UFDC Home  myUFDC Home  Help 



Full Text  
FACTORS CONTROLLING PHOSPHORUS REMOVAL IN LARGE CONSTRUCTED WETLANDS IN SOUTH FLORIDA By MICHAEL JERAULD 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 2010 2010 Michael Jerauld To the people and places in this world in need of our mercy ACKNOWLEDGMENTS Without a doubt, my first and sincerest thanks are due to my parents. Only with their unending support has the success of my last 24 years been possible. I owe the same debt of gratitude to my extraordinary wife, Sarah, who forfeited countless evenings, weekends and getaways with me, to see my completion of this degree. I recognize also that this degree would not have been possible without the lavish love of my God. My advisor, Dr. James Jawitz, deserves no less than the highest praise, not only for his keen scientific eye, but for his unwavering patience and faultless advice as I progressed through my studies. His ability to unify criticism, instruction, and praise is unmatched in my experience. His influence magnified the value of all other aspects of my education at the University of Florida and all my future work will testify to the quality of his guidance. He has earned my deepest respect. Each of the members of my advisory committee contributed significantly and uniquely to my training. Dr. K. Ramesh Reddy generously offered his extensive understanding of wetland biogeochemistry though both his lecture course and his critiques of my work throughout my tenure here. It was a pleasure to work closely with Dr. Reddy on the project with which I began this degree. It was perhaps Dr. Mark Clark's passion and enthusiasm for wetlands that drew me into the field. I will long measure myself against his remarkable energy and his drive to affect real change in this world. Dr. Mike Annable provided me a firm foundation in wetland hydrology that was indispensable to my work. I would be amiss to exclude my excellent friends and colleagues, with whom I exchanged many insightful conversations over lunches eaten on the lawn. I am indebted to Rupesh Bhomia, for his overwhelming kindness (and excellent cooking); to R.J. Sindelar for being ever ready for racquetball and cycling; to Luke Gommermann for the constant assistance he provides to those around him; to Moshe Doron for many laughs and his unique perspective on the world; to Davie Kadyampakeni for his undying and infectious cheerfulness; and to Alex Cheesman for demonstrating his friendship by bearing too many jokes about "lift" and "torch". Each of these, and many more not mentioned here, deserves much credit for lightening the burden of graduate school. Much gratitude is also due Rajendra Paudel and Rupesh Bhomia, once more, for their GIS and soils analyses, respectively, that enabled much of this work. Finally, I was fortunate enough to interact with several professional wetland scientists, each of whom was instrumental in my graduate education. Dr. Robert Knight, of Wetland Solutions Inc., graciously shared his advice and expertise and his passion for treatment wetlands. Any success I achieve in the field will be due, in large part, to his mentoring. Mike Korvela and Dr. Mike Chimney, of the South Florida Water Management District went to great lengths to collate and provide the vast datasets investigated in this thesis and offered their invaluable advice without hesitation. TABLE OF CONTENTS page A C K N O W L E D G M E N T S ..............................................................................................................4 T A B L E O F C O N T E N T S ........................................................ .........................................6 LIST OF TA BLE S .................................................................................................. 9 LIST OF FIGURES .................................. .. .... ..... ................... 10 L IST O F A B B R E V IA T IO N S .... ..... ................................................................ ... .................. 11 A B S T R A C T ......... ....................... ............................................................ 12 CHAPTER 1 INTRODUCTION .................. ................... ...........................................14 Phosphorus C cycling in W wetlands ............................................................................ ............14 Methods ........................................ 18 S ite D e sc rip tio n ............................................................................................................... 1 8 D ata S o u rc e s .........................................................................................19 C a lcu latio n s .............................................................................2 0 Hydraulic flows .................................... ......................... ... ......... 20 Water column chemical and physical properties ...............................................21 W etland chem ical and physical properties............................................. 23 Wetland performance models..................... ...............23 T im eStep Selection ................................................................24 D ata S c re e n in g .................................................................................................................2 6 2 INFLUENCE OF WETTED AREA ON PHOSPHORUS DYNAMICS IN THE STORMWATER TREATMENT AREAS ......................................................................31 In tro d u ctio n ................... ................... ...................1.......... O bje ctiv e s ................... ...................3...................3.......... M eth o d s ........................................................................3 4 Calculation of the W etted A rea .............................................................. ............... 34 Statistical A nalyses................................................... 36 R results and D iscu ssion ................................................. ........ ............. ............... 36 Characterization of Elevation Distribution and Wetted Area in the STAs ...................36 Relative Wetted Area and Total Phosphorus Removal Performance .............................37 Relative Wetted Area and Total Phosphorus Mass Loading Rate ................................40 C onclu sions.......... ..........................................................43 3 ASSOCIATION BETWEEN LOADING RATE AND OUTFLOW CONCENTRATION IN THE STORMWATER TREATMENT AREAS ............................52 Intro du action .................. ........... ....................................................................................... 52 O bjectiv es ....................... ..........................55.......... M e th o d s ...........................................................................5 5 V vegetation .................................................. .. ....... .......................... ...............55 Outflow Concentration and Total Phosphorus Areal Loading Rate................................57 Results and Discussion ...................................... .......................... 57 Outflow Concentration Areal Loading Relationship ...................................... 57 Effect of vegetation type and cover ................................57 Effect of areal total phosphorus loading rate ........................................ ....58 Factors Controlling the Apparent Background Concentration ..... ............... ..60 Inflow phosphorus fractions .......................................... .... ........... 60 Longterm real loading ........ .................................... ...............61 S e a so n a lity ......................................................................................... 6 3 C onclusions......................................... .................... 65 4 MULTIPLE LINEAR REGRESSION TO DETERMINE FACTORS CONTROLLING PHOSPHORUS CONCENTRATION AND SETTLING RATE IN THE STORMWATER TREATMENT AREAS ....................................................... 72 In tro du ctio n ................... ...................7...................2.......... O bje ctiv e s ................... ...................7...................4.......... M methods .................... ................................... ..................... 74 M multiple Linear Regression ................................................................... 74 V ariab les C o n sid ered ................................................................................................. 7 6 kC* m odel term s ............................................................................. 77 Inflow phosphorus fractions ............................................................ ............77 W etland age and soil phosphorus ................................................................. 78 C calcium and pH ...........................................................79 W after tem p eratu re ....................................................... ................ .. .................... 80 Relative w etted area and w after depth..................................................................... ...80 R results and D discussion .......................................... ................................... ............... 81 Outflow Total Phosphorus Concentration: All Cells ................................................81 Outflow Total Phosphorus Concentration: Cells with Nonsignificant r'values ...........83 Outflow Total Phosphorus Concentration: Single Cell ................................................ 85 Total Phosphorus Areal Settling Rate: All Cells......................................................86 Total Phosphorus Areal Settling Rate: Single Cell ............... .................................... 88 Limitations and Future Application of the Multiple Regression Technique ................ 88 Conclusions................................................... .. .......... .... .. ..... ......... 90 5 C O N C L U SIO N S ............................................................................................ 103 Inflow Total Phosphorus Concentration........................ ............... ............... 104 Hydraulic and Total Phosphorus Loading Rates ................. ............. ....................105 Inflow Phosphorus Fractions .................... ............... ................. 105 W etland A ge and Soil Phosphorus ............................................ .................... .....106 C calcium and pH ................................................................................ 107 W after T em perature ............................................................................................ 107 7 Relative Wetted Area and Water Depth .................................................................108 APPENDIX: DATA SCREENING CRITERIA.......................................................................110 L IST O F R E F E R E N C E S .................................................................................. ..................... 112 B IO G R A PH IC A L SK E T C H ......................................................................... ........................ 117 8 LIST OF TABLES Table page 21 Average annual relative wetted area by water year for each cell in the Stormwater Treatm ent Areas ....................... ......... ...... ................. ..........45 22 Intraannual trends in relative wetted area........... ................................................46 23 Changes in the coefficients of correlation for different subsets of data. .........................47 31 Coefficients of correlation between monthly outflow total phosphorus (TP) concentration and monthly TP areal loading rate (ALR) within each cell .......................67 41 Coefficients of determination (r2) of multiple linear regression models explaining the monthly outflow total phosphorus concentration in all cells.........................................93 42 Estimates of parameters for the model explaining monthly outflow total phosphorus (TP) concentration in all cells. ............................................ ..........................................94 43 Coefficients of determination (r2) of multiple linear regression models explaining the monthly outflow total phosphorus concentration in all cells with nonsignificant r' v alu es .......................................................... ...................................9 5 44 Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in all cells with nonsignificant r'values........................................96 45 Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in STA 1W Cell 1. ............................................. ............................. 97 46 Coefficients of determination (r2) of multiple linear regression models explaining the monthly total phosphorus (TP) areal settling rate in all cells ........................................98 47 Estimates of parameters for the model explaining the monthly total phosphorus areal settling rate in all cells. .......................... ...... .........................................99 48 Estimates of parameters for the model explaining monthly total phosphorus (TP) real settling rate in STA 1W Cell 1.......................................... .......................... 100 LIST OF FIGURES Figure page 11 Phosphorus (P) cycle in surface flow wetlands.. ....................................... ...............28 12 Map showing the locations of the six Stormwater Treatment Areas (STAs), the Everglades Agricultural Area and the Everglades Protection Area in South Florida........29 13 Schematics of the configuration of the treatment cells within each Stormwater Treatm ent A rea (STA ) ................... .... .... .................... .. ........... ............... 30 21 Cumulative elevation distribution for each cell in the Stormwater Treatment Areas........48 22 Relative wetted area in each of the Stormwater Treatment Areas.......................... 49 23 Histogram of monthly relative wetted area (determined by the elevation distribution) values across all nonscreened months and all included cells. ........................................50 24 Monthly relative wetted area (determined by the elevation distribution) with respect to monthly hydraulic loading for all nonscreened months and all included cells. ...........51 31 When a wetland treats to background concentration (C*), the areal loading rate (ALR) does not affect the outflow concentration. .................................. .................68 32 Correlation between outflow total phosphorus (TP) concentration and TP mass loading rate with respect to submerged aquatic vegetation (SAV) coverage ....................69 33 Correlation between outflow total phosphorus (TP) concentration and TP areal loading rate (ALR) as a function of the average annual TPALR ...................................69 34 Outflow total phosphorus (TP) concentration with respect to the correlation coefficient between the TP areal loading rate (TPALR) and the outflow TP concentration s. .......................................................... ................. 70 35 Periodofrecord (POR) flowweighted mean outflow total phosphorus (TP) concentration as a function of the POR average annual TP areal loading rate..................70 36 Monthly average temperature (C) of inflow and outflow water in STA1W Cell 4........71 41 Changes in the coefficient of determination (r2) with increasing complexity of the model explaining monthly outflow TP concentration in all cells.................................101 42 Changes in the coefficient of determination (r2) with increasing model complexity of the model explaining the monthly total phosphorus settling rate in all cells.................102 LIST OF ABBREVIATIONS ALR Areal loading rate Ca MR Calcium mass retention CDF Cumulative distribution function CFW Central Flowway DIP Dissolved inorganic phosphorus DOP Dissolved organic phosphorus EAA Everglades Agricultural Area EAV Emergent aquatic vegetation NFW North Flowway P Phosphorus PIP Particulate inorganic phosphorus POP Particulate organic phosphorus POR Period of record PP Particulate phosphorus RWA Relative wetted area SAV Submerged aquatic vegetation SFWMD South Florida Water Management District SRP Soluble reactive phosphorus STA Stormwater treatment area TDP Total dissolved phosphorus TP Total phosphorus TPALR Total phosphorus areal loading rate WA Wetted area Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science FACTORS CONTROLLING PHOSPHORUS REMOVAL IN LARGE CONSTRUCTED WETLANDS IN SOUTH FLORIDA By Michael Jerauld August 2010 Chair: James Jawitz Major: Soil and Water Science The Florida Everglades is an oligotrophic, subtropical wetland extremely susceptible eutrophication from anthropogenic phosphorus (P) inputs. Six large treatment marshes (total effective treatment area 18,000 ha), called Stormwater Treatment Areas (STAs), have been constructed over the last fifteen years to remove P from agricultural runoff before it enters the Everglades. Because of the massive investment to build and maintain these wetlands, it is important to evaluate the factors that contribute to their performance. Naturally, each STA had a different performance record. This study attempted to account for those differences by investigating 1) the relative wetted area (RWA), 2) the relationship between the total P (TP) areal loading rate (ALR) and the outflow TP concentration and 3) the wetland characteristics co variant with the outflow TP concentration and the TP areal settling rate. The RWA was determined by subtracting the elevation distribution function from the average water level (stage). The RWA was less than 1.0 in 230 out of 1044 (22%) of months. The TPALR was not statistically different when calculated with the RWA rather than the nominal area; the nominal area was deemed sufficient for most loading rate calculations. Among months with substantial reflooding (n = 39), the TP areal settling rate was negatively correlated with the magnitude of the reflooding event (r = 0.605, p < 0.0001). In most of the STAs, the monthly outflow TP concentration was uncorrelated to the monthly TPALR. The vegetation type (submerged vs. emergent) and the magnitude of the loading rate were hypothesized to contribute to this uncoupling, but neither played a quantifiable role. The cause of this independence remains unclear. Possibly, the Damkohler number (areal settling rate divided by hydraulic loading rate) was sufficiently high in most STAs such that wetland factors other than the loading rate and settling rate tended to control the outflow TP concentration. Through multiple linear regression, five variables (inflow TP concentration, change in monthly wetted area, hydraulic residence time, wetland age and inflow Ca concentration) were found to explain 32% of the variability in the monthly outflow TP concentration. Six factors (hydraulic loading rate, inflow dissolved organic P fraction, inflow particulate P fraction, wetland age, outflow water temperature and TPALR) accounted for 51% of the variation in the monthly TP areal settling rate. The proportion of explained variability may be improved in future analyses by including variables not considered herein. Additional research is needed to confidently identify the factors that control the outflow TP concentration. CHAPTER 1 INTRODUCTION Phosphorus (P) is a naturally occurring, abundant element required by all forms of life. In the natural environment, the biosphere obtains P weathered from minerals. Typically, hydrologic flows transport P from uplands to aquatic environments, often via wetland ecotones. Excess P applied to the landscape (often as fertilizer) can join this migration and frequently contributes to eutrophication of downstream water bodies. In the landscape, mediating wetlands can serve as P sinks, dampening the transfer of P into aquatic systems (Richardson, 1999) and natural wetlands have long been used as receiving sites for point discharges of wastewater (Kadlec and Wallace, 2008). From this insight, it followed that wetlands could be employed to reduce P loads to downstream ecosystems, and many wetlands have been restored or constructed for the purpose of water treatment (Kadlec and Wallace, 2008). Treatment wetlands as a technology have progressed from being first a novelty, past trial pilot systems, to a point where optimization and diversity of application has become the focus (e.g. Mitsch et al., 1995; Kadlec and Knight, 1996; Higgins et al., 2000; Braskerud, 2002a; Turner et al., 2006; Vymazal, 2007; Kadlec and Wallace, 2008). Understanding and quantifying the internal P processing mechanisms are essential to maximizing treatment wetland P removal and retention. Phosphorus Cycling in Wetlands Phosphorus is found in wide variety of biological molecules beyond the notable examples including DNA and ATP (Turner and Newman, 2005). Phosphorus is also chemically active and may be found in myriad minerals in association with Ca, Mg, Fe, Al (Reddy et al, 1999). It is convenient to classify these many forms of P by their physical, chemical or operational characteristics. Many taxonomic schemes exist, but four broad P pools are commonly conceptualized: dissolved inorganic P (DIP), dissolved organic P (DOP), particulate inorganic P (PIP) and particulate organic P (POP) (Reddy and DeLaune, 2008). In each of these terms, "dissolved" and "particulate" distinguish particles that do and do not pass through a filter membrane (pore size typically, but not always, 0.45 [im), respectively. For methodological convenience, the following pools are often determined for watercolumn P: soluble reactive P (SRP), dissolved organic P (DOP) and particulate P (PP). These groups are somewhat operationally defined (i.e. the boundaries of these groups depend more on analytical technique than the chemical or physical properties of the compounds). Soluble reactive P is composed mainly of orthophosphate (P04P) (Reddy and DeLaune, 2008) but may include some readily hydrolysable organic P (Kadlec and Wallace, 2008). Particulate P encompasses all Pcontaining molecules larger than 0.45 im. Commonly, P is not limiting in wetlands, although the Florida Everglades are a well known exception. Phosphorus enters wetlands through surface water, groundwater, wet and dry atmospheric deposition, and biological transfers (e.g. guano production). Surface water flow dominates the P budget in many treatment wetlands (Kadlec and Wallace, 2008). Within wetlands, P is cycled between various storage compartments at rates which depend on the physical, chemical and biological conditions (Figure 11). The net effect of these P transformations determines the status of a given wetland as a source or sink of P. Wetlands have been successful as a P treatment technology because often the P processing results in net storage or P within the system. Four broad processes that contribute to P retention in wetlands: sorption to soil solids, sedimentation, coprecipitation and biological uptake. Various physical, chemical and biological characteristics inhibit or enhance each of these processes in a given wetland. The movements of P on and off charged sites on the surface of soil solids are called adsorption and desorption, respectively (Reddy and DeLaune, 2008). Froelich (1988) described two steps in the sorption process. First, following changes in pore water P concentration (e.g. in response to novel anthropogenic P loads) adsorption/desorption equilibria are reached within minutes or hours (Froelich, 1988). Reddy and DeLaune (2008) note "The balance between P adsorption and desorption maintains the equilibrium between solid phases and P in soil pore water." Second, absorption, the "solidstate diffusion of adsorbed phosphate from the surface into the interior of particles" occurs over days to months (Froelich, 1988). Generally, sorption is limited to SRP and some reactive components of DOP (Anderson and Magdoff, 2005). The net direction of the flux (on or off the soil) is controlled by the pore water P concentration and the affinity of the soil particles for P ions. In wetlands, the amount of P that can be adsorbed to the soils is often related to the amount of iron and aluminum in the soil (Lijklema, 1977). In South Florida wetlands, soil calcium is an important determinant of soil P sorption capacity (Reddy et al., 1998). Soil adsorption is not considered a sustainable P removal mechanism in treatment wetlands due to the relatively fast reaction time and the finite sorption capacity (Kadlec and Wallace, 2008). The inflow water to wetlands often contains suspended solids, some of which contain P. The aggregate of suspended solids often includes eroded soil particles, macrophyte detritus, and algae or other plankton cells (Stuck et al., 2001). Wetlands function to remove suspended particles primarily by reducing water velocity (by low elevation gradients and drag caused by dense plant stems) such that gravity allows the particles to settle out. This removal is enhanced by the trapping of particles within benthic litter and on biofilms (Schmid et al., 2005). Sedimentation of suspended solids can account for a significant portion of total phosphorus (TP) removal, and sustainability is constrained only by changes in bottom elevation (due to sediment accretion) that prevent surface water flows (Kadlec and Wallace, 2008). An additional pathway of removal is the precipitation of P with Ca, Fe, Al or Mg cations (Reddy and DeLaune, 2008). It may be difficult to quantify precipitation separately from adsorption, since the precipitates often form on the surfaces of soil particles. Reddy and DeLaune (2008) provided a thorough discussion of P coprecipitation in wetlands, particularly regarding the conditions that promote P coprecipitation with each of the identified cations. Generally, acidic conditions promote the coprecipitation of P with Fe and Al, and alkaline environments support the formation of PCa and PMg precipitates. In wetlands, both apatite (Ca5(C1)(PO4)3) and hydroxylapatite (Ca5(OH)(PO4)3) are notable precipitates (Reddy and D'Angelo, 1994). Co precipitation with Ca may be particularly important for P dynamics in the Everglades; Reddy et al. (1993) found a linear correlation between P and Ca accumulation in Everglades soils. Apparently promoted by the consumption of carbonate ions by submerged photosynthesizers, the precise mechanism of the PCa interaction may be either adsorption of P onto the surface of CaCO3 precipitates or the formation of mixed crystals during coprecipitation (Otsuki and Wetzel, 1972; Scinto, 1997). The P requirement of all organisms combined with the high productivity of wetland primary producers makes biological uptake an important mechanism of wetland P removal. Algae and other microorganisms can consume significant amounts of P very rapidly; for example, a diverse algal community reduced mesocosm water column P concentrations from 1,100 gg/L to 50 pg/L in 28 days (Havens et al., 1999). Plants and microorganisms typically can utilize only SRP directly. Other forms of P must first be hydrolyzed before they can be taken up biologically. In wetlands, much of this SRP is promptly transformed into PP (Noe et al., 2003). Many studies have investigated the P uptake potential of wetland plants (e.g. Reddy and DeBusk, 1985; Tanner, 1996). Nearly all of the P incorporated into microbial biomass and most of the macrophyteP is returned to the P cycle through decomposition (Reddy et al., 1995; Kadlec and Wallace, 2008). However, the anaerobic condition resulting from flooding slows decomposition and promotes organic matter accumulation. The P stored in refractory biomass compounds contributes to the longterm sustainable P removal by burial in accrued sediments. The conditions supporting and enhancing each of these P removal mechanisms vary from wetland to wetland depending in large part on soil characteristics, vegetation type and density and water column chemistry, including cation availability and the distribution of the TP pool among the various functional P forms. Maximizing P treatment in wetlands requires an understanding and quantification (and manipulation) of the relative contributions of each of these processes to net P removal. Methods Site Description Over the past 15 years, 6 treatment marshes, called Stormwater Treatment Areas (STAs) have been constructed in South Florida to capture P from agricultural runoff before it enters the Florida Everglades, an oligotrophic wetland susceptible to anthropogenic eutrophication (Chimney and Goforth, 2001). Many investigations have demonstrated P enrichment and the associated disruption of the existing ecosystems in the Everglades (see Reddy and DeLaune, 2008, for a comprehensive review of these works). In particular, most notable to the lay observer, is the shift from saw grass (Cladiumjamaicense Crantz) prairies to dense monotypic cattail (Typha spp.) stands. The Everglades Agricultural Area (EAA), comprising 280,000 ha and approximately 27% of the original Everglades expanse, is a rich producer of sugarcane and winter vegetable crops (Reddy and DeLaune, 2008), and the primary source of P to the Everglades and to the STAs (Pietro et al., 2009). The STAs are strategically located, spatially and hydrologically, between the EAA and the remnants of the Everglades, now broadly defined by the Everglades Protection Area (Figure 12). Altogether, the 6 STAs have a total footprint of over 26,000 ha, with more than 18,000 ha of effective treatment area, subdivided into 35 cells (Figure 13; Pietro et al., 2009). The balance of the land area is consumed by roads, levees, pump stations and other infrastructure. As of 2009, the STAs had retained over 1,200 metric tons of P since the inception of the Everglades Nutrient Removal Project (the forerunner to the STA project) in 1994 (Pietro et al., 2010). Over the same time period, flowweighted mean TP concentration was reduced from 0.143 mg/L to 0.040 mg/L (Pietro et al., 2010). This report took advantage of the water and P mass balance data compiled by Chimney (2009). It was therefore constrained to the cells for which data were reported in Chimney (2009). Excluded cells are indicated in Figure 13. Some currently subdivided cells were treated as combined larger units (e.g. Cell 1A and Cell 1B of STA5 were considered together as the North Flowway (NFW)). This was necessary to maintain longer records in cells that had been subdivided after startup or in cases where data were not available at the levee between subcells. Data Sources South Florida Water Management District (SFWMD) staff collect daily flow and weekly or biweekly water quality data for all STA cells. The data were retrieved from the publicly accessible, online database, DBHYDRO, maintained by SFWMD. Data from topographic surveys of each STA cell were provided by SFWMD. Vegetation coverage data, including maps generated from aerial images and field survey results were produced and collected, respectively, by various contractors hired by SFWMD. The data were made available to this project by SFWMD. Calculations Several physical and chemical characteristics of the STAs were important to multiple aspects of this study. For convenience, the methods employed to calculate each relevant parameter are described here. Hydraulic flows Daily inflow and outflow volumes were summed to determine the total flows for each period of interest: Q1= Ql, (11) ti t2 Q2 = Q2 (12) t\ where tl = starting date of the period of interest, t2 = end date of the period of interest, Q1 = total inflow volume in the period of interest[L3/T], Q2 = total outflow volume in the period of interest[L3/T], Qlt = inflow volume on day t [L3/T] and Q2t = outflow volume on day t [L3/T]. The hydraulic loading rate (q) is the rainfall equivalent [L/T] of the inflow volume: 01 q= (13) where A = wetland area [L2]. Chapter 2 provides a discussion of methods for calculating A. The nominal hydraulic residence time (z) is an estimate of the travel time [T] required for an average packet of water to pass through the wetland: Ah h r <(14) where h mean water depth [L]. where h = mean water depth [L]. Water column chemical and physical properties Weekly and biweekly composite samples were linearly interpolated to estimate daily TP concentrations. Similarly, weekly and biweekly grab samples were linearly interpolated to estimate daily SRP, total dissolved P (TDP) and Ca concentrations. These daily values were flowweighted to calculate average inflow and outflow concentrations for larger time steps. The same method of flowweighting was employed for all constituents: C1= QltClt Cl (15) 2C Q C2 C2= 2 (16) Zt Q2t where Cl = flowweighted mean inflow concentration in the period of interest [M/L3], C2 = flowweighted mean outflow concentration in the period of interest [M/L3], Clt = estimated inflow concentration on day t [M/L3] and C2t = estimated outflow concentration on day t [M/L3]. In cells with multiple inflow or outflow stations, concentration values were again flowweighted by station to estimate a single average value for each cell. Inflow and outflow concentrations of the P forms DOP and PP were calculated by difference using measured TP, TDP and SRP values: C1DOP= ClTDpClsRP ( 17) C2DOP= C2TDPC2SRP ( 18 ) CIpp= CITPCITDP ( 19 ) C2pp= C2TPC2TDP (110 ) where C1DOP = inflow DOP concentration [M/L3], Clpp = inflow PP concentration [M/L3], C1SRP = inflow SRP concentration [M/L3], C1TDP = inflow TDP concentration [M/L3], C1TP = inflow TP concentration [M/L3], C2DOP = outflow DOP concentration [M/L3], C2pp = outflow PP concentration [M/L3], C2SRP = outflow SRP concentration [M/L3], C2TDP = outflow TDP concentration [M/L3] and C2TP = outflow TP concentration [M/L3]. The relative proportions of each of these forms within the TP pool were calculated using the inflow and outflow concentrations: CISRP SRP1R T (111) C2SRP SRP2R= C2 (112) C2TP C1DOP DOPIR= C ( 113) C2DOP DOP2R= C2 (114) C1pp PP1R= C (115) CITP C2pp PP2R= (116) where SRP1R = inflow SRP fraction, SRP2R = outflow SRP fraction, DOP1R = inflow DOP fraction, DOP2R = outflow DOP fraction, PP1R = inflow PP fraction and PP2R = outflow PP fraction. The same procedure was used to calculate areal loading rates (ALR) for both TP and Ca: ALR= ltl (117) A Chapter 2 provides a discussion of methods for calculating A. The mass retention rate of Ca was calculated for each STA cell: ltC2 t2 CaMR= tl QtCttl QtC2t ( 118 ) A where CaMR = Ca mass retention rate [M/L2/T]. Weekly and biweekly inflow and outflow temperature and pH readings were linearly interpolated to estimate daily values of each. The daily values were averaged arithmetically (not flowweighted) to generate monthly figures. Wetland chemical and physical properties Wetland age for any given time period was calculated as the number of whole years between the date of the time period of interest and the initiation of operation. That is, the current age of each wetland was not reflected back to the beginning of the period of record (POR), but rather the actual age of the wetland at each time step was used. Soil TP concentrations from each sampling event were assumed to be spatially representative of each cell, so the values were arithmetically averaged to obtain a single value for each cell for each sampling event. The resulting value was applied to the entire year in which the sampling occurred. Annual values were linearly interpolated across years in which sampling events did not take place. Estimates were not extrapolated to years before the first sampling event or to years after the last sample collection in each cell. The daily average water depth was estimated by subtracting the elevation cumulative distribution function (CDF) from the daily mean water surface elevation (stage). A detailed description of the process by which the elevation CDF was obtained for each cell is available in Chapter 2. The resulting function describes the continuous cumulative distribution of depths in the wetland and was compartmentalized into 0.5 ft depth increments for convenience. The mean water depth was the areaweighted average of these depth increments. Wetland performance models The kC* model is commonly used to predict the outflow concentration of contaminants from wetlands (Kadlec and Knight, 1996): kA C2C*=(CC*) exp( ) (119) where A = wetland area [L2], C1 = inflow concentration [M/L3], C2 = outflow concentration [M/L3], C* = background concentration [M/L3], k = contaminant areal settling rate [L/T], and Q = flow rate [L3/T]. Equation ( 119 ) can be configured to include depth, such that it considers the volume rather than the surface area of the wetland (Kadlec and Knight, 1996): k, hA C2C*=(C1C*) exp ( ) (120) =(C1C*) exp (kvr) where kv = contaminant volumetric rate constant [T1]. Kadlec and Wallace (2008) noted that most wetland contaminant removal processes are "typically apportioned to wetland area to a greater extent than to wetland water volume." Further, they found that kv decreased with increasing depth, and advise that the areal settling rate is more appropriate for most situations. Thus, the areabased form of the kC* model (Equation ( 119 )) is considered throughout this document. The selection of k over kv has an additional implication. Because of the negative depth dependence of k, the increases in T associated with increases in depth do not result in commiserate improvements in performance. Therefore, only the changes in r associated with changes in q (Equation ( 14 )) effect changes in wetland performance, so q may be considered as a proxy for r. TimeStep Selection With respect to the ultimate purpose of the STAs, to reduce the P load to downstream ecosystems, it is the longterm outflow TP concentration that is justly of interest to regulators on behalf of the Everglades. Practice has shown that the longterm outflow TP concentrations varied among cells over the history of the project. Numerous studies have been undertaken to understand the root of these differences and to explore options for improving treatment effectiveness and decreasing outflow concentrations. In this and other studies comparing the STAs to one another, (POR) averages are valuable, but limiting in that the number of values is restrained by the number of cells (or STAs, depending on the design of the study). Frequently, the characteristics that potentially differentiate STA cells from one another also vary over time within cells (e.g. depth, Ca concentration, hydraulic loading, etc). By increasing the temporal resolution of the data, the number of data values available to explore correlative trends expands. In addition, many of the wetland biogeochemical mechanisms relevant to P processing occur at short time scales (minutes to days), suggesting that shorter periods of time averaging might elucidate valuable processlevel information. Conversely, the nonzero r observed in all wetlands dictate that at any instant, the outflow water is independent of the inflow water. Also, due to the imperfect hydraulics ubiquitous in wetlands, the outflow water at any instant comprises a mixture of water that entered the wetland a range of r previously, thereby preventing the simple comparison of inflow waters with waters exiting precisely one mean r later. These two phenomena necessitate an upper limit on the resolution of the time step. Kadlec and Wallace (2008) suggested an averaging period of at least three nominal r to ensure that the bulk of the outflow water considered in a given data value entered the wetland during that same period. For the reasons presented above, many analyses throughout this work aimed to assess the workings of the STAs in the "shortterm." Determining the appropriate time step required striking a balance between the aforementioned competing considerations for short versus long term averaging, while maintaining data manageability. All of the data included in this study were provided by SFWMD, either directly or through the publiclyaccessible, online database, DBHYDRO, which SFWMD populates and maintains. The flow and water quality data were organized by cell at a monthly time step so, given the data at hand, the minimum possible averaging period was 30 d by default. Following the prescription of Kadlec and Wallace (2008), the monthly r calculated from this data were examined. The average monthly r was 23 d. In only 44% of the nonscreened months was the r less than 10 d. The r was less than 30 d in only 77% of months. Together, these findings suggest that the monthly time step was too fine. The additional finding that 95% of months had r s less than 90 d suggested that a quarterly (90 d) averaging period was more appropriate. However, upon the initial conversion of select data to the quarterly time step, it was found that the correlations between the outflow TP concentration and both the inflow TP concentration and the TPALR were slightly stronger among the monthly data than the quarterly values. Ultimately, it is the physical connection between the inflow and outflow water, as measured by the solutes in that water, which must be maintained by selecting a sufficiently long averaging period. Because the 90 d time step failed to measurably strengthen the inflowoutflow relationship, the 30 d averaging period was selected. Throughout this work, each use of the phrase "shortterm" indicates the use of monthly data. Data Screening When analyzing data for relationships between known dependent variables and possible independent variables, it is common practice to screen data points generated by unusual operating conditions in the system of interest. For example, within treatment wetland science, convention calls for the exclusion of data from the "startup period," the time immediately after initiation of treatment, when temporary nutrient sinks (plant biomass expansion and soil sorption) may exaggerate observed treatment performance (Kadlec and Wallace, 2008). Also, some authors elect to omit periods of internal maintenance (e.g. Pietro et al., 2008; Juston and DeBusk, 2006). Within this analysis, the former etiquette was observed implicitly; the POR for each cell was selected to correspond with the data range included in Chimney (2009), who omitted startup years therein. Extremely low flow events created difficulties for the data analysis required by this study. For example, very low hydraulic loading causes the calculated r to become very large (exceeding tens of thousands of days in some cases). Likewise, exceptionally low flows (in or out) can generate extreme calculated TP mass removal values. In an effort to suppress such extravagant values (which, being outliers, have unique power to disrupt correlation and regression analyses), while simultaneously maintaining a large number of data points, all months in which neither the inflow nor outflow volume was at least 10% of the respective longterm average were screened. Of the original 1419 data months, 1050 (74%) remained after this criteria was applied. Some variables required additional special consideration. The appendix lists these variables and the conditions that resulted in the omission of individual values. POP D> OP DIP PIP A f PeriphytonP AEROBIC Figure 11. Phosphorus (P) cycle in surface flow wetlands. DIP = particulate inorganic P, DOP dissolved organic P, PIP = particulate inorganic P, POP = particulate organic P. Image: Reddy and DeLaune, 2008. Figure 12. Map showing the locations of the six Stormwater Treatment Areas (STAs), the Everglades Agricultural Area and the Everglades Protection Area in South Florida. Image: Pietro et al., 2008. STA1W SSTA1E s319 G3 0 Inflow Stations * Outflow Stations W Emergent Cell L SAV Cell : Flow direction S; STA3/4 Figure 13. Schematics of the configuration of the treatment cells within each Stormwater Treatment Area (STA). The dominant vegetation type in each cell is also indicated. Cells marked with "X" were not included in this study. Image: Pietro et al., 2010. G32S r STA2 ,, 4 211/ '\\ ,, ,,} STA6 ~t~]lU ~~lr ~~ ui r 2'.1 CHAPTER 2 INFLUENCE OF WETTED AREA ON PHOSPHORUS DYNAMICS IN THE STORMWATER TREATMENT AREAS Introduction Performance, measured by any metric (outflow concentration, concentration reduction, or settling rate) varied across the STAs, with some cells and STAs having removed P less effectively than their peers (Pietro et al., 2009). The total expenditures on the STA project are difficult to estimate, but the construction, monitoring and management of treatment wetlands with a combined footprint of over 26,000 ha (Pietro et al., 2009) is an enormous financial undertaking. As such, poor P removal performance by an STA is unaffordable by the SFWMD as well as unacceptable for the preservation of downstream ecosystems. As part of a larger diagnostic exercise to elucidate the controlling factors behind measured performance (and ultimately to advise SFWMD on management strategies to improve performance in trouble STAs), the relationship between TP areal loading rate (TPALR) and outflow TP concentration was explored. Terms A, C1, and Q from Equation ( 119 ) are related by: C Q ALR= (21) A where ALR = areal loading rate [M/L2/T]. It can easily be seen that in the kC* model, any change in A, C1, or Q that would cause ALR to increase, would correspondingly cause the outflow concentration (C2) to increase. In other words, the outflow concentration of a contaminant is directly proportional to the ALR of that contaminant. It is common for C1 and Q to vary over the daily, monthly and even annual operation of a treatment wetland. Less prevalent are changes in the treatment area. As a result, ALR is typically calculated for the nominal area (An), that is, the design or built wetland area. However, in certain circumstances, the actual wetted area (WA) may not be equal to An. All wetlands have a distribution of elevations arising from both macroscale ground slope and microscale topographic heterogeneity. The size of the STAs enhances this variation on both scales, and also precludes grading, which, in many constructed wetlands, minimizes topographic heterogeneity. Most wetlands, including the STAs, experience temporal variability in flow, driven primarily by regional rainfall patterns. During times of low flow, particularly during seasonal or regional drought, insufficient water may be delivered to the wetland to maintain the stage above the maximum ground elevation. When this occurs, WA will be less than An. The interaction of the spatial elevation distribution and the temporal stage distribution controls the WA in the STAs. Often, estimates of k are derived from Equation ( 119 ), when the other variables are known, so that treatment performance can be compared across wetlands with diverse inputs. Clearly, when WA < An, the use of A = An in Equation ( 21 ) would result in suppressed estimates of k. Some STA cells did experience periodic drydown conditions in whole or in part (Pietro et al., 2008) due to variability in weather conditions in the tributary basins, therefore previous estimates of k and ALR (of specific relevance to this study, kTp and TPALR) that incorporated An may have been inaccurate. It was hypothesized that poor performance (specifically as measured by the outflow TP concentration relative to the inflow TPALR), was, in some circumstances, an artifact of imprecise calculations. For example, hypothetical wetlands A and B have nominal TPALR of 1 g/m2/yr and observed outflow TP concentrations of 0.075 and 0.10 mg/L, respectively. Investigation reveals that only half of the nominal area of wetland B was actually flooded, due to topographic and stage variability. The revised TPALR in wetland B becomes 2 g/m2/yr, and it becomes evident that some portion of the "poor performance" (high outflow concentration relative to loading rate) of wetland B was a product of the data used in the ALR calculation. If the hypothesis held true, an extension of that logic would suggest that realized performance (absolute outflow concentration, regardless of inlet loading) could be improved by increasing the flooded area of candidate wetlands (while maintaining historical flow rates) through earthwork to reduce the topographic variability. Accurate estimation of the changes in WA over time is important beyond assuring correct k and ALR calculations. The drying of portions of cells is of particular concern for submerged aquatic vegetation (SAV) which may die if desiccated (Harwell, 2003). However, White et al. (2006) subjected SAV mesocosms to 1month periods of dry out (the condition of the vegetative communities upon reflooding was not reported), and found net treatment of TP (outflow concentration lower than inflow concentration) to resume 03 weeks following reflooding. Nonetheless, it may be difficult to maintain SAV in treatment cells that regularly or intermittently dry out. Additionally, reflooding of exposed sediments results in a flux of P out of the sediments into the water column (Olila et al., 1997, White et al., 2006; Bostic and White, 2007; Pietro et al., 2008 and 2009). Thus, the changing WA due to stage and topographic interaction would be expected to reduce treatment effectiveness. Objectives This chapter attempts to answer three primary questions: 1) Was the RWA, or changes in RWA, significantly correlated to kTp? 2) Did WA < An significantly alter the TPALR realized by any cell in the STAs? If so, does the use of A = WA, rather than A = An, in Equations ( 119 ) and ( 21 ) usefully improve the correlation between outflow TP concentration and TPALR? 