the relative abundance (RAjk) of a species in a group of plots; the relative frequency (RFjk) of a species in a group of plots; and the indicator value of each species to each group of plots, which is expressed as the percentage 100*(RAjk X RFjk). The indicator values range from 0 (no indicator) to 100 (perfect indicator) with a perfect indicator being faithful (always present) and exclusive to all plots in that group. The largest indicator value for a given species across all groups is recorded as the indicator value for that species (see Tables 3-1 and 3-2). A Monte Carlo test using 1000 randomized runs was then used to evaluate the statistical significance of the maximum indicator value for given species across all groups. The probability of type I error (i.e. the p-value) is the proportion of times, based on 1000 randomized runs, that the maximum indicator value from the randomized data set equals or exceeds the maximum indicator value from the actual data set. The null hypothesis being tested states that the maximum indicator value is no larger than would be expected by chance (the indicator value for the species would be 0), and there is no difference between groups (McCune and Mefford 1999). Statistical significance implies that the species is occurring at a significantly higher abundance and frequency than would be encountered by random chance.
24 Each of the seven analyses resulted in different p-values for species as indicators for a given cluster. The p-values were then summed across all species for each of the analyses, and used as a guide for choosing the optimum number of clusters (i.e., pruning of the cluster dendrogram). Once the optimum number of groups was determined, all groupings from the cluster analysis were removed from the secondary matrix except the optimal one. Multi-response permutation procedures Multi-response permutation procedures (MRPP) was chosen to test the hypothesis of no difference between groups. This nonparametric method was deemed more appropriate to the community analyses than its parametric equivalent, discriminant analysis and multivariate analysis of variance (MANOVA). MRPP supplements the indicator species analysis; where the indicator species analysis describes how well each species separates among the groups, the MRPP provides a test statistic (T) and its associated p-value, as well as a chance-corrected with-in group agreement (A) value (McCune and Grace 2002) for describing group differences. A-values range from 0 to 1, and are indicative of the amount of homogeneity that plots within groups have compared to what would be expected by chance (0). In this way, the A-value is representative of effect size (McCune and Grace 2002). For community analyses, higher A-values (those approaching 0.3) indicate that plots of the same group are not only significantly different, as indicated by the p-value, but are composed of similar species. For the freshwater tidal forest community data, MRPP methods were used to test the difference between forest stands (East vs. West), a-priori landscape grouping, and groups defined by the cluster analysis.
25 Nonmetric multidimensional scaling ordinations Indirect gradient analysis using nonmetric multidimensional scaling (NMS) is a method for assessing dimensionality and ordination that is designed to deal with scenarios inherent to this study. Specifically, NMS was chosen because it is best suited for imbalanced designs, non-normal data, and relationships that are non-linear. The software package PC-ORD was used to perform NMS ordinations based on Sorensen distances calculated from the primary matrix. The first NMS run utilized the autopilot mode in order to determine the appropriate number of axes to interpret, as well as determining correlations between the primary matrix and all environmental variables. A random number was generated for the starting configuration during this particular ordination. While in autopilot mode, the software package recommends dimensionality by comparing stress values among the best solutions for each of the 6 dimensional possibilities it investigates. Once the optimal dimensionality is determined, the autopilot mode does a final run with the appropriate dimensionality. While viewing ordination graphs, biplots of variables in the secondary matrix overlaid onto the ordinations of plots in species space, and correlations of the environmental variables to the axes can be output. By analyzing these correlations, insignificant environmental variables can be identified and removed (see insignificant environmental variables, above), thereby making interpretation easier. Subsequent NMS ordinations were run using data from the primary matrix, in conjunction with the secondary matrix containing only important environmental variables. These ordinations used a random starting configuration and were restricted to the 3-D dimensionality determined by autopilot, with 100 runs using real data. The Monte Carlo test used 100 runs of randomized data.
