Population Dynamics of Seagrass Communities in Biscayne Bay, Florida: Exploring the Influence of Salinity on Seascape Fragmentation

Transcription

1 University of Miami Scholarly Repository Open Access Theses Electronic Theses and Dissertations Population Dynamics of Seagrass Communities in Biscayne Bay, Florida: Exploring the Influence of Salinity on Seascape Fragmentation Clinton W. Stipek University of Miami, Follow this and additional works at: Recommended Citation Stipek, Clinton W., "Population Dynamics of Seagrass Communities in Biscayne Bay, Florida: Exploring the Influence of Salinity on Seascape Fragmentation" (2018). Open Access Theses This Open access is brought to you for free and open access by the Electronic Theses and Dissertations at Scholarly Repository. It has been accepted for inclusion in Open Access Theses by an authorized administrator of Scholarly Repository. For more information, please contact

2 UNIVERSITY OF MIAMI POPULATION DYNAMICS OF SEAGRASS COMMUNITIES IN BISCAYNE BAY, FLORIDA: EXPLORING THE INFLUENCE OF SALINITY ON SEASCAPE FRAGMENTATION By Clinton William Stipek A THESIS Submitted to the Faculty of the University of Miami in partial fulfillment of the requirements for the degree of Master of Science Coral Gables, Florida December 2018

3 2018 Clinton William Stipek All Rights Reserved

4 UNIVERSITY OF MIAMI A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science POPULATION DYNAMICS OF SEAGRASS COMMUNITIES IN BISCAYNE BAY, FLORIDA: EXPLORING THE INFLUENCE OF SALINITY ON SEASCAPE FRAGMENTATION Clinton William Stipek Approved: Diego Lirman, Ph.D. Associate Professor of Marine Biology and Ecology Elizabeth Babcock, Ph.D. Associate Professor of Marine Biology and Ecology Rolando Santos, Ph.D. Postdoctoral Associate, Florida International University Guillermo Prado, Ph.D. Dean of the Graduate School

5 STIPEK, CLINTON WILLIAM (M.S., Marine Biology and Ecology) Population Dynamics of Seagrass Communities (December 2018) In Biscayne Bay, Florida: Exploring the Influence of Salinity on Seascape Fragmentation Abstract of a thesis at the University of Miami. Thesis supervised by Professor Diego Lirman. No. of pages in text. (49) Seagrass communities display complex population dynamics that are presently poorly understood. While most studies of disturbance on seagrass habitats have focused on changes in biomass at small, quadrat-level scales, limited information is available on the impacts of disturbances on seagrass patch dynamics on a broad scale. In this study, a landscape approach based on remote sensing imagery and population modelling were applied to understand seagrass patch dynamics and forecast the fate of these important communities in Biscayne Bay, Miami, Florida, US. The seagrass communities of Biscayne Bay are especially influenced by the modifications of the Everglades watershed that has significantly changed the timing of freshwater deliveries and nearshore salinity patterns over the last 70 years. This research evaluated the historical influence of salinity on seagrass communities and how changes in salinity may cause seascape fragmentation by documenting the long-term population dynamics of seagrass habitats found adjacent and distant to freshwater canals. A positive outcome of this study is that the historical analysis covering > 70 years revealed remarkable persistence of seagrass cover even in habitats that were modified by the construction of freshwater canals. Fragmentation rates of the seagrass patches varied spatially and temporally but were higher in sites adjacent to canals compared to sites removed from these influences. Furthermore, there was a clear trend in mortality rates in relation to the size of seagrass patches, with the smallest

6 patches undergoing 57% annual mortality on average. Model simulations based on observed patch dynamics suggest that seagrass meadows can persist as long as the recruitment of new patches compensates for patch fragmentation. The combination of higher fragmentation rates and the higher mortality of smaller seagrass patches in habitats exposed to pulses of low salinity raises concern for the long-term persistence of seagrass meadows in nearshore urban habitats of Biscayne Bay that are presently targets of Everglades restoration. The combined remote sensing and population modelling approach used here provide evaluation and predictive tools that can be used by managers to track seagrass status and stress-response at seascape levels not available previously for the seagrasses of South Florida.

7 Acknowledgements Would like to extend thanks to my committee members: Dr. Lirman for his writing help and vision, Dr. Santos for his continual reliance and calmness, and Dr. Babcock for her statistical help and drive. Each one of them has committed time and effort and I am grateful for their guidance. Maria Estevanez, despite not being on my committee, has played an integral role in my pathway and provided continual direction and help. My friends and family have always been there for me through late nights and frustrations. Without any of these advisors, mentors or friends, I would not have been able to complete this project. iii

8 Table of Contents Chapter I. Introduction (pg. 1) Chapter II. Methods (pg. 5) Chapter III. Results (pg. 19) Chapter IV. Discussion (pg. 34) Chapter V. Conclusion (pg. 40) References (pg. 41) iv