3) Is there significant value added by calculating WA via the elevation distribution, as compared to simpler methods? The following hypotheses were tested to address the questions above: The longterm TP removal performance will be negatively correlated with variation in elevation within cells. The shortterm TP removal performance will be positively correlated with RWA, and negatively correlated with monthly ARWA, in months when ARWA was positive. The STAwide TPALR calculated by substituting A = WA into Equation ( 21 ) will be significantly higher than the TPALR calculated using A = An. The correlation between the monthly TPALR and the monthly outflow TP concentration will be usefully higher when Equation ( 21 ) is evaluated using A = WA derived from the stage area curve, than when Equation ( 21 ) is solved using either A = WA derived from the mean elevation or A = An. Methods Calculation of the Wetted Area Brown and Caldwell (1996) formalized the calculation of the relative wetted area (RWA): t2 RWA (t2t A A(t)dt (22) (tztdAn f tl where tl = start of time period, and t2 = end of time period. Note that RWA = WA /An. The RWA was calculated for each individual cell, and areaweighted averages were calculated at larger spatial scales. The cell areas reported in Pietro et al. (2009) were used for An. Approximations for the integral portion of Equation ( 22 ) can be made by any one of three methods. The simplest approach, employed by SFWMD (Pietro et al., 2009), uses the operational status of each cell as a proxy for inundation status on any given day, where online cells are fully inundated (WA = An) and offline cells are fully dry (WA = 0) and then the values of WA for all days in the period of interest are summed and then substituted for the integral portion of Equation ( 22 ). Alternatively, WA(t) can be estimated by (Pietro et al., 2010): 0 if hwi < hmean where h,, = average elevation of water surface (stage) on day i [L] and hmean = mean elevation of the ground surface [L]. Again, the WA values for all days in the period of interest are summed and applied to Equation ( 22 ). In some STAs, particularly those with wide elevation ranges or highly pulsed inflows, the methods of SFWMD may not accurately estimate the effective area. Finally, WA(t) may be estimated from a stagearea curve, a function that relates the flooded wetland area to the elevation of the water surface. This final method of WA calculation was selected for this analysis as it quantifies the extent of flooding when hmin < h, < hmax, as opposed to the flooded/dry dichotomy created by the two other methods. To construct the stagearea curve for each cell, topographic data from the most recent, or otherwise most reliable, survey of each STA were interpolated using the kriging method in ArcGIS 9.2 (ESRI, Redlands, CA). Extreme values (e.g. tops of levees, bottoms of ditches) included in the original survey data were excluded from interpolation. From the resulting continuous bathymetric map, ArcGIS returned tabulated frequencies of the elevations in each cell. These frequencies were transformed into cumulative elevation distributions, the vertical axes of which were multiplied by A,. Sixthorder polynomial curves were fit to the resulting points to estimate the stagearea curve: y = ah6 +bh5 +ch4 +dh3 + eh2 +h + g (24) where h = elevation, ft NGVD29, y = area of cell with elevation < x, ag = constants, unique for each cell. Sixthorder curves generally provided strong fits to the data, though the large number of unique coefficients for each cell was somewhat burdensome. Equation ( 24 ) was solved for h = hw,, returning the WA for each day i. The sum of the resulting values served as an estimate of the integral portion of Equation ( 22 ). The nature of these highorder polynomials dictates that they only describe the stagearea relationship over a specified domain, unique to each cell. This range of valid elevations (hvalues that produce frequency [y] values between 0 and An and lie within the range of surveyed elevations) is bounded by the minimum and maximum elevation within each cell. The effects of the method of RWA and TPALR calculation were examined. These terms are given the subscript "D" when they were based on the elevation distribution (stagearea curve), the subscript "M" when they were calculated from the mean elevation, and the subscript "N" when they relied on the nominal area. The practice of using operational status to estimate the inundation status was not considered here. Statistical Analyses All statistical analyses were conducted with SAS 9.1 (SAS Institute, Cary, NC). Non parametric tests were applied as necessary when data failed to meet assumptions of normality. Results and Discussion Characterization of Elevation Distribution and Wetted Area in the STAs The CDF (equivalent to the stagearea curve, normalized for area) of ground elevation varied among cells in the STAs (Figure 21). Qualitatively, the shape and spread of the distribution of elevation in each cell alone was not correlated with longterm P removal performance. For example, despite similar elevation CDFs, STA1W Cell 1 and STA3/4 Cell 1A have shown very different POR P mass removal effectiveness (7.6% and 44.9%, respectively; Chimney, 2009). Quantitatively, the standard deviation of the elevation in each STA was not correlated with the POR P mass removal effectiveness (r2 = 0.036). The annual RWAD was less than 100% for 74/118 cellyears but was lower than 90% for only 28 of 118 cellyears (Table 21). The POR average RWAD for all STAs was 96%. Most of the low RWAD cellyears are concentrated in water years 2007 and 2008 (Figure 22), a documented period of drought in South Florida (Pietro et al., 2009). Changes in RWAD were brought on by intraannual as well as interannual precipitation cycles. For most cells, the seasonal low RWAD occurs in May (Table 22), corresponding with the end of the South Florida dry season. Relative Wetted Area and Total Phosphorus Removal Performance The TP areal settling rate was poorly but significantly correlated to the RWAD across all cells and all nonscreened months (r=0.162, p<0.0001, n=943). The relative lack of variability in the explanatory term (RWAD was greater than 95% in 89% of months; Figure 23) obscured the influence of RWAD on kTp. Considering only months where RWAD < 1.0 (that is, months when hw < hmax) increased r to 0.308 (p<0.0001, n=247). The relationship between kTp and RWAD was strongest among all of the months when RWAM < 1.0, though this filter greatly reduced the number of included observations (r=0.412, p=0.0408, n=25). This suggests that, on the whole, the RWAD influenced the TP settling rate in the STAs (months with high RWAD were more likely to also have high kTp), but the overall impact across the operating history of the whole project was minimized by the relative infrequency of months with RWAD substantially less than 1.0. Likewise, there was a weak but significant relationship between the monthly outflow TP concentration and RWAD across all cells and all nonscreened months (r= 0.181, p<0.0001, n=989). The relationship was only marginally stronger among months where RWAD < 1.0 (r=  0.211, p=0.0003, n=285). Contrary to the findings for kTP, there was no correlation between outflow TP concentration and RWAD within months with RWAM < 1 (r=0.078, p=0.6775, n=31). Apparently the RWAD has little effect on the outflow TP concentration, relative to the variation caused by all other wetland processes. Several conditions may have contributed to the relationship between kTp and RWAD. First, in those cells where deviation from RWAD = 1.0 was augmented by highly variable topography (e.g. cells in STA5, Figure 21), at any given average depth, the variance on the distribution of depths would have been higher than in more 'flatbottomed' cells (e.g. cells in STA3/4, Figure 21). The effect of extreme depths on short and longterm performance has not been quantified. Second, dryout could have caused localized death of SAV or transition of herbaceous emergent aquatic vegetation (EAV) to less desirable woody shrubs, decreasing treatment capacity upon re flooding. The vegetation records in the STAs did not have sufficient temporal or spatial resolution to test this hypothesis. Finally, the absolute RWA may have had little influence on kTp, with changes in RWA primarily affecting the apparent settling rate through oxidation and rewetting of the soil. Nonzero changes in RWA can occur only when the RWA varies from 1.0, thus the relationship observed above may be only a vestige of a connection between changing RWA and settling rate. Interpretation of these correlation coefficients requires caution however. Although the RWAD is normalized to the An of each cell, the kTp values were not normalized to the cell means. Because the mean kTP was different in each cell, it is dangerous to compare points of equivalent RWAD if they came from different cells. Within cells, the strength and sign of the correlation between kTP and RWAD vary widely and the relationship was significant only in STA5 Central Flowway (CFW; Cells 2A and 2B) and STA6 Cell 3. Possibly, perennially high RWAD lends itself to higher kTP values. Additionally, the subsets of observations resulting from each of the sequential screening criteria were not necessarily representative subsamples of the larger sample of observations. For example, of the 25 months with RWAM < 1.0 and valid kTP values, approximately half came from STA5, which accounted for only 14% of the original 943 observations. The relationship between kTP and the change in monthly RWAD (ARWAD) tells a more interesting story. With all cells and all nonscreened months included, r= 0.074 (p=0.0241, n=932). Despite the expected negative coefficient, the strength of the correlation does not strongly support the hypothesis that kTp would decrease during reflooding months. The coefficient of correlation was slightly greater (as was the level of significance) with ARWAM (r= 0.090, p=0.0059, n=932), likely because the substantial drydown and reflooding events required to shift the stage past mean may have had more of an effect on the settling rate than small events that could cause ARWAD to be different from 0. As discussed above, the RWAD infrequently varied far from 1.0. As a result, ARWAD did not often vary far from 0. Removing from the correlation those cells that never experienced RWAD # 1 improved the strength and significance of the relationship between kTP and ARWAD only slightly (r= 0.088, p=0.0167, n=745). Again, kTp was slightly more closely associated with ARWAM (r = 0.105, p=0.0039, n=745). Of course, the deleterious effects of changing RWA (as explicated in the hypothesis) are expected only in reflooding months, or months with ARWAD > 0. Screening out all months that do not meet this criterion tripled the strength of the covariation (r= 0.313, p<0.0001, n=159). This implies that, in months when reflooding occurred, about 10% of the variability of kTp was explained by variability in the extent of the reflooding event. Finally, in the few months of substantial reflooding (ARWAD > 0.1), the kTP and ARWAD were remarkably covariant (r=  0.605, p<0.0001, n=39). The large amount of variability in kTP explained by ARWAD in these large reflood months, suggests that ARWAD may be a useful variable to include in a multiple linear regression analysis to determine the factors controlling kTP (Chapter 4). The previous discussion on the caution that must be applied when interpreting data of this nature applies to these results as well. Relative Wetted Area and Total Phosphorus Mass Loading Rate Determining the importance of WA < An on the TPALR in the STAs was a surprisingly complex task. In absolute terms, the TPALRD and TPALRM must be greater than TPALRn for a cell in any given month, year, or other period of interest where the WAD < An and WAM < An, respectively. It is the size of these differences and their implications for other assessments that require an accurate measure of the TPALR that are of interest. When the three methods of TPALR calculation were treated as repeated measures on each unique cellmonth (n=1000) the means were statistically, though not substantially, different (0.275 g/m2/mo, 0.273 g/m2/mo and 0.272 g/m2/mo for TPALRD, TPALRM and TPALRn, respectively. All pairs of means significantly different atp=0.05). The large number of observations boosted the significance of these trivial differences. This finding imparts nothing except the fact that the adjustment for WA makes very little impact on the longterm mean TPALR; the precision of flow (Q in Equation (21 )) measurements does not support the estimation of the loading rate to thousandths of a gram per m2 per month. Accordingly, when the repeated measure design was removed from the analysis and the means compared directly, the calculated TPALR was not significantly influenced by the method of WA calculation when all STA cells and all months with valid TPALR data (n=1000) were included (KurskalWallis X2=0.1261, df=2, p=0.9389). In fact, the TPALR adjusted for the WA (regardless of calculation method) was not statistically different from the TPALRn. The general tendency toward complete or nearly complete flooding, again, obscured statistical differences between the adjusted TPALR and the TPALRn because the WAcorrection has no effect in months when RWAD = RWAM = 1.0. To isolate the true effects of the WA calculation on the TPALR, the three TPALR ranksums were again compared including only months where RWAD < 1.0 (n=280). The KurskalWallis X2=0.1264 (df=2, p=0.9388) which failed to allow rejection of the null hypothesis that there is no effect of WA calculation on TPALR. Even considering only the months where RWAM < 1.0 (n=38) there was no effect of the method of WA calculation (X2=2.4048, df=2, p=0.3005). Finally, the TPALR ranksums were compared considering all months (regardless of WA value) only in those cells with POR mean WAD < 0.95 (n=209; X2=0.1985, df=2, p=0.9055). In none of these cases was there a significant difference between any pair of the TPALR ranksums. Apparently, the variation in monthly TPALR due to fluctuations in flow rate and inflow TP concentration overwhelmed the slight variance contributed by the shifting RWA. This suggests, that no correction for WA < An was necessary for a sufficiently accurate accounting of the monthly TPALR in the STAs. Months with low RWAD tended to coincide with months of low flow (Figure 24). Possibly, the adjustment of TPALR for WA < An was ineffectual in the monthly data because relatively low loading rates tended to dampen the impact of the adjustment in lowRWAD months. To estimate the annual TPALR, the sum of the 12 monthly loads is divided by the average of the 12 monthly WA values. (In adopting this calculation method, the dubious assumption that the water delivered to a cell in any given month was available for delivery in any other month, as would be the case if the water was metered out of a reservoir, was accepted). In this way, a lowRWA month contributes to the calculation (increasing the calculated TPALR) even if the mass load in that month was low or negligible. However, despite these considerations, the different methods of TPALR calculation did not lead to significant differences between the ranksums. Despite the lack of significant effect on the mean TPALR, adjustment for the calculated WA would still be warranted if the correction improved the strength of the correlation between the TPALR and the outflow TP concentration. The coefficients of correlation (and pvalues) between outflow TP concentration and TPALR are shown for each method of TPALR calculation over various subsets of the data in Table 23. When all months for all cells were included, the increases in the coefficient of correlation due to WAcorrection were trivial. The largest improvements in r were found among months with RWAD < 1.0 (in all cells) and among all months in cells with POR mean RWAD < 95% (STA2 Cell 2, STA5 NFW, STA5 CFW), though in both of these cases the relationship was so weak, regardless of TPALR calculation method, as to render the slight covariation valueless. Interestingly, within every subset of observations, TPALRM was a better covariable with outflow TP concentration than was TPALRD. Of course, the relationship between TPALR and outflow TP concentration was not uniform within each cell. The results and implications of the cellbycell analysis are presented in Chapter 3. Wetlands integrate, and thus dampen, shortterm loading effects through dynamic soil and macrophyte processes. For example, the vegetation in a healthy wetland may be able to assimilate a pulse of incoming P through biomass expansion. The excess P may be released over many more r than the mass of water that carried the initial pulse, as the plants die and decompose, and a shortterm time step (e.g. month) may fail to capture the effect of TPALR on outflow TP concentration. To that end, annual outflow TP concentration was compared to annual TPALRn, TPALRM and TPALRD. The value of examining these variables over a longer time step was immediately apparent from the improvement in rvalues compared to the whole population of cellmonths. Once again, the method of WA determination was insignificant: annual outflow TP concentration was more closely related to TPALRD (r=0.476, p<0.0001, n= 11) than it was to either TPALRM (r=.475, p<0.0001, n= 11) or TPALRn (r=0.466, p<0.0001, n= 11), though only trivially. Conclusions In the STAs, the interaction of intracell topography and timevariable stage occasionally resulted in the incomplete inundation of some cells, as revealed by the characterization study. When the actual WA is less than the nominal area, exposed soils and vegetation are subject to oxidation and desiccation. In addition, the value of the ALR calculated with An will not reflect the realized ALR of the system, potentially affecting expected relationships between ALR and wetland treatment performance indices. There existed a weak positive linear relationship between kTP and RWAD and a weak negative linear relationship between kTP and ARWAD. Both correlations increased in strength with sequential screening of "unimportant" months, e.g. months with RWAD = 1.0 or ARWAD = 0. However, the number of months among which these relationships were important was very limited. Neither the TPALR nor the strength of the correlation between TPALR and outflow TP concentration was meaningfully altered by adjusting the calculated TPALR for WA < An. These nondramatic results support important conclusions nonetheless. First, the expected biogeochemical consequences of dryout and rewetting appear to have been at work in the STAs. Fortunately, from a treatment point of view, RWA infrequently varied far from 1.0, so the bulk of the TP processing in the STAs was unaffected by dryout/reflooding events. Second, An was a satisfactory approximation of WA since neither the monthly nor annual TPALR was significantly different under either alternative calculation scheme. Finally, poor performance in certain cells (elevated outflow TP concentrations relative to inflow TPALR) was shown not to be an artifact of the TPALR calculation, validating the need for additional work to diagnose those factors contributing to performance in the STAs, as presented in the following chapters. Table 21. Average annual relative wetted area by water year for each cell in the Stormwater Treatment Areas. Area (ha) 2000 2001 2002 2003 2004 2005 2006 2007 2008 Average STA1E 1628 99% 98 98 Cell 3 238 97 93 95 Cell4N 261 99 99 99 Cell 4S 304 99 100 99 Cell 5 231 96 92 94 Cell 6 425 100 100 100 Cell 7 169 100 100 100 STA1W 2699 100 100 100 100 100 99 77 69 95 96 Cell 1 603 100 100 100 100 100 100 100 69 90 95 Cell 2 381 100 100 100 100 100 100 98 100 Cell 3 415 100 100 100 100 100 100 100 55 85 93 Cell 4 145 100 100 100 100 100 87 100 73 100 96 Cell 5 1155 100 100 100 100 100 80 95 100 97 STA2 2565 92 100 98 97 96 94 96 Cell 1 728 79 100 100 100 100 97 96 Cell2 919 94 99 94 93 89 87 92 Cell 3 919 100 100 100 100 100 100 100 STA3/4 6695 99 99 97 98 Cell 1A 1230 99 98 90 96 Cell 1B 1412 99 100 100 100 Cell 2A 1029 100 99 99 99 Cell2B 1171 99 100 100 100 Cell 3A 871 97 93 99 96 Cell 3B 982 99 99 98 98 STA5 1663 84 91 95 97 91 73 69 72 84 CFW 832 79 88 93 95 94 65 63 75 81 NFW 832 89 93 97 99 87 82 76 69 87 STA6 352 93 97 94 93 70 74 87 Cell 3 99 87 94 88 88 51 51 77 Cell5 253 96 99 96 95 78 83 91 Table 22. Intraannual trends in relative wetted area. STA1E Cell 3 Cell 4N Cell 4S Cell 5 Cell 6 Cell 7 STAW Cell 1 Cell 2 Cell 3 Cell 4 Cell 5 STA2 Cell 1 Cell 2 Cell 3 STA3/4 Cell 1A Cell 1B Cell 2A Cell 2B Cell 3 STA5 CFW NFW STA6 Cell 3 Cell 5 Grand Total No. Yr. Jan 100% 2 100 2 100 2 100 2 100 2 100 2 100 97 9 93 7 100 9 89 9 100 8 100 97 6 100 6 91 6 100 100 3 100 3 100 3 100 3 100 3 100 81 8 76 8 87 84 6 70 6 89 97 Feb 100 100 100 99 100 100 100 94 91 100 89 100 95 97 98 94 100 100 100 100 100 100 99 76 68 83 82 69 87 95 Mar 99 99 100 99 97 100 100 91 89 100 89 93 88 98 100 95 100 99 100 100 100 100 97 76 73 80 86 75 90 95 May 88 62 95 98 65 100 100 91 89 96 89 84 92 87 81 78 100 91 73 100 97 100 89 65 60 69 56 23 68 87 Jul 100 100 99 100 100 100 100 100 99 100 100 100 100 96 90 97 100 100 100 100 100 100 100 92 91 92 94 84 97 98 Aug 100 100 100 100 100 100 100 100 100 100 100 100 100 98 98 96 100 100 100 100 100 100 100 94 95 93 100 100 100 99 Oct 100 100 100 100 100 100 100 100 100 100 100 100 100 98 100 94 100 99 99 99 99 99 99 96 96 96 100 100 100 99 Nov 100 100 100 100 100 100 100 99 99 100 94 100 100 98 100 94 100 98 97 97 100 98 99 93 94 93 98 97 99 98 Dec 100 100 100 100 100 100 100 98 97 100 91 100 100 98 100 95 100 100 100 100 100 100 100 90 89 90 93 88 94 98 Table 23. Changes in the coefficients of correlation for different subsets of data. 1 2 3 4 5 Months when Months when All months in cells All cells RWAD < 1.0 RWAM < 1.0 with mean RWAD < 0.95 No. data months 945 248 25 187 0.048 TPALRn 0.266 (<0.0001) 0.099(0.1214) 0.493(0.0122) 0048 (0.5154) 0.067 TPALRD 0.271 (<0.0001) 0.119(0.0614) 0.486 (0.0137) 0.067 (0.3589) 0.083 TPALRM 0.273 (<0.0001) 0.133 (0.0370) 0.516 (0.0083) 0.083 (0.2567) STA1E /, ..***" '" " I * i / , I ,' :~ *# STAiW k/l '; : ji : I r I /I 12 13 14 Feet of elevation NGVD 15 16 17 7 Cell 3 Cell 4N *** Cell 4S S Cell 5 Cell 6  Cell 7 8 9 10 11 12 13 Feet of elevation NGVD  Cell 1 Cell 2A Cell 2B *.. Cell 3  Cell4  Cell5A Cell 5B STA3/4 ' f t I 1]_ 8 9 10 11 Feet of elevation NGVD ... Cell 1  Cell 2 STA5 12 13 14 7 8 9 10 11 Feet of elevation NGVD  Cell3 Cell 1A . Cell 1B .....Cell 2B Cell 3A 12 13 14 Cell 2A  Cell 3B 9 10 11 12 13 14 15 16 7 8 9 10 11 12 13 14 Feet of elevation NGVD Feet of elevation NGVD Cell 1A . Cell 1B  Cell 2A **** Cell 2B Cell 3 Cell 5 Figure 21. Cumulative elevation distribution for each cell in the Stormwater Treatment Areas. Note that the range of each horizontal axis is the same. 10 11 100% ................. ................. ................. ........   90% ....0 80% 70% 1 STA1E (100) ...O..* STA1W (97) 60% A STA2 (96)  STA3/4 (98) 50% O0 STA5 (84)  STA6 (87) Drought 50% Figure 22. Relative wetted area in each of the Stormwater Treatment Areas. Note that the vertical axis does not extend to 0. Error bars have been omitted to increase clarity. 100% 80% 60% 40% 20% 0% 0% 20% 40% 60% 80% 100% Relative wetted area Figure 23. Histogram of monthly relative wetted area (determined by the elevation distribution) values across all nonscreened months and all included cells. 100% .o o o oo 00 *o S40% 00 0%  20% 0% '1 1' ''' 0 5 10 15 20 25 30 35 Hydraulic loading rate (m/mo) Figure 24. Monthly relative wetted area (determined by the elevation distribution) with respect to monthly hydraulic loading for all nonscreened months and all included cells. CHAPTER 3 ASSOCIATION BETWEEN LOADING RATE AND OUTFLOW CONCENTRATION IN THE STORMWATER TREATMENT AREAS Introduction Several species of SAV, macrophytes with the bulk of their biomass suspended in the water column, are common in wetlands and other water bodies in South Florida (Dierberg et al., 2002). Because of their growth habit, the mechanisms of P removal in SAV systems include two distinct differences from those in EAV wetlands. First, roots of emergent plants can access only porewater nutrients (with diffusion or mass transfer of nutrients to the sediment mediating uptake from the water column), whereas SAV obtains nutrients directly from the water column through the shoots and leaves (Graneli and Solander, 1988). This means that SAV can uptake SRP very quickly, particularly in the shortterm (Pietro et al., 2006). Second, during photosynthesis, SAV removes carbon dioxide and bicarbonate from the water column, which raises the system pH, and drives the system towards calcium carbonate (CaCO3) supersaturation, promoting CaCO3 precipitation (McConnaughey et al., 1994). Several studies have proposed coprecipitation with CaCO3 as a mechanism of P removal in a variety of systems (Scinto, 1997 reviews some of these works). It has been reasoned that SAV is more suited for P removal in treatment wetlands than EAV, a hypothesis supported by studies at mesocosm (Dierberg et al., 2002) prototype (Nungesser and Chimney, 2001) and fieldscales (Juston and DeBusk, 2006). SAV systems have generally been successful within the STAs, and now comprise over half (ca. 10,000 ha) of the STA treatment area, often in downstream positions within serial flow trains (Pietro et al., 2010). Most of the cells with the lowest longterm outflow concentrations are dominated by SAV; of the nine cells with longterm flowweighted average outflow TP concentrations below 0.030 mg/L, six are designated SAV. Though elevated relative to the levels in the historical, unimpacted Everglades, it has been suggested by way of internal profile studies that these low outflow concentrations approach C* in some SAV cells (i.e. the concentration profile reaches a plateau some fraction of the way through the cell, beyond which no additional treatment is observed; Pietro et al., 2010). The alternative P treatment mechanisms at work in SAV may be responsible for the reduction of TP to C*, possibly because, given nonlimiting calcium and light for photosynthesis, the co precipitation process is not subject to 'saturation', as soils (via sorption) and microbial and plant biomass may be, nor is it ratelimited by biotic uptake. Excessively loading a wetland with a high settling rate conceptually results in poor performance (e.g. elevated outflow concentrations, lowered percent removal). Conversely, even wetlands with low settling rates can conceptually perform well at sufficiently low loading rates. The interaction of the settling rate and loading rate determines the realized treatment performance for a wetland (assuming the inflow concentration is well above C*). It is convenient to employ the Damkohler number, which captures the "treatment potential" of a wetland by combining the settling rate with the loading rate (Kadlec and Wallace, 1996): k Da (31) q where Da = Damkohler number. By inserting Equation ( 31 ) into Equation ( 119 ), the kC* model suggests that at very high Da, C* controls the outflow concentration: C2C*=(CiC*) exp (Da) (32) Internal profile studies are reliable for determining whether a wetland is treating to C*. They are however, laborious and expensive both in the field and the laboratory. Potentially, the strength of the correlation between the outflow concentration and the TPALR may be used as a proxy to estimate if wetland effluent is at C*. Since C* is independent of the shortterm TPALR (Equations ( 119 ) and ( 21 )), in situations where the outflow TP concentration approaches C*, the correlation between the outflow TP concentration and the TPALR will be weak or absent. For example, internal profiles were estimated for a hypothetical wetland with a very high Da under four different loading scenarios, each with a different ALR (Figure 31). In each scenario, the outflow concentration was equal to C* and outflow concentration and ALR were uncorrelated. The term 'apparent background concentration' is adapted from Kadlec and Knight (1996) to describe the outflow TP concentrations from those cells with weak shortterm outflow concentrationTPALR interactions, even when those concentrations were above the range of commonly described background TP concentrations for South Florida systems of 616 [g/L (Juston and DeBusk, 2006; Kadlec and Wallace, 2008). The hypothesis that alternative P removal mechanisms elevate Da such that SAV cells tend to produce apparent TP background concentrations (while EAV cells are less likely to do so) may be tested by comparing the correlations between outflow TP concentration and TPALR within SAV cells to those in EAV cells. It should be noted, however, that even for well performing cells, the outflow concentration varied, both from cell to cell and over time. In conjunction with the previous discussion, this implies that the apparent background concentration was not fixed. In a plugflow wetland, the real P background concentration is the point of equilibrium between P availability and biogeochemical P demand. Atmospheric deposition, internal hydraulics (the degree of mixing within a wetland), TP fractionation, and the internal loading (particularly of PP and DOP) are known to contribute to the apparent background concentration (Kadlec and Wallace, 2008). Some of these factors can be easily measured (TP fractionation) and others can be inferred from easily quantifiable factors (e.g. temperature can be a proxy for seasonal changes in biomass production and senescence, both of which contribute to autochthonous P loads). In STA cells that produce outflow TP concentrations independent of the TPALR, understanding and quantifying the factors that control the apparent background concentration are important for increasing both realized performance and predictions of that performance. Objectives This chapter addresses four primary questions: 1) Were SAV cells in the STAs operating at apparent background concentrations, as determined by independence of outflow TP concentration from the TPALR? 2) Were the outflow TP concentration data uncorrelated to the TPALR in all SAV cells, or only a specific subset? 3) In the case that only certain cells are exempt from this correlation, what was the underlying cause if not the dominant vegetation? 4) What determined the outflow TP concentration in cells operating at an apparent background concentration? The above questions were answered by testing the following hypotheses: The outflow TP concentration data will be more poorly correlated to the TPALR data in SAV cells relative to EAV cells. The outflow TP concentration data will covary weakly with TPALR data in most SAV cells, and will not covary at all with TPALR in some SAV cells. The strength of the covariation will be negatively correlated to the percent cover of SAV within cells. In those cells operating at an apparent C*, the outflow concentration will be a function of one or more of the variables: TP fractionation, longterm TPALR and temperature. Methods Methods regarding the calculation or preparation of TPALR, r, TP fraction, and temperature data may be found in Chapter 1. Vegetation Each STA cell has been classified by dominant vegetation type (SAV/EAV) by SFWMD (Pietro et al. 2008). While the classifications generally capture the dominant vegetation types, and reflect the plant communities targeted by SFWMD for each cell, they do not imply 100% coverage by the indicated vegetation class. The term 'Mixed' was applied to those units composed of cells of different designations, but for which water quality data was only examined at the larger scale (e.g. the North Flowway of STA5 consists of an EAV (Cell 1A) and a SAV (Cell 1B) designated cell, but flow data were only available at the inflow and outflow of the flowway). To enhance the precision of the vegetation classification, the percent SAV cover was estimated for each cell in this study using tabulations of vegetation cover based on aerial imagery and vegetation field survey data provided by the SFWMD. The detailed (often specieslevel) designations from the map tabulations were combined into three distinct groups: EAV, SAV and open water. For this study, the SAV and open water coverages tabulated from vegetation maps were combined and assumed to approximately represent total SAV cover. The field survey data available to this project were collected by several different contractors and each employed a unique reporting system. Briefly, the coverage or abundance data at each survey point, plot or transect were normalized to a 0 to 1 scale (1 representing 100% cover). The relative coverages of each SAV species (including algae, when reported) were summed at each survey location. The nature of field surveys allowed for the distinction of SAV from open water, so open water did not contribute to the total SAV cover at each site. The values of percent SAV cover were averaged over all survey locations to estimate the total relative SAV cover for each cell. The spatial distributions of the survey points were checked to verify representative sampling within each cell. This study did not attempt to address changes in SAV cover over time. In cells where more than one estimate of SAV cover was available, the values were simply averaged, with equal weight given to survey and map data. Outflow Concentration and Total Phosphorus Areal Loading Rate The Pearson productmoment correlation coefficient (r) between monthly outflow TP concentration and monthly TPALRM (the best regressor of the three possible TPALR calculations; see Chapter 2) was determined for the POR for each cell. As r describes only the strength of the linear relationship between two variables, scatter plots were produced for each cell and visually inspected for nonlinearity. It was noted that the significant (p = 0.05) positive rvalues in several cells resulted entirely from the presence of outliers in the upperright quadrant (high outflow concentrations corresponding to high loading events). The practical significance of these relationships defined by outliers is discussed below. However, in an effort to capture the "typical" behavior of each cell, r was recalculated after excluding the month with the largest TPALR value in each cell, and termed r' for convenience. The bulk of this study relies on r' because approximately onehalf of the significant r values were artifacts of single outliers (Table 31). Results and Discussion Outflow Concentration Areal Loading Relationship Effect of vegetation type and cover The magnitude and significance of r'was not clearly determined by the vegetation designation reported for each cell in Pietro et al. (2008) (Table 31). Of the 7 SAVclassified cells examined, r'was nonsignificant in 6 (86%). Similarly, 9 (75%) of the included 12 EAV cells had nonsignificant r'. Unexpectedly, one EAV cell (STA2 Cell 1) reported a weak but significant negative covariance between outflow TP concentration and TPALR. Possible causes for this counterintuitive relationship were unclear. As a result, data from Cell 1 of STA2 were excluded from further analyses in this chapter. Three of the four (75%) 'mixed' units had non significant r '. Broadly, EAV and SAV cells were not overwhelmingly differentiated by the strength of r' (Table 31), though the caveats associated with the vegetation designations (see Methods of this chapter) depreciate the power of this assessment. To avoid the difficulties and inaccuracies associated with using the SFWMDassigned vegetation classifications, r' was plotted against estimates of the actual percent cover of SAV. The cells clustered into three distinct groups, (Figure 32). Group 1 contains cells with low to moderate SAV coverage and low r'values. The low to midSAV coverage cells, with relatively high r'values comprise Group 2. Group 3 consists of highSAV cells with nonsignificant r'. Immediately apparent from Figure 32 are the strictly nonsignificant r'values in all highSAV cells. Secondly, the unexpected tendency for nonsignificant r'values in low and moderate SAV cells is noteworthy. Together these observations do not support any conclusions about the influence of SAV P processing mechanisms on the uncoupling the shortterm outflow TP concentration from the TPALR in SAV cells, but do suggest that this method may be inappropriate for testing that hypothesis. Also, importantly, they make clear that some other factor (or factors) contributed to the disassociation of the outflow TP concentration from the monthly TPALR. Effect of areal total phosphorus loading rate Equation ( 119 ) indicates that, at sufficiently light loads, the outflow concentration from a wetland will converge on the background concentration, C*, and thus approach independence from the loading rate. This was supported by the findings of Qian and Richardson (1997) (and reinforced by Richardson and Qualls, 1999) that showed that below POR TPALR of about 1.0 g/m2/yr, the longterm average outflow concentrations from a large number of North American wetlands were fairly invariant with respect to changes in TPALR. Likewise, among SAV and select EAV cells with annual TPALR at or below 2.0 g/m2/yr, Juston and DeBusk (2006) noted "no significant relationships (p>0.05) were identified in the slopes of PALR relationships using either 1 or 2 year average ALRs, thus suggesting no evidence of association of ALR with TP concentrations in this range." It follows that the areal loads in cells in Groups 1 and 3 were possibly too low to force the outflow TP concentration above the apparent background concentration. A oneway analysis of variance was performed on the monthly TPALR to compare each of the three groups. The Fvalue was 17.19 (df=2, p<0.0001). A StudentNewman Keuls test (p=0.05) found that the mean monthly TPALR was significantly higher in Group 2 (0.42 g/m2/mo) than in either Group 1 or Group 3. The mean monthly TPALR were similar in Groups 1 and 3 (0.21 and 0.22 g/m2/mo, respectively). The implication, in the spirit of Richardson and Qualls (1999), is that the TPALR became influential on the outflow TP concentration once it exceeded some threshold. All cells with average annual TPALRM less than about 2.0 g/m2/yr showed no significant relationship between shortterm outflow TP concentration and the TPALRM (Figure 33), as expected from the work of Juston and DeBusk (2006). The presence of significant, positive r'values only in cells loaded with greater than 2.0 g/m2/yr does suggest that the magnitude of the loading rate contributed to the strength of the TPALRoutflow TP concentration relationship. The fact that some cells also loaded above 2.0 g/m2/yr had nonsignificant correlations between outflow TP concentration and TPALRM indicates the action of an additional factor (or factors) restraining r' even under large loads. Moreover, this additional factor was very likely not the alternative mechanisms of SAV systems, as the highTPALRM, lowr' group contained both EAV and SAV cells. A mention must be made of the substantial difference between r and r' in a few cells (Table 31). The extreme influence that the outliers had on the correlation between the outflow TP concentration and the TPALR justifies their exclusion for analytical purposes. Nonetheless, these data imply that, had these cells experienced higher and more variable loading rates on a monthly basis, than was observed during their recorded operation, the outflow TP concentration may have actually been influenced by the TPALR. The nearuniversal tendency for the month of maximum load to increase the strength and significance of r suggests that, regardless of the vegetation type or the shortterm outflow concentrationTPALR relationship under 'typical' operating conditions, extreme pulses of P were very likely to produce aboveaverage outflow concentrations. Factors Controlling the Apparent Background Concentration Having established that, in most cells of the STAs, the shortterm outflow TP concentration did not depend strongly on the monthly TPALR, the challenge of identifying the factors that did determine the shortterm outflow concentration arises. Curiously, the POR FWM outflow TP concentrations from those cells with nonsignificant r'values showed the same range and variability as those cells with moderate to strong TPALRoutflow TP concentration relationships (Figure 34). This chapter considers two possible controls on the longterm mean outflow TP concentration and one possible regulator of the shortterm outflow concentration. A more thorough exploration of potential factors controlling monthly outflow TP concentrations is presented in Chapter 4. Inflow phosphorus fractions The most chemo and bioavailable P forms comprise the SRP fraction of the TP pool (Kadlec and Wallace, 2008; Reddy and DeLaune, 2008). In treatment wetlands, this often leads to preferential removal of SRP relative to the TP aggregate (Dierberg et al., 2002; Chimney, 2007). Conversely, DOP is thought to be generally less bioavailable and has been found to be less effectively removed than the bulk TP in treatment wetlands (Reddy and DeLaune, 2008; Chimney, 2007), though this may be due as much to internal production of DOP as to non treatment of influent DOP (Pinney et al., 2000). Therefore, it is proposed that the composition of the inflow TP potentially contributed to the apparent background concentration in the STAs, with relative increases in SRP lowering C* and relative increases in DOP elevating C*. Although the monthly outflow TP concentration from the set of cells of interest was not correlated to the TPALR, it was weakly but measurably influenced by the inflow TP concentration: r = 0.353 (p <0.0001, n = 597 considering monthly data among all the cells of interest). The positive relationship also existed within the monthly data for 4 of the 18 included individual cells (STA1E Cell 7, STA2 Cell 3, STA3/4 Cell 1B, STA6 Cell 5). The presence of this covariance required that the effects of the TP fractionation on the outflow concentration be assessed within the effects of the inflow TP concentration. After the variability due to the inflow TP concentration was removed, neither the relative size of the SRP fraction nor the DOP fraction had any significant effect on the outflow TP concentration (SRP: Fvalue = 0.12, p = 0.7298, df = 1; DOP: Fvalue = 0.03, p = 0.8722, df=l). The effect of the inflow DOP concentration was significant (Fvalue = 6.89, p = 0.0089, df = 1), but the additional variability explained was minor (approx. 