26 Classification and Regression Tree The statistical program S-Plus 2000 Professional Release 3 (Mathsoft 2000) with the TreesPlus add-in (De’ath 2002) was used classify plots into communities (clusters) by using only the soil properties data (i.e., without species data). The classification and multivariate regression tree approach was chosen as the final step in choosing how many communities (clusters) to describe due to its predictive and descriptive ability to model community composition with environmental correlates. It was also chosen for its ability to handle interactions (correlations) among variables because only the single best predictor is selected at each branch, while different predictors are still free to be selected at other branches of the tree (Urban 2002). Environmental variables are first rank-transformed. Recursive splitting of the data minimizing the amount of within-partition heterogeneity for each side of the split is then performed. After growing a tree of n-1 leaves (where n = the number of plots), the appropriate number of leaves was chosen using the 1-SE method (Therneau and Atkinson 1997) based on cross-validation. The model took the form: Y = X1 + X2 + X3 + X4 + X5 + X6 +X7 + X8 +X9 + X10 where Y = cluster X1 = landscape position X2 = organic matter content X3 = bulk density X4 = Ph X5 = phosphorous concentration X6 = calcium concentration X7 = magnesium concentration X8 = copper present X9 = electrical conductivity X10 = sodium concentration.
27 After models had been run for each scenario (number of clusters), the crossvalidation standard errors were compared. Interpretability of each tree was also assessed based on whether the tree gave a good representation of the corresponding number of clusters.
CHAPTER 3 RESULTS Introduction Statistical approaches were used to determine how many freshwater tidal forest communities exist in the 2 stands sampled in the Savannah River floodplain, as well as an aid in describing them. Plots were first agglomerated based on their relative species compositions by using a cluster analysis, which was followed by indicator species analyses for various numbers of groups (i.e., clusters). Multi-response permutation procedures (MRPP) was used to test for differences in community makeup for various numbers of groups, differences in a-priori designation of a plot based upon general landscape position, as well as broad-scale site differences. Nonmetric multidimensional scaling (NMS) was used to determine trends in soil characteristics (through biplots overlays) and species importance values as they relate to individual plots. As a final step, or cross-validation step, in determining the appropriate number of communities, classification and regression trees (CART) analysis was used to recreate communities based solely on environmental parameters. Once the appropriate number of communities was determined, they were then named based upon their respective indicator species, as determined by significantly high relative abundance and relative frequency of a species in each community.
28
29 Exploratory Data Analyses Cluster Analysis Cluster analysis was one of the many tools used to determine that four communities comprise the tidal forest of the Savannah River floodplain. This analysis alone is minimally informational. However, it is perhaps the single-most useful step in determining and describing community compositions in a multi-step process and is the first step in most statistical analyses of community makeup. The clustering routine agglomerated sampling plots based upon the relative species makeup and Sorensen distances computed from the primary matrix, resulting in a dendrogram with only 3.52% chaining using the flexible beta linkage method. The resultant dendrogram (Figure 3-1) depicts plot associations for all levels of grouping. The dendrogram was pruned at the point where 50% of the information was remaining; this pruning is the key step in determining how many communities exist. As noted previously, an entire suite of statistical analyses were carried out on several groupings to determine where to prune. The starting point for each of the routines was determining the group membership based upon this cluster analysis. An option in PC-ORD v. 4.27 (McCune and Mefford 1999) allows each plot to be color coded according to some grouping variable (in the secondary matrix). Color coding in Figure 3-1 shows how landscape position (Figure 2-2) can be used as an arbitrary guide to assessing community makeup, even when the site has never been visited. The plots labeled in black are relatively distant from creeks and drainages, and comprise practically all plots in the “Shrub” community. Similarly, though not as strong a relationship, the plots labeled in grey are proximal to either the main Savannah River or a large distributary; they comprise over half of the plots in the “Water Oak – Swamp bay”
30 community. Community designations will be described in detail in the following sections. Indicator Species Analysis An example of one of the seven Monte Carlo runs of the indicator species analysis is presented in Table 3-1. P-values were summed across all species for each of the seven analyses. The lowest total p-values were 0.1667 and 0.1657, found in cluster sizes 4 and 2, respectively (Figure 3-2). The number of significant indicator species (α < 0.05) for each analysis (cluster size) were also tallied and used as an aid for choosing the appropriate cluster size (Figure 3-2). With cluster sizes of 5 or more the average p-value increases sharply, while the number of significant indicators drops dramatically, indicating that 5 or more distinct communities probably do not exist in the freshwater tidal forests of the Savannah River floodplain. Although cluster sizes 3 and 2 resulted in the highest number of significant indicator species and had low total p-values, the cluster size of 4 was chosen due to the fact that it has a very low total p-value, a high number of significant indicators, and still allows detailed interpretation in further analyses. Later analyses, including NMS ordinations and CART, further supported the choice of 4 clusters (i.e., communities) (following sections). Monte Carlo results from testing the significance of no difference in species indicator value [(RAjk* RFjk) * 100] between groups based on 4 clusters and 1000 runs of randomized data are presented in Table 3-1. Nine significant indicators were identified: tag alder (Alnus serrulata; ALSE), dahoon holly (Ilex cassine; ILCA), virginia willow (Itea virginica; ITEA), fetterbush (Leucothoe racemosa;LERA), wax myrtle (Myrica cerifera;
MYCE), water tupelo (NYAQ), swamp tupelo (NYBI), swamp bay (Persea palustris; PEPA), and water oak (Quercus nigra; QUNI).