9 List of Figures 1. Map of the sites that were sampled during this experiment (pg. 6) 2. Salinity patterns within Biscayne Bay (pg. 7) 3. Hand delineated polygons drawn over the seagrass patches in GIS (pg. 9) 4. Seagrass fracturing (pg. 12) 5. Seagrass merging (pg. 13) 6. Changes displayed by the seagrass patches (pg. 14) 7. Mortality rates and the associated size classes (pg. 19) 8. Annual fragmentation rates across all time steps (pg. 20) 9. Average fragmentation rates by adjacent and distant sites (pg. 20) 10. Linear regression of yearly fragments versus time steps (pg. 21) 11. Linear regression of yearly fragments versus percent cover (pg. 21) 12. Linear regression of recruits versus total patches (pg. 22) 13. Linear regression of recruits versus percent cover (pg. 22) 14. Average Recruitment rate (pg. 22) 15. Linear regression of lambda values versus time steps (pg. 23) 16. High fragmentation, low recruitment projection - proportion (pg. 25) 17. High fragmentation, average recruitment projection - proportion (pg. 25) 18. High fragmentation, high recruitment projection - proportion (pg. 26) 19. Low fragmentation, low recruitment projection - proportion (pg. 27) 20. Low fragmentation, average recruitment projection - proportion (pg. 27) 21. Low fragmentation, high recruitment projection - proportion (pg. 28) 22. Proportional Changes by high/low fragmentation scenario s (pg. 28) 23. High fragmentation, low recruitment projection patch abundance (pg. 29) 24. High fragmentation, average recruitment projection patch abundance (pg. 30) 25. High fragmentation, high recruitment projection patch abundance (pg. 30) 26. Low fragmentation, low recruitment projection patch abundance (pg. 31) 27. Low fragmentation, average recruitment projection patch abundance (pg. 31) 28. Low fragmentation, high recruitment projection- patch abundance (pg. 32) 29. Temporal seagrass cover trends (pg. 33) v

10 List of Tables 1. Sites and the time series for analysis (pg. 7) 2. Size classes and the associated size (pg. 10) 3. Transition matrix utilized in this study (pg. 14) 4. Leslie matrix (pg. 15) 5. Annual lambda values (pg. 23) 6. Average lambda values for adjacent and distant sites (pg. 24) 7. Proportion of population by time steps (pg. 24) 8. High and low fragmentation patch abundance (pg. 32) 9. Percent cover and temporal change (pg. 33) vi

11 List of Equations 1. Transition matrix projection pg. (14) 2. Merging (pg. 15) 3. Fragmentation (pg. 16) 4. Recruitment (pg. 16) vii

12 Chapter I. Introduction Submerged aquatic vegetation communities composed of seagrasses and macroalgae create productive ecosystems in shallow coastal waters around the world (Cuhna & Santos, 2009; Montefalcone, 2009; van der Heide et al., 2012). These communities provide a wide range of essential ecological and economic services valued at US $3.8 trillion per year (Orth et al., 2006; Barbier et al., 2011; Cuvillier et al., 2017). While serving as key habitat to species such as green sea turtles and manatees, seagrass communities also provide important ecosystem services such as carbon sequestration (Hemminga & Duarte, 2000; Duarte et al., 2008; Lirman et al., 2014). Furthermore, they facilitate trophic transfers to nearby habitats, such as salt marshes, mangroves, and coral reefs (Davis et al., 2009; Hyndes et al., 2014). Seagrasses also provide the essential habitat for the growth and development of juvenile and adult populations of pelagic and demersal species such as gag, blue and nassau groupers and queen conch (Gillanders et al., 2003; Mumby et al., 2004; Olds et al., 2012; Williams et al., 2016). Between 1980 and 2006 seagrass communities have been disappearing at a rate of 110 km 2 per year globally and this rate is expected to accelerate (Waycott et al., 2009). Declines in seagrass communities have been especially magnified near populated coastlines and coastal development has severely reduced seagrass abundance (Kemp et al., 1983; Giesen et al., 1990; Duarte, 2002; Lotze et al., 2006; Ondiviela et al., 2014). One example of nearshore modifications impacting seagrass communities can be found in Florida Bay, Florida. From and again in 2015, there was a rapid mass mortality of the seagrass Thalassia testudinum in which > 4000 hectares of dense seagrass beds were lost (Robblee et al., 1991; Zieman et al., 1999; Hall et al., 2016). 1

13 2 Florida Bay is a shallow lagoon that has experienced a significant reduction in freshwater input from the Florida Everglades due to the instillation of a South Florida Water Management canal system (Robblee et al., 1991). The modification of the Everglades watershed has resulted in a reduction in the amount of freshwater reaching the coastal bays, and Florida Bay now commonly experiences hypersalinity on a seasonal basis. While the cause of these die-off s is still not well understood, factors such as extreme salinity, high temperature, nutrient levels, and hypoxia are thought to have had an influence (Smith et al., 1989; Zieman et al., 1999; Hall et al., 2016). Freshwater inputs and salinity patterns are also key drivers of seagrass abundance and distribution in Biscayne Bay, Florida. Biscayne Bay is a shallow coastal lagoon highly influenced by the quantity and timing of freshwater deliveries (Zink et al., 2017). During the late 1800s, anthropogenic influences began to alter the flow of freshwater from the Everglades into Biscayne Bay (Sklar, 2002). From the early 1900s until the 1920s, there were three canals that cut through the Everglades and assisted in the drainage of Lake Okeechobee for agricultural and flood prevention purposes (Herman, 2014). These canals were responsible for the drainage of 1.5 million acres with an average annual removal of water of 1 foot/year (Herman, 2014). In 1948, more canals were constructed which further diverted water from Lake Okeechobee into Biscayne Bay. These canals, completed by 1963 (Herman, 2014) have reduced the amount of freshwater reaching the bay and changed the delivery method. While the historic salinity patterns were dominated by the slow discharge of freshwater through overland flows and groundwater, freshwater is presently discharged into littoral habitats through pulsed releases from canals. This creates environments near canals that experience very drastic

