Influence of net freshwater supply on salinity in Florida Bay

Transcription

1 WATER RESOURCES RESEARCH, VOL. 36, NO. 7, PAGES , JULY 2000 Influence of net freshwater supply on salnty n Florda Bay Wllam K. Nuttle and James W. Fourqurean Southeast Envronmental Research Center, Bologcal Scences, Florda Internatonal Unversty, Mam Bernard J. Cosby and Joseph C. Zeman Department of Envronmental Scences, Unversty of Vrgna, Charlottesvlle Mchael B. Robblee Bologcal Resources Dvson, U.S. Geologcal Survey, Florda Internatonal Unversty, Mam Abstract. An annual water budget for Florda Bay, the large, seasonally hypersalne estuary n the Everglades Natonal Park, was constructed usng physcally based models and long-term (31 years) data on salnty, hydrology, and clmate. Effects of seasonal and nterannual varatons of the net freshwater supply (runoff plus ranfall mnus evaporaton) on salnty varaton wthn the bay were also examned. Partcular attenton was pad to the effects of runoff, whch are the focus of ambtous plans to restore and conserve the Florda Bay ecosystem. From 1965 to 1995 the annual runoff from the Everglades nto the bay was less than one tenth of the annual drect ranfall onto the bay, whle estmated annual evaporaton slghtly exceeded annual ranfall. The average net freshwater supply to the bay over a year was thus approxmately zero, and nterannual varatons n salnty appeared to be affected prmarly by nterannual fluctuatons n ranfall. At the annual scale, runoff apparently had lttle effect on the bay as a whole durng ths perod. On a seasonal bass, varatons n ranfall, evaporaton, and runoff were not n phase, and the net freshwater supply to the bay vared between postve and negatve values, contrbutng to a strong seasonal pattern n salnty, especally n regons of the bay relatvely solated from exchanges wth the Gulf of Mexco and Atlantc Ocean. Changes n runoff could have a greater effect on salnty n the bay f the seasonal patterns of ranfall and evaporaton and the tmng of the runoff are consdered. One model was also used to smulate spatal and temporal patterns of salnty responses expected to result from changes n net freshwater supply. Smulatons n whch runoff was ncreased by a factor of 2 (but wth no change n spatal pattern) ndcated that ncreased runoff wll lower salnty values n eastern Florda Bay, ncrease the varablty of salnty n the South Regon, but have lttle effect on salnty n the Central and West Regons. 1. Introducton Florda Bay, a broad (2000 km 2 ), shallow (approxmately 1 m) estuarne lagoon nestled between the south Florda manland and the Florda Keys (Fgure 1), occupes a large porton of Everglades Natonal Park and s contguous wth the Florda Keys Natonal Marne Sanctuary. It s bounded by the mangrove wetlands of the manland, the open marne systems of the Gulf of Mexco, and the slands that compose the Florda Keys. Bay waters support a valuable recreatonal fshery wthn the bay tself [Tlmant, 1989] and a commercal shrmp fshery n the Gulf of Mexco [Costello and Allen, 1966]. Begnnng n 1987, sudden and extensve de-off n the sea grass beds that cover 95% of the bottom sgnaled a rapd, general declne n the ecologcal health of the bay [Robblee et al., 1991; Fourqurean et al., 1993; Phlps et al., 1995]. Increased turbdty followed de-off n the grass beds [Boyer et al., 1999], and recurrent blooms of cyanobactera n the wnters of and decmated the sponge populaton [Butler et al., 1995]. The resultng changes n water qualty and the long-term Copyrght 2000 by the Amercan Geophyscal Unon. Paper number 1999WR /00/1999WR900352$09.00 structural changes n the bay s ecosystems have also affected the health of adjacent coastal systems, such as the coral reefs of the Florda Keys. Ecologcal declne n Florda Bay s wdely consdered to be the result of long-term regonal water management n south Florda. Although there s general agreement about the nature and extent of the mpacts of water management n the extensve wetlands of the Everglades, whch le mmedately upstream of Florda Bay, the chan of cause and effect lnkng water management to sea grass de-off and plankton blooms n the bay has not yet been fully establshed. Because of management practces, dscharge of freshwater drectly nto the Atlantc Ocean and farther north nto the Gulf of Mexco has ncreased up to a factor of 10, whle the dscharge of freshwater nto Florda Bay and along the southwest coast of Florda has decreased by an unknown but mportant amount [Lght and Dneen, 1994]. Because we do not know the senstvty of the Florda Bay ecosystem, prmarly the extensve sea grass communtes, to varatons n freshwater runoff, we cannot tell what benefts restorng the hstorcal runoff would have. Even wthout ths knowledge, plans for water management and ecosystem restoraton n south Florda [U.S. Army Corps of Engneers, 1998] are progressng, based, at least n part, on the assump- 1805