Distance (Objective Function) 8E-03
6.7E-01
100
75
1.3E+00
2E+00
2.7E+00
25
0
Information Remaining (%)
Shrub
Water Tupelo Swamp tupelo – Tag Alder
31
NE1 NE4 NE2 NW1 NE3 SE3 NW2 SE2 SE4 SE1 NW4 W11 W16 W7 W5 NW3 SW4 SW1 W4 W12 W2 W10 SW2 SW3 W1 W15 W8 W14 W3 W6 W9
50
Water Oak – Swamp bay
Landscape Position proximal to main channel/distributary associated with creeks/drainages distant from creeks/drainages
Figure 3-1. Cluster dendrogram based on results of cluster analysis on matrix of 31 plots X 20 species. Plot names on the left correspond to those depicted in figures 1-1 and 2-2. Landscape position of each plot corresponds to those depicted in figure 2-2. Pruning of the dendrogram is indicated by the /, and community names are given for each of the four groups based upon indicator species analysis.
32
3.0
3.5
9
4.0 8
Total p
Number of Significant Indicator Species
10
4.5 7
5.0 6 8
7
6
5
4
3
2
Number of Clusters Total p # Indicators
Figure 3-2. Summary of the 7 indicator species analyses. P-values are based on Monte Carlo randomization, then averaged over all species for each cluster size (x axis) (see table 3-1). Blue circles denote cluster sizes with lowest average pvalues. Table 3-1. Monte Carlo results of species indicator value (IV) between the 4 groups. See Appendix B for species abbreviations. IV from randomized groups Species Observed IV Mean Std Dev p ACRU ALSE CEOC FRAX ILCA ITEA LERA MYCE NYAQ NYBI TADI VINU ILDE LYLU PEPA VIDE LIST QUNI COFF CACA
27.7 41.6 14.3 36.2 80.2 54.5 82.1 79.3 47.9 44.3 34.0 21.4 21.2 36.4 56.8 24.7 27.6 52.5 38.3 24.7
32.0 30.9 15.5 32.8 24.5 18.3 25.3 26.2 29.6 32.9 32.0 15.7 17.4 17.0 29.3 16.7 26.9 30.6 21.7 21.9
4.71 4.71 8.47 3.97 9.22 9.15 7.98 7.81 5.60 4.74 6.83 8.85 9.26 9.52 9.21 8.80 9.64 7.21 10.64 9.64
0.822 0.013 0.466 0.194 0.001 0.008 0.001 0.001 0.002 0.021 0.328 0.172 0.242 0.072 0.013 0.201 0.379 0.011 0.069 0.318
* * * * * * *
* *
33 After determining that 4 groups were going to be interpreted (through all methods of analyses), the indicator species analysis was used to determine which species are indicative of each of the 4 clusters (communities). Further, these species were used to provide names to the communities (see Figure 3-1). Significant indicator species pertaining to clusters are presented in Table 3-2. Cluster 1 has the following significant indicators: dahoon holly (ILCA), virginia willow (ITEA), fetterbush (LERA), and wax myrtle (MYCE); Cluster 2 has water tupelo (NYAQ) as an indicator; Cluster 3 has swamp tupelo (NYBI) and tag alder (ALSE) as indicators; and Cluster 4 has water oak (QUNI) and swamp bay (PEPA) as significant indicators. Table 3-2. Indicator values for species in each of 4 clusters. Numbers in parentheses indicate number of plots included in each cluster (community). Colored fields correspond to significant indicators (see table 3-1). Cluster Species 1 (11) 2 (4) 3 (9) 4 (7) Average ACRU ALSE CEOC FRAX ILCA ITEA LERA MYCE NYAQ NYBI TADI VINU ILDE LYLU PEPA VIDE LIST QUNI COFF CACA averages:
26 22 14 15 80 55 82 79 38 3 14 21 1 36 0 25 13 0 1 0 26
9 33 0 35 1 0 0 2 48 10 5 0 0 0 1 0 0 21 4 2 9
27 42 5 14 1 0 5 4 1 44 34 2 4 0 25 4 13 22 38 18 15
28 1 0 36 0 0 0 0 9 42 24 0 21 0 57 0 28 53 1 25 16
22 25 5 25 20 14 22 21 24 25 19 6 6 9 21 7 13 24 11 11 17
34 Multi-response Permutation Procedures Testing for differences between groups was accomplished using multi-response permutation procedures (MRPP) on the primary matrix, with groups defined by categorical variables in the secondary matrix. For each test, a p-value and an A-value are reported. The p-value reported corresponds to the hypothesis of no difference between groups. When statistically significant differences were found between groups, multiple comparisons were done for further investigation. The A-value is the chance-corrected within-group agreement (see chapter 2). A=0 when heterogeneity (species importance values) within groups is what would be expected by random chance. As A approaches 1, the homogeneity within a group is maximized and importance values for individual species are identical for each of the plots within the group. Note, however, that importance values do not need to be identical for all species within a plot. In this case of maximum homogeneity the corresponding δ value is equal to zero. Groups were based upon the following three criteria: 1.