14 3 drops in salinity (reaching 0 ppt in some instances) over a matter of hours. Thus, areas with previously low variation in salinity are now exposed to high variability (Lirman et al., 2008). These changes in the salinity regime have already been linked to changes in the abundance and distribution of seagrasses and associated fauna (Lirman et al., 2008, 2014; Santos et al., 2014; Zink et al., 2017, 2018). The Comprehensive Everglades Restoration Plan (CERP) is presently being implemented to improve the quality and quantity of fresh water delivered into the coastal bays of South Florida, with unknown ecological consequences on the components of the coastal ecosystems (Santos et al., 2014). To document and predict the impacts of CERP, there is a pressing need to develop a hierarchy of models and indicators that evaluate status and trends of key ecosystem indicators like seagrasses at multiple spatial and temporal scales. Historically, the impacts of human and natural disturbances on seagrass meadows have been commonly characterized at small, quadrat-level scales, with limited attention paid to the influence of the disturbance on seagrass seascape dynamics (Vidondo et al., 1997; Abadie et al., 2018). With the documentation of widespread declines and reports of localized mass seagrass mortality, there is an increasing need to evaluate response patterns at scales beyond the quadrat level (Vidondo et al., 1997; Hall et al., 2016). Seagrass patches vary widely in size, from < 1 m 2 to hectares of continuous seagrass cover. Because of their clear boundaries (seagrass patches are commonly surrounded by sediments or rubble), seagrass patches are ideal candidates for studies of patch dynamics at seascape scales. This study examined the long-term dynamics of seagrass beds in Biscayne Bay, Florida, by analyzing the historical response of seagrass patches of different sizes within

15 4 distinct salinity environments over > 70 years to evaluate the long-term impacts of seascape fragmentation on ecosystem resilience. We hypothesized that the rates of fragmentation and the long-term population dynamics of seagrass habitats were influenced by the discharge of freshwater from canals, with areas closer to canals having higher fragmentation and patch extinction rates, compared to areas isolated from these influences. These patterns of fragmentation will likely have a major influence on the long-term persistence of seagrass meadows in Biscayne Bay.

16 Chapter II. Methods Study Area This project expanded prior seascape studies completed by Santos et al. (2014). The hydrology into Biscayne Bay has been modified since the 1900s by construction of canals that has altered the delivery, quantity and quality of freshwater (Herman, 2014). The hydrologic salinity patterns are now dominated by canals as the main method of freshwater rather than the historic method of sheetflow (Santos et al., 2011). Areas near the canals are exposed to extreme oscillations in salinity levels and this pattern is heightened during the wet season (July to October) when the freshwater can be released in pulses to help alleviate flooding (Santos et al., 2011). Study Design Six sites were selected along the western shoreline of Biscayne Bay, Florida where the impacts of CERP are concentrated (Figure 1). The sites considered adjacent were located in proximity to canals with the highest freshwater discharge rates within Biscayne Bay. These sites were: Snapper Creek (SC), Black Point Canal (BP), and Convoy Point (CP) which had a mean distance of 0.54 (S.D.= ± 0.10) km from the canals (Santos et al., 2014). Paired distant sites were randomly selected at distances > 1 km 2 from a canal (Santos et al., 2014). These sites were: Chicken Key (CK), Black Point Lagoon (BL), and Turkey Point (TP) which had a mean distance of 2.77 (± 0.94) km from the canals. Each site encompassed a 500-m buffer around a location selected along the shoreline as described by Santos et al. (2014). Salinity data was obtained for to evaluate differences in salinity between adjacent and distant sites (Figure 2). Once these sites were selected, historical aerial photos were assessed over nine historical 5

17 6 periods, 6-13 years apart from based on the availability and quality of aerial imagery (Table 1). Figure 1. Location of study sites for this project. The sites with a green outline are distant from canals (Black Point Lagoon, Chicken Key and Turkey Point) and the sites with a blue outline (Black Point Canal, Convoy Point and Snapper Creek) are adjacent to canals.

18 7 Figure 2. Salinity patterns at the survey sites between Displaying the average (± SD), maximum and minimum values for salinity (ppt). Table 1. Time steps for aerial imagery available for this study. Time Step BL BP CK CP SC TP X X X X X X X X X X N/A X X X X X X N/A X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X In the study by Santos et al. (2014), seagrass maps were created by using blackand-white aerial photographs obtained from various local agencies. These images were processed to standardize the resolution, optical properties, and the area of sampling. Images were geo-rectified using a United States Geological Survey topographic map as a spatial reference. The resolution of all aerial photographs was then re-sampled to 1-m pixel size, and a histogram equalization and convolution filtering technique was applied to control for the contrast and textural optical variability among years and sites. The majority of the aerial photographs had 1-m pixel size, but the most recent years ( ) had 0.35-m pixel size. This caused the more recent years to be re-sampled to 1-m pixel size for standardization of the mapping procedure. The 1-m resolution provided