1%). The three variables, inflow DOP concentration, inflow DOP fraction and inflow SRP fraction, had a significant effect on the monthly outflow TP concentration after accounting for the inflow TP concentration in only 2, 3, and 4 cells, respectively. In addition, there was no relationship found between the outflow TP concentration and any of these three variables at either annual or POR averaging periods. In summary, both across cells and within cells, the composition of the inflow TP played a minor role, at best, in the determination of the outflow concentration from the cells where the outflow TP concentration was not related to the TPALR. Longterm areal loading Sustained high annual areal P loading is known to increase the soil P concentration. In South Florida, a wellstudied example of this phenomenon is Water Conservation Area 2A (WCA2A), where decades of P loading have resulted in elevated soil P concentrations immediately downstream from the inflows. Reddy and DeLaune (2008) provide an excellent review of the literature on WAC2A. As part of a positive feedback cycle with increasing soil P, primary productivity of wetlands tends to increase with sustained P loads (Lowe and Keenan, 1997; Kadlec and Knight, 1996). As a result, the Pprocessing "biomachine" grows in response to increased P loads. The Psequestering power (in terms of g P/m2/yr) grows as well, until biomass expansion becomes limited by space, light or nutrient(e.g. N) availability. Biomass production, senescence and decomposition are known to export dissolved nutrients to the water column (Pinney et al., 2000; Qualls and Richardson, 2002, Reddy and DeLaune, 2008). It is conceivable that increasing the size of the "biomachine" would increase the autochthonous P production and export. Thus, STA cells with high longterm TPALR may be expected to operate with relatively higher apparent background concentrations, even if in the short term, the outflow TP concentration is independent of the TPALR. It has been previously established that the expected positive relationship between outflow TP concentration and TPALR exists in the STAs when longterm (POR) averaging periods are considered; among all STA cells, the POR outflow TP concentration was positively, nonlinearly correlated to the POR TPALR through a power function (Pietro et al., 2009). The objective here is to assess if that relationship holds when considering only those cells in which the shortterm TPALR did not directly control the monthly outflow TP concentration. Among all the cells operating with nonsignificant r', the POR FWM outflow TP concentration was significantly linearly correlated to the average annual TPALR (r = 0.703, p=0.0011, n = 18; Figure 35). A power function (Pietro et al. 2009) described the data only slightly more accurately (r2 = 0.508; Figure 35). The shape of the data cloud in Figure 35 was remarkably similar to the distribution of the wide variety of wetlands studied by Richardson and Qualls (1999), including an apparent breakpoint near 1.0 g P/m2/yr. Unclear, for both the STA cells and the wetlands documented by Richardson and Qualls (1999), is why some wetlands are able to produce low outflow concentrations even under high areal loads (e.g. STA1E Cell 4N). At an annual scale, the data demonstrated a similar relationship, although the nonlinearity observed in the POR data was absent (Figure 36). Even among cells where the monthly outflow concentration was independent of the monthly TPALR, years with higher areal loads tended to also have higher outflow TP concentrations. This may be attributable to the aforementioned effect of the size of the "biomachine." The nonlinearity in the POR data may develop as the effects of consistent light or heavy loading compound over years of operation. Seasonality The bulk of the longterm sustainable P removal in wetland is biologically mediated by microorganisms, algae and submerged and emergent macrophytes (Kadlec and Knight, 1996; Kadlec, 1997; Dierberg at al., 2002). The P demand and production from these biomass compartments varies as their size and activity fluctuate in response to environmental factors (e.g. temperature and insolation) (Kadlec and Wallace, 2008). Seasonal cycles are common in treatment wetlands, but are minimized, particularly for P, in subtropical wetlands with a year round growing season (Kadlec and Wallace, 2008). Nonetheless, a distinct intraannual temperature cycle is observed in South Florida (Figure 36) that may lend itself to seasonal fluctuations in biological processing sufficient to alter treatment performance. If so, the impacts ought to be seen as an annual cyclical variation in the outflow concentration, particularly in those cells producing effluent at an apparent background concentration. Water temperature is known to control microbial activity (Reddy and DeLaune, 2008) and serves as a good proxy for the average insolation, a driver of primary productivity (Best and Visser, 1987), so any annual cyclical variation in outflow TP concentration may be a reflection of the annual water temperature cycle (Figure 36). Considering only the cells with nonsignificant r'values, the monthly outflow TP concentration did not significantly vary with either the monthly inflow or outflow water temperature (r < 0.064, n = 578). Likely, the celltocell variation in outflow TP concentration devalued this assessment. The monthly outflow TP concentration was significantly (p = 0.05) correlated to the monthly inflow water temperature in only 3 cells (positively in STA3/4 Cell 3 and STA5 CFW and negatively in STA5 NFW), and to the outflow water temperature in only 2 cells (positively in STA6 Cell 3 and negatively in STA3/4 Cell 2B). No nonlinearity was found between outflow TP concentration and water temperature in a visual inspection of scatter plots within each cell. That the relationship was significant in relatively few cells and that the conditions of the covariances were inconsistent (some positively and some negatively correlated), suggest that the water temperature in the STAs did not play a critical role in determining the apparent background TP concentration in the STAs. Water temperature was thought to be a satisfactory surrogate for seasonality because many other variables are expected to be strongly correlated to it (e.g. insolation, seasonal biomass changes). However, a flaw is recognized in this approach: because of the sinusoidal nature of temperature over time, each average monthly temperature occurs twice per year (except for the annual maximum and minimum temperature). The environmental conditions may be quite different at the two manifestations of a given temperature. For example, the average water temperature tended to be about 25 C in both April and October in STA1W Cell 4 (Figure 36), but this fails to capture the obvious biological differences between the two months (e.g. spring flush vs. fall senescence). In none of the 18 cells with low r'values was the outflow TP concentration significantly correlated to the ordinal month value (e.g. January = 1, February = 2...). Additionally, no regular nonlinear patterns were observed in scatter plots of outflow TP concentration vs. time in months. A more powerful signalprocessing method (e.g. Fourier transform) might be applied to the monthly outflow TP concentration data to better effect. Conclusions Although the POR outflow TP concentration was generally a function of the longterm average annual TPALR among all cells of the STAs (Pietro et al., 2009), as is common in treatment wetlands (Qian and Richardson, 1997), in a majority of cells the monthly outflow TP concentration was statistically independent of the monthly TPALR. Celltocell variability in the strength of the shortterm outflow concentrationTPALR relationship was not explained by the dominant vegetation type, and unsatisfactorily justified by the longterm TPALR. In the short term, the outflow TP concentration from those cells operating at an apparent background concentration was not determined by the relative fractionation of the inflow TP pool nor did it vary with season, as approximated by the water temperature. Among these cells in the longterm, the annual average TPALR accounted for approximately 51% variation in the POR FWM outflow TP concentration. Two useful implications arise from this study. First, the apparent uncoupling of the monthly outflow TP concentration from the TPALR suggests large Da in most STA cells, independent of the vegetation type. Equation ( 32 ) demonstrates that in wetlands with high Da, the realized outflow TP concentration will be controlled primarily by the apparent background concentration. Therefore, the ability for kC* to accurately model outflow TP concentrations in the STAs may depend heavily on accurate estimates of C*. Second, the notable lack of powerful descriptive variables for the shortterm outflow TP concentration resulting from this work highlights the limits of both the currently available data and the body of knowledge of wetland P processing at this time. If minimizing the outflow concentrations from the STAs remains a political and managerial priority in South Florida, smallscale (e.g. mesocosm) research ought to be applied to identify additional potential controllers of the shortterm outflow TP concentration, and largescale (e.g. field) studies will be needed to define the interactions of these variables under operational conditions. Table 31. Coefficients of correlation between monthly outflow total phosphorus (TP) concentration and monthly TP areal loading rate (ALR) within each cell before (r) and after (r') exclusion of the month of maximum TPALR. Nonsignificant (p = 0.05) coefficients are reported as 0.0. Cell STA2 Cell 1 STA1E Cell STA1E Cell STA1E Cell STA1E Cell STA1E Cell STA1E Cell STA1W Cell STA3/4 Cell STA3/4 Cell STA3/4 Cell STA3/4 Cell STA3/4 Cell STA5 CFW STA5 NFW STA2 Cell 2 STA2 Cell 3 STA6 Cell 3 STA6 Cell 5 STA1W Cell STA1W Cell STA1W Cell STA1W Cell Vegetation Designation EAV SAV SAV EAV EAV SAV EAV SAV EAV EAV EAV SAV Mixed Mixed Mixed EAV SAV EAV EAV SAV EAV Mixed EAV 0.284 0.0 0.0 0.0 0.0 0.859 0.648 0.085 0.0 0.0 0.433 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.622 0.433 0.524 0.742 0.692 0.293 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.292 0.422 0.591 0.707 C Inflow 0 A) ALR= 1.0 HLR ALR .o cone. ,B) ALR = 1.0 m/yr mg/L g/m2/yr t 0 C) ALR = 0.75 c D0 A 15 0.067 1.00 *" D) ALR = 0.38 o I**.. B 20 0.050 1.00 .****..........::. C 15 0.050 0.75 D 15 0.025 0.38 0% 20% 40% 60% 80% 100% Distance through wetland Figure 31. When a wetland treats to background concentration (C*), the areal loading rate (ALR) does not affect the outflow concentration. Group 2 Group 3 0% 20% 40% 60% 80% 100% Percent SAV cover Figure 32. Correlation between outflow total phosphorus (TP) concentration and TP mass loading rate with respect to submerged aquatic vegetation (SAV) coverage. Non significant correlations were assigned an r'value of 0. *Group 1 AGroup 2 0 Group 3 m 6 0 1I0 1 POR average annual TPALRM (g/m2/yr) Figure 33. Correlation between outflow total phosphorus (TP) concentration and TP areal loading rate (ALR) as a function of the average annual TPALR. Nonsignificant correlations were assigned an r'value of 0. Group 1 " ~' "" ~"' 0.25 S0.2 SE = 0.15 o. 0.1 o0 0 0 0.05 I. * Group 1 AGroup 2 o Group 3 A A I evalue Figure 34. Outflow total phosphorus (TP) concentration with respect to the correlation coefficient between the TP areal loading rate (TPALR) and the outflow TP concentrations. y = 0.0292x0 9777 R2 = 0.5067 y = 0.0344x 0.0004 R2= 0.4946 * U U POR annual average TPALR (g/m2/yr) Figure 35. Periodofrecord (POR) flowweighted mean outflow total phosphorus (TP) concentration as a function of the POR average annual TP areal loading rate. 0.25 0.2 0.15 0.1 0.05 4 o Low r'value High rvalue y = 0.0165x + 0.0327 R2= 0.2143 0 2 4 6 8 10 12 14 16 Annual TPALR (g/m2/yr) Figure 36. Annual flowweighted mean (FWM) outflow total phosphorus (TP) concentration as a function of the annual TP areal loading rate (ALR). Data from cells with high r' values are differentiated. The regression line considers data from low r'value cells only. o Inflow Outflow o oo J F M A M J J A S O N D Figure 37. Monthly average temperature (oC) of inflow and outflow water in STA1W Cell 4. 0 0) Sj o r So 00 E n 0 6 li f 13 E E D3 40C 30C 20C 10C CHAPTER 4 MULTIPLE LINEAR REGRESSION TO DETERMINE FACTORS CONTROLLING PHOSPHORUS CONCENTRATION AND SETTLING RATE IN THE STORMWATER TREATMENT AREAS Introduction The cardinal performance metric for the STAs is outflow TP concentration. Many studies evaluating the STAs have investigated the outflow TP concentration with respect to various wetland characteristics, such as TPALR or vegetation type (e.g. Dierberg et al., 2002; Juston and DeBusk, 2006; Pietro et al., 2009). Importantly, the enforced and proposed regulations on the STAs also consider, primarily, the effluent concentration (Pietro et al, 2008). These regulations obligate STA managers to minimize the outflow TP concentrations. Ensuring the conditions for high Davalues (high settling rates and low loading rates) is critical to that effort. Chapter 3 of this work showed that the shortterm outflow TP concentration was controlled more by C* than by TPALR or Da in most STA cells. Therefore, it is important to understand the wetland factors that contribute to the C*. Interwetland and temporal variations in outflow TP concentration have been well studied (Kadlec and Wallace, 2008). Likewise, fluctuations in outflow TP concentration have been documented in the STAs (e.g. Pietro et al., 2008). A stated or implied assumption in almost all treatment wetland work, and adopted here as well, is that the outflow TP concentration is a function of various wetland characteristics. That is, the difference in the outflow TP concentration observed for two different wetlands ought to be due to the variation of some features) between those wetlands. Similarly, the fluctuations of outflow concentration with time within a wetland are assumed to have resulted from changes in attributes of that wetland. For example, holding everything else constant, a wetland with a TP load comprised primarily of SRP may be expected to demonstrate a lower outflow concentration than an identical wetland receiving mainly DOP. In wetland science of course, it is impossible to "hold everything else constant," and so the effects of a single wetland characteristic are easily confounded. Even wetlands in sidebyside studies show some disparities in outflow concentrations (Kadlec and Wallace, 2008). Such differences are often ascribed to "stochastic variability," a lumped error term containing all of the variance due to unaccountedfor wetland properties (Kadlec and Wallace, 2008). The objective of this study was to partition the variability in the shortterm outflow TP concentration, to the maximum extent possible, to quantifiable wetland characteristics. In addition, the outflow concentration was correlated to TPALR in a few STA cells (Chapter 3), implying that Da was limiting performance. Therefore, the factors controlling the TP areal settling rate were also explored here. Being unable to isolate the effects of single variables in the field, this study attempts to do so via multiple regression using the vast datasets available for the STAs. Considering the entire pool of nonscreened months from all STA cells, the differences in means from cell to cell accounted for approximately 30% and 20% of the variation in monthly outflow TP concentrations and TP settling rates, respectively. From a managerial perspective, identifying the combination of variables responsible for these celltocell differences (as well as the factors responsible for the substantial intracell variability) may possibly allow managers to better regulate the performance of the STAs by manipulating key attributes in all or some wetland cells. In addition, understanding which macroscale wetland characteristics control the outflow concentration and the kvalue in the STAs would provide guidance for future work on the illumination of poorlyunderstood processlevel P dynamics in all subtropical surfaceflow treatment wetlands. Objectives The objective of this chapter was to assess the relative influence of a wide variety of factors on the outflow TP concentration and the TP areal settling rate in the STAs. Multiple regression methods were implemented as an initial evaluation. It is recognized that the methods employed herein were relatively elementary, but were necessary to provide guidance for future exercises in this vein. This chapter specifically addresses the question: can variation in STA performance be adequately explained by variations in the characteristics of the wetlands that comprise the STAs? Of course, the wetland properties open to investigation were limited to those for which appropriately spatially and temporallyresolved, quantitative data were available. After the first inquiry, a second question naturally follows: which factors did contribute to variation in STA performance, and are they subject to manipulation by STA managers? This chapter tested the following hypothesis: variability in the monthly outflow TP concentration and TP areal settling can be significantly explained by a subset of these factors: inflow TP concentration, inflow TP composition (fractionation), TPALR, r, soil TP concentration, mean water depth, wetland age, inflow Ca concentration, Ca ALR, Ca areal retention, water column pH, and water temperature. Methods Multiple Linear Regression Multiple regression considers linear models of the form: y, = Po + x+ 2Xi 2 + P3X31 + ... + PnXni + El (41 ) where yi = the ith observation of the dependent variable of interest, xl,Xn, = ith values of independent variables related to y, Po = estimated value ofy when xl through Xn = 0, 1iPn = parameters (coefficients) estimating the effect of each xiXn on y, and F, = ith residual. Multiple regression procedures seek to select values of PoPn which minimize the sum of the squares of the residuals. Multiple regression analyses were used to determine the most influential independent variables (xiXn in Equation (41 )) on the dependent variables monthly outflow TP concentration and monthly TP areal settling rate (kTP). Because the absolute value of kTp was not considered here, but rather the relative differences from cell to cell and month to month (Pietro et al., 2009), for simplicity, it was calculated with C* = 0: kTp = (ln() (42) where C1 = inflow concentration [M/L3], C2 = outflow concentration [M/L3], and Qi = flow rate in [L3/T], Q2 = flow rate out [L3/T], and WAD = wetted area, determined by elevation distribution [L2]. The assumption of a constant C*value for all STA cells (in this case C*=0) presents a special difficulty. This and previous chapters function on the recognition of spatially and temporally variable C*, meaning that the use of any constant C* is inappropriate. Unfortunately, the value (or function) that should be used for C* for each STA cell is yet unclear, though this chapter attempts to clarify the matter. The multiple regressions were performed using SAS 9.1 (SAS Institute, Cary, NC). Built in variableselection procedures were not viable for this dataset; these routines include only observations (months) for which a value was present for every possible factor. For example, the missing inflow Ca concentration value in STA1W Cell 5 in March 2005 would remove all the data for that month from consideration even if the inflow Ca concentration was not selected for the final model. After the extensive screening process (see Methods, Chapter 1), only about 15% of months had values for each variable. Since it was unlikely that all of the variables would prove to be useful in the final model, manual incremental construction of the model was deemed more appropriate. The regression model building approach was similar for both the outflow TP concentration and the TP areal settling rate. In all instances throughout this chapter, the adjusted coefficient of determination (r2) is reported, to enforce parsimony and appropriately penalize the measure of fit for the number of dependent variables included in each model. The manual variable selection procedure was intended to be straightforward. The r2 was determined for each univariate model. Next, bivariate models were tested using the top regressor (variable producing the highest r2) from the singlevariable models and iteratively inserting each other variable. Trivariate models were tested by iteratively introducing each remaining variable into the best (highest r2) bivariate model. This process was repeated until the marginal gains in explanatory power (increase in r2 due to each additional variable) were deemed trivial. Each cell had a different POR mean outflow TP concentration (Figure 34) and settling rate (not shown). The mean value for each cell may be thought of as an estimate of each monthly value in each cell. The variability in the cell means explained a portion of the overall variability in the monthly values. It was possible to model the effects of each cell on the dependent parameters. At each level of complexity, the model was also tested with the cell effects to assess the extent that each additional variable accounted for celltocell differences in outflow concentration and settling rate. Variables Considered The inclusion of specific variables for consideration was based primarily on data availability with a focus on wetland attributes commonly thought to influence P processing. Some wetland characteristics that may be commonly expected to influence wetland P processing were necessarily omitted for lack of quantitative data in the monthly timescale (e.g. plant biomass). Altogether, 21 quantitative variables were tested in each regression model. kC* model terms A great deal of research has supported the validity of the kC* model for application to a variety of wetlands (Kadlec and Knight, 1996; Kadlec and Wallace, 2008). As discussed previously (see Chapters 1 and 2), the model captures some intrinsic relationships between several wetland parameters. For example, as described by the model, increases in the inflow TP concentration result in elevated outflow TP concentrations. The critical model variables inflow TP concentration and q were included in the set of variables for regression. The inflow TP concentration multiplied by q yields of course the TPALR. The intuitive notion that the outflow TP concentration ought to be correlated to the TPALR has been confirmed by many studies of longterm data (e.g. Qian and Richardson, 1997; Pietro et al., 2009; Kadlec and Wallace, 2008), so the nominal (based on the findings of Chapter 2) TPALR was also included in the regression analysis. Wetlands integrate continuous loads and so outflow concentrations may reflect past operating conditions as well as recent inflow events (Juston and DeBusk, 2006; Reddy and DeLaune, 2008), so both the 1year and 2year rolling average TPALRN were incorporated in the variable set. Inflow phosphorus fractions Within wetlands, not all Pcontaining compounds are treated equally. Soluble reactive P is typically preferentially removed because it is highly bioavailable (Havens et al., 1999, Dierberg et al., 2002, Pietro et al,. 2006). Particulate P removal may be high if the wetland conditions promote settling and retention of the suspended solids (e.g Braskerud, 2002a,b). When water velocities or other turbations inhibit settling, the P associated with particles has to be enzymatically cleaved before biota can access it, so a portion of the PP may progress through the wetland, escaping removal (Dierberg and DeBusk, 2008). Finally, DOP may contain molecules that resist uptake or transformation in the wetland environment. In fact, previous studies of the Everglades and the STAs have demonstrated poor sequestration of DOP (Proctor et al., 1999; Chimney, 2007). For these reasons, the nature of the inflow TP pool was expected to affect the treatment of that TP, and so the relative proportions of each fraction were included in the regression variable set. Wetland age and soil phosphorus A small but growing number of treatment wetlands have been operated for P removal for time periods longer than a decade (Kadlec and Wallace, 2008), so the longevity of wetlands for sustained, effective P removal remains a matter of investigation. Two well known examples of treatment wetlands with longterm performance data, Houghton Lake in Michigan and the Orlando Easterly Wetland in Florida, provide some insight. Following mesocosm and pilot studies, full scale discharge of wastewater into the Houghton Lake wetland began in 1978 (Kadlec, 1993). Nearly 30 years later, the former sedge meadow was transformed into Typha marsh and was still providing positive P removal (Kadlec and Wallace, 2008). Kadlec and Wallace (2008) note that although Richardson and Marshall (1986) predicted P "saturation" of the wetland following a total loading of 1.01.5 g/m2, the wetland still provided P mass removal exceeding 80% after cumulative loading of 63 g/m2. However, the percent removal was slightly lower and much more variable in the most recent 10 years of operation than in the previous 20 years. This variation was attributed to changes in site hydrology and hydraulics (Kadlec and Wallace, 2008). Similarly, the Orlando Easterly Wetland successfully polished tertiary wastewater from the city of Orlando, Florida, from 1988 until 2003 (Sees and Jackson, 2001; Kadlec and Wallace, 2008). After 10 years of operation, a strong seasonal trend developed in the outflow TP concentration, resulting in unacceptably poor P removal effectiveness in some winter months (Wang et al., 2006). A series of studies identified that both sediment P recycling (White et al., 2002) and hydraulic inefficiencies (Martinez and Wise, 2003) contributed to the observed reduction in P removal performance. A rejuvenation project in 20022003 that included sediment removal and earthwork to increase hydraulic efficiency immediately boosted P removal performance (Wang et al., 2006). The changes in performance after longterm operation observed in the two preceding examples justify the inclusion of wetland age in the multiple regression exercises. As Wang et al. (2006) alluded to by their consideration of a host of studies to diagnose the problems experienced by the Orlando Easterly Wetland, wetland age is a "lumped" parameter, integrating the effects of many wetland characteristics that change with time (e.g. soil P, internal hydraulics and vegetation characteristics). Because soil P, in particular, is known to influence the P concentration of the overlying water (White et al., 2003; Reddy and DeLaune, 2008), it was included in the regression analyses in an effort to isolate the integrated effects of age. Data for internal hydraulics and vegetation in the STAs were not available at a temporal scale useful for this analysis. Calcium and pH Phosphorus is well known to coprecipitate with Ca in aquatic systems and wetlands (e.g. House and Donaldson, 1986; Diaz et al., 1994; Hartley et al., 1997; Scinto, 1997). In their review of the literature on the topic, Reddy and DeLaune (2008) state "retention of inorganic phosphorus by precipitation will be significant in waters with high Ca2+ and alkalinity." Indeed, in Everglades marsh soils, the accumulation of P was correlated with the accumulation of Ca, providing possible evidence of CaP interactions (Reddy et al., 1993). Therefore, inflow Ca concentration, Ca ALR, and Ca mass removal (M/L2/T) were added to regression analysis to investigate the effects of possible differences in Ca supply and processing among the STAs. In addition, inflow and outflow pH were included in recognition of the pH dependency of Ca precipitates (Diaz et al., 1994). Water temperature The rates of many microbial processes relevant to the wetland P processing are temperature dependent. For example, rates of decomposition of Everglades histosols increased dramatically with increasing temperatures (Volk, 1973). Indeed, the release of soluble P from an organic wetland soil was strongly dependent on temperature (Kadlec and Reddy, 2001). By these bases it may be expected that P removal in the STAs would also show some temperature dependence, but previous studies have suggested relatively little influence of temperature on measured P removal in wetlands (Kadlec and Reddy, 2001; Kadlec and Wallace, 2008). Nonetheless, inflow and outflow water temperature were incorporated in the multiple regression to increase the comprehensiveness of this study. Relative wetted area and water depth The biogeochemical influences of RWAD are discussed in Chapter 2. In that chapter it was found that the RWAD was very nearly 1.0 in the STAs, and therefore no significant biogeochemical impacts of the wetted area were detected. However, RWAD was included in the regression exercise in the event that it offered some explanatory power when combined with other wetland characteristics. The findings of Chapter 2 also indicated that, though infrequent, drawdown/reflood events were important determinants of the outflow TP concentration. They were therefore also included in the multiple regression analysis. Water depth influences the wetland characteristics and processes such as vegetation types, internal hydraulics and oxygen diffusion (Kadlec and Wallace, 2008). Studies on the integrated effect of water depth on P removal in wetlands associated with the STA project have produced unclear results; at high P concentrations, depth inhibited removal for cattails (Chimney et al., 2004) and SAV (Jorge et al., 2002), but did not produce a pronounced deleterious effect at low P concentrations. Despite this, the individual biogeochemical effects of water depth in wetlands call for its inclusion in the multiple regression investigation. Results and Discussion Outflow Total Phosphorus Concentration: All Cells The results of the variableselection procedure are presented in Table 41 and Figure 41. Altogether, only about 32% of the total variability in the monthly outflow TP concentration was explained by the final 5variable model. The relatively poor explanatory power of this model was unexpected since factors known to influence treatment wetland outflow concentration (e.g. inflow TP concentration) (Kadlec and Wallace, 2008) were included in the selection process. However, this finding was congruent with the very weak connection found between the TPALR and the outflow TP concentration in most cells (demonstrated in Chapter 3). The independent variables included in the final model were (in order of selection) inflow TP concentration, AWAD, r, wetland age and inflow Ca concentration. When added to these five, one other variable, inflow pH, showed a significant effect on the outflow TP concentration. However, it did not increase the strength of the model fit to the data, and was excluded for parsimony. Estimates of the parameters (coefficients) for each explanatory factor in the final fivevariable model are shown in Table 42. These values may be interpreted as the number of units of change in the outflow TP concentration reflecting one unit of change in the subject variable, holding the other wetland characteristics constant. The precise values of these estimates are somewhat sampledependent (i.e. they would change with the inclusion of an additional data year) and change with the introduction or removal of other variables in the model. The relative magnitudes (within parameters of similar units) and signs (positive or negative) are of more value for interpretation. At every level of complexity, the inclusion of the cell effects contributed a significant portion of explanatory power to the model. In fact, this additional variability explained in the outflow TP concentration appeared to be additive with the variance accounted for by the included wetland attributes. Therefore, the wetland characteristics accounting for observed differences in outflow concentration across cells were evidently not considered in this study. The reduction in r2 after the inclusion of the fifth variable was a result of marked colinearity between the cell effects and the wetland characteristics, penalized by the adjusted r2. Among the five variables in the outflow TP concentration model, several findings were of interest. First, the relationship between inflow concentration and outflow concentration was positive, as expected from Equation ( 119 ). However, the inflow concentration accounted for only 20% of the variability in the outflow concentration, in agreement with the poor outflow TP concentrationTPALR correlations found in Chapter 3, and exposing the relative influence of other wetland attributes. Second, as suggested in Chapter 2, the positive coefficient associated with changes in WA confirmed that months of reflooding tended to have higher outflow concentrations. Third, increases in inflow Ca concentration slightly decreased the outflow TP concentration, an expected relationship based on many studies of the CaP interactions in South Florida wetlands (e.g. Reddy et al., 1993; Dierberg et al., 2002). However, within this dataset, the effect of the varying inflow Ca concentrations was minor relative to the influence of the inflow TP concentration; the relative impact of increasing the inflow Ca concentration 1.0 mg/L was 3 orders of magnitude less than the effect of a 1.0 mg/L increase in the inflow TP concentration. Fourth, outflow concentrations tended to increase with wetland age. This finding, if it can be satisfactorily corroborated, has undesirable implications for the ability of the STAs to produce low outflow concentrations over time. For example, P removal in a large municipal treatment wetland in Orlando, Florida started to flag after 13 years of operation resulting in a massive rejuvenation project including muck removal and other significant earthworks (Wang et al., 2006). Indeed, sediment removal projects have also been undertaken in STA1W, the longest running STA, though not directly in response to declining P removal (Pietro et al., 2008). If ultimately necessary in additional STAs as they age, maintenance requirements of this nature will undoubtedly reduce the cost effectiveness of the STA project. However, wetland age is likely a lumped term, containing simultaneously the effects soil P concentration and other soil characteristics, plant biomass P concentration and possibly unknown others. Effectively combating any deleterious effects associated with aging STAs will require the evaluation of the isolated effects of each of these factors. Finally, the positive correlation between r and outflow TP concentration was surprising. This suggested that months with longer residence times produced relatively higher outflow concentrations. While contrary to the logic of ratebased reaction models of P reduction (e.g. the kC* model), an explanation may lie in the findings of Chapter 3. It was determined that, in the majority of STA cells, the outflow TP concentration was independent of the TPALR and therefore, it was proposed, expressed the apparent background concentration. In this case then, residence times were generally sufficient to accommodate all potential P removal. Thus, it is hypothesized that any extension of the r beyond that needed to achieve maximum treatment of inflow P simply expanded the opportunity for autochthonous P production, resulting in the slight positive relationship between the outflow TP concentration and r. Outflow Total Phosphorus Concentration: Cells with Nonsignificant r'values The variable selection procedure resulted in a trivariate model that explained about 54% of the variability in the monthly outflow TP concentration from cells with nonsignificant r'values. The final model included the following factors: change in RWAD, 1year rolling average TPALR, and wetland age. The number of variables included was limited to three because no additional factor explained a significant (p=0.05) portion of the variability in the dependent term. The regression model considering only the cells with nonsignificant r'values was different from the model for all STA cells in one considerable way. As expected, the inflow TP concentration was no longer included in the model. The presence of this term in the model for all cells was due only to the inclusion of those four cells (Cells 14 of STA1W) that had previously demonstrated a correlation between inflow and outflow TP concentration. Two important interpretations may be drawn from the three variables selected for this model. First, the movement of ARWAD to the forefront reinforces earlier findings of its significance in the STAs. Though infrequent, large reflood events appear to be one of the most powerful and predictable biogeochemical processes in the STAs. Second, the inclusion of both the 1year rolling average TPALR and wetland age indicates that the longterm operation of these STAs plays a measurable role in their shortterm performance. Both of these terms had positive coefficients meaning that the monthly outflow TP concentration tended to increase both in response to large loads throughout the preceding year and as the wetlands aged. Evidently, even though these wetlands assimilate P quickly enough to disassociate the monthly outflow from the monthly inflow concentration, the apparent background concentration is still subject to the influence of longerterm loading. A wetland achieving apparent background concentrations could presumably receive additional loads without experiencing reciprocal elevations in outflow concentration. That notion is countered by the apparent dependence of the background concentration on longterm loads. Therefore, additional loads may be expected to eventually increase the outflow concentration, even if inflow and outflow concentrations remained independent. Outflow Total Phosphorus Concentration: Single Cell This and other works have clearly established the celltocell variability in performance in the STAs (e.g. Juston and DeBusk, 2006; Pietro et al., 2008). The multiple regression exercises in this chapter presumed that variation in a particular wetland characteristic would have the same relative effect on treatment performance regardless of whether the change was observed between two different cells, or across time in a single cell. Because the results of the multicell regressions for outflow TP concentration were not wholly satisfying (i.e. the final adjusted r2 values were not above 0.54 and most included variables explained trivial proportions of the variance), it was hoped that additional clarity would be gained by regressing the data from a single cell. Cell 1 from STA1W was selected for this exercise because it had the largest number of data months (n=70) that had observations for each of the 21 tested variables. The final model had an adjusted r2 of about 0.54 and included only two independent variables, inflow TP concentration and wetland age (Table 45). When added to these two factors, no additional variable explained a significant portion of the variability in outflow TP concentration. This singlecell regression provided little additional information. A significant correlation between inflow and outflow TP concentrations in STA1W Cell 1 was established in Chapter 3, so the inclusion of inflow TP concentration was expected in the final model. Not surprisingly, the inflow TP concentration term contributed much more to this singlecell regression than it did in the previous exercise that included cells with both significant and nonsignificant r'values; the univariate regression model including only inflow TP concentration had an adjusted r2 of 0.38 for STA1W Cell 1 data only and an adjusted r2 of 0.20 for data from all cells. Serious management implications arise when wetland age appears in the regression model with a positive coefficient, but these are discussed previously in this chapter, as well as Chapter 5. Total Phosphorus Areal Settling Rate: All Cells The results of the variable selection procedure are presented in Table 43 and Figure 42. Altogether, the final sixvariable model explained about 51% of the variability in the monthly TP areal settling rate. The first variable added to the model, qN, accounted for the first 40% while, in total, the next five variables, inflow DOP fraction, inflow PP fraction, outflow temperature and TPALRN, increased the r2 of the model only about 0.10. The importance and influence of qN was reflected in the relatively high r2 of the singlevariable models with TPALRN and Ca ALRN. Variableinclusion was stopped after six factors because the addition of the last term, TPALRN, increased the explanatory power of the model by less than 1%. The r2 of the final model (0.512) was unsatisfactory, considering the effort invested in collection and compilation of the extensive dataset. The declining disparity with increasing model complexity between the r2values of the models with and without the inclusion of the cell effects suggests that the six wetland attributes in the final model explained a portion of the variability due to celltocell differences. Each additional factor accounted for a small portion of this variance, and the settling rate differences across cells were apparently not due to a single wetland characteristic. Estimates of the parameters (coefficients) for each explanatory factor in the final six variable model are shown in Table 44. As with the outflow concentration model, these values may be interpreted as the number of units of change in the TP areal settling rate reflecting one unit of change in the subject variable, holding the other wetland characteristics constant. As before, the relative magnitudes (within parameters of similar units) and signs (positive or negative) are of primary interest for interpretation. The set of variables selected for the final model was noteworthy for several reasons. First, as noted earlier, qN was the primary determinant of the monthly settling rate. Months with high hydraulic loading tended to have higher kTpvalues, holding all else constant. As with the relatively poor relationship between inflow and outflow TP concentration, this is likely a result of the Chapter 3 finding that the outflow TP concentration was independent of the TPALR in most cells. The implication that follows from that independence, that the outflow concentration reflected an apparent background concentration, and thus that maximum potential removal occurred in most months, means that the magnitude of the areal mass removal ought to have been based more or less strictly on the loading rate. This was evidenced by the inclusion of both qN and TPALRN in the final model. Second, the negative coefficients associated with both the inflow DOP and PP fractions confirm, primarily, that of the three P fractions, the STAs most effectively remove SRP. Possibly the independent effects of DOP and PP were such that the information contributed to the model by these two fractions was more useful than simply providing the SRP fraction (SRP = TP PP  DOP). Third, Pietro et al. (2009) found a negative relationship between POR outflow TP concentration and POR kTpvalue, with lower concentrations generally produced from cells with higher settling rates. That interaction may have been expressed in these data through the age term. While increasing wetland age tended to increase the outflow TP concentration, older cells had a greater likelihood of having lower kTpvalues. As with the outflow concentration model, age likely contains the combined effects of a variety of wetland features that must be separated for a thorough interpretation and assessment of the implications for the STAs. Finally, that the TP settling rate was positively related to the outflow water temperature was