2 1806 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY Fgure 1. Broad, shallow ( 30 cm) seagrass-covered mudbanks (shaded area) restrct water exchange between Florda Bay and the coastal ocean. Freshwater enters the bay as runoff through Taylor Slough and as dffuse flow from the wetlands to the east. Regonal dscharge through Shark Slough (nset) also nfluences salnty n the Gulf of Mexco on the western boundary of the bay. Locatons are shown for sources of data on ranfall (ponts n the bay), pan evaporaton (Flamngo), and runoff (Taylor Sough Brdge and the control structures S18C and S197 on the C111 canal). ton that the ecologcal health of Florda Bay wll be restored by ncreasng the freshwater runoff to the bay to as near to hstorc levels as possble. Salnty s an ntermedate lnk n the chan of cause and effect that connects water-management actvtes to the structure and functons of the bay s ecosystem. In Florda Bay, salnty vares markedly n tme and space (Fgure 2). Hypersalne condtons ( 40) (salnty values gven n practcal salnty unts) n one part of the bay frequently coexst wth more estuarne condtons ( 30) n another. At some nteror locatons, salnty regularly fluctuates between hypersalne and nearly freshwater condtons [Frankovch and Fourqurean, 1997]. Only wthn the confnes of a few, semenclosed basns along the north shore of the bay do salnty fluctuatons closely follow changes n canal dscharge. The degree to whch water management and, consequently, runoff from south Florda nfluence salnty fluctuatons n Florda Bay cannot be ascertaned wthout a detaled analyss. Therefore we have used salnty, hydrology, and clmate data from 1965 through 1995 to nvestgate how the annual water balance and the varatons n freshwater fluxes have nfluenced the salnty n Florda Bay. 2. Background 2.1. Factors Affectng Estuarne Salnty Varaton n estuarne salnty can be attrbuted to the ntensty of the two-way exchange between the estuary and the coastal ocean, the net supply of freshwater that flows through the estuary to the coastal ocean, and the salnty of the coastal ocean at the estuary s mouth. The two-way exchange between estuary and ocean s drven by several physcal processes, ncludng densty dfferences, astronomcal tdes, and wnd. The net freshwater supply s the sum of runoff and drect ranfall mnus any evaporaton from the estuary. The patterns of salnty n estuares result from a dynamc steady state n whch the advectve flux of salt nto or out of the estuary, whch s drven by the net freshwater supply, s balanced by a dspersve flux from the two-way water exchange created by tdes and other hydrodynamc mxng processes. In a classcal estuary a postve net freshwater supply, usually from heavy runoff delvered by rver dscharge, dlutes the salnty n the estuary to below that of ocean water. Salnty

3 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY 1807 ranges from zero at the head of the estuary to the salnty of the coastal ocean near the mouth. Other estuares may experence hypersalne condtons. Many coastal bays and lagoons, lke Shark Bay, Western Australa [Smth and Atknson, 1983]; Laguna Madre, Texas, Unted States of Amerca [Smth, 1988]; and Lagoa de Araruama, Brazl [Kjerfve et al., 1996], have hgher average salntes near ther mouths than the coastal ocean for most or all of the year. Because these nverse estuares have a negatve freshwater supply caused by evaporaton rates hgher than both ranfall and runoff rates, salt concentrates to hypersalne condtons. Salntes n these nverse estuares can range from zero near freshwater dscharge to a greater-than-coastal salnty n the man body of the estuary. Seasonally hypersalne estuares form a thrd class of estuary characterzed by ther epsodc hypersalnty [Larger et al., 1997]. These estuares experence lmted exchange wth the coastal ocean, and net freshwater supply fluctuates on the postve and negatve sde of zero n response to clmatc varatons. Estuares n ths class are found n both temperate, Medterranean clmates (e.g., Tomales, Msson, and San Dego Bays, Calforna, Unted States of Amerca [Larger et al., 1997]) and tropcal, monsoonal areas (e.g., northern Australa [Wolansk, 1986], Kenya [Ktheka, 1998], and Sr Lanka [Arulananthan et al., 1995]). Snce the net annual freshwater balance of seasonally hypersalne estuares s close to zero, small perturbatons n the freshwater supply may lead