Stand. Two groups of plots were made based upon their broad-scale placement within the landscape. The eastern stand, composed of 16 plots, and the western stand, composed of 15 plots, are located off the Little Back River and the main channel of the Savannah River, respectively (Figure 1-1).
2.
Floodplain physiography. Three groups were made based upon their proximity to the river, as well as the size of the channel supplying tidewater (see a priori grouping, chapter 2). The 3 groups consist of plots that are: a. Proximal to the main river channel or a large distributary (n=4). b. Associated with tidal creeks and drainages (n=17). c. Backswamp sites relatively far removed (distant) from tidal creeks and drainages and, therefore, from the main channels (n=10).
3.
Cluster. The 4 groups of plots based on the cluster analysis. Groups were simply given numbers as identifiers. Sizes of each group are listed in table 3-2. Although
35 only results from the 4-cluster analysis are presented, MRPP was performed for clusters of size 2 and 3. Results showed that all groupings are statistically different based upon the MRPP analyses. Further, all multiple comparisons are also statistically significant (Table 3-3). Table 3-3. MRPP results for groups of plots. General Mulitiple Criteria comparisons Stand
p-value
A-value
0.00042660 0.11826623
Region river vs creek river vs distant creek vs distant
0.00000142 0.01100517 0.00005802 0.00004074
0.23889437 0.06810518 0.37853775 0.19628380
1 vs 2 1 vs 3 1 vs 4 2 vs 3 2 vs 4 3 vs 4
0.00000000 0.00018304 0.00000747 0.00001505 0.00010362 0.00088010 0.00014051
0.40428521 0.17175219 0.38497760 0.38976488 0.30341608 0.31202930 0.14952343
Cluster
NMS Ordinations Autopilot The NMS autopilot run with the primary matrix and all environmental data indicated that a 3-dimentional solution was optimal. The probability that a similar final stress could have been obtained by chance (i.e., the Monte Carlo p-value) is 0.0196 for the 3-D solution based upon 50 runs with randomized data. The final ordination for the autopilot mode completed 82 iterations while analyzing the 3-D solution, resulting in a stress value of 6.62099. This is well within the acceptable range (Kruskal 1964), especially when statistics of this sort are applied to ecological community data (McCune and Grace 2002). Rotating the ordination graph allows correlations to be seen between the environmental data (secondary matrix variables) and plot-species ordination via biplot
36 overlays. One such rotation allowed for all significant environmental variables (r2 ≥0.392) to be seen on one graph. Although other rotations change strengths of environmental correlations to the axes (by changing species-plot placement and, therefore, how the biplots are oriented along axes), no rotations resulted in correlations ≥0.392 for any “insignificant” variables. Therefore, the rotation showing high correlations (Table 3-4) for all significant variables was used as the basis for removal of insignificant environmental variables (Table 2-2). Table 3-4. Pearson’s coefficients of determination (r2) and Kendal’s tau values of environmental variables to axes for NMS ordination using autopilot mode. These correlations were used as the basis for removal of insignificant (r2≤0.392) variables. Axis 1 r tau 0.248 0.295 0.096 -0.371 0.012 -0.019 0.170 0.277 0.041 -0.170 0.051 0.140 0.048 -0.194 0.192 0.258 0.012 -0.101 0.256 0.277 0.005 0.135 0.061 0.153 0.047 -0.245 0.013 -0.138 0.081 -0.308 0.114 -0.241 0.139 -0.269 0.058 0.187 0.030 -0.183 0.324 0.342 0.362 0.385 0.051 0.187 2