expected. Kadlec and Wallace (2008) reviewed the matter and their conclusion closely matched the result of this study; the TP settling rate tends to be slightly dependent on temperature, but temperature explains very little variance in the shortterm kTpvalue. Total Phosphorus Areal Settling Rate: Single Cell As was done for outflow TP concentration, a multiple regression model of kTp was assembled for the monthly data from STA1W Cell 1. The final model included only the terms water depth and wetland age and had an adjusted r2 of only 0.18 (Table 48). The selection of depth as the first factor in the model was not expected; this was the only final model in this study in which it appeared. Apparently, however, this was another expression of the dependence of kTp on q that was observed in the exercise that included all cells. In STA1W Cell 1, water depth and q were highly collinear (r=0.686, p<0.0001, n=94). It is unclear why depth proved to be a better regressor than q. Once again wetland age was a significant contributor of explained variance. The negative coefficient indicated that the monthly calculated settling rate declined as this cell aged, though the poor fit of this model (low r2) undermines the strength of any conclusions drawn from it. Note the discussion above in this chapter, as well as in Chapter 5 on the management implications of aging wetlands. Limitations and Future Application of the Multiple Regression Technique Multiple regression is attractive for analysis of the deterministic factors in wetlands because it partitions explanatory power among many variables, allowing the isolation of the effects of particular wetland characteristics. The STAs are exceptionally well suited for this technique because of the breadth and quality of the data available. However, the ability to assess a large number of factors simultaneously introduces a significant limitation to the technique. Multiple regression can only consider observations for which valid data are available for every variable included in the model being tested. This can (and in the case of the dataset considered here, did) result in loss of potentially valuable information. As model complexity increased, so did the likelihood of encountering a variable, within a given observation, for which the value had been screened. Also, the regression techniques as applied here were limited to assessing linear relationships between the dependent and independent variables. Nonlinear relationships commonly exist among wetland data at different time scales (e.g. between POR outflow TP concentration and TPALR in Pietro et al., 2009). Potentially this study failed to capture non linear, but important, relationships within this dataset. As mentioned previously, this investigation was intended to serve as a first step toward identifying wetland characteristics important to P treatment. Future researchers may contemplate several considerations to advance this technique. The inclusion of additional wetland attributes as model variables not considered in this study should be a first priority. The "lowhanging fruit" consist of both unique additional variables (e.g. herbicide application or dissolved oxygen) and further permutations of the data included here (e.g. cumulative PP areal loading). Quite possibly though, data for key variables are not currently being collected altogether or at useful spatial and temporal resolutions. Undoubtedly, any call for additional data collection at the massive scale of the STAs would need to be well defended by findings in the literature and potentially even lab or mesocosmlevel studies. Second, new and existing data should be investigated at additional time scales. Some factors (e.g. soil P or plant biomass) may not vary sufficiently over short (monthly) time scales for their effects on P treatment to be captured by regression techniques. Quarterly, annual and POR averaging periods are suggested, acknowledging that each reduction in temporal resolution will reduce the available number of observations. Finally, it is recommended to use multiple regression to investigate possible interactions between wetland characteristics. For example, hypothetically, the effect of depth on P treatment could differ between vegetation types. Including interaction terms complicates regression efforts, and it is suggested that such an investigation be initiated with expected interactions (e.g. Ca concentration and pH), and expanded to include unforeseen interactions if necessary. Conclusions Within the data available for this analysis, fluctuations in the monthly outflow TP concentration and the monthly TP areal settling rate were relatively poorly explained by any of the 21 wetland characteristics tested. In combination, the five variables inflow TP concentration, AWAD, r, wetland age and inflow Ca concentration accounted for about 32% of the variability in the monthly outflow TP concentration. About 51% of the variability in the monthly TP areal settling rate was explained by the linear model containing the six parameters qN, inflow DOP fraction, inflow PP fraction, wetland age, outflow water temperature and TPALRN. In both cases, the bulk of the explained variability was accounted for by single wetland characteristics; inflow TP concentration most strongly controlled outflow TP concentration and the settling rate was most closely related to the qN. In both models, each of the remaining variables contributed meagerly to the r2. The guiding hypothesis, that the tested variables would significantly explain the variance in the outflow TP concentration and the TP areal settling rate was rejected, in spirit if not in letter, by the findings of this chapter. The relatively large amounts of unexplained variance and the distributions of the resolved variability in both tested dependent variables support several important conclusions. First, while Chapter 3 demonstrated the importance of identifying appropriate values or functions for C* for each STA cell, this chapter revealed the challenge of doing so. Second, it is likely that additional wetland characteristics not quantified in this study were and are influential on the shortterm outflow TP concentration and the TP settling rate. Although fully accounting for all the wetland attributes that cause these performance indices to vary (e.g. r2 = 1.0) is the conceptual goal, all the error accumulating from measurement, inflowoutflow lag due to positive r, and the limitations of multiple linear regression will manifest as unexplained variance. The relative proportions of the unresolved variability in this study contributed by missing variables and other sources of error remain undetermined. Third, the shortterm measured outcomes of wetland P processing, which are known to be quite complex, may in fact depend on such an array of wetland attributes that even a majority of the variability may not be explained by only a handful of measured traits. The small individual contributions of most of the variables examined herein attests to this conclusion. In this case, the data collection and processing required to substantially or fully explain the variability in, say, outflow TP concentration, would almost certainly surpass the usefulness of this information. Readers interested in practical applications of the interpretations in this chapter regarding the potential to improve TP removal performance may find the conclusions of this chapter unsatisfying, particularly with respect to the shortterm outflow TP concentration. First, the factor that explained the majority of the variance in the monthly outflow concentration (inflow TP concentration) is not reasonably subject to manipulation by STA managers. Conceivably, inflow concentrations could be lowered by reducing P losses from upstream sources, but this approach falls outside the realm of improving wetland treatment effectiveness. Second, though the longterm outflow TP concentration and kTpvalue were negatively associated (Pietro et al., 2009), attempts to increase the settling rate by increasing q (the primary driver of the settling rate) should not be expected to depress the outflow concentration. Finally, the relative lack of influence (with regard to improvements in r2) of each of the additional variables that contributed to the outflow TP concentration and the settling rate suggests that modifying any of them would be unlikely to result in costeffective reductions in outflow concentrations. Table 41. Coefficients of determination (r2) of multiple linear regression models explaining the monthly outflow total phosphorus concentration in all cells. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by "X". The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 4 5 6 Inflow TP 0.198 X X X X X concentration TPALRA 0.070  0.287 0.304 nominal 1yr rolling 0.014  0.279 0.301  average TPALR 2yr rolling   0.260 average TPALR Inflow SRPA 0.004  fraction Inflow DOPA  fraction Inflow PPA 0.010 0.225 0.268 0.276  fraction Inflow CaA 0.021 0.215 0.266 0.290 0.323 X concentration Ca ALR      nominal Ca areal mass     removal rate Wetland age 0.032 0.212 0.258 0.301 X X qA nominal     A  0.224 0.281 X X X Mean water 0.003  depth Soil TP       concentration Inflow pH 0.004 0.212 0.258 0.289 0.303 0.323 Outflow pH   0.245    Inflow water  0.250  temperature Outflow water   0.243   temperature WADA 0.004 0.205 0.253 0.289 Change in WAD 0.028 0.247 X X X X ATPALR = TP areal loading rate; SRP = soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = calcium; q = hydraulic loading rate; r = hydraulic residence time; WAD = wetted area, determined by the elevation distribution Table 42. Estimates of parameters for the model explaining monthly outflow total phosphorus (TP) concentration in all cells. Variable Units Parameter estimate (0) Probability (p) that 0 = 0 Outflow TP concentration mg/L   Inflow TP concentration mg/L 0.4387 <0.0001 Change in wetted area % 0.2443 <0.0001 Hydraulic residence time d 0.0004 <0.0001 Wetland age yr 0.0051 <0.0001 Inflow Ca concentration mg/L 0.0004 <0.0001 Table 43. Coefficients of determination (r2) of multiple linear regression models explaining the monthly outflow total phosphorus (TP) concentration in all cells with nonsignificant r'values. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by "X". The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 Inflow TP concentration 0.056 0.477 TPALRA nominal 0.014 0.445 1yr rolling average TPALR  0.523 X 2vr rolling average TPAT.R    I SRPA .fra ct Inflow SRPA fraction Inflow DOA fraction Inflow PPA fraction Inflow CaA concentration Ca ALR nominal Ca areal mass removal rate Wetland age q nominal A Mean water depth Soil TP concentration Inflow pH Outflow pH Inflow water temperature Outflow water temperature WADA Change in WAD ATPALR = TP areal loading rate; SRP calcium; q = hydraulic loading rate; r = distribution 0.032 0.019 0.042 0.016 0.093 0.470 0.443 0.537 0.116 0.490 0.133 0.032 0.425 0.459 0.026 0.028 0.024 0.014 0.437 X X soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = hydraulic residence time; WAD = wetted area, determined by the elevation Table 44. Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in all cells with nonsignificant r'values. Variable Units Parameter estimate (0) Probability (p) that 0 = 0 Outflow TP concentration mg/L   Change in wetted area % 0.2080 <0.0001 1yr rolling average g/m2/yr 0.0365 0.0378 TPALR Wetland age yr 0.0018 0.0068 Table 45. Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in STA1W Cell 1. Variable Units Parameter estimate (0) Probability (p) that 0 = 0 Outflow TP concentration mg/L   Inflow TP concentration mg/L 0.4254 <0.0001 Wetland age yr 0.0149 <0.0001 Table 46. Coefficients of determination (r2) of multiple linear regression models explaining the monthly total phosphorus (TP) areal settling rate in all cells. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by "X". The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 4 5 6 Inflow TP 0.030 0.397 0.446   0.506 concentration TPALRA 0.277 0.395 0.448 0.474 0.489 0.512 nominal 1yr rolling 0.041 0.416 0.447 0.478 0.484  average TPALR 2yr rolling 0.044 0.393    average TPALR Inflow SRPA 0.028 0.422 0.461  fraction Inflow DOPA 0.017 0.432 X X X X fraction Inflow PPA 0.026 0.401 0.469 X X X fraction Inflow CaA 0.008 0.411   concentration Ca ALR 0.346    nominal Ca areal mass  0.401 0.462   removal rate Wetland age  0.421 0.458 0.481 0.503 X q nominal 0.391 X X X X X TA 0.117 0.387 0.437 Mean water 0.036  0.439  depth Soil TP  0.389 0.443   concentration Inflow pH 0.026 0.393 0.448 0.476 0.490 Outflow pH      Inflow water 0.038 0.396 0.445  0.489 0.504 temperature Outflow water 0.058 0.416 0.456 0.486 X X temperature WADA  0.402 0.447  0.488  Change in WAD 0.003 0.395 0.446 0.478 0.495 0.511 ATPALR = TP areal loading rate; SRP = soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = calcium; q = hydraulic loading rate; r = hydraulic residence time; WAD = wetted area, determined by the elevation distribution Table 47. Estimates of parameters for the model explaining total phosphorus areal settling rate in all cells. Variable Units Parameter estimate (3) Probability (p) that 0 = 0 TP areal settling rate m/yr   Hydraulic loading rate /m 6.6 m/mo 6.6940 <0.0001 nominal Inflow DOP fraction % 33.7914 <0.0001 Inflow PP fraction % 13.2197 0.0012 Wetland age yr 1.8512 <0.0001 Outflow water OC 0.6224 0.0064 temperature TP areal loading rate g/m2/yr 11.0970 0.0002 nominal ATP = total phosphrous; DOP = dissolved organic P; PP = particulate P Table 48. Estimates of parameters for the model explaining monthly total phosphorus (TP) areal settling rate in STA1W Cell 1. Variable Units Parameter estimate (0) Probability (p) that 0 = 0 TP areal settling rate m/yr   Water depth m 50.5583 0.0001 Wetland age yr 1.7433 0.0221  .  0 ,.[71 * TP C 1 + WAD I ICell ei lects * With Without T Age Ca C Number of variables included in model Figure 41. Changes in the coefficient of determination (r2) with increasing complexity of the model explaining monthly outflow TP concentration in all cells. The models are shown with and without the effects of the periodofrecord cell means. TP Cl = inflow TP concentration, AWAD = change in monthly wetted area, T = nominal hydraulic residence time, Age = wetland age, Ca Cl = inflow calcium concentration. c 0.6 ... ... 0.5 6 = 0.4 SCell effects 0.3 S* With 0.2  Without 0.1 0 q + DOPIR + PP1R + Temp2 + Age + TPALR , 0 1 2 3 4 5 6 Number of variables included in model Figure 42. Changes in the coefficient of determination (r2) with increasing model complexity of the model explaining the monthly total phosphorus areal settling rate in all cells. The models are shown with and without the effects of the periodofrecord cell means. q = hydraulic loading rate, DOP1R = inflow dissolved organic P fraction, PP1R = inflow particulate P fraction, Temp2 = inflow water temperature, Age = wetland age, TPALRN = nominal total phosphorus areal loading rate. CHAPTER 5 CONCLUSIONS This work attempted to address several important questions regarding the operation and performance of the Stormwater Treatment Areas in South Florida. The management of these extensive constructed wetlands is a massive expenditure for the SFWMD and is a major contributor to the reduction of P loads to the Florida Everglades. The presence of marked elevation gradients and timevariable water levels in some STA cells raised concern that estimates of the hydraulic and TP areal loading rates were inaccurate due to incomplete flooding of the nominal treatment areas. It was demonstrated that the occasions when the relative wetted area was substantially less than 100% were infrequent, and did not significantly increase the TPALR in any of the STAs, suggesting that the nominal treatment area was satisfactory for loading rate calculations. The outflow TP concentration did show a positive correlation to the magnitude of reflooding, within the months with large re flood events, an effect confirmed to be significant by multiple regression modeling. Maintaining flooding as to reduce rewetting events would prevent these occasional pulses of outflow P but this phenomenon was important in less than 5% of nonscreened months. It was also shown that poor performance in certain cells (elevated outflow TP concentrations relative to inflow TPALR) was shown not to be an artifact of the TPALR calculation, validating the need for additional work to diagnose those factors contributing to performance in the STAs. The monthly outflow TP concentration was shown to be uncorrelated to the monthly TPALR (r') in most cells. This independence was not a function of vegetation type or the magnitude of the TPALR. It was hypothesized that the Damkohler number (Da = k/q) was sufficiently high as to remove this correlation, implying that the outflow concentration was controlled primarily by the background concentration. The monthly outflow TP concentration from the cells with nonsignificant r' was not determined by the composition of the inflow TP pool nor by the outflow water temperature. More extensive efforts to isolate the factors controlling the apparent background concentrations were justified because accurate estimates of C* appeared necessary for the successful employment of the kC* model in wetlands (i.e. most STA cells) with high Davalues. Regression Variables Each of the 21 factors was specifically included because either mathematical reasoning (in the case of those terms related by the kC* model) or previously documented studies suggested their relevance to wetland P cycling (e.g. Ca, pH and temperature). Therefore, it is important to consider the conditions that led to inclusion or rejection of each in the models assembled in this work. Inflow Total Phosphorus Concentration The kC* model (Kadlec and Knight, 1996), introduced in Chapter 2, has been found by many investigators to satisfactorily describe pollutant reduction in wetlands. When using that model to consider the outflow TP concentration, the inflow TP concentration is an important input term (which makes sound intuitive sense). Chapter 3 showed that in the majority of the cells in the STAs, the outflow TP concentration in any given month was unrelated to the conditions of the inflow water in that month. This situation suggests that P removal in the most of the STAs is not limited by residence time (discussed further in Chapter 3), but does imply that the inflow TP concentration ought not to be useful in describing the shortterm outflow TP concentration. Indeed, the inflow TP concentration was only important to the regression model that considered outflow TP concentration data from cells that were known to have positive correlations between inflow and outflow TP concentrations. Clearly, the shortterm inflow TP concentration will not be an important determinant of the outflow TP concentration in any wetland system whose P reduction is not limited by residence time. Hydraulic and Total Phosphorus Loading Rates The hydraulic (q) and total phosphorus loading rates came to bear in interesting ways in the final regression models established in Chapter 4. First, both the monthly q and TPALRN contributed positively to the monthly areal settling rate when all cells of the STAs were considered together, suggesting that kTp is, in the STAs, a measure of P removal power rather than efficiency. This idea is in line with the previously discussed notion that the observed outflow TP concentrations are limited more by the apparent background concentration than by insufficient contact time. In this case, a hypothetical additional packet of water (thus additional P) does not increase the outflow concentration, but does lead to increased P mass removed. Second, the 1year rolling average TPALR was the second most important factor in determining the monthly outflow TP concentration from those cells with nonsignificant r'values. That is, even when the shortterm outflow concentration was apparently independent of the shortterm inputs, the longterm loading still influenced the shortterm outflow concentration. This is simply a different perspective on the wellunderstood connection between longterm loading and performance (e.g. Qian and Richardson, 1997; Pietro et al. 2009), but deserves consideration by treatment wetland managers. Increasing loading rates to maximize mass removal in response to the independence of shortterm inflow and outflow concentrations may result in a longterm increase in outflow concentrations, visible even in shortterm data. Inflow Phosphorus Fractions Despite appearing in the final regression model for kTp (data from all cells) the composition of the inflow TP pool did little to affect the measured performance of the STAs. Possibly, the internal transformation and production of the various forms was so substantial as to make the initial P composition irrelevant. This hypothesis gains some support from the finding that residence time did not limit P reduction in most STAs. Regardless of the inflow P composition, additional contact time would not support additional removal of the lessbioavailable DOP and PP. Therefore, the outflow DOP pool, for example, is likely not composed primarily of the same DOP particles that entered the wetland. Wetland Age and Soil Phosphorus Wetland age was the only term included in all the final regression models for both outflow TP concentration and kTp. In all cases, the sign on the coefficient indicated decreasing performance (i.e. increasing outflow TP concentrations and decreasing kTpvalues) with increasing age. Of course, there are precedents for such a scenario including the Orlando Easterly Wetland in central Florida (Wang et al., 2006) and the impacted zones of WAC2A in south Florida (DeBusk et al., 2004). Wetland age was such a prominent factor in determining STA performance probably because it is a lumped term, containing simultaneously multiple wetland characteristics that change over time, such as soil P and plant biomass and tissue P. Though wetland managers should anticipate changes in performance as treatment wetlands age, this information is only of real value if the individual variables pooled in the age term can be extracted and evaluated individually. For example, soil P, a wetland parameter expected to change as the wetland ages, was included in the regression analyses, but was not selected for any of the final models. (This analysis should not be taken as conclusive, as the soil P data were ill suited for evaluation at a monthly timestep.) Therefore, some other characteristic of the STAs was changing with time and caused the broad age term to contribute more information to the regression than the soil P term. Calcium and pH Generally, none of the calcium terms included in the multiple regression exercises (inflow Ca concentration, Ca ALR, and Ca mass removal) contributed any explanatory power regarding either outflow TP concentration or TP areal settling rate. The sole exception was the selection of inflow Ca concentration as the fifth variable in the regression model for outflow TP concentration among all cells. Likewise, neither inflow nor outflow pH was selected for inclusion in any model. The chemistry of Ca and P interactions (and the influence of pH) is moderately well understood (e.g. Diaz et al., 2004; Scinto, 1997) and associations between Ca and P have been demonstrated in the field (Reddy et al., 1993). Connecting the processlevel dynamics to the fieldscale effects has been challenging. Two factors may have contributed to the relative lack of influence of Ca on P observed in this study. First, the amount of P removed from the water column due to interaction with Ca actually may be minimal relative to biological and physical processes. The probability of this being the case is unclear; CaP interaction is apparently important in some wetlands (Reddy et al., 1993) but has not been quantified in the STAs. Second, the data possibly failed to capture the CaP association. The relatively small proportions of variance explained by most of the considered factors suggest that some of the tested datasets were not suited to analysis at a monthly timestep. Water Temperature Kadlec and Wallace (2008) suggest an Arrhenius temperature coefficient of about 1.005 for total P in warm climate wetlands. (Arrehenius coefficients greater than 1.0 indicate improvement in removal performance with increases in temperature and values less than 1.0 denote loss of performance with temperature increases. Values very close to 1.0 imply relatively little temperature dependence.) This value for TP indicates that P removal tends to improve very slightly with increases in water temperature. A temperature term (both inflow and outflow water temperature were considered) was included in the final regression model only for the monthly TP areal settling rate that included all cells. Most likely, the weak temperature dependence of P processing expected for the STAs was obscured by other wetland characteristics. Relative Wetted Area and Water Depth The RWAD was expected to influence P processing primarily by altering the realized loading rates applied to the STAs. Chapter 2 established that the nominal wetted area provided a satisfactory estimate of the actual wetted area for most applications in the STAs. Therefore, the relative unimportance of RWAD in the regression analyses was anticipated; RWAD was not selected for any of the final models assembled in this study. Of much more interest from the outset was ARWAD, since great biogeochemical action was expected to associate with drydown and reflood events. Indeed, ARWAD was the second variable included in the regression model for outflow TP concentration from all cells and was the factor of primary importance in the explaining variance in the outflow TP concentration from cells with nonsignificant r'values. Because large reflood events were fairly infrequent in the STAs (see Chapter 2), these findings indicate that the flushing of P associated with reflood events is one of the most powerful and predictable biogeochemical processes active in the STAs. Depth affects the biogeochemistry (and thus P processing) of surfaceflow wetlands by regulating oxygen diffusion and changing the hydraulics. However, it was selected for only one regression model developed in this study, where it explained about 14% of the variability in the monthly kTpvalue of STA1W Cell 1. A correlation between q and depth aligned that model with the dependence of the settling rate on the hydraulic loading rate found among all cells. It appears that the other biogeochemical effects of water depth were apparently relatively unimportant to STA performance, based on the monthly data. The longerterm role of influencing vegetation communities played by water depth was likely not captured by the monthly data. Future Work Following the results presented in this work, two major avenues of investigation remain. First, C* needs further elucidation in the STAs. Chapter 3 suggests that the monthly outflow TP concentration from the STAs was more strongly controlled by C* than kTp. This implies the need for accurate estimates of C* when modeling P removal performance in the STAs. Although the outflow concentrations from those cells operating at apparent background concentrations did very with time, it remains unclear whether a function or value will best predict C*. Second, multiple linear regression was successful in identifying some wetland characteristics that were uniquely important to P processing in the STAs. However, the overall proportions of variance in the dependent variables (monthly outflow TP concentration and monthly TP areal settling rate) that were explained by the regression efforts were unsatisfactory. It is recommended that future multiple regression efforts for the STAs consider nonlinearity among the data, as well include additional variables not tested in this study. In particular, 12 of the 21 tested variables concerned water chemistry and hydrology and hydraulics accounted for another five terms. Only one term (soil P) was related to the soil properties and no vegetation data was incorporated. Owing to the relative importance of plants and soil to wetland functions, it is likely that variables regarding these characteristics would be particularly helpful in explaining observed STA performance. APPENDIX DATA SCREENING CRITERIA Variable Inflow TP concentration Outflow TP concentration Inflow SRP concentration Outflow SRP concentration Inflow DOP concentration Outflow DOP concentration Inflow PP concentration Outflow PP concentration Inflow SRP fraction Outflow SRP fraction Inflow DOP fraction Outflow DOP fraction Inflow PP fraction Outflow PP fraction Areal TP settling rate (k) TP areal loading rate  nominal TP areal loading rate  distribution adjusted TP areal loading rate  mean adjusted Inflow Ca concentration Not reported if: Inflow volume = 0 Outflow volume = 0 Inflow volume = 0 Outflow volume = 0 Inflow SRP concentration or inflow TDP concentration blank Inflow volume < 0 Inflow SRP concentration > inflow TPD concentration Outflow SRP concentration or outflow TDP concentration blank Outflow volume < 0 Outflow SRP concentration > outflow TDP concentration Inflow TP concentration or inflow TDP concentration blank Inflow volume < 0 Inflow TP concentration > inflow TPD concentration Outflow TP concentration or outflow TDP concentration blank Outflow volume < 0 Outflow TP concentration > outflow TDP concentration Inflow SRP concentration or inflow TP concentration blank or < 0 Inflow SRP concentration > inflow TP concentration Outflow SRP concentration or outflow TP concentration blank or < 0 Outflow SRP concentration > outflow TP concentration Inflow SRP concentration or inflow TDP concentration blank or < 0 Inflow SRP concentration > inflow TDP concentration Outflow SRP concentration or outflow TDP concentration blank or < 0 Outflow SRP concentration > outflow TDP concentration Inflow TP concentration or inflow TDP concentration blank or < 0 Inflow TP concentration < inflow TDP concentration Outflow TP concentration or outflow TDP concentration blank or < 0 Outflow TP concentration < outflow TDP concentration Inflow volume, inflow TP concentration, outflow TP concentration or WAD blank or < 0 Inflow TP mass blank Inflow volume < 0 Inflow TP mass blank Inflow volume or WAD < 0 Inflow TP mass blank Inflow volume or WAN < 0 Inflow Ca mass or inflow volume blank or < 0 Inflow Ca concentration <= 1 Outflow Ca concentration Ca areal loading rate  nominal Ca areal loading rate  distribution adjusted Ca areal mass removal rate Hydraulic loading rate  nominal Hydraulic loading rate  distribution adjusted Hydraulic residence time Outflow Ca mass or outflow volume blank or < 0 Outflow Ca concentration <= 1 Inflow Ca mass blank Inflow volume < 0 Inflow Ca mass blank Inflow volume or WAD < 0 Inflow Ca mass, outflow Ca mass or WAD blank or < 0 Inflow volume blank or < 0 Inflow volume or WAD blank or < 0 Inflow volume or outflow volume < 0 Hydraulic residence time < 0 Hydraulic residence time > 150 LIST OF REFERENCES Anderson, B.H. and F.R. Magdoff. 2005. Relative movement and soil fixation of soluble organic and inorganic phosphorus. J. Environ. Qual. 34:22282233. Braskerud, B.C. 2002a. Design considerations for increased sedimentation in small wetlands treating agricultural runoff Wat. Sci. and Tech. 45:7785. Braskerud, B.C. 2002b. Factors affecting phosphorus retention in small constructed wetlands treating agricultural nonpoint source pollution. Ecol. Eng. 19:4161. Best, E.P.H. and H.W.C. Visser. 1987. Seasonal growth of the submerged macrophyte Ceratophyllum demersum L. in mesotrophic Lake Vechten in relation to insolation, temperature and reserve carbohydrates. Hydrobiologia 148:231243. Bostic, E.M. and J.R. White. 2007. Soil phosphorus and vegetation influence on wetland phosphorus release after simulated drought. Soil Sci. Soc. Am. J. 71:238244. Chimney, M.J. 2007. The role of largescale treatment wetlands in Everglades restoration: design expectations, operational history and realized performance. International Society of Limnology, Montreal, Canada. Chimney, M.J. 2009. Appendix 58: Water Budgets and Total Phosphorus Mass Balance Budgets and Treatment Performance in STA Treatment Cells and Flowways. Appendix to: Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2009. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, Florida. Chimney, M.J. and G. Goforth. 2001. Environmental impacts to the Everglades ecosystem: A historical perspective and restoration strategies. Wat. Sci. and Tech. 44:93100. Chimney, M. J., M.K. Nungesser, C. Combs, B. Gu, E. FogartyKellis, J.M. Newman, W. Wagner and L. Wenkert. 2004. Everglades Consolidated Report Chapter 4B: STA Optimization and Advanced Treatment Technologies. South Florida Water Management District, West Palm Beach, Florida. DeBusk, T.A., K.A. Grace, F.E. Dierberg, S.D. Jackson, M.J. Chimney and B.H. Gu. 2004. An investigation of the limits of phosphorus removal in wetlands: A mesocosm study of a shallow periphytondominated treatment system. Ecol. Eng. 23:114. Diaz, O.A., K.R. Reddy and P.A. Moore. 1994. Solubility of inorganic phosphorus in stream water as influenced by pH and calcium concentration. Water Res. 28:17551763. Dierberg, F.E. and T.A. DeBusk. 2008. Particulate phosphorus transformations in south Florida stormwater treatment areas used for Everglades protection. Ecol. Eng. 34:100115. Dierberg, F.E., T.A. DeBusk, S.D. Jackson, M.J. Chimney and K. Pietro. 2002. Submerged aquatic vegetationbased treatment wetlands for removing phosphorus from agricultural runoff: Response to hydraulic and nutrient loading. Water Res. 36:14091422. Froelich, P.N. 1988. Kinetic control of dissolved phosphate in natural rivers and estuaries a primer on the phosphate buffer mechanism. Limnol. Oceanogr. 33:649668. Graneli, W. and D. Solander. 1988. Influence of aquatic macrophytes on phosphorus cycling in lakes. Hydrobiologia 170:245266. Hartley, A.M., W.A. House, M.E. Callow and B.S.C. Leadbeater. 1997. Coprecipitation of phosphate with calcite in the presence of photosynthesizing green algae. Water Res. 31:2261 2268. Harwell, M.C. and K.E. Havens. 2003. Experimental studies on the recovery potential of submerged aquatic vegetation after flooding and desiccation in a large subtropical lake. Aquat. Bot. 77:135151. Havens, K.E., T.L. East, A.J. Rodusky and B. Sharfstein. 1999. Littoral periphyton responses to nitrogen and phosphorus: An experimental study in a subtropical lake. Aquat. Bot. 63:267290. Higgins, J.P., S. Hurd and C. Weil. 2000. The use of engineered wetlands to treat recalcitrant wastewaters. J. Environ. Sci. Health, Part A: Environ. Sci. Eng. 35:13091334. House, W.A. and L. Donaldson. 1986. Adsorption and coprecipitation of phosphate on calcite. J. Colloid Interface Sci. 112:309324. Jackson, J.A. and M. Sees. 2001. Rerating capacity of a constructed wetland treatment system. Wat. Sci. and Tech. 44:435440. Jorge J., J.M. Newman, M.J. Chimney, G. Goforth, T. Bechtel, G. Germain, M.K. Nungesser, D. Rumbold, J. Lopez, L. Fink, B. Gu, R. Bearzotti, D. Campbell, C. Combs, K. Pietro, N. Iricanin and R. Meeker. 2002. Everglades Consolidated Report Chapter 4: Stormwater Treatment Areas and Advanced Treatment Technologies. South Florida Water Management District, West Palm Beach, Florida. Juston, J. and T.A. DeBusk. 2006. Phosphorus mass load and outflow concentration relationships in stormwater treatment areas for Everglades restoration. Ecol. Eng. 26:206223. Kadlec, R.H. 1993. Natural wetland treatment at Houghton Lake: the first fifteen years. In Proceedings of the WEF 66th Annual Conference, Anaheim, CA. WEF, Alexandria, VA. 7384. Kadlec, R.H. 1997. An autobiotic wetland phosphorus model. Ecol. Eng. 8:145172. Kadlec, R.H. and K.R. Reddy. 2001. Temperature effects in treatment wetlands. Water Environ. Res. 73:543557. Kadlec, R.H. and S.D. Wallace. 2008. Treatment wetlands. CRC Press, Boca Raton. Lijklema, L. 1977. The role of iron in the exchange of phosphate between water and sediments. In H.L. Golterman (ed.) Interactions between sediments and fresh water. B.V. Publishers, The Hauge, The Netherlands. Lowe, E.F. and L.W. Keenan. 1997. Managing phosphorusbased, cultural eutrophication in wetlands: A conceptual approach. Ecol. Eng. 9:109118. Martinez, C.J. and W.R. Wise. 2003. Hydraulic analysis of Orlando Easterly Wetland. J. Env. Eng. 129:553560. Mcconnaughey, T.A., J.W. Labaugh, D.O. Rosenberry, R.G. Striegl, M.M. Reddy, P.F. Schuster and V. Carter. 1994. Carbon budget for a groundwaterfed lake calcification supports summer photosynthesis. Limnol. Oceanogr. 39:13191332. Mitsch, W.J., J.K. Cronk, X.Y. Wu, R.W. Nairn and D.L. Hey. 1995. Phosphorus retention in constructed freshwater riparian marshes. Ecol. Appl. 5:830845. Noe, G.B., L.J. Scinto, J. Taylor, D.L. Childers and R.D. Jones. 2003. Phosphorus cycling and partitioning in an oligotrophic Everglades wetland ecosystem: A radioisotope tracing study. Freshwat. Biol. 48:19932008. Nungesser, M.K. and M.J. Chimney. 2001. Evaluation of phosphorus retention in a South Florida treatment wetland. Wat. Sci. and Tech. 44:109115. Olila, O.G., K.R. Reddy and D.L. Stites. 1997. Influence of draining on soil phosphorus forms and distribution in a constructed wetland. Ecol. Eng. 9:157169. Otsuki, A. and R.G. Wetzel. 1972. Coprecipitation of phosphate with carbonates in a marllake. Limnol. Oceanogr. 17:763767. Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2008. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2009. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pietro, K.C., M.J. Chimney and A.D. Steinman. 2006. Phosphorus removal by the Ceratophyllum/periphyton complex in a south Florida (USA) freshwater marsh. Ecol. Eng. 27:290300. Pietro, K., G. Germain, R. Bearzotti, and N. Iricanin. 2010. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pinney, M.L., P.K. Westerhoff and L. Baker. 2000. Transformations in dissolved organic carbon through constructed wetlands. Water Res. 34:18971911. Proctor, L.M., G. Shaefer, W.M. Landing. 1999. Bioavailability of dissolved organic phosphorus in the Florida Everglades: Evaluation with native bacterial and reference enzymes. In Proceedings of the 6th Symposium on the Biogeochemistry of Wetlands, 1114 July 1999. Fort Lauderdale, Florida. Qian, S.S. and C.J. Richardson. 1997. Estimating the longterm phosphorus accretion rate in the Everglades: A Bayesian approach with risk assessment. Water Resour. Res. 33:16811688. Quails, R.G. and C.J. Richardson. 2003. Factors controlling concentration, export, and decomposition of dissolved organic nutrients in the everglades of Florida. Biogeochemistry 62:197229. Reddy K.R. and E.M. D'Angelo 1994. Soil processes regulating water quality in wetlands. In W.J. Mitsch (ed.) Global wetlands: Old world and new. Elsevier, Amsterdam, The Netherlands. Reddy, K.R. and W.F. DeBusk. 1985. Nutrient removal potential of selected aquatic macrophytes. J. Environ. Qual. 14:459462. Reddy, K.R. and R.D. DeLaune. 2008. Biogeochemistry of wetlands. CRC Press, Boca Raton, Florida. Reddy, K.R., R.D. DeLaune, W.F. DeBusk and M.S. Koch. 1993. Longterm nutrient accumulation rates in the Everglades. Soil Sci. Soc. Am. J. 57:11471155. Reddy, K.R., O.A. Diaz, L.J. Scinto and M. Agami. 1995. Phosphorus dynamics in selected wetlands and streams of the Lake Okeechobee basin. Ecol. Eng. 5:183207. Reddy, K.R., Y. Wang, W.F. DeBusk, M.M. Fisher and S. Newman. 1998. Forms of soil phosphorus in selected hydrologic units of the Florida Everglades. Soil Sci. Soc. Am. J. 62:1134 1147. Reddy, K.R., R.G. Wetzel, and R.H. Kadlec. 2005. Biogeochemistry of phosphorus in wetlands. In J.T. Sims, A.N. Sharpley (eds.) Phosphorus: Agriculture and the environment. Soil Science Society of America, Madison, Wisconsin. Richardson, C.J. 1999. The role of wetlands in storage, release and cycling of phosphorus on the landscape: A 25year retrospective. p. 4768. In K.R. Reddy, G.A. O'Connor and C.L. Schelske (eds.) Phosphorus biogeochemistry in subtropical ecosystems. CRC Press LLC, Boca Raton. Richardson, C.J. and P.E. Marshall. 1986. Processes controlling movement, storage, and export of phosphorus in a fen peatland. Ecol. Monogr. 56:279302. Scinto, L.J. 1997. Phosphorus cycling in a periphytondominated freshwater wetland. Dissertation. University of Florida. Schmid, B.H., U. Stephan and M.A. Hengl. 2005. Sediment deposition in constructed wetland ponds with emergent vegetation: Laboratory study and mathematical model. Wat. Sci. and Tech. 51:307314. Stuck, J.D., F.T. Izuno, K.L. Campbell, A.B. Bottcher and R.W. Rice. 2001. Farmlevel studies of particulate phosphorus transport in the Everglades Agricultural Area. Trans. ASAE 44:1105 1116. Tanner, C.C. 1996. Plants for constructed wetland treatment systems A comparison of the growth and nutrient uptake of eight emergent species. Ecol. Eng. 7:5983. Turner, B.L. and S. Newman. 2005. Phosphorus cycling in wetland soils: The importance of phosphate diesters. J. Environ. Qual. 34:19211929. Turner, B.L., S. Newman and J.M. Newman. 2006. Organic phosphorus sequestration in subtropical treatment wetlands. Environ. Sci. Technol. 40:727733. Volk, B.G. 1973. Everglades histosol subsidence 1: CO2 evolution as affected by soil type, temperature, and moisture. Soil Crop Sci. Soc. Fla. Proc. 32:132135. Vymazal, J. 2007. Removal of nutrients in various types of constructed wetlands. Sci. Total Environ. 380:4865. Wang, H.G., J.W. Jawitz, J.R. White, C.J. Martinez and M.D. Sees. 2006. Rejuvenating the largest municipal treatment wetland in Florida. Ecol. Eng. 26:132146. White, J.R., K.R. Reddy, W.F. DeBusk, W.R. Wise, T. Crisman. 2002. Phosphorus removal capacity of the Orlando wetland treatment system: final report. Soil and Water Science Department, University of Florida, Gainesville, Florida. BIOGRAPHICAL SKETCH Mike Jerauld's passion for the environment, which inspired the undertaking of this advanced degree, was fostered by his experiences fishing, camping and traveling as a child with his parents. His formal education in the field began with the Jupiter Environmental Research and Field Studies Academy at Jupiter High School, in South Florida. Impressed with the urgency of the need to respond to the everdeclining condition of the natural world, Mike earned his Bachelor's degree in Environmental Science from the University of Florida in 2004. His undergraduate studies introduced and developed his awareness of the intimate connection between human and environmental wellbeing. In 2008, he enrolled in the Soil and Water Science Department at the University of Florida to pursue a Master of Science that would enable him to employ wetlands for the treatment of wastewater. The quality of all aspects of Mike's term in Gainesville was marked best, perhaps, by the four national titles won by the University of Florida football and men's basketball teams, collectively, during his tenure there. In 2009, he married his extraordinary wife, Sarah. After 19 years of education, he now plans to leave academia and try his hand in "the real world." PAGE 1 FACTORS CONTROLLING PHOSPHORUS REMOVAL IN LARGE CONSTRUCTED WETLANDS IN SOUTH FLORIDA By MICHAEL JERAULD A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010 1 PAGE 2 2010 Michael Jerauld 2 PAGE 3 To the people and places in this world in need of our mercy 3 PAGE 4 ACKNOWLEDGMENTS Without a doubt, my first and sincerest thanks are due to my parents. Only with their unending support has the success of my last 24 year s been possible. I owe the same debt of gratitude to my extraordinar y wife, Sarah, who forfeited countless evenings, weekends and getaways with me, to see my completion of this degree. I recognize also th at this degree would not have been possible without the lavish love of my God. My advisor, Dr. James Jawitz, deserves no le ss than the highest praise, not only for his keen scientific eye, but for his unwavering patie nce and faultless advice as I progressed through my studies. His ability to unify cr iticism, instruction, and praise is unmatched in my experience. His influence magnified the value of all other as pects of my education at the University of Florida and all my future work will testify to the quality of his guidance. He has earned my deepest respect. Each of the members of my advisory comm ittee contributed signifi cantly and uniquely to my training. Dr. K. Ramesh Reddy generously offered his extensive understanding of wetland biogeochemistry though both his l ecture course and his critique s of my work throughout my tenure here. It was a pleasure to work closely with Dr. Reddy on the project with which I began this degree. It was perhaps Dr. Mark Clarks passion and enthusia sm for wetlands that drew me into the field. I will long measure myself agains t his remarkable energy and his drive to affect real change in this world. Dr. Mike Anna ble provided me a firm foundation in wetland hydrology that was indispensible to my work. I would be amiss to exclude my excellent fr iends and colleagues, with whom I exchanged many insightful conversations over lunches eaten on the lawn. I am indebted to Rupesh Bhomia, for his overwhelming kindness (and excellent cooking) ; to R.J. Sindelar for being ever ready for 4 PAGE 5 racquetball and cycling; to Luke Gommermann for the constant assistance he provides to those around him; to Moshe Doron for many laughs and hi s unique perspective on the world; to Davie Kadyampakeni for his undying and infectious cheerfulness; and to Alex Cheesman for demonstrating his friendship by bearing too many j okes about lift and tor ch. Each of these, and many more not mentioned here, deserves much credit for lightening the burden of graduate school. Much gratitude is also due Rajendra Pa udel and Rupesh Bhomia, once more, for their GIS and soils analyses, respectively, that enabled much of this work. Finally, I was fortunate enough to interact with several pr ofessional wetla nd scientists, each of whom was instrumental in my graduate education. Dr. Robert Knight, of Wetland Solutions Inc., graciously shared his advice and expertise and hi s passion for treatment wetlands. Any success I achieve in the field will be due, in large part, to his mentoring. Mike Korvela and Dr. Mike Chimney, of the South Florida Water Ma nagement District went to great lengths to collate and provide the vast datasets investigated in this thesis a nd offered their invaluable advice without hesitation. 