19 8 adequate pixel size to delineate the smaller patches observed. Re-sampling the aerial photographs helped smooth the image and reduce the noise and salt-pepper effects (Santos et al., 2014). Seagrass maps were originally created by hand-digitizing and delineating individual seagrass patches. This process was standardized by setting all photographs to a 50% contrast level and a 1:2500 scale, with a minimum mapping unit of 20 m 2. These seagrass patches were then manually digitized because the optical properties varied significantly within, and among, photographs (Cuvillier et al., 2017). The contrast level further helped to interpret the patches with dense coverage (>50% cover) which facilitated the digitization process and reduced misclassification. There were three photograph interpreters who were trained to follow a set of digitization rules to limit variability among observers/interpreters. The lead interpreter also analyzed all of the preliminary maps for quality assessment. The current study enhanced the post-processing analysis of the data presented above to study the population dynamics of seagrass communities within Biscayne Bay. ArcGIS PRO was utilized for further examination of the maps produced by Santos et al. (2014). Polygons were drawn over the manually digitized seagrass patches and distinguished into five different size classes based on their area (Figure 3; Table 2). Population models based on size rather than age are particularly useful in describing the dynamics of clonal invertebrates and plants (Cuff & Hardman, 1980; Hughes, 1984; McMurray et al., 2010).

20 Figure 3. Polygons delineated over the seagrass patches from the aerial photographs. 9

21 10 Table 2. Seagrass patch sizes used in this study to track population dynamics. Size Class Area (m 2 ) m m m m 2 5 >2000 m 2 Determination of Patch Fate The fate of the seagrass patches selected per size class were followed from one time period to the next (t0-t1). Five patches from each size class were randomly selected for each site every time interval, as this was the highest number of patches, on average, that would provide equal representation across all five size categories. To evaluate the fate of each patch between time steps, the GIS map for t1 was juxtaposed on top of the map for t0 with different colors for the associated patches of each time step. Only two time steps were overlaid at one time to improve clarity and effectiveness when analyzing the selected polygons. The fate of each of the five selected polygons per size class was recorded as either growth, shrinkage, fragmentation, merging, or extinction using the following rules: 1. For a fate to be classified as growth, a patch from the next time period (t1) had to be in contact with only the original patch and show a greater area. The change in area and size class was recorded for each patch that showed growth. 2. Shrinkage was recorded in the opposite way of growth; the t1 patch had to show a decrease in area and be the only patch in contact with the t0 patch. 3. Fragmentation was recorded if the original patch in t0 divided into more than one new patch in t1. If a patch exhibited fragmentation, the number of new patches and their associated sizes were recorded (Figure 4). Some of these t1 patches were also in contact

22 11 with other patches from the t0 step and were thus associated with two t0 patches. For a t1 patch to be associated with the selected t0 patch for analysis, 50% of its area had to be in the parent patch. If the t1 patch was not in contact with any patches other than the parent, then it did not matter how much of the area was within the t0 patch. 4. Merging was recorded if the t0 patch formed a patch in the t1 period with one, or more patches from the t0 period (Figure 5). Similarly to fragmentation, there were some instances when a patch from t0 was associated with more than one patch in t1. Again, the parent patch association was applied and if the t0 patch that merged into the t1 patch was associated with another t1 patch and contributed less than 50% of its area to the parent patch, it was not included in the analysis. 5. Extinction (mortality) was recorded when the original patch disappeared between t0 and t1.

23 Figure 4. Example of fragmentation patterns documented between 1973 (t0) and 1985 (t1) for TP. None of the fragmented polygons were in contact with any other polygons from the t0 population. 12

24 Figure 5. Example of patch merging between 1973 (t0) and 1985 (t1) for TP. All polygons that merged were only in contact with one t1 polygon. 13

25 14 Tracking Changes The data were used to develop a transition matrix with the fates of the selected patches recorded as proportions (Table 3; Figure 6) (Caswell, 1989). The matrix model used here can be described as: N(t+1) = A N(t) (Equation 1) where A is a Leslie matrix describing the probabilities of transition between size classes and N(t) is the population vector that describes the number of individuals in each size category at time t (Hughes & Tanner, 2000). This model accounts for recruitment by adding a recruitment vector (R) as a proportion of the new patches in t1 to the first row in the matrix (Table 4) (Hughes & Tanner, 2000; Stubben & Milligan, 2007). Table 3. Visualization of the Leslie matrix which defines the contribution of patches in one size class at time t0 to patches in each size class at time t1. Some proportion of the patches in a size class patch may grow (G), shrink (S) or stay constant (C). R is recruitment. t 0-t C +R S + R S + R S + R S + R 2 G C S S S 3 G G C S S 4 G G G C S 5 G G G G C Figure 6. Fates of seagrass patches over time. Patches can grow (green arrows), shrink (red arrows), or stay in the same size category (black arrows). Complete mortality and recruitment are not captured in this diagram but are incorporated into the model.

26 15 Table 4. Example of a Leslie matrix with the probabilities of the size classes growing or shrinking to a larger or smaller size class between intervals recorded in this study, along with recruitment as 0.02 times the number of patches. t 0-t Growth or shrinkage contributed a proportion to the larger or smaller size class. If two of the five size 1 patches in t0 grew to size 3 in t1, 0.4 would be recorded as growth from size 1 to size 3 (Table 4). If three of the size 3 class patches from t0 shrunk to size 1 in t1, 0.6 would be recorded as shrinkage. If there were only three t0 size 5 patches and one of them shrunk to a size 4 in t1, it would contribute 0.33 as shrinkage. Merging was distinguished as a proportion of the patches that came together to form a new patch in t1 with x representing the number of patches that merged to form the t1 patch and y as the number of patches within the size class being analyzed (Equation 2). If one of five selected size 2 patches in t0 merged with three other patches to form a size 5 patch in t1, the proportion it contributed to the size 5 category would be (1/4) * (1/5). For fragmentation, the number of patches that were created from the initial patch were recorded as a proportion of the initial population size with z representing the number of fragments of that respective size class being produced and w being the number of patches within the size class being analyzed (Equation 3). Therefore, if there were three initial size 5 patches in t0 which generated six new size 3 patches in t1, the proportion of size 5 patches shrinking to size 3 patches was 2. To standardize fragmentation for comparisons over time, they were put into a yearly value by dividing the total number of fragments observed by the respective number of years in the time step. For example, if there 60 total fragments from the time period , there would be a rate of 6 fragments a year. Merging = (1/x)(1/y) (Equation 2)