to large changes n the salnty of the estuary. Dversons of freshwater runoff for urban or agrcultural use, as n the Colorado Rver Estuary, Mexco, can drastcally change the salnty regme. Small clmactc varatons can also have large mpacts on salnty: For example, a multdecadal trend of decreasng ranfall has changed the Casamance Rver Estuary n Senegal from a seasonally hypersalne estuary to a permanently nverse estuary [Debenay et al., 1994]. In Laguna Madre, Texas, prolonged and ntense hypersalne condtons assocated wth droughts may trgger the brown tde phenomenon by changng the structure of the plankton communty [Rhudy et al., 1999]. Florda Bay s a seasonally hypersalne estuary (Fgure 2a). In the bay a network of broad, shallow mud banks and the lack of densty stratfcaton lmt the magntude of tdally drven and baroclnc exchange flows. The nfluence of the south Florda clmate s evdent n runoff, ranfall, wnd-drven tdes, and the salnty of the coastal ocean. Wet and dry perods fluctuate seasonally and from year to year. Tdes and currents n the bay are partcularly nfluenced by the sustaned wnds assocated wth the passage of fronts characterstc of the subtropcal wnter weather [Wang et al., 1994]. Patterns of local runoff from the Everglades drectly affect salnty n the bay (Fgure 2b). Varatons n runoff from all of south Florda, ncludng Lake Okeechobee, nfluence the salnty of the Gulf of Mexco along ts border wth Florda Bay thereby ndrectly lnkng regonal patterns of runoff to salnty varatons n the bay Regonal Patterns n Florda Bay Florda Bay lacks the clearly defned upstream/downstream axs that, n most estuares, organzes spatal varatons n salnty. However, several analyses of water-qualty parameters, for example, salnty, nutrent, chlorophyll, etc., have shown a consstent pattern. For example, Boyer et al. [1997] dentfed three zones of smlar water qualty n Florda Bay: a core regon, a western regon, and an eastern regon. Other authors have suggested dvdng the bay nto smlar zones based on Fgure 2. Salnty vares wdely n tme and space n Florda Bay. (top) Temporal patterns n the Central Regon reflect the nfluence of sources of varaton operatng on seasonal and nterannual tmescales. (mddle) Spatal patterns n mean salnty from February to March 1994 and (bottom) range of salnty varaton from 1990 to 1994 over the whole bay reveal the nfluence of exchange wth ocean waters and the localzed effect of runoff nto the bay. bank morphology and dynamcs [Wanless and Tagett, 1989], benthc mollusk communtes [Turney and Perkns, 1972], salnty and ntrogen [Fourqurean et al., 1993], and benthc plant communtes [Zeman et al., 1989]. These schemes all suggest that the prmary axs of dfferentaton runs from northeast to southwest, and most schemes nclude a separate, dstnct regon (of varyng sze) n the upper central part of the bay adjacent to the Everglades. On the bass of the work summarzed above, for ths study we dvded the bay nto four regons (Fgure 3) that dffer n ther proxmty to the Gulf of Mexco, areas of water flow through the Florda Keys, and sources of freshwater runoff from the manland. In the Central Regon, broad, shallow banks (Fgure 1) restrct exchange wth the Gulf and the Atlantc, and there s lttle freshwater runoff. Resdence tmes are hgh and hypersalne condtons are frequent and persstent (Fgure 2a). The East Regon resembles the Central Regon wth ts lmted oceanc exchange and long resdence tmes; however, t receves most of the bay s freshwater runoff prmarly from the C111 canal and Taylor Slough (Fgure 1). Salnty n the East Regon vares wdely between nearly fresh and hypersalne condtons (Fgure 2b). In the South and West Regons, salnty varatons are less extreme (Fgure 2b). Greater exchange wth the Gulf of Mexco and the Atlantc

4 1808 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY Fgure 3. Subtdal banks and slands dvde Florda Bay nto 44 basns used to categorze the locaton of salnty observatons. On a larger scale, studes of water qualty, sedment type, faunal communtes, and benthc plant communtes generally dvde the bay nto four regons of smlar character. Ths study uses data from an aggregaton of basns wthn each of these four regons (hatched area) to characterze the spatal and temporal varatons n Florda Bay. Ocean and a lack of drect runoff result n salntes nearer to that of the coastal waters. 