5 PAGE 6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4TABLE OF CONTENTS ............................................................................................................. ....6LIST OF TABLES ...........................................................................................................................9LIST OF FIGURES .......................................................................................................................10LIST OF ABBREVIATIONS ........................................................................................................1 1ABSTRACT ...................................................................................................................... .............12 CHAPTER 1 INTRODUCTION ............................................................................................................... ..14Phosphorus Cycling in Wetlands ............................................................................................14Methods ..................................................................................................................................18Site Description ...............................................................................................................18Data Sources ....................................................................................................................19Calculations .....................................................................................................................20Hydraulic flows ........................................................................................................20Water column chemical and physical properties ......................................................21Wetland chemical and physical properties ...............................................................23Wetland performance models ...................................................................................23TimeStep Selection ........................................................................................................24Data Screening .................................................................................................................262 INFLUENCE OF WETTED AREA ON PHOSPHORUS DYNAMICS IN THE STORMWATER TREATMENT AREAS .............................................................................31Introduction .................................................................................................................. ...........31Objectives .................................................................................................................... ...........33Methods ..................................................................................................................................34Calculation of the Wetted Area .......................................................................................34Statistical Analyses .......................................................................................................... 36Results and Discussion ........................................................................................................ ...36Characterization of Elev ation Distribution and We tted Area in the STAs .....................36Relative Wetted Area and Total P hosphorus Removal Performance ..............................37Relative Wetted Area and Total P hosphorus Mass Loading Rate ..................................40Conclusions .............................................................................................................................433 ASSOCIATION BETWEEN LOADING RATE AN D OUTFLOW CONCENTRATION IN THE STOR MWATER TREATMENT AREAS ............................52 6 PAGE 7 Introduction .................................................................................................................. ...........52Objectives .................................................................................................................... ...........55Methods ..................................................................................................................................55Vegetation .................................................................................................................... ....55Outflow Concentration and Total Phosphorus Areal Loading Rate ................................57Results and Discussion ........................................................................................................ ...57Outflow ConcentrationAr eal Loading Relationship .....................................................57Effect of vegetation type and cover ..........................................................................57Effect of areal total phosphorus loading rate ...........................................................58Factors Controlling the Appare nt Background Concentration ........................................60Inflow phosphorus fractions .....................................................................................60Longterm areal loading ...........................................................................................61Seasonality ...............................................................................................................63Conclusions .............................................................................................................................65 4 MULTIPLE LINEAR REGRESSION TO DETERMINE FACTORS CONTROLLING PHOSPHORUS CONCENTRATIO N AND SETTLING RATE IN THE STORMWATER TREATMENT AREAS .............................................................................72Introduction .................................................................................................................. ...........72Objectives .................................................................................................................... ...........74Methods ..................................................................................................................................74Multiple Linear Regression .............................................................................................74Variables Considered .......................................................................................................76kC* mode l terms .....................................................................................................77Inflow phosphorus fractions ............................................................................................77Wetland age and soil phosphorus ....................................................................................78Calcium and pH ...............................................................................................................79Water temperature ...........................................................................................................80Relative wetted area and water depth ..............................................................................80Results and Discussion ........................................................................................................ ...81Outflow Total Phosphorus Concentration: All Cells .......................................................81Outflow Total Phosphorus Concentra tion: Cells with Nonsignificant r values ...........83Outflow Total Phosphorus Concentration: Single Cell ...................................................85Total Phosphorus Areal Se ttling Rate: All Cells .............................................................86Total Phosphorus Areal Se ttling Rate: Single Cell .........................................................88Limitations and Future Application of the Multiple Regression Technique ...................88Conclusions .............................................................................................................................90 5 CONCLUSIONS ................................................................................................................ ..103Inflow Total Phosphorus Concentration ........................................................................104Hydraulic and Total Phosphorus Loading Rates ...........................................................105Inflow Phosphorus Fractions .........................................................................................105Wetland Age and Soil Phosphorus ................................................................................106Calcium and pH .............................................................................................................107Water Temperature ........................................................................................................107 7 PAGE 8 Relative Wetted Area and Water Depth ........................................................................108APPENDIX: DATA SCREENING CRITERIA ..........................................................................110LIST OF REFERENCES .............................................................................................................112BIOGRAPHICAL SKETCH .......................................................................................................117 8 PAGE 9 LIST OF TABLES Table page 21 Average annual relative wetted area by wa ter year for each cell in the Stormwater Treatment Areas. ................................................................................................................4522 Intraannual trends in relative wetted area. ........................................................................4623 Changes in the coefficients of corre lation for different subsets of data. ...........................4731 Coefficients of correlation between monthly outflow total phosphorus (TP) concentration and monthly TP areal lo ading rate (ALR) within each cell. .......................6741 Coefficients of determination ( r2) of multiple linear regression models explaining the monthly outflow total phosphorus concentration in all cells. ............................................9342 Estimates of parameters for the model explaining monthly outfl ow total phosphorus (TP) concentration in all cells. ...........................................................................................9443 Coefficients of determination ( r2) of multiple linear regression models explaining the monthly outflow total phosphorus concentr ation in all cells with nonsignificant r values. ................................................................................................................................9544 Estimates of parameters for the m odel explaining outflow total phosphorus (TP) concentration in all cel ls with nonsignificant r values. ...................................................9645 Estimates of parameters for the mode l explaining outflow total phosphorus (TP) concentration in STA1W Cell 1. ......................................................................................9746 Coefficients of determination ( r2) of multiple linear regression models explaining the monthly total phosphorus (TP) area l settling rate in all cells ............................................9847 Estimates of parameters for the model explaining the monthly total phosphorus areal settling rate in all cells. ................................................................................................... ...9948 Estimates of parameters for the m odel explaining monthly total phosphorus (TP) areal settling rate in STA1W Cell 1. ...............................................................................100 9 PAGE 10 LIST OF FIGURES Figure page 11 Phosphorus (P) cycle in surface flow wetlands.. ...............................................................2812 Map showing the locations of the si x Stormwater Treatment Areas (STAs), the Everglades Agricultural Area and the Everglades Protection Area in South Florida ........2913 Schematics of the configuration of th e treatment cells within each Stormwater Treatment Area (STA) .......................................................................................................3021 Cumulative elevation dist ribution for each cell in the St ormwater Treatment Areas.. ......4822 Relative wetted area in each of the Stormwater Treatment Areas .....................................4923 Histogram of monthly relative wetted area (determined by the el evation distribution) values across all nonscreened months and all included cells. ..........................................5024 Monthly relative wetted ar ea (determined by the elevation distribution) with respect to monthly hydraulic loading for all nonsc reened months and all included cells. ...........5131 When a wetland treats to background c oncentration (C*), the areal loading rate (ALR) does not affect the outflow concentration. .............................................................6832 Correlation between outflow total ph osphorus (TP) concentration and TP mass loading rate with respect to submerge d aquatic vegetation (SAV) coverage ....................6933 Correlation between outfl ow total phosphorus (TP) c oncentration and TP areal loading rate (ALR) as a functi on of the average annual TPALR ......................................6934 Outflow total phosphorus (TP) concentration with respect to the correlation coefficient between the TP areal loading rate (TPALR) and the outflow TP concentrations. ............................................................................................................... ....7035 Periodofrecord (POR) flowweighted mean outflow total phosphorus (TP) concentration as a function of the POR average annual TP areal loading rate ..................7036 Monthly average temperature (C) of in flow and outflow water in STA1W Cell 4. .......7141 Changes in the coefficient of determination ( r2) with increasing complexity of the model explaining monthly outflow TP concentration in all cells ....................................10142 Changes in the coefficient of determination ( r2) with increasing model complexity of the model explaining the monthly total phosphorus settling rate in all cells ...................102 10 PAGE 11 LIST OF ABBREVIATIONS ALR Areal loading rate Ca MR Calcium mass retention CDF Cumulative distribution function CFW Central Flowway DIP Dissolved inorganic phosphorus DOP Dissolved organic phosphorus EAA Everglades Agricultural Area EAV Emergent aquatic vegetation NFW North Flowway P Phosphorus PIP Particulate inorganic phosphorus POP Particulate organic phosphorus POR Period of record PP Particulate phosphorus RWA Relative wetted area SAV Submerged aquatic vegetation SFWMD South Florida Wate r Management District SRP Soluble reactive phosphorus STA Stormwater treatment area TDP Total dissolved phosphorus TP Total phosphorus TPALR Total phosphorus areal loading rate WA Wetted area 11 PAGE 12 12 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science FACTORS CONTROLLING PHOSPHORUS REMOVAL IN LARGE CONSTRUCTED WETLANDS IN SOUTH FLORIDA By Michael Jerauld August 2010 Chair: James Jawitz Major: Soil and Water Science The Florida Everglades is an oligotrophic, subtropical wetland extremely susceptible eutrophication from anthropogenic phosphorus (P) inputs. Six large treatment marshes (total effective treatment area 18,000 ha), called Stormw ater Treatment Areas (STAs), have been constructed over the last fifteen years to remove P from agricultural runoff before it enters the Everglades. Because of the massive investment to build and maintain these wetlands, it is important to evaluate the factor s that contribute to th eir performance. Naturally, each STA had a different performance record. This study attempted to account for those differences by investigating 1) the relative wett ed area (RWA), 2) the relationship between the total P (TP) areal loading rate (ALR) and the outflow TP concentr ation and 3) the wetland characteristics covariant with the outflow TP concentration and the TP areal settling rate. The RWA was determined by subtracting the elevation distribution function from the average water level (stage). The RWA was less than 1.0 in 230 out of 1044 (22%) of months. The TPALR was not statistically different when calculated with the RWA rather than the nominal area; the nominal area was deemed suffici ent for most loading ra te calculations. Among months with substa ntial reflooding ( n = 39), the TP areal settling rate was negatively correlated with the magnitude of the reflooding event ( r = 0.605, p < 0.0001). PAGE 13 In most of the STAs, the monthly outflow TP concentration was uncorrelated to the monthly TPALR. The vegetation type (submerged vs. emergent) and the magnitude of the loading rate were hypothesized to contribute to this uncoupling, but neither played a quantifiable role. The cause of this independence remains unclear. Possibly, the Damkhler number (areal settling rate divided by hydrauli c loading rate) was sufficiently high in most STAs such that wetland factors other than the lo ading rate and settling rate te nded to control the outflow TP concentration. Through multiple linear regression, five variab les (inflow TP concentration, change in monthly wetted area, hydraulic residence time, wetland age and inflow Ca concentration) were found to explain 32% of the variability in the monthly outflow TP concentration. Six factors (hydraulic loading rate, inflow dissolved orga nic P fraction, inflow particulate P fraction, wetland age, outflow water temperature and TPALR) accounted for 51% of the variation in the monthly TP areal settling rate. The proportion of explained variability may be improved in future analyses by including variab les not considered herein. Additional research is needed to confidently identify the factors that control the outflow TP concentration. 13 PAGE 14 CHAPTER 1 INTRODUCTION Phosphorus (P) is a naturally occurring, abundant element required by all forms of life. In the natural environment, the bi osphere obtains P weathered fro m minerals. Typically, hydrologic flows transport P from uplands to aquatic environments, often via wetland ecotones. Excess P applied to the landscape (often as fertilizer) can jo in this migration and frequently contributes to eutrophication of downstream wate r bodies. In the landscape, medi ating wetlands can serve as P sinks, dampening the transfer of P into aquatic systems (Richardson, 1999) and natural wetlands have long been used as receiving sites for point discharges of wastewater (Kadlec and Wallace, 2008). From this insight, it follo wed that wetlands could be employed to reduce P loads to downstream ecosystems, and many wetlands have been restored or constructed for the purpose of water treatment (Kadlec a nd Wallace, 2008). Treatment wetlands as a technology have progressed from being first a novelty, past trial pi lot systems, to a point where optimization and diversity of application has beco me the focus (e.g. Mitsch et al., 1995; Kadlec and Knight, 1996; Higgins et al., 2000; Braskerud, 2002a; Turner et al., 2006; Vymazal, 2007; Kadlec and Wallace, 2008). Understanding and quantifying the internal P processing mechanisms are essential to maximizing treatment wetland P removal and retention. Phosphorus Cycling in Wetlands Phosphorus is found in wide variety of biol ogical molecules beyond the notable examples including DNA and ATP (Turner and Newman, 2005). Phosphorus is also chemically active and may be found in myriad minerals in association with Ca, Mg, Fe, Al (Reddy et al, 1999). It is convenient to classify these many forms of P by their physical, chemical or operational characteristics. Many taxonomic schemes ex ist, but four broad P pools are commonly conceptualized: dissolved inorgani c P (DIP), dissolved organic P (DOP), particulate inorganic P 14 PAGE 15 (PIP) and particulate organic P (POP) (Reddy and DeLaune, 2008). In each of these terms, dissolved and particulate distinguish par ticles that do and do not pass through a filter membrane (pore size typically, but not always, 0.45 m), respectiv ely. For methodological convenience, the following pools are often determin ed for watercolumn P: soluble reactive P (SRP), dissolved organic P (DOP) and particulate P (PP). These groups are somewhat operationally defined (i.e. the boundaries of thes e groups depend more on analytical technique than the chemical or physical properties of th e compounds). Soluble reac tive P is composed mainly of orthophosphate (PO4P) (Reddy and DeLaune, 2008) but may include some readily hydrolysable organic P (Kadlec and Wallace, 2008) Particulate P encompasses all Pcontaining molecules larger than 0.45 m. Commonly, P is not limiting in wetlands, alt hough the Florida Everglades are a wellknown exception. Phosphorus enters wetlands th rough surface water, groundwater, wet and dry atmospheric deposition, and biological tran sfers (e.g. guano production). Surface water flow dominates the P budget in many treatment we tlands (Kadlec and Wallace, 2008). Within wetlands, P is cycled between various storag e compartments at rates which depend on the physical, chemical and biological conditions ( Figure 11 ). The net effect of these P transformations determines the status of a given wetland as a source or sink of P. Wetlands have been successful as a P treatment technology because often the P processing results in net storage or P within the system. Four broad processes that contribute to P retenti on in wetlands: sorption to soil solids, sedimentation, co precipitation and biological uptake. Various physical, chemical and biological characteristics i nhibit or enhance each of these processes in a given wetland. The movements of P on and off charged site s on the surface of so il solids are called adsorption and desorption, respectively (Re ddy and DeLaune, 2008). Froe lich (1988) described 15 PAGE 16 two steps in the sorption process. First, following changes in pore water P concentration (e.g. in response to novel anthropogenic P loads) adsorption/desorption e quilibria are reached within minutes or hours (Froelich, 1988). Reddy and De Laune (2008) note The balance between P adsorption and desorption maintains the equilib rium between solid phases and P in soil pore water. Second, absorption, the solidstate diffu sion of adsorbed phosphate from the surface into the interior of particles occurs over days to months (Froelich, 1988). Generally, sorption is limited to SRP and some reactive components of DOP (Anderson and Magdoff, 2005). The net direction of the flux (on or off the soil) is controlled by the pore water P concentration and the affinity of the soil particles for P ions. In wetland s, the amount of P that can be adsorbed to the soils is often related to the amount of iron and aluminum in the soil (Lijklema, 1977). In South Florida wetlands, soil calcium is an important determinant of soil P sorption capacity (Reddy et al., 1998). Soil adsorption is not considered a sustainable P removal mechanism in treatment wetlands due to the relatively fast reaction tim e and the finite sorption capacity (Kadlec and Wallace, 2008). The inflow water to wetlands often contains suspended solids, some of which contain P. The aggregate of suspended solids often includes eroded soil particles, macrophyte detritus, and algae or other plankton cells (Stuck et al., 2001). Wetlands function to remove suspended particles primarily by reducing wa ter velocity (by low elevation gradients and drag caused by dense plant stems) such that gravity allows the particles to settle out. This removal is enhanced by the trapping of particles within benthic litter and on biofilms (Schmid et al., 2005). Sedimentation of suspended solids can account for a significant portion of total phosphorus (TP) removal, and sustainability is constrained only by changes in bo ttom elevation (due to sediment accretion) that prevent surface wate r flows (Kadlec and Wallace, 2008). 16 PAGE 17 An additional pathway of removal is the precip itation of P with Ca, Fe, Al or Mg cations (Reddy and DeLaune, 2008). It may be difficult to quantify precipita tion separately from adsorption, since the precipitates often form on the surfaces of soil particles. Reddy and DeLaune (2008) provided a thorough discussion of P cop recipitation in wetlands, particularly regarding the conditions that promote P coprecipitation wi th each of the identified cations. Generally, acidic conditions promote the cop recipitation of P with Fe and Al, and alkaline environments support the formation of PCa and PMg pr ecipitates. In wetlands, both apatite (Ca5(Cl)(PO4)3) and hydroxylapatite (Ca5(OH)(PO4)3) are notable precipitates (R eddy and DAngelo, 1994). Coprecipitation with Ca may be pa rticularly important for P dynami cs in the Everglades; Reddy et al. (1993) found a linear correlation between P and Ca accumulation in Everglades soils. Apparently promoted by the consumption of carb onate ions by submerged photosynthesizers, the precise mechanism of the PCa interaction may be either adsorption of P onto the surface of CaCO3 precipitates or the formati on of mixed crystals during coprecipitation (Otsuki and Wetzel, 1972; Scinto, 1997). The P requirement of all organisms comb ined with the high productivity of wetland primary producers makes biological uptake an important mechanism of wetland P removal. Algae and other microorganisms can consume significant amounts of P very rapidly; for example, a diverse algal community reduced me socosm water column P concentrations from 1,100 g/L to 50 g/L in 28 days (Havens et al., 1999). Plants and microor ganisms typically can utilize only SRP directly. Other forms of P must first be hydrolyzed before they can be taken up biologically. In wetlands, much of this SRP is promptly transformed into PP (Noe et al., 2003). Many studies have investigated the P uptake pot ential of wetland plants (e.g. Reddy and DeBusk, 1985; Tanner, 1996). Nearly all of the P incorporated into microbial biomass and most of the 17 PAGE 18 macrophyteP is returned to the P cycle thr ough decomposition (Reddy et al., 1995; Kadlec and Wallace, 2008). However, the anaerobic conditi on resulting from flooding slows decomposition and promotes organic matter accumulation. The P stored in refractory biomass compounds contributes to the longter m sustainable P removal by bur ial in accrued sediments. The conditions supporting and enhancing each of these P removal mechanisms vary from wetland to wetland depending in large part on soil characteristics, vegetation type and density and water column chemistry, incl uding cation availabil ity and the distribution of the TP pool among the various functional P forms. Maximizing P treatment in wetlands requires an understanding and quantification (and manipulation) of the relative co ntributions of each of these processes to net P removal. Methods Site Description Over the past 15 years, 6 treatment marshes, called Stormwater Treatment Areas (STAs) have been constructed in South Florida to captur e P from agricultural runoff before it enters the Florida Everglades, an oligotrophic wetland susceptible to anthropogenic eutrophication (Chimney and Goforth, 2001). Many investigations have demonstrated P enrichment and the associated disruption of the existing ecosystem s in the Everglades (see Reddy and DeLaune, 2008, for a comprehensive review of these works). In particular, most notable to the lay observer, is the shift from saw grass ( Cladium jamaicense Crantz) prairies to dense monotypic cattail ( Typha spp.) stands. The Everglades Agricu ltural Area (EAA), comprising 280,000 ha and approximately 27% of the original Everglades expanse, is a rich producer of sugarcane and winter vegetable crops (Reddy and DeLaune, 2008), and the primary source of P to the Everglades and to the STAs (Pietro et al., 2009). 18 PAGE 19 The STAs are strategically located, spatially and hydrologically, between the EAA and the remnants of the Everglades now broadly defined by the Everglades Protection Area ( Figure 12 ). Altogether, the 6 STAs have a total foot print of over 26,000 ha, with more than 18,000 ha of effective treatment area, subdivided into 35 cells ( Figure 13 ; Pietro et al., 2009). The balance of the land area is consumed by roads, levees, pu mp stations and other in frastructure. As of 2009, the STAs had retained over 1,200 metric tons of P since the inception of the Everglades Nutrient Removal Project (the forerunner to the STA project) in 1994 (Pietr o et al., 2010). Over the same time period, flowweighted mean TP concentrat ion was reduced from 0.143 mg/L to 0.040 mg/L (Pietro et al., 2010). This report took advantage of the water and P mass balance data compiled by Chimney (2009). It was therefore constraine d to the cells for which data were reported in Chimney (2009). Excluded cells are indicated in Figure 13 Some currently subdivided cells were treated as combined larger units (e.g. Cell 1A and Cell 1B of STA5 were considered together as the North Flowway (NFW)). This was necessary to main tain longer records in cells that had been subdivided after startup or in cases where data were not available at the levee between subcells. Data Sources South Florida Water Management District ( SFWMD) staff collect daily flow and weekly or biweekly water quality data for all STA cells The data were retrie ved from the publiclyaccessible, online database, DBH YDRO, maintained by SFWMD. Data from topographic surveys of ea ch STA cell were provided by SFWMD. Vegetation coverage data, incl uding maps generated from aer ial images and field survey results were produced and collected, respectivel y, by various contractor s hired by SFWMD. The data were made available to this project by SFWMD. 19 PAGE 20 Calculations Several physical and chemical characteristics of the STAs were important to multiple aspects of this study. For convenience, the me thods employed to calculate each relevant parameter are described here. Hydraulic flows Daily inflow and outflow volumes were summed to determine the total flows for each period of interest: Q 1= Q 1t t 2t1 ( 11 ) Q 2= Q 2t t 2t1 ( 12 ) where t1 = starting date of the period of interest, t2 = end date of the period of interest, Q 1 = total inflow volume in the period of interest[L3/T], Q 2 = total outflow volume in the period of interest[L3/T], Q 1t = inflow volume on day t [L3/T] and Q 2t = outflow volume on day t [L3/T]. The hydraulic loading rate ( q) is the rainfall equivalent [L/T] of the inflow volume: q= Q 1 A ( 13 ) where A = wetland area [L2]. Chapter 2 provides a discussion of methods for calculating A The nominal hydraulic residence time ( ) is an estimate of the travel time [T] required for an average packet of water to pass through the wetland: = Ah Q 1 = h q ( 14 ) where h = mean water depth [L]. 20 PAGE 21 Water column chemical and physical properties Weekly and biweekly composite samples were linearly interpolated to estimate daily TP concentrations. Similarly, weekly and biweekly grab samples were linearly interpolated to estimate daily SRP, total dissolved P (TDP) and Ca concentrations. These daily values were flowweighted to calculate average inflow and out flow concentrations for larger time steps. The same method of flowweighting wa s employed for all constituents: C 1= Q 1tC 1t t 2t1 Q 1t t2t1 ( 15 ) C 2= Q 2tC 2t t 2t1 Q 2t t2t1 ( 16 ) where C 1 = flowweighted mean inflow concentr ation in the period of interest [M/L3], C 2 = flowweighted mean outflow concentra tion in the period of interest [M/L3], C 1t = estimated inflow concentration on day t [M/L3] and C 2t = estimated outflow concentration on day t [M/L3]. In cells with multiple inflow or outflow stations concentration values were again flowweighted by station to estimate a single average value for each cell. Inflow and outflow concentrations of th e P forms DOP and PP were calculated by difference using measured TP, TDP and SRP values: C 1DOP= C 1TDPC 1SRP ( 17 ) C 2DOP= C 2TDPC 2SRP ( 18 ) C 1PP= C 1TPC 1TDP ( 19 ) C 2PP= C 2TPC 2TDP ( 110 ) where C 1DOP = inflow DOP concentration [M/L3], C 1PP = inflow PP concentration [M/L3], C 1SRP = inflow SRP concentration [M/L3], C 1TDP = inflow TDP concentration [M/L3], C 1TP = inflow TP concentration [M/L3], C 2DOP = outflow DOP concentration [M/L3], C 2PP = outflow PP 21 PAGE 22 concentration [M/L3], C 2SRP = outflow SRP concentration [M/L3], C 2TDP = outflow TDP concentration [M/L3] and C 2TP = outflow TP concentration [M/L3]. The relative proportions of each of these forms within the TP pool were calculated using the inflow and outflow concentrations: SRP1R= C 1SRPC 1TP ( 111 ) SRP2R= C 2SRPC 2TP ( 112 ) DOP1R= C 1DOPC 1TP ( 113 ) DOP2R= C 2DOPC 2TP ( 114 ) PP1R= C 1PPC 1TP ( 115 ) PP2R= C 2PPC 2TP ( 116 ) where SRP1R = inflow SRP fraction, SRP2R = outflow SRP fraction, DOP1R = inflow DOP fraction, DOP2R = outflow DOP fraction, PP1R = inflow PP fraction and PP2R = outflow PP fraction. The same procedure was used to calculate ar eal loading rates (ALR) for both TP and Ca: ALR= Q 1tC 1t t 2t1A ( 117 ) Chapter 2 provides a discussion of methods for calculating A The mass retention rate of Ca was calculated for each STA cell: CaMR= Q 1tC 1t t 2t1Q 2tC 2t t 2t1A ( 118 ) where CaMR = Ca mass retention rate [M/L2/T]. 22 PAGE 23 Weekly and biweekly inflow and outflow temperature and pH readings were linearly interpolated to estimate daily values of each. The daily values were averaged arithmetically (not flowweighted) to generate monthly figures. Wetland chemical and physical properties Wetland age for any given time period was cal culated as the numb er of whole years between the date of the time period of interest an d the initiation of operation. That is, the current age of each wetland was not reflected back to the beginning of the period of record (POR), but rather the actual age of the we tland at each time step was used. Soil TP concentrations from each sampling event were assumed to be spatially representative of each cell, so th e values were arithmetically aver aged to obtain a single value for each cell for each sampling event. The resulting valu e was applied to the entire year in which the sampling occurred. Annual values were linearly interpolated across years in which sampling events did not take place. Estimates were not ex trapolated to years before the first sampling event or to years after the last sample collection in each cell. The daily average water depth was estimated by subtracting the elevation cumulative distribution function (CDF) from the daily mean water surface elevation (stage). A detailed description of the process by wh ich the elevation CDF was obtaine d for each cell is available in Chapter 2. The resulting function descri bes the continuous cumulative distribution of depths in the wetland and was compartmenta lized into 0.5 ft depth increments for convenience. The mean water depth was the areaweighted av erage of these depth increments. Wetland performance models The kC* model is commonly used to predict the outflow concentra tion of contaminants from wetlands (Kadlec and Knight, 1996): 23 PAGE 24 C2C*=( C1C*) exp ( k A Q ) ( 119 ) where A = wetland area [L2], C1 = inflow concentration [M/L3], C2 = outflow concentration [M/L3], C = background concentration [M/L3], k = contaminant areal settling rate [L/T], and Q = flow rate [L3/T]. Equation ( 119 ) can be configured to include depth, such that it considers the volume rather than the surface area of the wetland (Kadlec and Knight, 1996): C2C*=( C1C*) exp ( A Q ) =( C1C*) exp (kv ) ( 120 ) where kv = contaminant volumetric rate constant [T1]. Kadlec and Wallace (2008) noted that most wetland contaminant removal processes are typically apportioned to wetland area to a great er extent than to wetland water volume. Further, they found that kv decreased with increasing depth, and advise that the areal settling rate is more appropriate for most situations. Thus, the areabased form of the kC* model (Equation ( 119 ) ) is considered throughout this document. The selection of k over kv has an additional implication. Because of the negative depth dependence of kv, the increases in associated with increases in depth do not result in commiserate improvements in performance. Therefore, only the changes in associated with changes in q (Equation ( 14 ) ) effect changes in wetland performance, so q may be considered as a proxy for TimeStep Selection With respect to the ultimate purpose of the STAs, to reduce the P load to downstream ecosystems, it is the longterm outfl ow TP concentration that is jus tly of interest to regulators on behalf of the Everglades. Practice has shown that the longterm outflow TP concentrations varied 24 PAGE 25 among cells over the history of the project. Nu merous studies have been undertaken to understand the root of these differences and to explore options fo r improving treatment effectiveness and decreasing outflow concentrati ons. In this and other studies comparing the STAs to one another, (POR) averages are valuable but limiting in that the number of values is restrained by the number of cells (or STAs, depending on the design of the study). Frequently, the characteristics that potentially differentiate STA cells from one another also vary over time within cells (e.g. depth, Ca con centration, hydraulic lo ading, etc). By incr easing the temporal resolution of the data, the number of data values available to explore correlative trends expands. In addition, many of the wetland biogeochemical m echanisms relevant to P processing occur at short time scales (minutes to days), suggesti ng that shorter periods of time averaging might elucidate valuable pro cesslevel information. Conversely, the nonzero observed in all wetlands dictate that at any instant, the outflow water is independent of the inflow water. Al so, due to the imperfect hydraulics ubiquitous in wetlands, the outflow water at any instant compri ses a mixture of water that entered the wetland a range of previously, thereby preventing the simple comparison of inflow waters with waters exiting precisely one mean later. These two phenomena ne cessitate an upper limit on the resolution of the time step. Kadlec and Wallace ( 2008) suggested an averaging period of at least three nominal to ensure that the bulk of the outflow water considered in a given data value entered the wetland during that same period. For the reasons presented above, many analyses throughout this work aimed to assess the workings of the STAs in the shortterm. De termining the appropriate time step required striking a balance between the aforementioned comp eting considerations fo r shortversus longterm averaging, while maintaining data manageabili ty. All of the data included in this study were 25 PAGE 26 provided by SFWMD, either directly or th rough the publiclyaccessibl e, online database, DBHYDRO, which SFWMD populates and maintains. The flow and water quality data were organized by cell at a monthly time step s o, given the data at hand, the minimum possible averaging period was 30 d by default. Followi ng the prescription of Kadlec and Wallace (2008), the monthly calculated from this data were examined. The average monthly was 23 d. In only 44% of the nonscreened months was the less than 10 d. The was less than 30 d in only 77% of months. Together, these findings suggest th at the monthly time step was too fine. The additional finding that 95% of months had s less than 90 d suggested that a quarterly (90 d) averaging period was more appropriate. However, upon the initial conversion of select data to the quarterly time step, it was f ound that the correlations between the outflow TP concentration and both the inflow TP concentration and the TP ALR were slightly stronger among the monthly data than the quarterly values. Ultimately, it is the physical connection between the inflow and outflow water, as measured by the solutes in that water, which must be maintained by selecting a sufficiently long averaging period. Because the 90 d time step failed to measurably strengthen the inflowoutflow relationshi p, the 30 d averaging period was selected. Throughout this work, each use of the phrase shortterm i ndicates the use of monthly data. Data Screening When analyzing data for relationships betw een known dependent variables and possible independent variables, it is common practice to screen data points generated by unusual operating conditions in the system of interest. For example, within treatment wetland science, convention calls for the exclusion of data from the startup period, th e time immediately after initiation of treatment, when temporary nutri ent sinks (plant biomass expansion and soil sorption) may exaggerate observed treatment performance (Kadlec and Wallace, 2008). Also, some authors elect to omit periods of internal maintenance (e. g. Pietro et al., 2008; Juston and 26 PAGE 27 DeBusk, 2006). Within this analysis, the former etiquette was observed implicitly; the POR for each cell was selected to corre spond with the data range included in Chimney (2009), who omitted startup years therein. Extremely low flow events created difficulties for the data analysis required by this study. For example, very low hydraulic loading causes the calculated to become very large (exceeding tens of thousands of days in some cases). Li kewise, exceptionally low flows (in or out) can generate extreme calculated TP mass removal valu es. In an effort to suppress such extravagant values (which, being outliers, have unique power to disrupt correlation a nd regression analyses), while simultaneously maintaining a large number of data points, all months in which neither the inflow nor outflow volume was at least 10% of the respective longterm average were screened. Of the original 1419 data months, 1050 (74%) remained after this criteria was applied. Some variables required a dditional special considerati on. The appendix lists these variables and the conditions that resulted in the omissi on of individual values. 27 PAGE 28 Figure 11. Phosphorus (P) cycle in surface flow wetlands. DIP = particulate inorganic P, DOP = dissolved organic P, PIP = particulate i norganic P, POP = part iculate organic P. Image: Reddy and DeLaune, 2008. 28 PAGE 29 29 Figure 12. Map showing the loca tions of the six Stormwater Treatment Areas (STAs), the Everglades Agricultural Area and the Ever glades Protection Area in South Florida. Image: Pietro et al., 2008. PAGE 30 Figure 13. Schematics of the configuration of the treatment cells within each Stormwater Treatment Area (STA). The dominant vegetatio n type in each cell is also indicated. Cells marked with X were not included in this study. Image: Pietro et al., 2010. 