Organic matter Bulk density pH P concentration P present K concentration K present Ca concentration Ca present Mg concentration Mg present Zn concentration Zn present Mn concentration Mn present Cu concentration Cu present Fe concentration Fe present Electrical conductivity Na concentration Na present
Axis 2 r tau 0.651 -0.643 0.395 0.572 0.205 0.313 0.037 -0.092 0.448 0.535 0.035 -0.196 0.332 0.435 0.404 -0.465 0.161 0.260 0.471 -0.514 0.014 -0.105 0.140 -0.209 0.276 0.366 0.007 0.112 0.280 0.424 0.103 0.220 0.392 0.550 0.068 -0.148 0.280 0.402 0.578 -0.563 0.572 -0.527 0.074 -0.200 2
Axis 3 2
r 0.071 0.002 0.075 0.019 0.019 0.001 0.001 0.065 0.043 0.083 0.089 0.004 0.003 0.087 0.023 0.195 0.020 0.017 0.046 0.055 0.087 0.074
tau 0.217 -0.138 -0.145 -0.084 -0.187 0.071 -0.069 0.129 0.071 0.148 0.144 0.032 -0.108 -0.022 -0.092 -0.314 -0.231 -0.140 -0.226 0.139 0.084 0.213
37 Subsequent ordinations Additional ordinations were run to fit a 3-D solution based on Sorensen distances computed from data in the primary matrix with overlays of only the 8 important soil properties. Monte Carlo results based on 100 runs of randomized data give a p-value = 0.0099. Ninety three iterations were used in the final solution, resulting in a final stress of 6.62099 and a final instability of 0.00001. Axes 2 and 3 represent the largest proportion of variance explained by the ordinations (Table 3-5), and the plots separate into relatively concise groups of similar communities (clusters) when viewing these axes. Therefore, most ordinations that follow will show Axes 2 and 3 (See Appendix-E for correlations of species and soil constituents to Axes 2 and 3). It follows that ordinations portraying species’ importance in plots, as well as overlays of variables from the secondary matrix can be interpreted easiest when viewing these 2 axes. It is important to realize that, when viewing ordination graphs, the axes are not a single variable, nor are they necessarily a summation of variables that have been measured. Rather, they are best thought of as a synthesis of variables, both measured and not measured, representing the relative differences between the plots that have been sampled. In these ordination graphs large symbols correspond to higher importance values for a given species as well as larger values for soil constituents. Table 3-5. Proportion of variance represented by axes based on the r2 distance between distance in the NMS ordination space and distance in the original space. r2 Axis Increment Cumulative 1 0.025 0.025 2 0.407 0.433 3 0.527 0.960
38 Indicator species importance within plots. A graphical representation of the ordinations is useful for ease of depicting how the plots separate out in species space, as well as perceiving relative importance of select species within plots. The indicator species for Cluster 1 are represented in Figures 3-3 and 3-4. Dahoon holly and Virginia willow are most important in plots of Cluster 1, with very little representation in other plots (Figure 3-3). Likewise, fetterbush and wax myrtle are also highly important in plots of Cluster 1, but they are also represented a bit more in plots of Cluster 3 (Figure 3-4). Water tupelo is well represented in Clusters 1 and 2 (Figure 3-5) even though statistical tests indicate that it is a significant indicator for Cluster 2 only. The plots of Cluster 3 indicate swamp tupelo as one of the significant indicators, yet it is clear that it is also relatively abundant in plots of Cluster 4 (Figure 3-6A). Similarly, tag alder (Figure 3-6B) is an indicator species for plots in Cluster 3. Tag alder is also the most prevalent of all species encountered in this study, but only reaches maximum importance in plots of Cluster 3. Water oak (Figure 3-7A) is an indicator for the plots of Cluster 4, yet the ordination shows water oak as being most abundant in W10, a member of Cluster 3. Likewise, W10 has the largest importance value for swamp bay (Figure 3-7B), yet swamp bay is an indicator for plots in Cluster 4.