27 16 Fragmentation = (z/w) (Equation 3) To evaluate the influence of size and salinity environment (i.e., adjacent vs distant) on the patch mortality rates, a two-way ANOVA was used, where the response variable, mortality, was the fraction of the five patches in each site, time step and size class that went extinct. Recruitment The recruitment values were added into transition probabilities from each larger size class to size 1 (Table 3). To estimate recruitment, the t1 size patches that were not in contact with any patches from t0 were identified and counted as recruits. This definition is supported by Campbell (2003) who stated that recruitment was introduced into the meadow as pieces of unattached rhizome that settled on the seafloor, implying that new patches were not in spatial contact with any prior patches. The recruits identified were divided by the total number of patches from the t0 time step to obtain the proportion of recruits that the t0 population was supplying (Equation 4). Recruit Proportion = (Recruits (t1 size 1 patches)/t0 population (total patches)) (Equation 4) Linear regression was used to evaluate the relationship between the number of recruits as the response (y) variable and total number of seagrass patches as well as percent seagrass coverage as the predictor (x) variables. Matrix Analysis Once the Leslie matrices were completed by site and time step, the popbio package in R was used for the analysis of the population dynamics and to calculate lambdas (eigenvalues) and the stable size distributions (eigenvectors). A lambda >

28 17 1 indicates the population is growing while a lambda < 1 indicates population shrinkage. However, the lambda values were associated with the time step and were not yearly. To correct for this, the values obtained for the time steps were converted to a yearly rate by taking the root of the lambda to however many years the time step was. For example, lambda value for the Black Point Lagoon (BL) time step of would be raised to the (1/6) power, as 6 was the total number of years for this time step. The stable stage distribution (eigenvector) was also calculated for each transition matrix. Population Projections The transition and population structure information collected here were used to run population projections based on different recruitment and fragmentation scenarios to evaluate the long-term impacts of these key demographic processes under different salinity environments. The function pop.projection was applied to project the changes of the transition matrices into the future using the Leslie matrix multiplied by the respective population vector (Stubben & Milligan, 2007). These population vectors, built by time step by site, were multiplied by the associated transition matrix for 17 intervals, with each interval representing ~5 years (Hughes & Tanner, 2000). The function stage.vector.plot was then used to visualize the results of the pop.projection to identify when the population converged to steady state distributions (Stubben & Milligan, 2007). Fragmentation Scenarios Population projections were run under high and low fragmentation, and high, average and low recruitment conditions. The transition values used to represent high and low fragmentation conditions were determined based on the transition data collected in this study by selecting one site/time interval that displayed fragmentation values above

29 18 and below the global averages. These scenarios were then run with low, average, and high recruitment values selected as just described for the fragmentation values. All scenarios were run with equal proportions of size classes as starting conditions. A chisquared goodness of fit test was utilized to determine if the final proportions of size classes were significantly different from the initial proportions in each of the scenarios simulated. Additional Analysis Percent cover of seagrass was examined by site to evaluate changes over time. Seagrass percent cover was calculated for each site and time period using the polygons delineated as part of the mapping methods. Linear regression was used to evaluate cover patterns, with cover as the response (y) variable and fragmentation, recruitment and time as the predictor (x) variables.

30 Chapter III. Results Patch Mortality Rates The two-factor ANOVA used to evaluate the role of patch size and salinity (adjacent vs. distant) on patch extinction showed that patch size played a significant role on the mortality rate with the smallest size classes (1 and 2) having significantly higher mortality than the largest size classes (4 and 5) (ANOVA, p<0.05; Figure 7). No significant effects of salinity on mortality were found (ANOVA, p>0.05), and no significant interactions between the two factors were documented (ANOVA, p>0.05). Mortality Rate Size 1 Size 2 Size 3 Size 4 Size 5 Size Classes Adjacent Distant Figure 7. Average mortality rates in relation to patch size. Values represent average (± SD) annual mortality (extinction) rates for all time periods combined. Fragmentation On average, sites adjacent to canals had higher (but not significantly different, t- test, p=0.09) rates of yearly fragmentation than distant sites (Figure 8). For the adjacent sites BP, CP, and SC, the average number of patches produced per year were 6.4, 6.6 and 5.7 respectively. For the distant sites BL, CK and TP, the average number of patches produced per year were 5.7, 2.7 and 3.4 respectively. When the sites were grouped into adjacent and distant and averaged over all time steps, the adjacent sites had an average of 6.2 patches created per year compared to 4.0 for the distant sites (Figure 9). 19

31 20 25 Annual Fragmentation Rate Time Steps Adjacent Distant Figure 8. Fragmentation rates over time. Values are annual fragmentation rates (± SD) for all distant and adjacent sites combined Annaul Fragmentation Rate Adjacent Distant Adjacent Distant Figure 9. Yearly fragmentation rates (SD±) for all time periods combined by distant and adjacent sites. No significant relationship was found between fragmentation rates and time (linear regression, p>0.05; Figure 10). Similarly, the apparent relationship between seagrass cover and fragmentation was not significant (linear regression, p>0.05; Figure 11).