30 PAGE 31 CHAPTER 2 INFLUENCE OF WETTED AREA ON PHOSPHORUS DYNAMICS IN THE STORMWATER TREATMENT AREAS Introduction Performance, measured by any metric (outfl ow concentration, concentration reduction, or settling rate) varied across the STAs, with some cells and STAs having removed P less effectively than their peers (P ietro et al., 2009). The total expe nditures on the STA project are difficult to estimate, but the construction, monitoring and management of treatment wetlands with a combined footprint of over 26,000 ha (Pie tro et al., 2009) is an enormous financial undertaking. As such, poor P removal performance by an STA is unaffordable by the SFWMD as well as unacceptable for the preservation of downstream ecosystems. As part of a larger diagnostic exercise to elucidate the control ling factors behind measured performance (and ultimately to advise SFWMD on management st rategies to improve performance in trouble STAs), the relationship between TP areal loading rate (TPALR) and outflow TP concentration was explored. Terms A C1, and Q from Equation ( 119 ) are related by: ALR = C 1* Q A ( 21 ) where ALR = areal loading rate [M/L2/T]. It can easily be seen that in the kC* model, any change in A C1, or Q that would cause ALR to increase, would correspondingly cause the outflow concentration ( C2) to increase. In other words, the outflow concentration of a contaminant is directly proportional to the ALR of that contaminant. It is common for C1 and Q to vary over the daily, monthl y and even annual operation of a treatment wetland. Less preval ent are changes in the treatment ar ea. As a result, ALR is typically calculated for the nominal area ( An), that is, the design or built wetland area. However, in certain 31 PAGE 32 circumstances, the actual wetted area (WA) may not be equal to An. All wetlands have a distribution of elevations arising from both macroscale ground slope and microscale topographic heterogeneity. The size of the STAs enha nces this variation on both scales, and also precludes grading, which, in many constructed wetlands, minimizes topographic heterogeneity. Most wetlands, including the STAs experience temporal variability in flow, driven primarily by regional rainfall patterns. During times of low flow, particularly during seasonal or regional drought, insufficient water may be delivered to the wetland to maintain the stage above the maximum ground elevation. When this occurs, WA will be less than An. The interaction of the spatial elevation distribution and the temporal stage distribution controls the WA in the STAs. Often, estimates of k are derived from Equation ( 119 ) when the other variables are known, so that treatment performance can be co mpared across wetlands with diverse inputs. Clearly, when WA < An, the use of A = An in Equation ( 21 ) would result in suppressed estimates of k. Some STA cells did experience periodic drydown conditions in whole or in part (Pietro et al., 2008) due to variability in weather conditions in the tributary basi ns, therefore previous estimates of k and ALR (of specific relevance to this study, kTP and TPALR) that incorporated An may have been inaccurate. It was hypothesized th at poor performance (specifically as measured by the outflow TP concentration relative to the in flow TPALR), was, in so me circumstances, an artifact of imprecise calculations. For example, hypothetical wetlands A and B have nominal TPALR of 1 g/m2/yr and observed outflow TP concentrations of 0.075 and 0.10 mg/L, respectively. Investigation reveal s that only half of the nominal area of wetland B was actually flooded, due to topographic and stage variability The revised TPALR in wetland B becomes 2 g/m2/yr, and it becomes evident that some porti on of the poor performance (high outflow 32 PAGE 33 concentration relative to loadi ng rate) of wetland B was a product of the data used in the ALR calculation. If the hypothesis held tr ue, an extension of that logi c would suggest that realized performance (absolute outflow concentration, rega rdless of inlet loading) could be improved by increasing the flooded area of candi date wetlands (while maintain ing historical flow rates) through earthwork to reduce th e topographic variability. Accurate estimation of the changes in WA over time is important beyond assuring correct k and ALR calculations. The drying of portions of cells is of pa rticular concern for submerged aquatic vegetation (SAV) which may die if desi ccated (Harwell, 2003). However, White et al. (2006) subjected SAV mesocosms to 1month periods of dry out (the condition of the vegetative communities upon reflooding was not reported), and found net treatment of TP (outflow concentration lower than inflow concentration) to resume 03 weeks following reflooding. Nonetheless, it may be difficult to maintain SAV in treatment cells that regularly or intermittently dry out. Additionally, reflooding of exposed sediments re sults in a flux of P out of the sediments into the water column (Olila et al., 1997, White et al., 2006; Bostic and White, 2007; Pietro et al., 2008 and 2009). Thus, the changing WA due to stage and topographic interaction would be expected to reduce treatment effectiveness. Objectives This chapter attempts to answer three primar y questions: 1) Was the RWA, or changes in RWA, significantly correlated to kTP? 2) Did WA < An significantly alter the TPALR realized by any cell in the STAs? If so, does the use of A = WA, rather than A = An, in Equations ( 119 ) and ( 21 ) usefully improve the correlation between out flow TP concentration and TPALR? 3) Is there significant value added by calculating WA vi a the elevation distribution, as compared to simpler methods? The following hypotheses were tested to address the questions above: 33 PAGE 34 The longterm TP removal performance will be ne gatively correlated with variation in elevation within cells. The shortterm TP removal performance will be positively correlated with RWA, and negatively correlated with monthly RWA, in months when RWA was positive. The STAwide TPALR calculated by substituting A = WA into Equation ( 21 ) will be significantly higher than the TPALR calculated using A = An. The correlation between the monthly TPALR and th e monthly outflow TP concentration will be usefully higher when Equation ( 21 ) is evaluated using A = WA derived from the stagearea curve, than when Equation ( 21 ) is solved using either A = WA derived from the mean elevation or A = An. Methods Calculation of the Wetted Area Brown and Caldwell (1996) formalized the calc ulation of the relative wetted area (RWA): RWA = 1 (t2t1) A dt t 2t1 ( 22 ) where t1 = start of time period, and t2 = end of time period. Note that RWA WA / An. The RWA was calculated for each individual cell, and areaweighted averages were calculated at larger spatial scales. The cell areas reported in Pietro et al. (2009) were used for An. Approximations for the integral portion of Equation ( 22 ) can be made by any one of three methods. The simplest approach, employed by SFWM D (Pietro et al., 2009), uses the operational status of each cell as a proxy for inundation stat us on any given day, where online cells are fully inundated (WA = An) and offline cells are fully dry (WA = 0) and then the values of WA for all days in the period of interest are summed and then substituted for the integral portion of Equation ( 22 ) Alternatively, WA(t) can be estimated by (Pietro et al., 2010): 34 PAGE 35 WA(t) = Anif hwi hmean0if hwi < hmean ( 23 ) where hwi = average elevation of water surface (stage) on day i [L] and hmean = mean elevation of the ground surface [L]. Again, the WA values for al l days in the period of interest are summed and applied to Equation ( 22 ) In some STAs, particularly those with wide elevation ranges or highly pulsed inflows, the methods of SFWMD ma y not accurately estimate the effective area. Finally, WA(t) may be estimated from a stagearea curve, a function that relates the flooded wetland area to the elevation of the water surf ace. This final method of WA calculation was selected for this analysis as it qu antifies the extent of flooding when hmin < hw < hmax, as opposed to the flooded/dry dichotomy created by the tw o other methods. To c onstruct the stagearea curve for each cell, topographic da ta from the most recent, or othe rwise most reliable, survey of each STA were interpolated using the kriging method in ArcGIS 9.2 (ESRI, Redlands, CA). Extreme values (e.g. tops of levees, bottoms of ditches) included in the original survey data were excluded from interpolation. From the resulti ng continuous bathymetric map, ArcGIS returned tabulated frequencies of the elevations in each cell. These frequencies were transformed into cumulative elevation distributions, the vertical axes of which were multiplied by An. Sixthorder polynomial curves were fit to the resulting poin ts to estimate the stagearea curve: y = ah6 + bh5 + ch4 + dh3 + eh2 +fh + g ( 24 ) where h = elevation, ft NGVD29, y = area of cell with elevation x ag = constants, unique for each cell. Sixthorder curves generally provided strong fits to the data, though the large number of unique coefficients for each cell was somewhat burdensome. Equation ( 24 ) was solved for h = hwi, returning the WA for each day i The sum of the resulting valu es served as an estimate of the integral portion of Equation ( 22 ). 35 PAGE 36 The nature of these highorder polynomials dict ates that they only describe the stagearea relationship over a specified domain, unique to eac h cell. This range of valid elevations ( hvalues that produce frequency [y ] values between 0 and An and lie within the range of surveyed elevations) is bounded by the minimum and maximum elevation within each cell. The effects of the method of RWA and TPAL R calculation were examined. These terms are given the subscript D when they were based on the elevation di stribution (stagearea curve), the subscript M when they were calculated from the mean elevation, and the subscript N when they relied on the nominal area. The pr actice of using operation al status to estimate the inundation status was not considered here. Statistical Analyses All statistical analyses were conducted w ith SAS 9.1 (SAS Institute, Cary, NC). Nonparametric tests were applied as necessary when data failed to meet assumptions of normality. Results and Discussion Characterization of Elevation Distri bution and Wetted Area in the STAs The CDF (equivalent to the stagearea curv e, normalized for area) of ground elevation varied among cells in the STAs (Figure 21). Qualitatively, the shape and spread of the distribution of elevation in each cell alone was not correlat ed with longterm P removal performance. For example, despite similar elevation CDFs, STA1W Cell 1 and STA3/4 Cell 1A have shown very different POR P mass remo val effectiveness (7.6% and 44.9%, respectively; Chimney, 2009). Quantitatively, the standard deviation of the elevati on in each STA was not correlated with the POR P mass removal effectiveness ( r2 = 0.036). The annual RWAD was less than 100% for 74/118 cellyears but was lower than 90% for only 28 of 118 cellyears ( Table 21 ). The POR average RWAD for all STAs was 96%. Most of 36 PAGE 37 the low RWAD cellyears are concentrated in wa ter years 2007 and 2008 (Figure 22), a documented period of drought in Sout h Florida (Pietro et al., 2009). Changes in RWAD were brought on by intraannual as well as interannual precipitation cycles. For most cells, the seasonal low RWAD occurs in May ( Table 22 ), corresponding with the end of the South Florida dry season. Relative Wetted Area and Total Phosphorus Removal Performance The TP areal settling rate was poorly but significantly corre lated to the RWAD across all cells and all nonscreened months ( r =0.162, p<0.0001, n=943). The relative lack of variability in the explanatory term (RWAD was greater than 95% in 89% of months; Figure 23 ) obscured the influence of RWAD on kTP. Considering only months where RWAD < 1.0 (that is, months when hw < hmax) increased r to 0.308 ( p<0.0001, n=247). The relationship between kTP and RWAD was strongest among all of the months when RWAM < 1.0, though this filter greatly reduced the number of included observations ( r =0.412, p=0.0408, n=25). This suggests that, on the whole, the RWAD influenced the TP settling rate in the STAs (months with high RWAD were more likely to also have high kTP), but the overall impact across the operating history of the whole project was minimized by the relative infrequency of months with RWAD substantially less than 1.0. Likewise, there was a weak but significant re lationship between the monthly outflow TP concentration and RWAD across all cells and al l nonscreened months ( r = 0.181, p<0.0001, n=989). The relationship was only margina lly stronger among months where RWAD < 1.0 ( r = 0.211, p=0.0003, n=285). Contrary to the findings for kTP, there was no correlation between outflow TP concentration and RWAD within months with RWAM < 1 ( r =0.078, p=0.6775, n=31). Apparently the RWAD has little effect on the outflow TP co ncentration, relative to the variation caused by all other wetland processes. 37 PAGE 38 Several conditions may have contri buted to the relationship between kTP and RWAD. First, in those cells where deviation from RWAD = 1.0 was augmented by hi ghly variable topography (e.g. cells in STA5, Figure 21), at any given average depth, the variance on the distribution of depths would have been higher than in more flatbottomed cells (e.g. cells in STA3/4, Figure 21). The effect of extreme depths on shorta nd longterm performance has not been quantified. Second, dryout could have caused localized death of SAV or transition of herbaceous emergent aquatic vegetation (EAV) to less desirable woody shrubs, decreasing treatment capacity upon reflooding. The vegetation records in the STAs di d not have sufficient temporal or spatial resolution to test this hypothesis. Finally, the absolute RWA ma y have had little influence on kTP, with changes in RWA primarily affecting the appare nt settling rate through oxidation and rewetting of the soil. Nonzero changes in RWA can occur only when the RWA varies from 1.0, thus the relationship observed above may be only a vestige of a connection between changing RWA and settling rate. Interpretation of these correlation coefficients requires caution however. Although the RWAD is normalized to the An of each cell, the kTP values were not normalized to the cell means. Because the mean kTP was different in each cell, it is dange rous to compare points of equivalent RWAD if they came from different cells. Within cells, the strength and sign of the correlation between kTP and RWAD vary widely and the relationship wa s significant only in STA5 Central Flowway (CFW; Cells 2A and 2B) and ST A6 Cell 3. Possibly, perennially high RWAD lends itself to higher kTP values. Additionally, the subsets of obs ervations resulting from each of the sequential screening criteria were not necessarily representative subsamples of the larger sample of observations. For example, of the 25 months with RWAM < 1.0 and valid kTP values, 38 PAGE 39 approximately half came from STA5, which accounted for only 14% of the original 943 observations. The relationship between kTP and the change in monthly RWAD ( RWAD) tells a more interesting story. With all cells a nd all nonscreened months included, r = 0.074 ( p=0.0241, n=932). Despite the expected negative coefficien t, the strength of th e correlation does not strongly support th e hypothesis that kTP would decrease during reflooding months. The coefficient of correlation was slightly greater (as was the level of significance) with RWAM ( r = 0.090, p=0.0059, n=932), likely because the substantia l drydown and reflooding events required to shift the stage past hmean may have had more of an e ffect on the settling rate than small events that could cause RWAD to be different from 0. As discussed above, the RWAD infrequently varied far from 1.0. As a result, RWAD did not often vary far from 0. Removing from the correlation those cells that never experienced RWAD 1 improved the strength and significance of the re lationship between kTP and RWAD only slightly ( r = 0.088, p=0.0167, n=745). Again, kTP was slightly more closely associated with RWAM ( r = 0.105, p=0.0039, n=745). Of course, the deleterious effects of cha nging RWA (as explicated in the hypothesis) are expected only in reflooding months, or months with RWAD > 0. Screening out all months that do not meet this criterion tripled the strength of the covariation (r = 0.313, p<0.0001, n=159). This implies that, in months when refloodi ng occurred, about 10% of the variability of kTP was explained by variability in the extent of the reflo oding event. Finally, in the few months of substantial reflooding ( RWAD > 0.1), the kTP and RWAD were remarkably covariant ( r = 0.605, p<0.0001, n=39). The large amount of variability in kTP explained by RWAD in these large reflood months, suggests that RWAD may be a useful variable to include in a multiple linear regression analysis to determine the factors controlling kTP (Chapter 4). The previous 39 PAGE 40 discussion on the caution that must be applied when interpreting data of th is nature applies to these results as well. Relative Wetted Area and Total Phosphorus Mass Loading Rate Determining the importance of WA < An on the TPALR in the STAs was a surprisingly complex task. In absolute terms, the TPALRD and TPALRM must be greater than TPALRn for a cell in any given month, year, or ot her period of interest where the WAD < An and WAM < An, respectively. It is the size of these differences and their implications for other assessments that require an accurate measure of th e TPALR that are of interest. When the three methods of TPALR calculation were treated as repeated measures on each unique cellmonth (n=1000) the means were statisticall y, though not substantially, different (0.275 g/m2/mo, 0.273 g/m2/mo and 0.272 g/m2/mo for TPALRD, TPALRM and TPALRn, respectively. All pairs of m eans significantly different at p=0.05). The large number of observations boosted the significan ce of these trivial differences. This finding imparts nothing except the fact that the adjustment for WA ma kes very little impact on the longterm mean TPALR; the precision of flow ( Q in Equation ( 21 ) ) measurements does not support the estimation of the loading rate to thousandths of a gram per m2 per month. Accordingly, when the repeated measure design was removed from the analysis and the means compared directly, the calculated TPALR was not signifi cantly influenced by the method of WA calculation when all STA cells and all months with valid TPALR data ( n=1000) were included (KurskalWallis =0.1261, df=2, p=0.9389). In fact, the TPALR adjusted for the WA (regardless of calculation method) was not statistically different from the TPALRn. The general tendency toward complete or nearly complete flooding, again, obscured sta tistical differences between the adjusted TPALR and the TPALRn because the WAcorrection has no effect in months when RWAD = RWAM = 1.0. To isolate the true effects of the WA calculation on the TPALR, the three TPALR ranksums 40 PAGE 41 were again compared incl uding only months where RWAD < 1.0 ( n=280). The KurskalWallis =0.1264 (df=2, p=0.9388) which failed to allow rejection of the null hypothesis that there is no effect of WA calculation on TPALR. Even considering only the months where RWAM < 1.0 ( n=38) there was no effect of the method of WA calculation ( =2.4048, df=2, p=0.3005). Finally, the TPALR ranksums were compared cons idering all months (regardless of WA value) only in those cells with POR mean WAD 0.95 ( n=209; =0.1985, df=2, p=0.9055). In none of these cases was there a significant difference between any pair of the TPALR ranksums. Apparently, the variation in monthly TPALR due to fluctuations in flow rate and inflow TP concentration overwhelmed the slight variance cont ributed by the shifting RWA. This suggests, that no correction for WA < An was necessary for a sufficiently accurate accounting of the monthly TPALR in the STAs. Months with low RWAD tended to coincide with months of low flow ( Figure 24 ). Possibly, the adjustment of TPALR for WA < An was ineffectual in the monthly data because relatively low loading rates tended to dampen the impact of the adjustment in lowRWAD months. To estimate the annual TPALR, the sum of the 12 monthly loads is divided by the average of the 12 monthly WA values. (In adopting this calculation method, the dubious assumption that the water delivered to a cell in any given month was available for delivery in any other month, as would be the case if the water was metered out of a reservoir, was accepted). In this way, a lowRWA month contributes to the calculation (incr easing the calculated TPALR) even if the mass load in that month was low or negligible. However, despite these considerations, the different methods of TPALR calculation did no t lead to significant differences between the ranksums. 41 PAGE 42 Despite the lack of significant effect on the mean TPALR, adjustment for the calculated WA would still be warranted if the correction im proved the strength of the correlation between the TPALR and the outflow TP concentration. Th e coefficients of correlation (and pvalues) between outflow TP concen tration and TPALR are show n for each method of TPALR calculation over various s ubsets of the data in Table 23 When all months for all cells were included, the increases in the coefficient of corr elation due to WAcorrect ion were trivial. The largest improvements in r were found among months with RWAD < 1.0 (in all cells) and among all months in cells with POR mean RWAD < 95% (STA2 Cell 2, STA5 NFW, STA5 CFW), though in both of these cases th e relationship was so weak, rega rdless of TPALR calculation method, as to render the slight covariation valu eless. Interestingly, within every subset of observations, TPALRM was a better covaria ble with outflow TP c oncentration than was TPALRD. Of course, the relationship between TPAL R and outflow TP concentration was not uniform within each cell. The results and implica tions of the cellbycell analysis are presented in Chapter 3. Wetlands integrate, and thus dampen, shortterm loading effects through dynamic soil and macrophyte processes. For example, the vegeta tion in a healthy wetland may be able to assimilate a pulse of incoming P through biomass expansion. The excess P may be released over many more than the mass of water that carried the initial pulse, as the plants die and decompose, and a shortterm time step (e.g. mont h) may fail to capture the effect of TPALR on outflow TP concentration. To th at end, annual outflow TP concen tration was compared to annual TPALRn, TPALRM and TPALRD. The value of examining these va riables over a longer time step was immediately apparent from the improvement in rvalues compared to the whole population of cellmonths. Once again, the method of WA dete rmination was insignific ant: annual outflow 42 PAGE 43 43 TP concentration was more closely related to TPALRD ( r =0.476, p<0.0001, n=111) than it was to either TPALRM ( r =.475, p<0.0001, n=111) or TPALRn ( r =0.466, p<0.0001, n=111), though only trivially. Conclusions In the STAs, the interaction of intracell topography and timevariable stage occasionally resulted in the incomplete inundation of some cells, as revealed by th e characterization study. When the actual WA is less than the nominal ar ea, exposed soils and vegetation are subject to oxidation and desiccation. In addition, the value of the ALR calculated with An will not reflect the realized ALR of the system potentially affecting expected relationships between ALR and wetland treatment performance indices. There existed a weak positive linear relationship between kTP and RWAD and a weak negative lin ear relationship between kTP and RWAD. Both correlations increased in streng th with sequential screening of unimportant months, e.g. months with RWAD = 1.0 or RWAD = 0. However, the number of months among which these relationships were important was very limited. Neither the TPALR nor the strength of the correlation between TPALR and outflow TP concentration was meani ngfully altered by adjusting the calculated TPALR for WA < An. These nondramatic results support important co nclusions nonetheless. First, the expected biogeochemical consequences of dr yout and rewetting appear to have been at work in the STAs. Fortunately, from a treatment point of view, RW A infrequently varied far from 1.0, so the bulk of the TP processing in the STAs was una ffected by dryout/reflooding events. Second, An was a satisfactory approximation of WA since neither the monthly nor annual TPALR was significantly different under eith er alternative calculation sche me. Finally, poor performance in certain cells (elevated outflow TP concentrations relative to inflow TPALR) was shown not to be PAGE 44 44 an artifact of the TPALR calculation, validating the need for additional work to diagnose those factors contributing to performance in the ST As, as presented in the following chapters. PAGE 45 Table 21. Average annual relative wetted area by water year for each cell in the Stormwater Treatment Areas. Area (ha) 2000 2001 2002 2003 2004 2005 2006 2007 2008 Average STA1E 1628 99% 98 98 Cell 3 238 97 93 95 Cell 4N 261 99 99 99 Cell 4S 304 99 100 99 Cell 5 231 96 92 94 Cell 6 425 100 100 100 Cell 7 169 100 100 100 STA1W 2699 100 100 100 100 100 99 77 69 95 96 Cell 1 603 100 100 100 100 100 100 100 69 90 95 Cell 2 381 100 100 100 100 100 100 98 100 Cell 3 415 100 100 100 100 100 100 100 55 85 93 Cell 4 145 100 100 100 100 100 87 100 73 100 96 Cell 5 1155 100 100 100 100 100 80 95 100 97 STA2 2565 92 100 98 97 96 94 96 Cell 1 728 79 100 100 100 100 97 96 Cell 2 919 94 99 94 93 89 87 92 Cell 3 919 100 100 100 100 100 100 100 STA3/4 6695 99 99 97 98 Cell 1A 1230 99 98 90 96 Cell 1B 1412 99 100 100 100 Cell 2A 1029 100 99 99 99 Cell 2B 1171 99 100 100 100 Cell 3A 871 97 93 99 96 Cell 3B 982 99 99 98 98 STA5 1663 84 91 95 97 91 73 69 72 84 CFW 832 79 88 93 95 94 65 63 75 81 NFW 832 89 93 97 99 87 82 76 69 87 STA6 352 93 97 94 93 70 74 87 Cell 3 99 87 94 88 88 51 51 77 Cell 5 253 96 99 96 95 78 83 91 45 PAGE 46 46 Table 22. Intraannual trends in relative wetted area. No. Yr. Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec STA1E 100% 100 99 95 88 96 100 100 100 100 100 100 Cell 3 2 100 100 99 82 62 94 100 100 100 100 100 100 Cell 4N 2 100 100 100 100 95 95 99 100 100 100 100 100 Cell 4S 2 100 99 99 100 98 99 100 100 100 100 100 100 Cell 5 2 100 100 97 81 65 85 100 100 100 100 100 100 Cell 6 2 100 100 100 100 100 100 100 100 100 100 100 100 Cell 7 2 100 100 100 100 100 100 100 100 100 100 100 100 STAW 97 94 91 89 91 98 100 100 100 100 99 98 Cell 1 9 93 91 89 89 89 98 99 100 100 100 99 97 Cell 2 7 100 100 100 100 96 100 100 100 100 100 100 100 Cell 3 9 89 89 89 89 89 91 100 100 100 100 94 91 Cell 4 9 100 100 93 80 84 91 100 100 100 100 100 100 Cell 5 8 100 95 88 87 92 100 100 100 100 100 100 100 STA2 97 97 98 96 87 92 96 98 99 98 98 98 Cell 1 6 100 98 100 100 81 85 90 98 100 100 100 100 Cell 2 6 91 94 95 90 78 88 97 96 97 94 94 95 Cell 3 6 100 100 100 100 100 100 100 100 100 100 100 100 STA3/4 100 100 99 97 91 96 100 100 100 99 98 100 Cell 1A 3 100 100 100 93 73 87 100 100 100 99 97 100 Cell 1B 3 100 100 100 100 100 100 100 100 100 99 97 100 Cell 2A 3 100 100 100 97 97 100 100 100 100 99 100 100 Cell 2B 3 100 100 100 100 100 100 100 100 100 99 98 100 Cell 3 3 100 99 97 95 89 95 100 100 100 99 99 100 STA5 81 76 76 70 65 80 92 94 96 96 93 90 CFW 8 76 68 73 64 60 77 91 95 96 96 94 89 NFW 8 87 83 80 76 69 83 92 93 96 96 93 90 STA6 84 82 86 79 56 75 94 100 100 100 98 93 Cell 3 6 70 69 75 64 23 50 84 100 100 100 97 88 Cell 5 6 89 87 90 85 68 85 97 100 100 100 99 94 Grand Total 97 95 95 92 87 93 98 99 99 99 98 98 PAGE 47 Table 23. Changes in the coefficients of correlation for different subsets of data. 1 2 3 4 5 All cells Months when RWAD < 1.0 Months when RWAM < 1.0 All months in cells with mean RWAD < 0.95 No. data months 945 248 25 187 TPALRn 0.266 (<0.0001) 0.099 (0.1214) 0.493 (0.0122) 0.048 (0.5154) TPALRD 0.271 (<0.0001) 0.119 (0.0614) 0.486 (0.0137) 0.067 (0.3589) TPALRM 0.273 (<0.0001) 0.133 (0.0370) 0.516 (0.0083) 0.083 (0.2567) 47 PAGE 48 48 0% 20% 40% 60% 80% 100% 7891011121314 Feet of elevation NGVD Cell 3 Cell 5 0% 20% 40% 60% 80% 100% 7891011121314 Feet of elevation NGVD Cell 1A Cell 1B Cell 2A Cell 2B Cell 3A Cell 3B 0% 20% 40% 60% 80% 100% 7891011121314 Feet of elevation NGVD Cell 1 Cell 2A Cell 2B Cell 3 Cell 4 Cell 5A Cell 5B 0% 20% 40% 60% 80% 100% 1011121314151617 Feet of elevation NGVD Cell 3 Cell 4N Cell 4S Cell 5 Cell 6 Cell 7 0% 20% 40% 60% 80% 100% 7891011121314 Feet of elevation NGVD Cell 1 Cell 2 Cell 3 0% 20% 40% 60% 80% 100% 910111213141516 Feet of elevation NGVD Cell 1A Cell 1B Cell 2A Cell 2B STA1E STA1W STA2 STA3/4 STA5 STA6 Figure 21. Cumulative elevation distribution for each cell in th e Stormwater Treatment Areas. Note that the range of each horizontal axis is the same. PAGE 49 50% 60% 70% 80% 90% 100% Drough t ST A1 E ( 1 0 0 ) ST A1 W ( 9 7 ) ST A2 ( 9 6 ) ST A3 / 4 ( 9 8 ) ST A5 ( 8 4 ) ST A6 ( 8 7 ) Figure 22. Relativ e wetted area in each of th e Stormwater Treatm ent Areas. Note that the vertical axis does not ex tend to 0. Err o r bars hav e been om itted to inc r ease clar ity. 49 PAGE 50 50 0% 20% 40% 60% 80% 100% 0%20%40%60%80%100% Relative wetted area Figure 23. Histogram of monthly relative wetted area (determine d by the elevation distribution) values across all nonscreened months and all included cells. PAGE 51 0% 20% 40% 60% 80% 100% 05101520253035Relative wetted areaHydraulic loading rate (m/mo) Figure 24. Monthly relative wetted area (determined by the elevati on distribution) with respect to monthly hydraulic loading for all nonsc reened months and all included cells. 51 PAGE 52 CHAPTER 3 ASSOCIATION BETWEEN LOADING RATE AND OUTFLOW CONCENTRATION IN THE STORMWATER TREATMENT AREAS Introduction Several species of SAV, macr ophytes with the bulk of thei r biomass suspended in the water column, are common in wetlands and other water bodies in South Florida (Dierberg et al., 2002). Because of their growth habit, the mechan isms of P removal in SAV systems include two distinct differences from those in EAV wetlands. First, roots of emergent plants can access only porewater nutrients (with diffusion or mass transfer of nutrients to the sediment mediating uptake from the water column), whereas SAV obtains nu trients directly from the water column through the shoots and leaves (Granli and Solander, 198 8). This means that SAV can uptake SRP very quickly, particularly in the s hortterm (Pietro et al., 2006). Second, during photosynthesis, SAV removes carbon dioxide and bicarbonate from the wa ter column, which raises the system pH, and drives the system towards calcium carbonate (CaCO3) supersaturation, promoting CaCO3 precipitation (McConnaughey et al ., 1994). Several studies have pr oposed coprecipitation with CaCO3 as a mechanism of P removal in a variety of systems (Scinto, 1997 reviews some of these works). It has been reasoned that SAV is more suited for P removal in treatment wetlands than EAV, a hypothesis supported by studies at mesocosm(Dierber g et al., 2002) prototype(Nungesser and Chimney, 2001) and fieldscales (Juston and DeBusk, 2006). SAV systems have generally been successful wi thin the STAs, and now comprise over half (ca. 10,000 ha) of the STA treatment area, often in downstream positions within serial flowtrains (Pietro et al., 2010). Most of the cells with the lowest longterm outflow concentrations are dominated by SAV; of the nine cells with longterm flowweighted average outflow TP concentrations below 0.030 mg/L, six are designated SAV. 52 PAGE 53 Though elevated relative to the levels in the hi storical, unimpacted Everglades, it has been suggested by way of internal pr ofile studies that these low outfl ow concentrations approach C* in some SAV cells (i.e. the concentration prof ile reaches a plateau some fraction of the way through the cell, beyond which no additional treatment is observed; Pietro et al., 2010). The alternative P treatment m echanisms at work in SAV may be responsible for the reduction of TP to C*, possibly because, given nonlimiting cal cium and light for photosynthesis, the coprecipitation process is not subject to saturation as soils (via sorption) and microbial and plant biomass may be, nor is it ratelimited by biotic uptake. Excessively loading a wetland with a high sett ling rate conceptually results in poor performance (e.g. elevated outflow concentratio ns, lowered percent remo val). Conversely, even wetlands with low settling rates can conceptually perform well at sufficiently low loading rates. The interaction of the settling rate and loading rate determ ines the realized treatment performance for a wetland (assuming the inflow con centration is well above C*). It is convenient to employ the Damkhler number, which capture s the treatment potential of a wetland by combining the settling rate with the lo ading rate (Kadlec and Wallace, 1996): Da = k q ( 31 ) where Da = Damkhler number. By inserting Equation ( 31 ) into Equation ( 119 ) the kC* model suggests that at very high Da, C* controls the outflow concentration: C2C*=( C1C*) exp ( Da) ( 32 ) Internal profile studies are reliable for de termining whether a wetland is treating to C*. They are however, laborious and expensive both in the field and the laboratory. Potentially, the strength of the correla tion between the outflow concentrati on and the TPALR may be used as a proxy to estimate if wetland effluent is at C*. Si nce C* is independent of the shortterm TPALR 53 PAGE 54 (Equations ( 119 ) and ( 21 ) ), in situations where the outflo w TP concentration approaches C*, the correlation between the outflow TP concentration and the TPALR will be weak or absent. For example, internal profiles were estimated for a hypothetical wetland with a very high Da under four different loading scenarios, each with a different ALR (Figure 31). In each scenario, the outflow concentration was equal to C* and outflow concentration and ALR were uncorrelated. The term apparent background concen tration is adapted from Kadlec and Knight (1996) to describe the outflow TP concentrations from those cells with weak shortterm outflow concentrationTPALR interactions, even when those concentrations were above the range of commonly described background TP concentrations for South Florida systems of 616 g/L (Juston and DeBusk, 2006; Kadlec and Wallace, 2008). The hypothesis that alternative P removal mechanisms elevate Da such that S AV cells tend to produce apparent TP background concentrations (while EAV cells are less likely to do so) may be tested by comparing the correlations between outflow TP concentration and TPALR within SAV cells to those in EAV cells. It should be noted, however, that even for we ll performing cells, the outflow concentration varied, both from cell to cell a nd over time. In conjunction with the previous discussion, this implies that the apparent background concentr ation was not fixed. In a plugflow wetland, the real P background concentration is the point of equilibrium between P availability and biogeochemical P demand. Atmospheric deposition, internal hydraulics (t he degree of mixing within a wetland), TP fractionation, and the inte rnal loading (particularly of PP and DOP) are known to contribute to the apparent backgr ound concentration (Kadlec and Wallace, 2008). Some of these factors can be eas ily measured (TP fractionation) a nd others can be inferred from easily quantifiable factors (e.g. temperature can be a proxy for seasonal changes in biomass 54 PAGE 55 production and senescence, both of which contribute to autochthonous P loads) In STA cells that produce outflow TP concentrations independent of the TPALR, understanding and quantifying the factors that control the apparent background concentration are important for increasing both realized performance and predictions of that performance. Objectives This chapter addresses four primary questions: 1) Were SAV cells in the STAs operating at apparent background concentrations, as de termined by independence of outflow TP concentration from the TPALR? 2) Were the outf low TP concentration data uncorrelated to the TPALR in all SAV cells, or only a specific subset ? 3) In the case that only certain cells are exempt from this correlation, what was the under lying cause if not the do minant vegetation? 4) What determined the outflow TP concentratio n in cells operating at an apparent background concentration? The above quest ions were answered by testing the following hypotheses: The outflow TP concentration data will be more poorly correlated to the TPALR data in SAV cells relative to EAV cells. The outflow TP concentration data will covary weakly with TPALR data in most SAV cells, and will not covary at all with TPALR in some SAV cells. The strength of the covariation wi ll be negatively correlated to the percent cover of SAV within cells. In those cells operating at an apparent C*, the out flow concentration will be a function of one or more of the variables: TP fractiona tion, longterm TPALR and temperature. Methods Methods regarding the calculati on or preparation of TPALR, TP fraction, and temperature data may be found in Chapter 1. Vegetation Each STA cell has been classified by do minant vegetation type (SAV/EAV) by SFWMD (Pietro et al. 2008). While the cl assifications generally capture the dominant vegetation types, 55 PAGE 56 and reflect the plant communities targeted by SFWMD for each cell, they do not imply 100% coverage by the indicated vegetation class. Th e term Mixed was applied to those units composed of cells of different designations, bu t for which water quality data was only examined at the larger scale (e.g. the Nort h Flowway of STA5 consists of an EAV(Cell 1A) and a SAV(Cell 1B) designated cell, but flow data were on ly available at the infl ow and outflow of the flowway). To enhance the precision of the vegetati on classification, the percent SAV cover was estimated for each cell in this study using tabulati ons of vegetation cover based on aerial imagery and vegetation field survey data provided by the SFWMD. The detailed (often specieslevel) designations from the map tabula tions were combined into thre e distinct groups: EAV, SAV and open water. For this study, the SAV and open wa ter coverages tabulated from vegetation maps were combined and assumed to approximately repr esent total SAV cover. The field survey data available to this project were collected by se veral different contractors and each employed a unique reporting system. Briefly, th e coverage or abundance data at each survey point, plot or transect were normalized to a 0 to 1 scale (1 representing 100% cover). Th e relative coverages of each SAV species (including algae, when reporte d) were summed at each survey location. The nature of field surveys allowed for the distinct ion of SAV from open water, so open water did not contribute to the total SAV cover at each site. The values of percent SAV cover were averaged over all survey locations to estimate the total relative SAV cover for each cell. The spatial distributions of the survey points were checked to verify representative sampling within each cell. This study did not attempt to address changes in SAV cover over time. In cells where more than one estimate of SAV cover was availa ble, the values were simply averaged, with equal weight given to survey and map data. 56 PAGE 57 Outflow Concentration and Total Phosphorus Areal Loading Rate The Pearson productmoment correlation coefficient ( r ) between monthly outflow TP concentration and monthly TPALRM (the best regressor of the three possible TPALR calculations; see Chapter 2) was determined for the POR for each cell. As r describes only the strength of the linear relationship between two variables, scatter plot s were produced for each cell and visually inspected for nonlinearity. It was noted that the significant ( p = 0.05) positive r values in several cells resulted entirely from the presence of outliers in the upperright quadrant (high outflow concentrations corresponding to high lo ading events). The pract ical significance of these relationships defined by outliers is discussed below. However, in an effort to capture the typical behavior of each cell, r was recalculated after excluding the month with the largest TPALR value in each cell, and termed r for convenience. The bulk of this study relies on r because approximately onehalf of the significant r values were artifact s of single outliers ( Table 31 ). Results and Discussion Outflow ConcentrationA real Loading Relationship Effect of vegetation type and cover The magnitude and significance of r was not clearly determined by the vegetation designation reported for each cell in Pietro et al. (2008) ( Table 31 ). Of the 7 SAVclassified cells examined, r was nonsignificant in 6 (86%). Sim ilarly, 9 (75%) of the included 12 EAV cells had nonsignificant r Unexpectedly, one EAV cell (STA2 Cell 1) reported a weak but significant negative covariance between outflow TP con centration and TPALR. Possible causes for this counterintuitive relationship were unclear As a result, data from Cell 1 of STA2 were excluded from further analyses in this chapter. Three of the four (75%) mixed units had nonsignificant r Broadly, EAV and SAV cells were not overwhelmingly differentiated by the 57 PAGE 58 strength of r ( Table 31 ), though the caveats asso ciated with the vege tation designations (see Methods of this chapter) depreciate the power of this assessment. To avoid the difficulties and inaccuracies associated with using the SFWMDassigned vegetation classifications, r was plotted against estimates of the actual percen t cover of SAV. The cells clustered into three distinct groups, ( Figure 32 ). Group 1 contains cells with low to moderate SAV coverage and low r values. The lowto midSAV coverage cells, with relatively high r values comprise Group 2. Group 3 consists of highSAV cells with nonsignificant r Immediately apparent from Figure 32 are the strictly nonsignificant r values in all highSAV cells. Secondly, the unexpected tendency for nonsignificant r values in lowand moderateSAV cells is noteworthy. Togeth er these observations do not s upport any conclusions about the influence of SAV P processing mechanisms on the uncoupling the shortterm outflow TP concentration from the TPALR in SAV cells but do suggest that this method may be inappropriate for testing that hypothesis. Also, importantly, they make clear that some other factor (or factors) contributed to the disassociation of the outfl ow TP concentration from the monthly TPALR. Effect of areal total phosphorus loading rate Equation ( 119 ) indicates that, at sufficiently light loads, the outflow concentration from a wetland will converge on the background concentr ation, C*, and thus approach independence from the loading rate. This was supported by the findings of Qian and Richardson (1997) (and reinforced by Richardson and Qualls, 1999) th at showed that below POR TPALR of about 1.0 g/m2/yr, the longterm average outflow concentra tions from a large number of North American wetlands were fairly invariant with respect to changes in TPALR. Likewise, among SAV and select EAV cells with annual TPALR at or below 2.0 g/m2/yr, Juston and DeBusk (2006) noted no significant relationships ( p>0.05) were identified in the slopes of PALR relationships using 58 PAGE 59 either 1 or 2 year average ALRs, thus suggesti ng no evidence of association of ALR with TP concentrations in this range. It follows that the areal loads in cells in Groups 1 and 3 were possibly too low to force the outflow TP concentration a bove the apparent background concentration. A oneway analysis of vari ance was performed on th e monthly TPALR to compare each of the three groups. The Fvalue was 17.19 (df=2, p<0.0001). A StudentNewmanKeuls test ( p =0.05) found that the mean monthly TPAL R was significantly higher in Group 2 (0.42 g/m2/mo) than in either Group 1 or Group 3. The mean monthly TPALR were similar in Groups 1 and 3 (0.21 and 0.22 g/m2/mo, respectively). The implication, in the spirit of Richardson and Qualls (1999), is that the TPALR became influential on the outflow TP concentration once it exceeded some thresh old. All cells with av erage annual TPALRM less than about 2.0 g/m2/yr showed no significant relationshi p between shortterm outflow TP concentration and the TPALRM ( Figure 33 ), as expected from the work of Juston and DeBusk (2006). The presence of significant, positive r values only in cells loaded with greater than 2.0 g/m2/yr does suggest that the magnitude of the loading rate contributed to the strength of the TPALRoutflow TP concentration relationship. The fact that some cells also loaded above 2.0 g/m2/yr had nonsignificant correlations betw een outflow TP con centration and TPALRM indicates the action of an additiona l factor (or factors) restraining r even under large loads. Moreover, this additional factor was very likely not the alternative mechanisms of SAV systems, as the highTPALRM, lowr group contained both EAV and SAV cells. A mention must be made of the substantial difference between r and r in a few cells ( Table 31 ). The extreme influence that the outliers had on the correlation between the outflow TP concentration and the TPALR justifies their exclusion for analytical purposes. Nonetheless, these data imply that, had these cells experien ced higher and more vari able loading rates on a 59 PAGE 60 monthly basis, than was observed during their recorded operation, the outflow TP concentration may have actually been influenced by the TPALR. The nearuniversal tendency for the month of maximum load to increase the strength and significance of r suggests that, regardless of the vegetation type or the shortterm outflow concentrationTPALR re lationship under typical operating conditions, extreme pulses of P were very likely to produce aboveaverage outflow concentrations. Factors Controlling the Apparent Background Concentration Having established that, in most cells of the STAs, the shortterm outflow TP concentration did not depend strongly on the monthly TPALR, the challenge of identifying the factors that did determine the shortterm outflow concentrati on arises. Curiously, the POR FWM outflow TP concentrations from those cells with nonsignificant r values showed the same range and variability as those cells with moderate to strong TPALRoutflow TP concentration relationships ( Figure 34 ). This chapter considers two possi ble controls on the longterm mean outflow TP concentration and one possible regulator of the shor tterm outflow concentration. A more thorough exploration of poten tial factors controlling monthly outflow TP concentrations is presented in Chapter 4. Inflow phosphorus fractions The most chemoand bioavailable P forms comprise the SRP fraction of the TP pool (Kadlec and Wallace, 2008; Reddy and DeLaune, 2008). In treatment wetlands, this often leads to preferential removal of SR P relative to the TP aggregat e (Dierberg et al., 2002; Chimney, 2007). Conversely, DOP is thought to be generall y less bioavailable and has been found to be less effectively removed than the bulk TP in treatment wetlands (Reddy and DeLaune, 2008; Chimney, 2007), though this may be due as much to internal production of DOP as to nontreatment of influent DOP (Pinney et al., 2000). Th erefore, it is proposed that the composition of 60 PAGE 61 the inflow TP potentially contributed to the appa rent background concentration in the STAs, with relative increases in SRP lowering C* and re lative increases in DOP elevating C*. Although the monthly outflow TP concentration fr om the set of cells of interest was not correlated to the TPALR, it was weakly but measurably influenced by the inflow TP concentration: r = 0.353 ( p <0.0001, n = 597 considering monthly data among all the cells of interest). The positive relationshi p also existed within the monthly data for 4 of the 18 included individual cells (STA1E Cell 7, STA2 Cell 3, STA3/4 Cell 1B, STA6 Cell 5). The presence of this covariance required that the effects of the TP fractionation on the outflow concentration be assessed within the effects of the inflow TP concentration. After the variability due to the inflow TP concentration was removed, neither th e relative size of the SRP fraction nor the DOP fraction had any significant effect on the outflow TP concentration (SRP: F value = 0.12, p = 0.7298, df = 1; DOP: F value = 0.03, p = 0.8722, df=1). The effect of the inflow DOP concentration was significant ( F value = 6.89, p = 0.0089, df = 1), but the additional variability explained was minor (approx. 