39 Cluster 1 2 3 4
NW4 SE4 SE1 NW2 SE2 NW1
W5 W7
W11 SE3
NE2
NE4 NE1
W16
W6
NE3
Axis 3
W14
W9
W3
W10
NW3
W15 W1
W12
SW1
W2 SW2 SW3 W4
W8
SW4
Axis 2
A
W7
W11
NE2
NE4 NE1
NE3 W14
Axis 3
SE3
W16
W6
Cluster 1 2 3 4
NW4 SE4 SE1 NW2 SE2 NW1
W5
W9
W3
W10
NW3
W15 W1
W12 W8
SW1
W2 SW2 SW3 W4
SW4
Axis 2
B Figure 3-3. NMS ordination. Relative importance values of dahoon holly (A) and Virginia willow (B). Note the size of the triangle depicts the importance of species in plots.
40
W5
SE4
SE1 NW2 SE2 NW1
W7
W11
SE3
NE4
NE2
NE1
W16
W6
NE3 W14
Axis 3
Cluster 1 2 3 4
NW4
W9
W3
W10
NW3
W15 W1
W12
SW1
W2 SW2 SW3 W4
W8
SW4
Axis 2
A
W7
SE2
W11
NE2
NE4 NE1 NE3
W14
Axis 3
SE3
NW1
W16
W6
Cluster 1 2 3 4
NW4 SE4 SE1 NW2
W5
W9
W3
NW3
W10 W15 W1
W12 W8
SW1
W2 SW2 SW3 W4
SW4
Axis 2
B Figure 3-4. NMS ordination. Relative importance values of fetterbush (A) and wax myrtle (B), as indicated by the size of triangles.
41
W7
SE2
W11
W14
Axis 3
SE3
W16
W6
Cluster 1 2 3 4
NW4 SE4 SE1 NW2
W5
NW1 NE2
NE4 NE3
NE1
W9
W3
W10
NW3
W15 W1
W12 W8
SW1
W2 SW2 SW3 W4
SW4
Axis 2
Figure 3-5. NMS ordination. Relative importance values of water tupelo, as indicated by the size of triangles.
42
W5
Cluster 1 2 3 4
NW4
SE4
SE2 SE1 NW2
W7
NW1 W11 W16
SE3
NE2
W6
Axis 3
W14
NE4 NE3
NE1
W9
W3
W10 NW3 W1
W15 W12
SW1 SW2 W2
W8
SW3
SW4
W4
Axis 2
A W5
SE4 SE2 W7
SE1
W11 SE3
Axis 3
NW2 NW1 NE2
NE4 NE3
W16
W6
Cluster 1 2 3 4
NW4
NE1
W9
W14
W3 W10 NW3 W1
W15 W12 W8
SW1
SW2 W2
SW3
SW4
W4
Axis 2
B Figure 3-6. NMS ordination. Relative importance values of swamp tupelo (A) and tag alder (B), as indicated by the size of triangles.