3. Methods The man objectves of ths study were (1) to establsh the annual and seasonal water balances and net freshwater supples for Florda Bay and (2) to evaluate the degree to whch the amounts of and varatons n ranfall and runoff contrbute to the observed salnty varatons n Florda Bay. The nfluence on salnty of any of the components of the net freshwater supply could not be demonstrated smply by searchng for correlatons wth salnty. Most factors controllng estuarne salnty share clmate as a common source of varaton, and each can be expected to exhbt smlar patterns of varaton. The only way to understand the nfluence of a partcular component of the net freshwater supply on salnty n Florda Bay was to solate that component and quantfy drectly ts effect on salnty. Ths approach requred spatally and temporally extensve measurements of freshwater fluxes and salnty n the bay, salnty models that ncorporated dfferent temporal and spatal scales, and a framework for nterpretng the smulated and observed salnty varatons. We assembled ranfall, runoff, and evaporaton data and a database of publshed and unpublshed salnty measurements n Florda Bay (see the appendx) that spans the 31 years from 1965 through We used the annual and monthly means of these data to establsh the yearly and seasonal water balances for Florda Bay. We then used the data wth physcally based, mass balance salnty models n comparatve analyses to examne the effects of spatal and temporal varatons n net freshwater supply on the observed salnty varatons Framework for the Comparatve Analyses In general, physcally based models treat salnty, S, asa functon of coastal ocean salnty, S ocn, exchange fluxes wth the coastal ocean, Q T, and the fluxes of freshwater (ranfall, Q P, runoff, Q R, and evaporaton, Q E ), all of whch vary n tme and space: S j,k f S ocn j,k, Q T j,k, Q E j,k, Q P j,k, Q R j,k R j,k, (1) where the superscrpt and subscrpts dentfy locaton (), year ( j), and month (k). The resdual errors, R, are the dfferences between measured and smulated salnty and represent nose n the data and salnty varatons not explaned by the processes or assumptons nherent n the model formulatons. We employed models that dffered n ther spatal and temporal resoluton and compared the successes of the models n reproducng observed salnty varatons n order to draw nferences about the mportance for salnty varatons of (1) nterannual and seasonal varatons n net freshwater supply and (2) locaton wthn the bay. We used a measure of model effcency, eff, to assess how successfully each model reproduced the patterns of varaton n observed salnty data: eff j k R j,k 2, (2) n Var S j,k where R are the resdual errors, Var (S) s the total varance of the salnty measurements, and n s the number of observatons. A model s effcency score (untless) can be broadly nterpreted as the proporton of the varance n the data ex-

5 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY 1809 planed by the model. In ths sense, model effcency s smlar to the coeffcent of determnaton r 2. In contrast to r 2 the effcency score can take on negatve values f, for example, the model produces a based estmate of the data or f fluctuatons n the model are out of phase wth fluctuatons n the data. If the effcency score was zero, then the model explaned the varaton n the data no better than dd the mean of the data. If the effcency score was 100%, then the resduals were zero, and the model explaned all of the varance n the data. Any measure of model success s most useful f comparatvely appled. That s, by examnng the ncrease (or decrease) n explanatory power between a null model and an alternatve, the power of the processes ncluded n the alternatve model to explan varance n the data can be assessed. The null model mplctly ncluded n eff was the mean of the observed salnty across all observatons and all regons n Florda Bay (.e., a model wth no temporal or spatal resoluton). We developed three alternatve salnty models that contaned ncreasng spatal and temporal complextes. By comparng the effcences of the alternatve models, we were able to estmate the relatve contrbutons of two temporal components (nterannual and seasonal varatons n freshwater fluxes) and one spatal component (locaton of the salnty measurements) to the overall varaton of salnty n Florda Bay. The frst type of model, a statc locaton model, only accounted for the effects on salnty of poston wthn the bay. Ths model was smply defned as the average of the observed salnty data, S av, for all months and years n each regon of the bay: S j,k S av R j,k. (3) The locaton model was appled to each of the four regons n Fgure 3. The model dd not contan a temporal component and could not explan any of the nterannual or seasonal varatons of salnty wthn any of the regons. The notaton denotes that the model smulated a salnty value S for each regon (), every year ( j), and every month (k), but the predcted salnty for all months and years n a regon was the same value. Because the model has no temporal component (only the spatal means are used), the resduals were expected to contan all of the temporal varance n the data, and the effcency was expected to be low. The second type of model, a steady state, spatally aggregated box model, quantfed the effects of locaton as well as the effects of long-term, nterannual varatons n ranfall and runoff on salnty wthn the bay: S j,k f S ocn, Q T, Q E, Q P j, Q R j R j,k. (4) The box model was mplemented usng annual tme steps and annual values of freshwater and exchange fluxes. Ranfall and runoff were unformly dstrbuted over the bay (no spatal component) and vared from year to year. Spatally explct but temporally constant estmates of evaporaton and dspersve exchange wth the coastal ocean were derved for the model durng calbraton. Ths model was appled to all four regons n the bay (Fgure 3). The resduals were expected to contan all of the seasonal varance, and the effcency was expected to ncrease over that of the locaton model. The thrd type of model, a dynamc, spatally explct model, smulated the effects on salnty of locaton and both nterannual and seasonal varatons n freshwater and exchange fluxes: S j,k f S ocn, Q T j,k, Q E k, Q P j,k, Q R j,k R j,k. (5) The dynamc model was based on the basn and bank geomorphology of Florda Bay and was drven by monthly values of ranfall, runoff, and evaporaton. Ranfall and evaporaton were appled unformly across the bay, but runoff was added at approprate locatons on the boundares of the bay. Hourly tdes generated advectve exchanges among 44 basns wthn the bay. We expected ths model, wth the greatest spatal and temporal complexty, to have the hghest effcency and explanatory power Salnty Data We drew our salnty observatons for ths study from an hstorcal salnty database for Florda Bay and the west coast of south Florda consstng of over 34,000 ndvdual observatons datng from 1947 (see the appendx). Ths database was assembled from the results of many feld studes and from systematc, water-qualty montorng programs ntated n response to the ecologcal crses of the late 1980s and early 1990s. Data from wthn Florda Bay are categorzed accordng to ther locaton n a grd of 44 numbered basns. Boundares of the basns follow the geometry of the system of anastomosng banks that physcally subdvde the bay (Fgure 3). Ths database provdes excellent spatal and temporal coverage begnnng wth 1989 when mountng concern about condtons n the bay resulted n the establshment of regular water-qualty montorng surveys. However, the data from before 1989 are dscontnuous and uneven wth the hghest number of observatons clustered n the East Regon of the bay. Because the data avalable from before 1965 are extremely spotty, they were not used n ths study. We aggregated the salnty data by subsamplng and processng data from the hstorcal database to assure an unbased samplng of the nterannual and seasonal salnty varatons and to provde a balanced representaton of the regonal varatons n the bay. Frst, the data were aggregated n tme by computng the ndvdual monthly average salnty n each basn for each month of record. Second, the data n each basn were screened to assure they consstently represented seasonal varatons by excludng calendar years wth data reported n fewer than 11 months. Thrd, n all but the East Regon, data from two adjacent basns were combned to provde the most contnuous salnty record possble (maxmum number of years) over the 31 years (Fgure 3). The resultng set of monthly averaged salnty data characterzed the nterannual and seasonal varatons n salnty n each regon of the bay for 1965 through 1995 (Fgures 4 and 5). The nne years from 1987 to 1995 contan 34 staton years (54%) of the data. We used ths data subset, the evaluaton perod subset, for our detaled comparsons of the model results because t provded the most complete temporal and spatal coverage of the bay. Our evaluaton of the nfluence of the net freshwater supply on salnty was prmarly based on our analyss of these data. We used the addtonal data n the complete 31-year record to evaluate the predctve ablty of the models. The dstrbuton of salnty data n the 9-year evaluaton perod was smlar to that of the complete 31-year record (Table 1) Freshwater Flux Data Runoff. Freshwater runoff nto Florda Bay was estmated as the sum of monthly volume dscharges n Taylor Slough and the C111 canal (Fgure 6a). Data for Taylor Slough are avalable for the entre perod from 1965 through 1995; data for the C111 canal are only avalable begnnng n 1970.