1%). The three vari ables, inflow DOP con centration, inflow DOP fraction and inflow SRP frac tion, had a significant effect on the monthly outflow TP concentration after accounting for the inflow TP concentration in only 2, 3, and 4 cells, respectively. In addition, there was no relationship found between the outflow TP concentration and any of these three variables at either annu al or POR averaging periods. In summary, both across cells and within cells, the composition of the inflow TP played a minor role, at best, in the determination of the outflow concentration from the cells where the outflow TP concentration was not related to the TPALR. Longterm areal loading Sustained high annual areal P loading is known to increase the soil P concentration. In South Florida, a wellstudied example of this phenomenon is Water Conservation Area 2A 61 PAGE 62 (WCA2A), where decades of P loading have re sulted in elevated soil P concentrations immediately downstream from the inflows. Reddy and DeLaune (2008) provide an excellent review of the literature on WAC2A. As part of a positive feedback cycle with increasing soil P, primary productivity of wetlands tends to increas e with sustained P loads (Lowe and Keenan, 1997; Kadlec and Knight, 1996). As a result, the Pprocessing biomachine grows in response to increased P loads. The Psequestering power (in terms of g P/m2/yr) grows as well, until biomass expansion becomes limited by space, light or nutrient(e.g. N) availability. Biomass production, senescence and decomposition are known to export dissolved nutrients to the water column (Pinney et al., 2000; Qualls and Richardson, 2002, Reddy and DeLaune, 2008). It is conceivable that increasing the size of the b iomachine would increase the autochthonous P production and export. Thus, STA cells with high longterm TPALR may be expected to operate with relatively higher apparent background concentrations, even if in the short term, the outflow TP concentration is independent of the TPALR. It has been previously established that the expected positive relationship between outflow TP concentration and TPALR exists in the STAs when longterm (POR) averaging periods are considered; among all STA cells, the POR outflow TP concentration wa s positively, nonlinearly correlated to the POR TPALR through a power func tion (Pietro et al., 2009). The objective here is to assess if that relationship holds when cons idering only those cells in which the shortterm TPALR did not directly control the monthly outflow TP concentration. Among all the cells operating with nonsignificant r the POR FWM outflow TP concentration was significantly linearly co rrelated to the average annual TPALR ( r = 0.703, p=0.0011, n = 18; Figure 35 ). A power function (Pietro et al. 2009) described the data only slightly more accurately ( r2 = 0.508; Figure 35 ). The shape of the data cloud in Figure 35 was 62 PAGE 63 remarkably similar to the distribution of the wi de variety of wetlands studied by Richardson and Qualls (1999), including an a pparent breakpoint near 1.0 g P/m2/yr. Unclear, for both the STA cells and the wetlands documented by Richardson and Qualls (1999), is why some wetlands are able to produce low outflow concentrations ev en under high areal loads (e.g. STA1E Cell 4N). At an annual scale, the data demonstrated a similar relationship, although the nonlinearity observed in the POR data was absent ( Figure 36 ). Even among cells where the monthly outflow concentration was independent of the monthly TPALR, years with higher areal loads tended to also have higher outflow TP concentrations. Th is may be attributable to the aforementioned effect of the size of the biomachine. The nonlinearity in the POR data may develop as the effects of consistent light or heavy loading compound over years of operation. Seasonality The bulk of the longterm sustainable P rem oval in wetland is biol ogically mediated by microorganisms, algae and submerged and em ergent macrophytes (Kadlec and Knight, 1996; Kadlec, 1997; Dierberg at al., 2002). The P demand and production from these biomass compartments varies as their si ze and activity fluctuat e in response to envi ronmental factors (e.g. temperature and insolation) (Kadlec and Wallace, 2008). Seas onal cycles are common in treatment wetlands, but are minimi zed, particularly for P, in s ubtropical wetlands with a yearround growing season (Kadlec and Wallace, 2008) Nonetheless, a distinct intraannual temperature cycle is observed in South Florida ( Figure 36 ) that may lend itself to seasonal fluctuations in biological proces sing sufficient to alter treatment performance. If so, the impacts ought to be seen as an annual cyc lical variation in the outflow con centration, particularly in those cells producing effluent at an apparent background concentration. Water temperature is known to control microbial activity (Reddy and DeLaune, 2008) and serves as a good proxy for the average insolation, a driv er of primary productivity (Best and Visser, 1987), so any annual 63 PAGE 64 cyclical variation in outflow TP concentra tion may be a reflection of the annual water temperature cycle ( Figure 36 ). Considering only the cells with nonsignificant r values, the monthly outflow TP concentration did not significantly vary with either the monthly inflow or outflow water temperature (r 0.064, n = 578). Likely, the celltocell vari ation in outflow TP concentration devalued this assessment. The monthly outfl ow TP concentration was significantly (p = 0.05) correlated to the monthly inflow water temperatur e in only 3 cells (positively in STA3/4 Cell 3 and STA5 CFW and negatively in STA5 NFW), and to the outflow water temperature in only 2 cells (positively in STA6 Cell 3 and negativel y in STA3/4 Cell 2B). No nonlinearity was found between outflow TP concentration and water temperature in a visual inspection of scatter plots within each cell. That the relationship was significant in relativ ely few cells and that the conditions of the covariances were inconsistent (some positively and some negatively correlated), suggest that the water temperature in the STAs did not play a critical role in determining the apparent background TP concentration in the STAs. Water temperature was thought to be a satisfactory surrogate for seasonality because many other variables are expected to be strongly correlated to it (e.g. insolation, seasonal biomass changes). However, a flaw is recognized in this approach: because of the sinusoidal nature of temperatur e over time, each average monthly temperature occurs twice per year (except for the a nnual maximum and minimum temperature). The environmental conditions may be quite differe nt at the two manife stations of a given temperature. For example, the average water temperature tended to be about 25 C in both April and October in STA1W Cell 4 ( Figure 36 ), but this fails to capt ure the obvious biological differences between the two mont hs (e.g. spring flush vs. fall sene scence). In none of the 18 cells 64 PAGE 65 with low r values was the outflow TP concentrati on significantly correlat ed to the ordinal month value (e.g. January = 1, February = 2) Additionally, no regular nonlinear patterns were observed in scatter plots of outflow TP c oncentration vs. time in months. A more powerful signalprocessing method (e.g. Fourier transform) might be applied to the monthly outflow TP concentration data to better effect. Conclusions Although the POR outflow TP concentration was generally a function of the longterm average annual TPALR among all cells of the ST As (Pietro et al., 2009), as is common in treatment wetlands (Qian and Richardson, 1997), in a majority of cells the monthly outflow TP concentration was statistically independent of th e monthly TPALR. Celltocell variability in the strength of the shortterm outflow concentra tionTPALR relationship was not explained by the dominant vegetation type, and uns atisfactorily justifie d by the longterm TPALR. In the shortterm, the outflow TP concentration from thos e cells operating at an apparent background concentration was not determined by the relative fractionation of the infl ow TP pool nor did it vary with season, as approximated by the water te mperature. Among these cells in the longterm, the annual average TPALR accounted for appr oximately 51% varia tion in the POR FWM outflow TP concentration. Two useful implications arise from this study. First, the apparent uncoupling of the monthly outflow TP concentration from the TP ALR suggests large Da in most STA cells, independent of the vegetation type. Equation ( 32 ) demonstrates that in wetlands with high Da, the realized outflow TP concen tration will be controlled primarily by the apparent background concentration. Therefore, the ability for kC* to accurately model outflow TP concentrations in the STAs may depend heavily on accurate estimates of C*. 65 PAGE 66 Second, the notable lack of powerful descriptiv e variables for the s hortterm outflow TP concentration resulting from this work highlights the limits of both the currently available data and the body of knowledge of wetland P processi ng at this time. If minimizing the outflow concentrations from the STAs remains a political and managerial priority in South Florida, smallscale (e.g. mesocosm) research ought to be applied to identif y additional potential controllers of the shortterm outfl ow TP concentration, a nd largescale (e.g. fiel d) studies will be needed to define the interactions of these variables under operational conditions. 66 PAGE 67 Table 31. Coefficients of correlation betw een monthly outflow total phosphorus (TP) concentration and monthly TP areal loadi ng rate (ALR) within each cell before ( r ) and after ( r ) exclusion of the month of ma ximum TPALR. Nonsignificant (p = 0.05) coefficients are reported as 0.0. Cell Vegetation Designation r r STA2 Cell 1 EAV 0.284 0.293 STA1E Cell 4N SAV 0.0 0.0 STA1E Cell 4S SAV 0.0 0.0 STA1E Cell 3 EAV 0.0 0.0 STA1E Cell 5 EAV 0.0 0.0 STA1E Cell 6 SAV 0.859 0.0 STA1E Cell 7 EAV 0.648 0.0 STA1W Cell 5 SAV 0.085 0.0 STA3/4 Cell 1A EAV 0.0 0.0 STA3/4 Cell 1B EAV 0.0 0.0 STA3/4 Cell 2A EAV 0.433 0.0 STA3/4 Cell 2B SAV 0.0 0.0 STA3/4 Cell 3 Mixed 0.0 0.0 STA5 CFW Mixed 0.0 0.0 STA5 NFW Mixed 0.0 0.0 STA2 Cell 2 EAV 0.0 0.0 STA2 Cell 3 SAV 0.0 0.0 STA6 Cell 3 EAV 0.0 0.0 STA6 Cell 5 EAV 0.622 0.0 STA1W Cell 4 SAV 0.433 0.292 STA1W Cell 1 EAV 0.524 0.422 STA1W Cell 2 Mixed 0.742 0.591 STA1W Cell 3 EAV 0.692 0.707 67 PAGE 68 A) ALR = 1.0 B) ALR = 1.0 C) ALR = 0.75 D) ALR = 0.38 C* 0%20%40%60%80%100%Water column concentrationDistance through wetland ScenarioHLR Inflow conc. ALR m/yrmg/Lg/m2/yr A150.0671.00 B200.0501.00 C150.0500.75 D150.0250.38 Figure 31. When a wetland treats to background concentration (C*), the areal loading rate (ALR) does not affect the outflow concentration. 68 PAGE 69 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0%20%40%60%80%100%r 'valuePercent SAV coverGroup 2 Group 1 Group 3 Figure 32. Correlation between outflow total phosphorus (TP) concentration and TP mass loading rate with respect to submerge d aquatic vegetation (SAV) coverage. Nonsignificant correlations were assigned an r value of 0. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 012345r 'valuePOR average annual TPALRM (g/m2/yr) 6 Group 1 Group 2 Group 3 Figure 33. Correlation between out flow total phosphorus (TP) concentration and TP areal loading rate (ALR) as a function of th e average annual TPALR. Nonsignificant correlations were assigned an r value of 0. 69 PAGE 70 70 0 0.05 0.1 0.15 0.2 0.25 00.20.40.60.81POR FWM outflow TP concentration (mg/L)r'value Group 1 Group 2 Group 3 Figure 34. Outflow total phosphorus (TP) concen tration with respect to the correlation coefficient between the TP areal loading rate (TPALR) and the outflow TP concentrations. y = 0.0344x 0.0004 R = 0.4946 y = 0.0292x0.9777R = 0.5067 0 0.05 0.1 0.15 0.2 0.25 0.1 1 10POR FWM outflow TP concentration ( mg/L)POR annual average TPALR (g/m2/yr) Figure 35. Periodofrecord (P OR) flowweighted mean outflow total phosphorus (TP) concentration as a function of the POR average annual TP areal loading rate. PAGE 71 y = 0.0165x + 0.0327 R = 0.2143 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0246810121416Annual FWM outflow TP concentration (mg/L)Annual TPALR (g/m2/yr) Low r'value High r'value Figure 36. Annual flowweighted mean (FWM) out flow total phosphorus (TP) concentration as a function of the annual TP areal loading rate (ALR). Data from cells with high r values are differentiated. The regre ssion line considers data from low r value cells only. 10C 20C 30C 40C JFMAMJJASOND Inflow OutflowFigure 37. Monthly average temper ature (C) of inflow and out flow water in STA1W Cell 4. 71 PAGE 72 CHAPTER 4 MULTIPLE LINEAR REGRESSION TO DETERMINE FACTORS CONTROLLING PHOSPHORUS CONCENTRATION AND SETTLING RATE IN THE STORMWATER TREATMENT AREAS Introduction The cardinal performance metric for the STAs is outflow TP concentration. Many studies evaluating the STAs have investigated the outfl ow TP concentration with respect to various wetland characteristics, such as TPALR or vegeta tion type (e.g. Dierberg et al., 2002; Juston and DeBusk, 2006; Pietro et al., 2009). Importantly, the enforced and proposed regulations on the STAs also consider, primarily, the effluent concentration (Pietro et al, 2008). These regulations obligate STA managers to minimize the outflow TP concentrations. Ensuring the conditions for high Davalues (high settling rates and low loading rates) is critical to that effort. Chapter 3 of this work showed that the shor tterm outflow TP concentration was controlled more by C* than by TPALR or Da in most STA cells. Therefore, it is important to understand the wetland factors that contribute to the C*. Interwetland and temporal variations in outfl ow TP concentration have been well studied (Kadlec and Wallace, 2008). Likewise, fluctuatio ns in outflow TP concentration have been documented in the STAs (e.g. Pietro et al., 2008). A stated or implied assumption in almost all treatment wetland work, and adopted here as well, is that the outflow TP concentration is a function of various wetland characteristics. That is, the difference in the outflow TP concentration observed for two different wetlands ought to be due to the variation of some feature(s) between those wetlands. Similarly, the fluctuations of outflow concentration with time within a wetland are assumed to have resulted from changes in attributes of that wetland. For example, holding everything else co nstant, a wetland with a TP lo ad comprised primarily of SRP 72 PAGE 73 may be expected to demonstrate a lower outfl ow concentration than an identical wetland receiving mainly DOP. In wetland science of course, it is impossible to hold everythi ng else constant, and so the effects of a single wetland char acteristic are easily confounded. Even wetlands in sidebyside studies show some disparitie s in outflow concentrations (Kadlec and Wallace, 2008). Such differences are often ascribed to stochastic vari ability, a lumped error term containing all of the variance due to unaccountedfor wetland properties (Kadlec and Wallace, 2008). The objective of this study was to partition the variabi lity in the shortterm out flow TP concentration, to the maximum extent possible, to quantifiable wetland character istics. In addition, the outflow concentration was correlated to TPALR in a few STA cells (Chapter 3), implying that Da was limiting performance. Therefore, the factors cont rolling the TP areal settling rate were also explored here. Being unable to isolate the effects of single va riables in the field, this study attempts to do so via multiple regression using the vast datasets available for the STAs. Considering the entire pool of nonscreened m onths from all STA cells, the differences in means from cell to cell accounted for approximately 30% and 20% of the variation in monthly outflow TP concentrations and TP settling rates, respectively. Fr om a managerial perspective, identifying the combination of variables responsib le for these celltocell differences (as well as the factors responsible for the s ubstantial intracell va riability) may possibly allow managers to better regulate the performance of the STAs by manipulating key attributes in all or some wetland cells. In addition, unders tanding which macroscale wetla nd characteristics control the outflow concentration and the kvalue in the ST As would provide guidance for future work on the illumination of poorlyunderstood processle vel P dynamics in all subtropical surfaceflow treatment wetlands. 73 PAGE 74 Objectives The objective of this chapter was to assess th e relative influence of a wide variety of factors on the outflow TP concentration and the TP areal settling rate in the STAs. Multiple regression methods were implemente d as an initial evaluation. It is recognized that the methods employed herein were relatively elementary, but were necessary to provide guidance for future exercises in this vein. This chapter specifically addresses the quest ion: can variation in STA performance be adequately explained by variations in the characteristics of the wetlands that comprise the STAs? Of course, the wetland properties open to i nvestigation were limited to those for which appropriately spatiallyand tempor allyresolved, quantitative data were available. After the first inquiry, a second question natu rally follows: which factors did contribute to variation in STA performance, and are they subjec t to manipulation by STA managers? This chapter tested the following hypothesis: variability in the monthly outflow TP concentration and TP areal settling can be significantly explained by a subset of these factors: inflow TP concentration, inflow TP composition (fractionation), TPALR, soil TP concentration, mean water dept h, wetland age, inflow Ca con centration, Ca ALR, Ca areal retention, water column pH and water temperature. Methods Multiple Linear Regression Multiple regression considers linear models of the form: yi = 0 + 1x1 i + 2x2 i + 3x3 i + + nxn i + i ( 41 ) where yi = the i th observation of the dependent variable of interest, x1 ixn i = i th values of independent variables related to y, 0 = estimated value of y when x1 through xn = 0, 1n = parameters (coefficients) estimating the effect of each x1xn on y, and i = i th residual. Multiple 74 PAGE 75 regression procedures seek to select values of 0n which minimize the sum of the squares of the residuals. Multiple regression analyses were used to determine the most influential independent variables (x1xn in Equation ( 41 ) ) on the dependent variables monthly outflow TP concentration and monthly TP areal settling rate (kTP). Because the absolute value of kTP was not considered here, but rather the re lative differences from cell to cell and month to month (Pietro et al., 2009), for simplicity, it was calculated with C* = 0: kTP = Q1+ Q22 WAD ln C2C1 ( 42 ) where C1 = inflow concentration [M/L3], C2 = outflow concentration [M/L3], and Q1 = flow rate in [L3/T], Q2 = flow rate out [L3/T], and WAD = wetted area, determined by elevation distribution [L2]. The assumption of a constant C*value for all STA cells (in this case C*=0) presents a special difficulty. This and previous chapters function on the recognition of spatially and temporally variable C*, m eaning that the use of any c onstant C* is inappropriate. Unfortunately, the value (or func tion) that should be used for C* for each STA cell is yet unclear, though this chapter atte mpts to clarify the matter. The multiple regressions were performed using SAS 9.1 (SAS Institute, Cary, NC). Builtin variableselection procedures were not viable for this dataset; these routines include only observations (months) for which a value was present for every possible factor. For example, the missing inflow Ca concentration value in STA1W Cell 5 in March 2005 would remove all the data for that month from consider ation even if the inflow Ca concentration was not selected for the final model. After the extensive screening process (see Methods, Chapter 1), only about 15% of months had values for each variable. Since it was unlikely that all of the variables would 75 PAGE 76 prove to be useful in the final model, manual incremental construction of the model was deemed more appropriate. The regression model building approach was similar for both the outflow TP concentration and the TP areal settling rate. In all instances throughout this chapter, the adjusted coefficient of determination ( r2) is reported, to enfo rce parsimony and appropr iately penalize the measure of fit for the number of dependent variables included in each model. The manual variable selecti on procedure was intended to be straightforward. The r2 was determined for each univariate model. Next, bivari ate models were tested using the top regressor (variable producing the highest r2) from the singlevariable models and iteratively inserting each other variable. Trivariate models were tested by iteratively introducing each remaining variable into the best (highest r2) bivariate model. This process was repeated until the marginal gains in explanatory power (increase in r2 due to each additional variable) were deemed trivial. Each cell had a different POR m ean outflow TP concentration ( Figure 34 ) and settling rate (not shown). The mean value for each cell may be thought of as an estimate of each monthly value in each cell. The variability in the cell m eans explained a portion of the overall variability in the monthly values. It was possible to m odel the effects of each cell on the dependent parameters. At each level of complexity, the model was also tested with the cell effects to assess the extent that each additional variable account ed for celltocell differences in outflow concentration and settling rate. Variables Considered The inclusion of specific variables for c onsideration was based primarily on data availability with a focus on wetland attributes commonly thought to influence P processing. Some wetland characteristics that may be comm only expected to influence wetland P processing were necessarily omitted for lack of quantitat ive data in the monthly timescale (e.g. plant biomass). Altogether, 21 quantitative variable s were tested in each regression model. 76 PAGE 77 kC* model terms A great deal of research has supported the vali dity of the kC* model for application to a variety of wetlands (Kadlec and Knight, 1996; Kadlec and Wallace, 2008). As discussed previously (see Chapters 1 and 2), the model captures some intrinsic relationships between several wetland parameters. For example, as desc ribed by the model, increas es in the inflow TP concentration result in elevated outflow TP con centrations. The critical model variables inflow TP concentration and q were included in the set of vari ables for regression. The inflow TP concentration multiplied by q yields of course the TPALR. The intuitive notion that the outflow TP concentration ought to be correlated to th e TPALR has been confirmed by many studies of longterm data (e.g. Qian and Richardson, 1997; Pietro et al., 2009; Ka dlec and Wallace, 2008), so the nominal (based on the findings of Chapter 2) TPALR was also included in the regression analysis. Wetlands integrate continuous loads and so outflow concentrations may reflect past operating conditions as well as recent inflow events (Jus ton and DeBusk, 2006; Reddy and DeLaune, 2008), so both the 1year and 2year rolling average TPALRN were incorporated in the variable set. Inflow phosphorus fractions Within wetlands, not all Pcontaining compounds are treated equally. So luble reactive P is typically preferentially removed be cause it is highly bioavailable (Havens et al., 1999, Dierberg et al., 2002, Pietro et al,. 2006). Particulate P removal may be high if the wetland conditions promote settling and retention of the suspended solids (e.g Braskerud, 2002a,b). When water velocities or other turbations inhibit settling, the P associated with particles has to be enzymatically cleaved before biota can access it, so a portion of the PP may progress through the wetland, escaping removal (Dierberg and DeBus k, 2008). Finally, DOP may contain molecules that resist uptake or transforma tion in the wetland environment. In fact, previous studies of the 77 PAGE 78 Everglades and the STAs have demonstrated poor sequestration of DOP (Proctor et al., 1999; Chimney, 2007). For these reasons, th e nature of the inflow TP pool was expected to affect the treatment of that TP, and so the relative proportions of each fraction were included in the regression variable set. Wetland age and soil phosphorus A small but growing number of treatment wetlands have been operated for P removal for time periods longer than a decade (Kadlec and Wallace, 2008), so the longevity of wetlands for sustained, effective P removal remains a matter of investigation. Two well known examples of treatment wetlands with longte rm performance data, Houghton Lake in Michigan and the Orlando Easterly Wetland in Florida, provide some insight. Following mesocosm and pilot studies, full scale discharge of wastewater into the Hought on Lake wetland began in 1978 (Kadlec, 1993). Nearly 30 year s later, the former sedge m eadow was transformed into Typha marsh and was still providing positive P rem oval (Kadlec and Wallace, 2008). Kadlec and Wallace (2008) note that although Richardson and Marshall (1986) predicted P saturation of the wetland following a total loading of 1.01.5 g/m2, the wetland still provided P mass removal exceeding 80% after cumulative loading of 63 g/m2. However, the percent removal was slightly lower and much more variable in the most recen t 10 years of operation than in the previous 20 years. This variation was attributed to changes in site hydrology and hydraulics (Kadlec and Wallace, 2008). Similarly, the Orlando Easterly Wetland successfully polished tertiary wastewater from the city of Orlando, Florida, from 1988 until 2003 (Sees and Jackson, 2001; Kadlec and Wallace, 2008). After 10 years of operation, a strong seas onal trend developed in the outflow TP concentration, resulting in unacceptably poor P re moval effectiveness in some winter months (Wang et al., 2006). A series of st udies identified that both sediment P recycling (White et al., 78 PAGE 79 2002) and hydraulic inefficiencies (Martinez and Wise, 2003) contributed to the observed reduction in P removal performance. A rejuvena tion project in 20022003 that included sediment removal and earthwork to increase hydrauli c efficiency immediately boosted P removal performance (Wang et al., 2006). The changes in performance after longt erm operation observed in the two preceding examples justify the inclusion of wetland age in the multiple regression exercises. As Wang et al. (2006) alluded to by their consideration of a host of studies to diagnose the problems experienced by the Orlando Easterly Wetland, wetla nd age is a lumped parameter, integrating the effects of many wetland characte ristics that change with time (e.g. soil P, internal hydraulics and vegetation characteristics). Because soil P, in particular, is known to influence the P concentration of the overlying water (White et al., 2003; Reddy and DeLaune, 2008), it was included in the regression analyses in an effort to isolate the integrated effects of age. Data for internal hydraulics and vegetation in the STAs were not available at a temporal scale useful for this analysis. Calcium and pH Phosphorus is well known to coprecipitate w ith Ca in aquatic systems and wetlands (e.g. House and Donaldson, 1986; Diaz et al., 1994; Hartle y et al., 1997; Scinto, 1997). In their review of the literature on the topi c, Reddy and DeLaune (2008) st ate retention of inorganic phosphorus by precipitation will be significant in waters with high Ca2+ and alkalinity. Indeed, in Everglades marsh soils, the accumulation of P was correlated with the accumulation of Ca, providing possible evidence of CaP interactions (Reddy et al., 1993). Therefore, inflow Ca concentration, Ca ALR, and Ca mass removal (M/L2/T) were added to regression analysis to investigate the effects of possible differences in Ca supply and processing among the STAs. In 79 PAGE 80 addition, inflow and outflow pH were included in recognition of the pH dependency of Ca precipitates (Diaz et al., 1994). Water temperature The rates of many microbial processes re levant to the wetland P processing are temperature dependent. For example, rates of de composition of Everglades histosols increased dramatically with increasing temp eratures (Volk, 1973). Indeed, the release of soluble P from an organic wetland soil was strongly dependent on temperature (Kadlec and Reddy, 2001). By these bases it may be expected that P removal in the STAs would also show some temperature dependence, but previous studies have suggested relatively little infl uence of temperature on measured P removal in wetlands (Kadl ec and Reddy, 2001; Kadlec and Wallace, 2008). Nonetheless, inflow and outflow water temperatur e were incorporated in the multiple regression to increase the comprehensiveness of this study. Relative wetted area and water depth The biogeochemical influences of RWAD are discussed in Chapter 2. In that chapter it was found that the RWAD was very nearly 1.0 in the ST As, and therefore no significant biogeochemical impacts of the wetted area were detected. However, RWAD was included in the regression exercise in the event that it offere d some explanatory power when combined with other wetland characteristics. The findings of Chapter 2 also indicated that, though infrequent, drawdown/reflood events were important determ inants of the outflow TP concentration. They were therefore also included in the multiple regression analysis. Water depth influences the wetla nd characteristics and processe s such as vegetation types, internal hydraulics and oxygen diffusion (Kadlec and Wallace, 2008) Studies on the integrated effect of water depth on P removal in wetlands associated with the STA project have produced unclear results; at high P concen trations, depth inhibited remova l for cattails (Chimney et al., 80 PAGE 81 2004) and SAV (Jorge et al., 2002), but did not produce a pronounced deleterious effect at low P concentrations. Despite this, the individual bioge ochemical effects of water depth in wetlands call for its inclusion in the multiple regression investigation. Results and Discussion Outflow Total Phosphorus Concentration: All Cells The results of the variableselec tion procedure are presented in Table 41 and Figure 41 Altogether, only about 32% of the total variability in the monthly outflow TP concentration was explained by the final 5variable model. The re latively poor explanatory power of this model was unexpected since factors known to influence treatment wetland outflow concentration (e.g. inflow TP concentration) (Kad lec and Wallace, 2008) were included in the selection process. However, this finding was congruent with the very weak connection found between the TPALR and the outflow TP concentration in most cells (demonstrated in Chapter 3). The independent variables included in the final model were (in order of selection) inflow TP concentration, WAD, wetland age and inflow Ca concentration. When added to these five, one other variable, inflow pH, showed a significant effect on the out flow TP concentration. However, it did not increase the strength of the model fit to the da ta, and was excluded for parsimony. Estimates of the parameters (coefficien ts) for each explanatory factor in the final fivevariable model are shown in Table 42 These values may be inte rpreted as the number of units of change in the outflow TP concentratio n reflecting one unit of change in the subject variable, holding the other wetland characteristics constant. The prec ise values of these estimates are somewhat sampledependent (i .e. they would change with the inclusion of an additional data year) and change with the introduction or remova l of other variables in the model. The relative magnitudes (within parameters of similar units) and signs (positive or negative) are of more value for interpretation. 81 PAGE 82 At every level of complexity, the inclusion of the cell effects c ontributed a significant portion of explanatory power to the model. In f act, this additional variability explained in the outflow TP concentration appeared to be a dditive with the variance accounted for by the included wetland attributes. Therefore, the we tland characteristics accounting for observed differences in outflow concentration across cells were evidently not considered in this study. The reduction in r2 after the inclusion of the fifth variable was a result of marked colinearity between the cell effects and the wetland character istics, penalized by the adjusted r2. Among the five variables in the outflow TP concentration model, several findings were of interest. First, the relationship between inflow concentration and outflow concentration was positive, as expected from Equation ( 119 ) However, the inflow concentration accounted for only 20% of the variability in the outflow concen tration, in agreement w ith the poor outflow TP concentrationTPALR correlations found in Chapter 3, and exposing the re lative influence of other wetland attributes. Sec ond, as suggested in Chapter 2, the positive coefficient associated with changes in WA confirmed that months of reflooding tended to have higher outflow concentrations. Third, increases in inflow Ca co ncentration slightly decreased the outflow TP concentration, an expected relatio nship based on many studies of th e CaP interactions in South Florida wetlands (e.g. Reddy et al ., 1993; Dierberg et al ., 2002). However, within this dataset, the effect of the varying inflow Ca concentratio ns was minor relative to the influence of the inflow TP concentration; the re lative impact of incr easing the inflow Ca concentration 1.0 mg/L was 3 orders of magnitude less than the eff ect of a 1.0 mg/L increase in the inflow TP concentration. Fourth, outflow concentrations tended to increase with wetland age. This finding, if it can be satisfactorily corroborated, has undesirable implicati ons for the ability of the STAs to produce 82 PAGE 83 low outflow concentrations over time. For exampl e, P removal in a large municipal treatment wetland in Orlando, Florida started to flag afte r 13 years of operation resulting in a massive rejuvenation project including muck removal a nd other significant earthworks (Wang et al., 2006). Indeed, sediment removal projects have also been undertaken in STA1W, the longest running STA, though not directly in response to declining P removal (Pietro et al., 2008). If ultimately necessary in additional STAs as they age, maintenance requirements of this nature will undoubtedly reduce the cost effectiveness of the STA project. However, wetland age is likely a lumped term, containing simultaneously the effects soil P concentration and other soil characteristics, plant biomass P concentration and possibly unknown others. Effectively combating any deleterious effects associated with aging STAs will require the evaluation of the isolated effects of each of these factors. Finally, the positive correlation between and outflow TP concentration was surprising. This suggested that months with longer resi dence times produced rela tively higher outflow concentrations. While contrary to the logic of rateb ased reaction models of P reduction (e.g. the kC* model), an explanation may lie in the findings of Chapter 3. It was determined that, in the majority of STA cells, the outflow TP concentration was independent of the TPALR and therefore, it was proposed, expre ssed the apparent background concentration. In this case then, residence times were generally sufficient to acco mmodate all potential P removal. Thus, it is hypothesized that any extension of the beyond that needed to achieve maximum treatment of inflow P simply expanded the opportunity for au tochthonous P production, resulting in the slight positive relationship between the outflow TP concentration and Outflow Total Phosphorus Concentration: Cells with Nonsignificant r values The variable selection procedur e resulted in a trivar iate model that explained about 54% of the variability in the monthly outflow TP c oncentration from cells with nonsignificant r values. 83 PAGE 84 The final model included the following factors: change in RWAD, 1year rolling average TPALR, and wetland age. The number of variab les included was limited to three because no additional factor explained a significant ( p=0.05) portion of the variability in the dependent term. The regression model considering on ly the cells with nonsignificant r values was different from the model for all STA cells in one considerable way. As expected, the inflow TP concentration was no longer included in the model. The presence of this term in the model for all cells was due only to the inclusion of those four cells (Cells 14 of STA1W) that had previously demonstrated a correlation between infl ow and outflow TP concentration. Two important interpretations may be drawn fr om the three variables selected for this model. First, the movement of RWAD to the forefront reinforces earlier findings of its significance in the STAs. Though infrequent, large re flood events appear to be one of the most powerful and predictable biogeochemical processe s in the STAs. Second, the inclusion of both the 1year rolling average TPAL R and wetland age indicates that the longterm operation of these STAs plays a measurable role in their s hortterm performance. Both of these terms had positive coefficients meaning that the monthly outflow TP concentration tended to increase both in response to large loads throughout the precedin g year and as the wetlands aged. Evidently, even though these wetlands assimilate P quick ly enough to disassociat e the monthly outflow from the monthly inflow concentration, the appare nt background concentration is still subject to the influence of longerterm loading. A wetland achieving apparent b ackground concentrations could presumably receive additional loads without experiencing reciprocal elevations in outflow concentration. That notion is countered by the apparent dependence of the background concentration on longterm loads. Therefor e, additional loads may be expected to eventually 84 PAGE 85 increase the outflow concentra tion, even if inflow and outflow concentrations remained independent. Outflow Total Phosphorus Concentration: Single Cell This and other works have clearly established the celltocell variability in performance in the STAs (e.g. Juston and DeBusk, 2006; Pietro et al., 2008). The multiple regression exercises in this chapter presumed that variation in a particular wetland ch aracteristic would have the same relative effect on treatment performance regardle ss of whether the change was observed between two different cells, or across time in a singl e cell. Because the results of the multicell regressions for outflow TP concentration were not wholly satisfying (i.e. the final adjusted r2values were not above 0.54 and most included variables explained trivial proportions of the variance), it was hoped that additional clarity w ould be gained by regressing the data from a single cell. Cell 1 from STA1W was selected for this ex ercise because it had the largest number of data months ( n=70) that had observations for each of the 21 tested variables. The final model had an adjusted r2 of about 0.54 and included only two independent variables, inflow TP concentration and wetland age ( Table 45 ). When added to these two factors, no additional variable explained a significant portion of th e variability in outflow TP concentration. This singlecell regression provi ded little additional informa tion. A significant correlation between inflow and outflow TP concentrations in STA1W Cell 1 was established in Chapter 3, so the inclusion of inflow TP concentration was expected in the final model. Not surprisingly, the inflow TP concentration term contributed much more to this singlecell regression than it did in the previous exercise that included cel ls with both significant and nonsignificant r values; the univariate regression model including only inflow TP concentration had an adjusted r2 of 0.38 for STA1W Cell 1 data only and an adjusted r2 of 0.20 for data from all cells. Serious 85 PAGE 86 management implications arise when wetland age appears in the regression model with a positive coefficient, but these are discussed previous ly in this chapter, as well as Chapter 5. Total Phosphorus Areal Settling Rate: All Cells The results of the variable sel ection procedure are presented in Table 43 and Figure 42 Altogether, the final sixvariable model explained about 51% of the variabili ty in the monthly TP areal settling rate. The first variable added to the model, qN, accounted for the first 40% while, in total, the next five variables, inflow DOP fraction, inflow PP fraction, outflow temperature and TPALRN, increased the r2 of the model only about 0.10. The importance and influence of qN was reflected in the relatively high r2 of the singlevariable models with TPALRN and Ca ALRN. Variableinclusion was stopped after six factors because the addition of the last term, TPALRN, increased the explanatory power of the model by less than 1%. The r2 of the final model (0.512) was unsatisfactory, considering the effort invested in collection a nd compilation of the extensive dataset. The declining disparity with increasing model complexity between the r2values of the models with and without the incl usion of the cell effects suggests that the six wetland attributes in the final model explained a portion of the variability due to celltocell differences. Each additional factor accounted for a small portion of this variance, and the se ttling rate differences across cells were apparently not due to a single wetland characteristic. Estimates of the parameters (coefficients) fo r each explanatory fact or in the final sixvariable model are shown in Table 44 As with the outflow concentration model, these values may be interpreted as the number of units of ch ange in the TP areal settling rate reflecting one unit of change in the subject variable, holding the other wetla nd characteristics constant. As before, the relative magnitudes (within paramete rs of similar units) and signs (positive or negative) are of primary in terest for interpretation. 