43
W5
Cluster 1 2 3 4
NW4
SE4 SE2 W7
SE1
NW2 NW1
W11 W16
W6
NE4 NE3
NE1
W9
W14
Axis 3
NE2
SE3
W3
W10 NW3 W15 W1
W12
SW1 SW2
W8
W2
SW4
SW3 W4
Axis 2
A W5 W7
SE1
Cluster 1 2 3 4
NW4
SE4
SE2 NW2 NW1
W11 W16
Axis 3
W6
SE3
NE2
NE4 NE3
NE1
W9
W14
W3
W10 NW3 W15 W1
W12 W8
SW1 SW2 SW3
W2
SW4
W4
Axis 2
B Figure 3-7. NMS ordination. Relative importance values of water oak (A) and swamp bay (B), as indicated by the size of triangles.
44
Soil properties. Although the secondary matrix has no affect on how the distance matrix is calculated for the NMS ordination, it is helpful to see how soil constituents are correlated, both to other constituents, and to plots. NMS graphs of plots in species space with biplot overlays of soil constituent values illustrate how most soil properties are closely correlated to other constituents, in both a positive and negative way (Figure 3-8). Organic matter content, electrical conductivity, concentration of Ca, concentration of Mg, and concentration of Na are all closely correlated with axis 2. At the same time, the values for bulk density, K present, and Cu present are also correlated to axis 2, but negatively correlated to the other soil parameters. With this rotation it is also easy to see that all soil constituents are orthogonal to Axis 1 and, for all practical purposes, constitute the majority of environmental differences associated with Axis 2. Note that lengths of vectors in Figure 3-8 indicate strengths of relationships to plots. Biplots of organic matter content, electrical conductivity, concentration of Ca, concentration of Mg, and concentration of Na are all associated with high values in the plots of Cluster 1. A view at axes 3 and 2 (Figure 3-9) allows the correlations to Axis 2 to be seen again, but not as strict a correlation as in the previous graph (Figure 3-8). Partial correlations to Axis 3 also exist. Viewing these 2 axes leads to a sense that the appropriate number of clusters to interpret may have been 2, as indicated by the larger solid line circles around the plots (Figure 3-9). This plot is an excellent demonstration of the 2 vs. 4 groups distinction that was indicated in the indicator species analysis (Figure 3-2). There are essentially 2 broader groups that are comprised of 2 subgroups each that can be distinguished as significantly different in the MRPP analysis (Table 3-3). The
45 negative correlation between Axis 2 and bulk density, potassium present, and copper present have little correlation to axis 3. Cluster
Organic Na conc Ec
1 2 3 4
Mg conc
Axis 2
Ca conc
bulk P present Cu present
Axis 1
Figure 3-8. Biplot of axis 2 vs 1. Strong correlations of all soil constituents to axis 2.
46
Cluster 1 2 3 4
NW4
W5
SE4 SE2
NW2
SE1
W7
NW1 W11 W16
SE3
NE4
NE2
NE3
Na conc
W6
Mg conc
NE1
Ec Organic
W14
Ca conc
Axis 3
W9 P present bulk Cu present
W3
W10
NW3
W15 W1
W12 W8
SW1 SW2 W2 SW3
SW4
W4
Axis 2
Figure 3-9. Biplot of axis 3 vs. 2. Correlations of soil constituents are split between axes 2 and 3. Larger circles with solid lines indicate the 2 broad groups while smaller circles with dashed lines encircle the 2 sub-groups. Classification and Regression Tree Analysis The Classification and Regression Tree (CART) analysis (Urban 2002) run in SPlus Release 3 (Mathsoft, Inc. 2000) was used to assist in the decision of how many clusters to interpret, as well as aid in the description of the clusters. To accomplish this, plots were classified into their respective cluster (for 2, 3, and 4 clusters) by using only the soil nutrient values (i.e., without species data). The best fit of the model (Figure 3-10)
47 explained 55% of the variation, which was accomplished when 4 clusters were classified, reinforcing the earlier analyses that 4 different tree communities exist in the study area and further, that these same 4 groups are communities are characterized, even predictable, under given sets of environmental conditions. Cluster 1: Shrub (11)
78%
Cluster 3: Swamp tupelo – Tag alder (9) Cluster 4: Water oak - Swamp bay (7)
Sodium concentration >353 mg/kg
3 dS/m
Shrub (11)