32 21 Figure 10. Annual fragmentation rates over the study period. Figure 11. Relationship between annual fragmentation rate and seagrass cover of all sites surveyed Recruitment The number of new patch recruits showed a significant positive relationship to the total number of patches in the seagrass meadows (linear regression, p<0.05; Figure 12). However, the number of recruits showed no relation to the seagrass cover of the sites (linear regression, p>0.05; Figure 13). No significant patters in recruitment rates were detected over time (linear regression, p>0.05) or between adjacent and distant sites (Figure 14).

33 22 Figure 12. Relationship between recruitment and the total number of patches in each site surveyed. Figure 13. Relationship between recruitment and seagrass percent cover Recruitment Rate Adjacent Distant Adjacent Distant Figure 14. Mean recruitment rates (SD±) by time step combined by distant and adjacent sites.

34 23 Lambda Values Lambda values varied, spatially and temporally without any clear patterns (Table 5). For adjacent sites, two of the three sites showed an average λλ 1 representing a growing population. Only one of the distant sites (BL, λλ =1.02) displayed λλ >1. The largest average λλ values were documented for the time step (λλ =1.03). The lowest average λλ values were documented for the and time interval (λλ =0.97). No temporal patterns in λλ values were recorded (linear regression, p>0.05; Figure 15). Table 5. Yearly λλ values for all sites and their associated time steps with observed recruitment. Sites in green are distant, blue are adjacent. Refer to Table 1 for the time frames associated with each site. Lambda Average Values 2009 BL BP CK CP SC 0.95 N/A TP N/A Average St. Dev Figure 15. Relationship between lambda values and time. Both adjacent and distant sites had a λλ 1 on three occasions (Table 6). The highest average λλ values for the adjacent sites was 1.08 (SD±=0.09) from

35 24 The highest average λλ for the distant sites was 1.07 (SD±=0.08), recorded during the time step. The lowest λλ recorded for adjacent sites was 0.95, displayed in The lowest λλ for distant sites was 0.92 recorded from No significant differences in average λλ were found between distant (mean λλ = 0.99 (SD = ± 0.069)) and adjacent distant (mean λλ = 1 (SD = ± 0.066)) (t-test, p>0.05). Table 6. Average (±) lambda values by time step for adjacent and distant sites. Sites Average Adjacent 0.99(0.04) 0.99(0.08) 1.02(0.04) 1.08(0.09) 0.96(0.02) 0.96(0.01) 1.05(0.1) 0.95(0.04) 1 (0.066) Distant 0.99(0.06) 1.02(0.02) 0.92(0.01) 0.98(0.04) 0.97(0.06) 1.07(0.08) 0.97(0.01) 1(0.13) 0.99 (0.069) Steady State Proportions The stable population proportions for all sites showed that size 3 composed the highest, on average, proportion of the population at 0.38 (Table 7). Size 1 had the next highest proportion at 0.23, with size 2 composing 0.21 of the population. Sizes 4 and 5 had the lowest proportions with values of 0.12 and 0.05 respectively. Table 7. Proportion of the population composed by each seagrass patch size class over time averaged across the six sites. Time Step Size 1 Average Size 2 Average Size 3 Average Size 4 Average Size 5 Average Average St. Dev Population Projections a. High Fragmentation Scenarios - Proportions All high fragmentation scenarios were run, over 17 intervals, each representing ~5 years, with an annual fragmentation rate of Low recruitment was set to 0.007,

36 25 average recruitment was 0.15 and high recruitment was 0.6. The stable size distribution for the high fragmentation/low recruitment scenario was 0.39, 0.43, 0.11, 0.01, and 0.06 for the respective size classes (Figure 16). The initial and final values were significant (chi 2 goodness of fit test, p<0.05). Figure 16. Changes in the proportion of size classes over time under high fragmentation and low recruitment conditions. The stacked columns show the starting and final proportions by patch size. The stable size distribution for the high fragment/average recruitment site was 0.55, 0.37, 0.01, 0.02 and 0.05 for the respective size classes (Figure 17). The final frequencies in each size class were significantly different from the initial frequencies (chi 2 goodness of fit test, p<0.05). Figure 17. Changes in the proportion of size classes over time under high fragmentation and average recruitment conditions. The stacked columns show the starting and final proportions by patch size.

37 26 The stable size distribution for the high fragment/high recruitment scenario was 0.66, 0.29, 0.01, 0.01 and 0.03 for the respective size classes (Figure 18). The final frequencies were significantly different from the initial frequencies (chi 2 goodness of fit test, p<0.05). Figure 18. Changes in the proportion of size classes over time under high fragmentation and high recruitment conditions. The stacked columns show the starting and final proportions by patch size. b. Low Fragmentation Scenarios Proportions All low fragmentation scenarios were run over the same number of intervals with an annual fragmentation rate of Low recruitment was set to 0.007, average recruitment was 0.15 and high recruitment was 0.6. The stable size distribution for the low fragmentation/low recruitment scenario was 0.06, 0.32, 0.41, 0.11, and 0.1 for the respective size classes (Figure 19). The final frequencies were significantly different from the initial frequencies (chi 2 goodness of fit test, p<0.05).