86 PAGE 87 The set of variables selected for the final m odel was noteworthy for several reasons. First, as noted earlier, qN was the primary determinant of the monthly settling rate. Months with high hydraulic loading tended to have higher kTPvalues, holding all else constant. As with the relatively poor relationship between inflow and outflow TP concentration, this is likely a result of the Chapter 3 finding that the outflow TP concentration was independent of the TPALR in most cells. The implication that follows from th at independence, that the outflow concentration reflected an apparent background concentrati on, and thus that maximum potential removal occurred in most months, means that the magni tude of the areal mass removal ought to have been based more or less strictly on the loading rate. This was evidenced by the inclusion of both qN and TPALRN in the final model. Second, the negative coefficients associated with both the inflow DOP and PP fractions confirm, primarily, that of the three P fractions, the STAs most effectively remove SRP. Possibly the independent effects of DOP and PP were such that the information contributed to the model by these two fractions was more useful than simp ly providing the SRP fraction (SRP = TP PP DOP). Third, Pietro et al. (2009) found a negativ e relationship between POR outflow TP concentration and POR kTPvalue, with lower concentrations generally produced from cells with higher settling rates. That interaction may have been expres sed in these data through the age term. While increasing wetland age tended to incr ease the outflow TP con centration, older cells had a greater likeli hood of having lower kTPvalues. As with the outflow concentration model, age likely contains the combined effects of a vari ety of wetland features th at must be separated for a thorough interpretation and assessmen t of the implications for the STAs. 87 PAGE 88 Finally, that the TP settling rate was positively related to the outflow water temperature was expected. Kadlec and Wallace (2008) review ed the matter and their conclusion closely matched the result of this study; the TP settl ing rate tends to be slightly dependent on temperature, but temperature explains ve ry little variance in the shortterm kTPvalue. Total Phosphorus Areal Settling Rate: Single Cell As was done for outflow TP concentra tion, a multiple regression model of kTP was assembled for the monthly data from STA1W Cell 1. The final model included only the terms water depth and wetland age and had an adjusted r2 of only 0.18 ( Table 48 ). The selection of depth as the first factor in the model was not expected; this was the only final model in this study in which it appeared. Apparently, however, th is was another expressi on of the dependence of kTP on q that was observed in the exercise that includ ed all cells. In STA1W Cell 1, water depth and q were highly collinear (r =0.686, p<0.0001, n=94). It is unclear why de pth proved to be a better regressor than q. Once again wetland age was a significan t contributor of explained variance. The negative coefficient indicated that the monthl y calculated settling rate declined as this cell aged, though the poor fit of this model (low r2) undermines the strength of any conclusions drawn from it. Note the discussion above in this chapter, as well as in Chapter 5 on the management implications of aging wetlands. Limitations and Future Application of the Multiple Regression Technique Multiple regression is attractive for analysis of the deterministic factors in wetlands because it partitions explanatory power among many variables, allowing the isolation of the effects of particular wetland characteristics. The STAs are exceptionally well suited for this technique because of the breadth and quality of the data available. However, the ability to assess a large number of factors simultaneously introdu ces a significant limitation to the technique. Multiple regression can only consider observations for which valid data are available for every 88 PAGE 89 variable included in the model being tested. This can (and in the case of the dataset considered here, did) result in loss of potentially valuable information. As model complexity increased, so did the likelihood of encountering a variable, within a given obser vation, for which the value had been screened. Also, the regression techniques as applie d here were limited to assessing linear relationships between the dependent and independent variables. N onlinear relationships commonly exist among wetland data at different time scales (e.g. between POR outflow TP concentration and TPALR in Piet ro et al., 2009). Potentially th is study failed to capture nonlinear, but important, relationshi ps within this dataset. As mentioned previously, this investigation wa s intended to serve as a first step toward identifying wetland characteristics important to P treatment. Future researchers may contemplate several considerations to advance this technique. The inclusion of additional wetland attributes as model variables not considered in this study should be a first priority. The lowhanging fruit consist of both unique additional variables (e.g. herbicide application or dissolved oxygen) and further permutations of the data included here (e.g. cumulative PP areal loading) Quite possibly though, data for key variables are not currently being collected altogether or at useful spat ial and temporal resolutions. Undoubtedly, any call for additional data collection at the massive scale of the STAs would need to be well defended by findings in the literature and potentially even labor mesocosmlevel studies. Second, new and existing data should be investigated at additional time scales. Some factors (e.g. soil P or plant biomass) may not vary sufficiently over short (monthly) time scales for their effects on P treatment to be captured by regression techniques. Quarterly, annual and 89 PAGE 90 POR averaging periods are sugge sted, acknowledging that each re duction in temporal resolution will reduce the available nu mber of observations. Finally, it is recommended to use multiple regr ession to investigate possible interactions between wetland characteristics. For example, hypot hetically, the effect of depth on P treatment could differ between vegetation type s. Including interaction terms complicates regression efforts, and it is suggested that such an investigation be initiated with expected interactions (e.g. Ca concentration and pH), and expanded to incl ude unforeseen interactions if necessary. Conclusions Within the data available for this analysis fluctuations in th e monthly outflow TP concentration and the monthly TP areal settling rate were rela tively poorly explained by any of the 21 wetland characteristics tested. In combination, the five variables inflow TP concentration, WAD, wetland age and inflow Ca concentration accounted for about 32% of the variability in the monthly outflow TP concentration. About 51% of the variability in the monthly TP areal settling rate was explained by the linear model containing the six parameters qN, inflow DOP fraction, inflow PP fraction, wetland age, outflow water temperature and TPALRN. In both cases, the bulk of the explained variability was accounted for by single wetland ch aracteristics; inflow TP concentration most strongly controlled outflo w TP concentration and the settling rate was most closely related to the qN. In both models, each of the re maining variables contributed meagerly to the r2. The guiding hypothesis, that the tested variables would significantly explain the variance in the outflow TP concentration and th e TP areal settling rate was rejected, in spirit if not in letter, by the find ings of this chapter. The relatively large amounts of unexplained vari ance and the distributions of the resolved variability in both tested dependent variables su pport several important co nclusions. First, while 90 PAGE 91 Chapter 3 demonstrated the importance of identifying a ppropriate values or functions for C* for each STA cell, this chapter revealed the challenge of doing so. Second, it is likely that additional wetland characteristics not quantified in this st udy were and are influential on the shortterm outflow TP concentration and the TP settling rate. Although fully account ing for all the wetland attributes that cause these pe rformance indices to vary (e.g. r2 = 1.0) is the conceptual goal, all the error accumulating from measurement, inflowoutflow lag due to positive and the limitations of multiple linear regression will manifest as unexplained variance. The relative proportions of the unresolved vari ability in this study contributed by missing variables and other sources of error remain undetermined. Third, th e shortterm measured outcomes of wetland P processing, which are known to be quite comple x, may in fact depend on such an array of wetland attributes that even a ma jority of the variability may not be explained by only a handful of measured traits. The small individual contribu tions of most of the variables examined herein attests to this conclusion. In this case, the data collection and processing required to substantially or fully explain the variability in, say, outflow TP concentration, would almost certainly surpass the usefulness of this information. Readers interested in practical applications of the interpretations in this chapter regarding the potential to improve TP removal performance may find the conclusions of this chapter unsatisfying, particularly with respect to the shortterm outflow TP concentration. First, the factor that explained the major ity of the variance in the monthl y outflow concentration (inflow TP concentration) is not reasonably subject to manipulation by STA managers. Conceivably, inflow concentrations could be lowered by re ducing P losses from upstream sources, but this approach falls outside the realm of improving wetland treatment effectiveness. Second, though the longterm outflow TP concentration and kTPvalue were negatively associated (Pietro et al., 91 PAGE 92 2009), attempts to increase th e settling rate by increasing q (the primary driver of the settling rate) should not be expected to depress the outfl ow concentration. Finally, the relative lack of influence (with regard to improvements in r2) of each of the additional variables that contributed to the outflow TP concentration and the settling rate suggests that modifying any of them would be unlikely to result in costeffective reductions in outflow concentrations. 92 PAGE 93 Table 41. Coefficients of determination ( r2) of multiple linear regression models explaining the monthly outflow total phosphor us concentration in all cells. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by X. The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 4 5 6 Inflow TP concentration 0.198 X X X X X TPALRAnominal 0.070 0.287 0.304 1yr rolling average TPALR 0.014 0.279 0.301 2yr rolling average TPALR 0.260 Inflow SRPA fraction 0.004 Inflow DOPA fraction Inflow PPA fraction 0.010 0.225 0.268 0.276 Inflow CaA concentration 0.021 0.215 0.266 0.290 0.323 X Ca ALR nominal Ca areal mass removal rate Wetland age 0.032 0.212 0.258 0.301 X X qA nominal A 0.224 0.281 X X X Mean water depth 0.003 Soil TP concentration Inflow pH 0.004 0.212 0.258 0.289 0.303 0.323 Outflow pH 0.245 Inflow water temperature 0.250 Outflow water temperature 0.243 WAD A 0.004 0.205 0.253 0.289 Change in WAD 0.028 0.247 X X X X ATPALR = TP areal loading rate; SRP = soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = calcium; q = hydraulic loading rate; = hydraulic residence time; WAD = wetted area, determined by the elevation distribution 93 PAGE 94 Table 42. Estimates of parameters for the m odel explaining monthly out flow total phosphorus (TP) concentration in all cells. Variable Units Parameter estimate ( ) Probability ( p) that = 0 Outflow TP concentration mg/L Inflow TP concentration mg/L 0.4387 <0.0001 Change in wetted area % 0.2443 <0.0001 Hydraulic residence time d 0.0004 <0.0001 Wetland age yr 0.0051 <0.0001 Inflow Ca concentration mg/L 0.0004 <0.0001 94 PAGE 95 Table 43. Coefficients of determination ( r2) of multiple linear regression models explaining the monthly outflow total phosphorus (TP) concen tration in all cells with nonsignificant r values. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by X. The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 Inflow TP concentration 0.056 0.477 TPALRAnominal 0.014 0.445 1yr rolling average TPALR 0.523 X 2yr rolling averag e TPALR Inflow SRPA fraction 0.032 Inflow DOPA fraction 0.019 Inflow PPA fraction 0.042 Inflow CaA concentration Ca ALR nominal Ca areal mass removal rate 0.016 Wetland age 0.093 0.470 0.537 qA nominal 0.443 A Mean water depth 0.116 0.490 Soil TP concentration 0.133 Inflow pH 0.032 0.425 0.459 Outflow pH 0.026 Inflow water temperature 0.028 Outflow water temperature 0.024 WAD A 0.014 Change in WAD 0.437 X X ATPALR = TP areal loading rate; SRP = soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = calcium; q = hydraulic loading rate; = hydraulic residence time; WAD = wetted area, determined by the elevation distribution 95 PAGE 96 Table 44. Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in all cel ls with nonsignificant r values. Variable Units Parameter estimate ( ) Probability ( p) that = 0 Outflow TP concentration mg/L Change in wetted area % 0.2080 <0.0001 1yr rolling average TPALR g/m2/yr 0.0365 0.0378 Wetland age yr 0.0018 0.0068 96 PAGE 97 Table 45. Estimates of parameters for the model explaining outflow total phosphorus (TP) concentration in STA1W Cell 1. Variable Units Parameter estimate ( ) Probability ( p) that = 0 Outflow TP concentration mg/L Inflow TP concentration mg/L 0.4254 <0.0001 Wetland age yr 0.0149 <0.0001 97 PAGE 98 Table 46. Coefficients of determination ( r2) of multiple linear regression models explaining the monthly total phosphorus (TP) areal settling rate in all cells. The column heading indicates the complexity (number of variables included) in the model. For a particular level of complexity, a given value is the r2 for the model containing that variable and all the variables indicated by X. The highest r2value in each column is boldfaced. Number of variables included in model Variable 1 2 3 4 5 6 Inflow TP concentration 0.030 0.397 0.446 0.506 TPALRAnominal 0.277 0.395 0.448 0.474 0.489 0.512 1yr rolling average TPALR 0.041 0.416 0.447 0.478 0.484 2yr rolling average TPALR 0.044 0.393 Inflow SRPA fraction 0.028 0.422 0.461 Inflow DOPA fraction 0.017 0.432 X X X X Inflow PPA fraction 0.026 0.401 0.469 X X X Inflow CaA concentration 0.008 0.411 Ca ALR nominal 0.346 Ca areal mass removal rate 0.401 0.462 Wetland age 0.421 0.458 0.481 0.503 X qA nominal 0.391 X X X X X A 0.117 0.387 0.437 Mean water depth 0.036 0.439 Soil TP concentration 0.389 0.443 Inflow pH 0.026 0.393 0.448 0.476 0.490 Outflow pH Inflow water temperature 0.038 0.396 0.445 0.489 0.504 Outflow water temperature 0.058 0.416 0.456 0.486 X X WAD A 0.402 0.447 0.488 Change in WAD 0.003 0.395 0.446 0.478 0.495 0.511 ATPALR = TP areal loading rate; SRP = soluble reactive P; DOP = dissolved organic P; PP = particulate P; Ca = calcium; q = hydraulic loading rate; = hydraulic residence time; WAD = wetted area, determined by the elevation distribution 98 PAGE 99 Table 47. Estimates of parameters for the mode l explaining total phosphorus areal settling rate in all cells. Variable Units Parameter estimate ( ) Probability ( p) that = 0 TP areal settling rate m/yr Hydraulic loading rate nominal m/mo 6.6940 <0.0001 Inflow DOP fraction % 33.7914 <0.0001 Inflow PP fraction % 13.2197 0.0012 Wetland age yr 1.8512 <0.0001 Outflow water temperature C 0.6224 0.0064 TP areal loading rate nominal g/m2/yr 11.0970 0.0002 ATP = total phosphrous; DOP = dissolved organic P; PP = particulate P 99 PAGE 100 Table 48. Estimates of parameters for the mode l explaining monthly tota l phosphorus (TP) areal settling rate in STA1W Cell 1. Variable Units Parameter estimate ( ) Probability ( p) that = 0 TP areal settling rate m/yr Water depth m 50.5583 0.0001 Wetland age yr 1.7433 0.0221 100 PAGE 101 TP C1 WAD Age Ca C1 ++++ 0 0.1 0.2 0.3 0.4 0.5 012345r2valueNumber of variables included in model With WithoutCell effectsFigure 41. Changes in the coefficient of determination ( r2) with increasing complexity of the model explaining monthly outflow TP concen tration in all cells. The models are shown with and without the effects of the periodofrecord cell means. TP C 1 = inflow TP concentration, WAD = change in monthly wetted area, = nominal hydraulic residence time, Age = wetland age, Ca C 1 = inflow calcium concentration. 101 PAGE 102 q DOP1RPP1RTemp2AgeTPALRN+++++ 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0123456r2valueNumber of variables included in model With WithoutCell effectsFigure 42. Changes in the coefficient of determination ( r2) with increasing model complexity of the model explaining the monthly total phospho rus areal settling rate in all cells. The models are shown with and without the eff ects of the periodofrecord cell means. q = hydraulic loading rate, DOP1R = inflow dissolved organic P fraction, PP1R = inflow particulate P fraction, Temp2 = inflow water temperature, Age = wetland age, TPALRN = nominal total phosphorus areal loading rate. 102 PAGE 103 CHAPTER 5 CONCLUSIONS This work attempted to address several im portant questions regarding the operation and performance of the Stormwater Treatment Areas in South Florida. The management of these extensive constructed wetlands is a massive expenditure for the SFWMD and is a major contributor to the reduction of P loads to the Florida Everglades. The presence of marked eleva tion gradients and timevariable water levels in some STA cells raised concern that estimat es of the hydraulic and TP areal loading rates we re inaccurate due to incomplete flooding of the nominal tr eatment areas. It was demonstrated that the occasions when the relative wetted area was s ubstantially less than 100% were infrequent, and did not significantly increase the TPALR in a ny of the STAs, suggesting that the nominal treatment area was satisfactory for loading rate calculations. The outflow TP concentration did show a positive correlation to the magnitude of reflooding, within the months with large reflood events, an effect confirmed to be signifi cant by multiple regression modeling. Maintaining flooding as to reduce rewetting events would prev ent these occasional pulses of outflow P but this phenomenon was important in less than 5% of nonscreened months. It was also shown that poor performance in certain cells (elevated outflow TP concentrati ons relative to inflow TPALR) was shown not to be an artifact of the TPAL R calculation, validating the need for additional work to diagnose those factors contributing to performance in the STAs. The monthly outflow TP concentration was shown to be uncorrelated to the monthly TPALR ( r ) in most cells. This independence was not a function of vegetation type or the magnitude of the TPALR. It was hypothesized that the Damkhler number (Da = k / q ) was sufficiently high as to remove this correlati on, implying that the outflow concentration was controlled primarily by the background concentration. The monthly outflow TP concentration 103 PAGE 104 from the cells with nonsignificant r was not determined by the composition of the inflow TP pool nor by the outflow water te mperature. More extensive efforts to isolate the factors controlling the apparent background concentrations were justified because accurate estimates of C* appeared necessary for the successful employment of the kC* model in wetlands (i.e. most STA cells) with high Davalues. Regression Variables Each of the 21 factors was specifically include d because either mathematical reasoning (in the case of those terms related by the kC* mode l) or previously documented studies suggested their relevance to wetland P cyc ling (e.g. Ca, pH and temperature). Therefore, it is important to consider the conditions that led to inclusion or rejection of each in the models assembled in this work. Inflow Total Phosphorus Concentration The kC* model (Kadlec and Kni ght, 1996), introduced in Chapter 2, has been found by many investigators to satisfactorily describe po llutant reduction in wetlands. When using that model to consider the outflow TP concentration, the inflow TP concentration is an important input term (which makes sound intuitive sense). Chapter 3 showed that in the majority of the cells in the STAs, the outflow TP concentra tion in any given month was unrelated to the conditions of the inflow water in that month. This situation suggests that P removal in the most of the STAs is not limited by reside nce time (discussed further in Chapter 3), but does imply that the inflow TP concentration ought not to be us eful in describing the shortterm outflow TP concentration. Indeed, the inflow TP concentrat ion was only important to the regression model that considered outflow TP concentration data from cells that were known to have positive correlations between inflow and outflow TP con centrations. Clearly, the shortterm inflow TP 104 PAGE 105 concentration will not be an important determin ant of the outflow TP concentration in any wetland system whose P reduction is not limited by residence time. Hydraulic and Total Phosphorus Loading Rates The hydraulic ( q) and total phosphorus loading rates came to bear in interesting ways in the final regression models established in Chapter 4. First, both the monthly q and TPALRN contributed positively to the monthly areal settling rate when all cells of the STAs were considered together, suggesting that kTP is, in the STAs, a measure of P removal power rather than efficiency. This idea is in line with th e previously discussed no tion that the observed outflow TP concentrations are limited more by the apparent background concentration than by insufficient contact time. In this case, a hypotheti cal additional packet of water (thus additional P) does not increase the outflow concentrati on, but does lead to increased P mass removed. Second, the 1year rolling averag e TPALR was the second most impor tant factor in determining the monthly outflow TP concentration fr om those cells with nonsignificant r values. That is, even when the shortterm outflow concentration was apparently independent of the shortterm inputs, the longterm loading still influenced the shortterm outflow concentration. This is simply a different perspective on the wellunderstood connection between longterm loading and performance (e.g. Qian and Richardson, 1997; Pietro et al. 2009), but deserves consideration by treatment wetland managers. Increasing loading ra tes to maximize mass removal in response to the independence of shortterm inflow and outflow concentrations may result in a longterm increase in outflow concentrations, visible even in shortterm data. Inflow Phosphorus Fractions Despite appearing in the final regression model for kTP (data from all cells) the composition of the inflow TP pool did little to affect the measured perfor mance of the STAs. Possibly, the internal transformation and producti on of the various forms was so substantial as to make the 105 PAGE 106 initial P composition irrelevant This hypothesis gains some support from the finding that residence time did not limit P reduction in most STAs. Regardless of the inflow P composition, additional contact time would not support additional removal of the lessbioavailable DOP and PP. Therefore, the outflow DOP pool, for example, is likely not composed primarily of the same DOP particles that entered the wetland. Wetland Age and Soil Phosphorus Wetland age was the only term included in all the final regression models for both outflow TP concentration and kTP. In all cases, the sign on the co efficient indicated decreasing performance (i.e. increasing outflow TP concentrations and decreasing kTPvalues) with increasing age. Of course, there are precedents for such a scenario including the Orlando Easterly Wetland in central Florida (Wang et al., 2006) an d the impacted zones of WAC2A in south Florida (DeBusk et al., 2004). Wetland age was su ch a prominent factor in determining STA performance probably because it is a lumped term, containing simultaneously multiple wetland characteristics that change over time, such as soil P and plant biomass and tissue P. Though wetland managers should anticipate changes in performance as treatment wetlands age, this information is only of real valu e if the individual variables po oled in the age term can be extracted and evaluated individually. For example, soil P, a wetland parameter expected to change as the wetland ages, was in cluded in the regression analyses but was not selected for any of the final models. (Thi s analysis should not be taken as conc lusive, as the soil P data were illsuited for evaluation at a monthly timestep.) Ther efore, some other char acteristic of the STAs was changing with time and caused the broad age term to contribute more information to the regression than the soil P term. 106 PAGE 107 Calcium and pH Generally, none of the calcium terms included in the multiple regression exercises (inflow Ca concentration, Ca ALR, and Ca mass removal) contributed any explanatory power regarding either outflow TP concentration or TP areal set tling rate. The sole exception was the selection of inflow Ca concentration as the fifth variab le in the regression model for outflow TP concentration among all cells. Likewise, neither inflow nor outflow pH was selected for inclusion in any model. The chemistry of Ca a nd P interactions (and th e influence of pH) is moderately well understood (e.g. Di az et al., 2004; Scinto, 1997) and associations between Ca and P have been demonstrated in the field (R eddy et al., 1993). Connec ting the processlevel dynamics to the fieldscale effects has been ch allenging. Two factors may have contributed to the relative lack of influence of Ca on P observe d in this study. First, the amount of P removed from the water column due to interaction with Ca actually may be minimal relative to biological and physical processes. The probability of this being the case is unclear ; CaP interaction is apparently important in some wetlands (Reddy et al., 1993) but has not been quantified in the STAs. Second, the data possibly failed to capture the CaP association. The relatively small proportions of variance expl ained by most of the considered factors suggest that some of the tested datasets were not suited to analysis at a monthly timestep. Water Temperature Kadlec and Wallace (2008) suggest an Arrhen ius temperature coefficient of about 1.005 for total P in warm climate wetlands. (Arrehe nius coefficients grea ter than 1.0 indicate improvement in removal performance with increases in temperature and values less than 1.0 denote loss of performance with te mperature increases. Values very close to 1.0 imply relatively little temperature dependence.) This value for TP indicates that P removal tends to improve very slightly with increases in water temperature. A temperature term (both in flow and outflow water 107 PAGE 108 temperature were considered) was included in th e final regression model only for the monthly TP areal settling rate that included all cells. Most likely, the weak temperature dependence of P processing expected for the STAs was obs cured by other wetland characteristics. Relative Wetted Area and Water Depth The RWAD was expected to influence P proces sing primarily by alte ring the realized loading rates applied to the STAs. Chapter 2 established that the nominal wetted area provided a satisfactory estimate of the actual wetted area for most applications in the STAs. Therefore, the relative unimportance of RWAD in the regression analyses was anticipated; RWAD was not selected for any of the final models assembled in this study. Of much more interest from the outset was RWAD, since great biogeochemical action was expected to associate with drydown and reflood events. Indeed, RWAD was the second variable incl uded in the regression model for outflow TP concentration from all cells a nd was the factor of primary importance in the explaining variance in the out flow TP concentration from cells with nonsignificant r values. Because large reflood events were fairly infrequent in the STAs (see Chapter 2), these findings indicate that the flushing of P associated with reflood events is one of the most powerful and predictable biogeochemical processes active in the STAs. Depth affects the biogeochemistry (and thus P processing) of surf aceflow wetlands by regulating oxygen diffusion and changing the hydraul ics. However, it was selected for only one regression model developed in th is study, where it explained about 14% of the variability in the monthly kTPvalue of STA1W Cell 1. A correlation between q and depth aligned that model with the dependence of the settling rate on the hydraulic loading rate f ound among all cells. It appears that the other biogeochemical effects of water de pth were apparently relatively unimportant to 108 PAGE 109 STA performance, based on the monthly data. The longerterm role of influencing vegetation communities played by water depth was lik ely not captured by the monthly data. Future Work Following the results presented in this work, two major avenues of investigation remain. First, C* needs further elucida tion in the STAs. Chapter 3 suggest s that the monthly outflow TP concentration from the STAs was mo re strongly controlled by C* than kTP. This implies the need for accurate estimates of C* when modeling P removal performance in the STAs. Although the outflow concentrations from those cells operati ng at apparent backgr ound concentrations did very with time, it remains unclear whether a function or value will best predict C*. Second, multiple linear regression was successful in iden tifying some wetland characteristics that were uniquely important to P processing in the STAs. However, the overall proportions of variance in the dependent vari ables (monthly outflow TP concentration and monthly TP areal settling rate) that were explained by the regres sion efforts were unsatisfactory. It is recommended that future multiple regression efforts for the STAs consider nonlinearity among the data, as well include additional variables not tested in this study. In particular, 12 of the 21 tested variables concerned water chemistry and hydrology and hydraulics accounted for another five terms. Only one term (soil P) wa s related to the soil prop erties and no vegetation data was incorporated. Owing to the relative impor tance of plants and soil to wetland functions, it is likely that variables regarding these char acteristics would be particularly helpful in explaining observed STA performance. 109 PAGE 110 110 APPENDIX DATA SCREENING CRITERIA Variable Not reported if: Inflow TP concentration Inflow volume = 0 Outflow TP concentration Outflow volume = 0 Inflow SRP concentration Inflow volume = 0 Outflow SRP concentration Outflow volume = 0 Inflow DOP concentration Inflow SRP concentration or in flow TDP concentration blank Inflow volume 0 Inflow SRP concentration > inflow TPD concentration Outflow DOP concentration Outflow SRP concentration or outflow TDP concentration blank Outflow volume 0 Outflow SRP concentration > outflow TDP concentration Inflow PP concentration Inflow TP concentration or inflow TDP concentration blank Inflow volume 0 Inflow TP concentration > inflow TPD concentration Outflow PP concentration Outflow TP concentration or outflow TDP concentration blank Outflow volume 0 Outflow TP concentration > outflow TDP concentration Inflow SRP fraction Inflow SRP concentration or infl ow TP concentration blank or 0 Inflow SRP concentration > inflow TP concentration Outflow SRP fraction Outflow SRP concentration or outflow TP concentration blank or 0 Outflow SRP concentration > outflow TP concentration Inflow DOP fraction Inflow SRP concentration or infl ow TDP concentration blank or 0 Inflow SRP concentration > inflow TDP concentration Outflow DOP fraction Outflow SRP concentration or out flow TDP concentration blank or 0 Outflow SRP concentration > outflow TDP concentration Inflow PP fraction Inflow TP concentration or infl ow TDP concentration blank or 0 Inflow TP concentration < inflow TDP concentration Outflow PP fraction Outflow TP concentration or outflow TDP concentration blank or 0 Outflow TP concentration < outflow TDP concentration Areal TP settling rate (k) Inflow volume, inflow TP concentr ation, outflow TP concentration or WAD blank or 0 TP areal loading rate nominal Inflow TP mass blank Inflow volume 0 TP areal loading rate distribution adjusted Inflow TP mass blank Inflow volume or WAD 0 TP areal loading rate mean adjusted Inflow TP mass blank Inflow volume or WAN 0 Inflow Ca concentration Inflow Ca mass or inflow volume blank or 0 Inflow Ca concentration <= 1 PAGE 111 Outflow Ca concentration Outflow Ca mass or outflow volume blank or 0 Outflow Ca concentration <= 1 Ca areal loading rate nominal Inflow Ca mass blank Inflow volume 0 Ca areal loading rate distribution adjusted Inflow Ca mass blank Inflow volume or WAD 0 Ca areal mass removal rate Inflow Ca mass, outflow Ca mass or WAD blank or 0 Hydraulic loading rate nominal Inflow volume blank or 0 Hydraulic loading rate distribution adjusted Inflow volume or WAD blank or 0 Hydraulic residence time Inflow volume or outflow volume 0 Hydraulic residence time 0 Hydraulic residence time 150 111 PAGE 112 LIST OF REFERENCES Anderson, B.H. and F.R. Magdoff. 2005. Relative m ovement and soil fixation of soluble organic and inorganic phosphorus. J. Environ. Qual. 34:22282233. Braskerud, B.C. 2002a. Design cons iderations for increased sedi mentation in small wetlands treating agricultural runoff. Wa t. Sci. and Tech. 45:7785. Braskerud, B.C. 2002b. Factors affe cting phosphorus retention in small constructed wetlands treating agricultural nonpoint source polluti on. Ecol. Eng. 19:4161. Best, E.P.H. and H.W.C. Visser. 1987. Seasonal growth of the submerged macrophyte Ceratophyllum demersum L in mesotrophic Lake Vechten in relation to insolation, temperature and reserve carbohydrates. Hydrobiologia 148:231243. Bostic, E.M. and J.R. White. 2007. Soil phosphor us and vegetation influence on wetland phosphorus release after simulated dr ought. Soil Sci. Soc. Am. J. 71:238244. Chimney, M.J. 2007. The role of largescale tr eatment wetlands in Everglades restoration: design expectations, operational history and realized performance. Inte rnational Society of Limnology, Montral, Canada. Chimney, M.J. 2009. Appendix 58: Water Budgets and Total Phosphorus Mass Balance Budgets and Treatment Performance in STA Treatment Cells and Flowways. Appendix to: Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2009. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, Florida. Chimney, M.J. and G. Goforth. 2001. Environmenta l impacts to the Everglades ecosystem: A historical perspective and restoration st rategies. Wat. Sci. and Tech. 44:93100. Chimney, M. J., M.K. Nungesser, C. Combs, B. Gu, E. FogartyKellis, J.M. Newman, W. Wagner and L. Wenkert. 2004. Everglades Consolid ated Report Chapter 4B: STA Optimization and Advanced Treatment Technologies. South Fl orida Water Management District, West Palm Beach, Florida. DeBusk, T.A., K.A. Grace, F.E. Dierberg, S.D. Jackson, M.J. Chimney and B.H. Gu. 2004. An investigation of the limits of phosphorus remova l in wetlands: A mesocosm study of a shallow periphytondominated treatment system. Ecol. Eng. 23:114. Diaz, O.A., K.R. Reddy and P.A. Moore. 1994. Solubility of inorganic phosphorus in stream water as influenced by pH and calci um concentration. Water Res. 28:17551763. Dierberg, F.E. and T.A. DeBu sk. 2008. Particulate phosphorus tran sformations in south Florida stormwater treatment areas used for Ev erglades protection. Ecol. Eng. 34:100115. 112 PAGE 113 113 Dierberg, F.E., T.A. DeBusk, S.D. Jackson, M.J. Chimney and K. Pietro. 2002. Submerged aquatic vegetationbased treatment wetlands for removing phosphorus from agricultural runoff: Response to hydraulic and nutrient loading. Water Res. 36:14091422. Froelich, P.N. 1988. Kinetic control of dissolved phosphate in natural rive rs and estuaries a primer on the phosphate buffer mechanism. Limnol. Oceanogr. 33:649668. Graneli, W. and D. Solander. 1988. Influence of aquatic macrophytes on phosphorus cycling in lakes. Hydrobiologia 170:245266. Hartley, A.M., W.A. House, M.E. Callow and B.S.C. Leadbeater. 1997. Coprecipitation of phosphate with calcite in the presence of photosynthesizing gr een algae. Water Res. 31:22612268. Harwell, M.C. and K.E. Havens. 2003. Experime ntal studies on the recovery potential of submerged aquatic vegetation after flooding and desi ccation in a large subt ropical lake. Aquat. Bot. 77:135151. Havens, K.E., T.L. East, A.J. Rodusky and B. Sharfstein. 1999. Littoral periphyton responses to nitrogen and phosphorus: An experimental study in a subtropical lake. Aquat. Bot. 63:267290. Higgins, J.P., S. Hurd and C. Weil. 2000. The use of engineered wetlands to treat recalcitrant wastewaters. J. Environ. Sci. Healt h, Part A: Environ. Sci. Eng. 35:13091334. House, W.A. and L. Donaldson. 1986. Adsorption and coprecipitation of phosphate on calcite. J. Colloid Interface Sci. 112:309324. Jackson, J.A. and M. Sees. 2001. Rerating capacity of a constructed wetland treatment system. Wat. Sci. and Tech. 44:435440. Jorge J., J.M. Newman, M.J. Chimney, G. Goforth, T. Bechtel, G. Germain, M.K. Nungesser, D. Rumbold, J. Lopez, L. Fink, B. Gu, R. Bearzotti, D. Campbell, C. Combs, K. Pietro, N. Iricanin and R. Meeker. 2002. Everglades Consolidated Re port Chapter 4: Stormwater Treatment Areas and Advanced Treatment Technologies. South Fl orida Water Management District, West Palm Beach, Florida. Juston, J. and T.A. DeBusk. 2006. Phosphorus mass lo ad and outflow concentration relationships in stormwater treatment areas for Ever glades restoration. Ecol. Eng. 26:206223. Kadlec, R.H. 1993. Natural wetland treatment at Houghton Lake: the first fifteen years. In Proceedings of the WEF 66th Annual Conference, Anaheim, CA. WEF, Alexandria, VA. 7384. Kadlec, R.H. 1997. An autobiotic wetland phosphorus model. Ecol. Eng. 8:145172. Kadlec, R.H. and K.R. Reddy. 2001. Temperature e ffects in treatment wetlands. Water Environ. Res. 73:543557. Kadlec, R.H. and S.D. Wallace. 2008. Treat ment wetlands. CRC Press, Boca Raton. PAGE 114 Lijklema, L. 1977. The role of iron in the exchan ge of phosphate between water and sediments. In H.L. Golterman (ed.) Interactions between se diments and fresh water. B.V. Publishers, The Hauge, The Netherlands. Lowe, E.F. and L.W. Keenan. 1997. Managing phosphorusbased, cultural eutrophication in wetlands: A conceptual approach. Ecol. Eng. 9:109118. Martinez, C.J. and W.R. Wise. 2003. Hydraulic analysis of Orlando Easterly Wetland. J. Env. Eng. 129:553560. Mcconnaughey, T.A., J.W. Labaugh, D.O. Rosenberry, R.G. Striegl, M.M. Reddy, P.F. Schuster and V. Carter. 1994. Carbon budget for a groundwaterf ed lake calcification supports summer photosynthesis. Limnol. Oceanogr. 39:13191332. Mitsch, W.J., J.K. Cronk, X.Y. Wu, R.W. Nair n and D.L. Hey. 1995. Phosphorus retention in constructed freshwater riparian marshes. Ecol. Appl. 5:830845. Noe, G.B., L.J. Scinto, J. Taylor, D.L. Child ers and R.D. Jones. 2003. Phosphorus cycling and partitioning in an oligotrophic Everglades we tland ecosystem: A radioisotope tracing study. Freshwat. Biol. 48:19932008. Nungesser, M.K. and M.J. Chimney. 2001. Evaluati on of phosphorus retenti on in a South Florida treatment wetland. Wat. Sci. and Tech. 44:109115. Olila, O.G., K.R. Reddy and D.L. Stites. 1997. Influence of draining on soil phosphorus forms and distribution in a construc ted wetland. Ecol. Eng. 9:157169. Otsuki, A. and R.G. Wetzel. 1972. Coprecipitation of phosphate with carbonates in a marllake. Limnol. Oceanogr. 17:763767. Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2008. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pietro, K., R. Bearzotti, G. Germain and N. Iricanin. 2009. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pietro, K.C., M.J. Chimney and A.D. Steinman. 2006. Phosphorus removal by the Ceratophyllum /periphyton complex in a south Florid a (USA) freshwater marsh. Ecol. Eng. 27:290300. Pietro, K., G. Germain, R. Bearzotti, and N. Iricanin. 2010. South Florida Environmental Report Chapter 5: STA Performance, Compliance and Optimization. South Florida Water Management District, West Palm Beach, FL. Pinney, M.L., P.K. Westerhoff and L. Baker. 2000. Transformations in dissolved organic carbon through constructed wetlands. Water Res. 34:18971911. 114 PAGE 115 115 Proctor, L.M., G. Shaefer, W.M. Landing. 1999. Bi oavailability of dissolved organic phosphorus in the Florida Everglades: Evaluation with native bacterial and reference enzymes. In Proceedings of the 6th Symposium on the Biogeochemistry of Wetlands, 1114 July 1999. Fort Lauderdale, Florida. Qian, S.S. and C.J. Richardson. 1997. Estimating the longterm phosphorus accretion rate in the Everglades: A Bayesian approach with risk assessment. Water Resour. Res. 33:16811688. Qualls, R.G. and C.J. Richardson. 2003. Fact ors controlling concentration, export, and decomposition of dissolved organic nutrients in the everglades of Florida. Biogeochemistry 62:197229. Reddy K.R. and E.M. DAngelo 1994. Soil proce sses regulating water quality in wetlands. In W.J. Mitsch (ed.) Global wetlands: Old world a nd new. Elsevier, Amsterdam, The Netherlands. Reddy, K.R. and W.F. DeBusk. 1985. Nutrient removal potential of selected aquatic macrophytes. J. Environ. Qual. 14:459462. Reddy, K.R. and R.D. DeLaune. 2008. Biogeochemi stry of wetlands. CRC Press, Boca Raton, Florida. Reddy, K.R., R.D. DeLaune, W.F. DeBusk and M.S. Koch. 1993. Longterm nutrient accumulation rates in the Everglades. Soil Sci. Soc. Am. J. 57:11471155. Reddy, K.R., O.A. Diaz, L.J. Scinto and M. Agami. 1995. Phosphorus dynamics in selected wetlands and streams of the Lake Okeechobee basin. Ecol. Eng. 5:183207. Reddy, K.R., Y. Wang, W.F. DeBusk, M.M. Fi sher and S. Newman. 1998. Forms of soil phosphorus in selected hydrologic units of the Florida Everglades. Soil Sci. Soc. Am. J. 62:11341147. Reddy, K.R., R.G. Wetzel, and R.H. Kadlec. 200 5. Biogeochemistry of phosphorus in wetlands. In J.T. Sims, A.N. Sharpley (eds.) Phosphorus: Agriculture and the envi ronment. Soil Science Society of America, Madison, Wisconsin. Richardson, C.J. 1999. The role of wetlands in st orage, release and cycling of phosphorus on the landscape: A 25year retrospective. p. 4768. In K.R. Reddy, G.A. O'Connor and C.L. Schelske (eds.) Phosphorus biogeochemistry in subtropical ecosystems. CRC Press LLC, Boca Raton. Richardson, C.J. and P.E. Marshall. 1986. Proces ses controlling movement storage, and export of phosphorus in a fen peatland. Ecol. Monogr. 56:279302. Scinto, L.J. 1997. Phosphorus cycling in a periphytondominated freshwater wetland. Dissertation. University of Florida. Schmid, B.H., U. Stephan and M.A. Hengl. 2005. Sediment deposition in constructed wetland ponds with emergent vegetation: La boratory study and mathematical model. Wat. Sci. and Tech. 51:307314. PAGE 116 116 Stuck, J.D., F.T. Izuno, K.L. Campbell, A.B. Bo ttcher and R.W. Rice. 2001. Farmlevel studies of particulate phosphorus transport in the Ever glades Agricultural Ar ea. Trans. ASAE 44:11051116. Tanner, C.C. 1996. Plants for constructed wetl and treatment systems A comparison of the growth and nutrient uptake of eight emergent species. Ecol. Eng. 7:5983. Turner, B.L. and S. Newman. 2005. Phosphorus cycling in wetland soils: The importance of phosphate diesters. J. En viron. Qual. 34:19211929. Turner, B.L., S. Newman and J.M. Newma n. 2006. Organic phosphorus sequestration in subtropical treatment wetlands. Environ. Sci. Technol. 40:727733. Volk, B.G. 1973. Everglades histosol subsidence 1: CO2 evolution as affected by soil type, temperature, and moisture. Soil Cr op Sci. Soc. Fla. Proc. 32:132135. Vymazal, J. 2007. Removal of nutrients in variou s types of constructe d wetlands. Sci. Total Environ. 380:4865. Wang, H.G., J.W. Jawitz, J.R. White, C.J. Ma rtinez and M.D. Sees. 2006. Rejuvenating the largest municipal treatment wetla nd in Florida. Ecol. Eng. 26:132146. White, J.R., K.R. Reddy, W.F. DeBusk, W.R. Wise, T. Crisman. 2002. Phosphorus removal capacity of the Orlando wetland treatment syst em: final report. Soil and Water Science Department, University of Fl orida, Gainesville, Florida. PAGE 117 117 BIOGRAPHICAL SKETCH Mike Jeraulds passion for the environment, which inspired the undertaking of this advanced degree, was fostered by his experiences fishing, camping and traveling as a child with his parents. His formal educati on in the field began with the J upiter Environmental Research and Field Studies Academy at Jupite r High School, in South Florida. Impressed with the urgency of the need to respond to the everdeclining condi tion of the natural wo rld, Mike earned his Bachelors degree in Environmental Science from the University of Florida in 2004. His undergraduate studies introduced and develope d his awareness of the intimate connection between human and environmental wellbeing. In 2008, he enrolled in the Soil and Water Science Department at the University of Florida to pursue a Master of Science that would enable him to employ wetlands for the treatment of wast ewater. The quality of all aspects of Mikes term in Gainesville was marked best, perhaps, by the four national titles won by the University of Florida football and mens basketball teams, coll ectively, during his tenure there. In 2009, he married his extraordinary wife, Sarah. After 19 years of education, he now plans to leave academia and try his hand in the real world. 