38 27 Figure 19. Changes in the proportion of size classes over time under low fragmentation and low recruitment conditions. The stacked columns show the starting and final proportions by patch size. The stable size distribution for the low fragment/average recruitment site was 0.21, 0.08, 0.33, 0.2 and 0.17 for the respective size classes (Figure 20). The final frequencies were not significantly different from the initial frequencies (chi 2 goodness of fit test, p>0.05). Figure 20. Changes in the proportion of size classes over time under low fragmentation and average recruitment conditions. The stacked columns show the starting and final proportions by patch size. The stable size distribution for the low fragment/high recruitment site was 0.43, 0.05, 0.14, 0.2 and 0.18 for the respective size classes (Figure 21). The final frequencies were not significantly different from the initial frequencies (chi 2 goodness of fit test, p>0.05).

39 28 Figure 21. Changes in the proportion of size classes over time under low fragmentation and high recruitment conditions. The stacked columns show the starting and final proportions by patch size. The representation of the proportional changes that each scenario underwent are represented in the following figure (22). Figure 22. Initial and final proportion for high and low fragmentation scenarios. The chi-square analyses conducted showed that there were significant differences between initial and final population size frequencies for all the high fragmentation/recruitment scenarios. The only low fragmentation scenario that displayed a significant difference in frequencies was the low recruitment scenario. Significant differences in population size proportions were documented between initial and final conditions for all of the low/high fragmentation scenarios (Figure 22).

40 29 c. High Fragmentation Scenario Patch Abundance The initial population vector was composed of 201 patches and the population was divided into 48 (size 1), 56 (size 2), 72 (size 3), 17 (size 4), and 7 (size 5) patches based on the average conditions recorded in this study. Fragmentation and recruitment rates are the same as described in the previous section. The population declined to zero at the end of the high fragmentation/low recruitment scenario (Figure 23). Figure 23. Changes in the abundance of seagrass patches of different sizes over time under high fragmentation and low recruitment conditions. The population declined to zero at the end of the high fragmentation/average recruitment scenario (Figure 24).

41 30 Figure 24. Changes in the abundance of seagrass patches of different sizes over time under high fragmentation and average recruitment conditions. The population declined to zero at the end of the high fragmentation/high recruitment scenario (Figure 25). Figure 25. Changes in the abundance of seagrass patches of different sizes over time under high fragmentation and high recruitment conditions. d. Low Fragmentation Scenario Patch Abundance The initial population vector was composed of the same composition as the high recruitment projections. Fragmentation and recruitment rates are the same as described in the previous section. The population consisted of 6, 30, 39, 10, and 9 patches for the respective size classes at the end of the low fragmentation/low recruitment scenario (Figure 26).

42 31 Figure 26. Changes in the abundance of seagrass patches of different sizes over time under low fragmentation and low recruitment conditions. The population consisted of 27, 10, 43, 26, and 22 patches for the respective size classes at the end of the low fragmentation/average recruitment scenario (Figure 27). Figure 27. Changes in the abundance of seagrass patches of different sizes over time under low fragmentation and average recruitment conditions. The population consisted of 38, 4, 13, 18, and 16 patches for the respective size classes at the end of the low fragmentation/high recruitment scenario (Figure 28).

43 32 Figure 28. Changes in the abundance of seagrass patches of different sizes over time under low fragmentation and high recruitment conditions. The initial and final population values for the high/low fragmentation scenarios are described in the following table (Table 8). Table 8. The initial and final population values for the high and low fragmentation scenarios. Low recruitment (LR), average recruitment (AR), and high recruitment (HR). High Fragmentation Low Fragmentation LR AR HR LR AR HR Initial Pop Final Pop Initial Pop Final Pop Initial Pop Final Pop Initial Pop Final Pop Initial Pop Final Pop Initial Pop Final Pop Seagrass Cover Five of the six sites surveyed showed a reduction on cover between initial and final surveys, with Convoy Point (CP) being the only site that had an increase in seagrass cover over the course of this study (Table 9). Four of the sites displayed at least a loss of 5% cover with Black Point Lagoon (BL) decreasing by 11%. Both distant and adjacent sites showed an overall decrease in total cover (Figure 29).

44 33 Figure 29. Percent cover over time by distant, adjacent and the average. Table 9. Percent seagrass cover for the study sites over time showing by site, time step and percent change by site over the entirety of this study. Time Step BL BP CK CP SC TP N/A N/A % Change

45 Chapter IV. Discussion Seagrass habitats around the world have experienced multiple disturbances that have resulted in accelerating rates of loss of these important coastal ecosystems (Waycott et al., 2007; Gretch et al., 2018). Of special concern are those seagrass habitats in close proximity to expanding urban centers, like the city of Miami, where human activities interact with natural disturbances to challenge the persistence of coastal habitats. Within this context, a positive outcome of our research is the documentation that nearshore seagrass meadows along the western margin of central Biscayne Bay, Florida, US, have experienced only limited declines in cover (average decline = 5.5%) over the 70 years examined. This is in direct contrast with the mass mortality of seagrasses reported for Florida Bay, Florida, US, in and more recently in (Robblee et al., 1991; Zieman et al., 1999; Hall et al., 2016). This limited decline is also noteworthy considering that the nearshore habitats of Biscayne Bay have been significantly modified over the past 5-6 decades due to the construction of a canal system that has altered the quantity, quality, and method of delivery of fresh water from the Florida Everglades (Sklar, 2002; Herman, 2014; Zink et al., 2017). These modifications have not only reduced the overall amount of fresh water reaching the Biscayne Bay shoreline but also created clear gradients in salinity that are created by the point discharge of fresh water from canals. During seasonal discharge periods, areas near canals experience drastic drops in salinity over a matter of hours (often reaching < 5 ppt), while areas removed from these influences have higher and more stable salinity. Prior research in this region has shown that these salinity patterns influence the abundance and distribution of submerged aquatic vegetation and associated fauna, but our historical seascape analyses 34

46 35 shows that seagrass cover has not been greatly affected by these changes and seagrass meadows appear to be both resistant and resilient to the modified salinity patterns (Santos et al., 2014). The fluctuation in lambda values documented here, alternating periods of population decline with periods of population growth, are evidence of these persistence and recovery patterns. While the limited historical loss of seagrasses cover is a positive result compared to the multiple reports of declines elsewhere, especially in areas influenced by anthropogenic pressures (Waycott et al., 2009; Marba et al., 2014), our seascape analyses combined with the development of a patch-based seagrass population model showed that fluctuating salinity found near the mouths of canals does have a negative impact on the fragmentation patterns of seagrass meadows, as seagrass fragmentation rates were found to be higher near fresh water canals. Thus, even if seagrass meadows in Biscayne Bay have retained areal coverage, they are showing signs of salinity-driven fragmentation. Reports from other systems (Farhig, 2003; Montefalcone et al., 2010) have shown that fragmented habitats are more susceptible to further disturbance and that they can decline rapidly (Livernois et al., 2017). The simulation scenarios completed here suggest that, under persistent high rates of fragmentation, seagrass populations may exceed a resistant threshold and decline rapidly. While recruitment of new seagrass patches through sexual or asexual reproduction may, to some extent, mitigate the impacts of high fragmentation, we found that populations under simulated high fragmentation scenarios can become extinct within 50 years. This study is the first to quantify the patch dynamics of seagrass communities in Florida and our seascape approach has revealed key aspects of the dynamics of seagrass

47 36 communities that further our understanding of how seagrass populations respond to differing environmental conditions. The extinction rate of seagrass patches was significantly affected by patch size, with the mortality rate of the smallest patches (57%) being an order of magnitude higher than that of the largest patches (< 5%). Similar sizebased patterns of mortality were documented for corals and sponges, which also showed higher mortality of the smaller size classes (Hughes, 1984; Cropper, 1999). Smaller seagrass patches have been shown in other studies to have higher susceptibility to physical disturbances (Vidondo et al., 1997; Sintes et al., 2005; Duarte et al., 2007; Kendrick et al., 2012; Livernois et al., 2017). The high mortality rate of smaller patches could be due to their lower biomass-to-perimeter ratio that may limit their anchoring capabilities as well as expose them to higher erosion rates along the patch perimeter (Santos et al., 2014). Larger patches have a more extensive root system with higher storage of below-ground biomass that can help increase their resilience and lead to lower mortality rates (Jarossy, 2016). Considering the large difference in extinction rates among patch sizes, any shift in population structure that reduces mean patch size (e.g., fragmentation) would clearly reduce the persistence of seagrass populations. This pattern was clearly captured as output of our simulated scenarios that showed rapid population declines under high fragmentation scenarios. In Biscayne Bay, 94% of the seagrass patches created by fragmentation were produced by the two largest size classes. The stability in the cover of the seagrass meadows recorded over the 70-year period of record, even when fragmentation rates were high in some time periods, is likely due to the fact that the larger patches were able to fragment and still remain within the large size classes that

48 37 provide a size refuge against extinction. Continued fragmentation would, eventually, lead to a reduction in the abundance of these large and stable patch sizes, resulting in the population declines observed in our simulations. The persistence of the simulated seagrass populations was directly related to the fragmentation rates that affected the proportion and size distribution of patches. Under high fragmentation scenarios, the abundance of larger patches declined quickly. Without the source of new small patches through fragmentation as these larger patches decline, the populations went to extinction within 50 years, regardless of recruitment rates. In contrast, under low fragmentation scenarios, the abundance of larger patches remains stable or even increases over time, leading to the persistence of the population. Thus, our model showed that, along with fragmentation rates, the abundance of larger size classes plays a major role in stabilizing the seagrass community. Of special concern would be scenarios in which both an increase in fragmentation and a decrease in percent cover co-occur, as these declines can cascade into reductions of macrofaunal species richness, biomass, diversity, composition, and habitat availability (Santos et al., 2014; Wlodarska-Kowalczuk et al., 2014; Jankowsak et al., 2019). The focus of seagrass monitoring studies has historically been at the quadrat-level (<1 m 2 ), yet seagrass communities often cover extensive areas (Robbins & Bell, 1994; Vidondo et al., 1997). Patterns documented at small scales may not extrapolate linearly to the entire landscape (Duarte & Sand-Jensen, 1990; Vidondo et al., 1997; Cunha & Santos, 2009). Using a landscape approach that measures the patterns (spatial and temporal) and processes that affect the dynamics of seagrass populations at multiple scales is thus necessary for a proper understanding and prediction of the response of