6 1810 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY Fgure 4. Monthly salnty values for 1965 to 1995 for each of four aggregated basns n Florda Bay show the combned nfluence of a strong seasonal cycle supermposed on nterannual varaton. The straght lnes ndcate the average salnty n each regon for the perod. Dscharge n Taylor Slough s measured as t crosses the man road through Everglades Natonal Park (Fgure 1). The flow down Taylor Slough dscharges nto a complex of ponds north of the bay and s dstrbuted from there nto the bay through several, smaller channels. Dscharge n the C111 canal s measured at the S18C and the S197 control structures (Fgure 1). The C111 canal conveys water from a regonal network of dranage canals. Most water that leaves the C111 canal dscharges nto the mangrove wetlands between the S18C and the S197 structures. Ths freshwater then flows south nto the East Regon of the bay. Durng nfrequent perods of extremely hgh flow, water s allowed to pass through the S197 structure and dscharge drectly nto the extreme eastern end of Florda Bay. These runoff data do not account for the net gan (or loss) of freshwater from precptaton and evaporaton over the area between the flow-montorng ponts and the coast. The contrbuton of (ungauged) groundwater flow to the coast s also not accounted for n these data. Evdence from natural groundwater tracers suggests that submarne groundwater dscharge nto Florda Bay contrbutes only slghtly to the net freshwater supply [Corbett et al., 1999] Ranfall. The avalable long-term ranfall records for land-based stes n south Florda do not provde relable estmates of ran fallng drectly onto the bay. Convectve storms form prmarly along the coast early n the wet season but do not form over the open water of the bay untl late n the wet season [Schomer and Drew, 1982]. Ths produces hgher ranfall measurements at manland statons just nland from the Florda Bay coast than actually occur n the bay. Therefore, to construct a long-term precptaton record for the bay, we had to correct for ths bas n the land-based records. We dd ths by correlatng land-based records wth recently avalable ranfall measurements from statons wthn the bay and usng ths correlaton to reconstruct ranfall for perods when no ranfall data for the bay were avalable. Ranfall s currently beng measured at several marnemontorng network statons mantaned n Florda Bay (D. Table 1. Summary of the Monthly Average Salnty Observatons n Each Regon of Florda Bay Fgure 5. Regonal salnty averaged by month over all 31 years llustrates smlartes and dfferences n the seasonal patterns of varaton. In all regons, salnty ncreases durng the dry season and decreases durng the wet season. The ampltude of seasonal varaton s greater n the Central and East Regons (bold and dashed lnes), whch are solated from actve exchange wth the coastal ocean, whch moderates the seasonal varaton n the South and West Regons (dotted and fne lnes). Basn Mean SD Mean SD East South Central West All basns SD s standard devaton.

7 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY 1811 Smth, Annual Data Reports: , Everglades Natonal Park, Homestead, Florda). We used the monthly totals for 1993 through 1996 from eght of these statons (Fgure 1) to estmate monthly bay-wde average ranfall for 1993 through We chose these statons because they reported at least 12 months of data wthn ths perod. We used lnear regresson to dentfy relatonshps between the monthly totals at each of the eght marne montorng statons and monthly ranfall amounts from long-term records (Natonal Weather Servce, Natonal Oceanc and Atmospherc Admnstraton (NOAA), data avalable from the Natonal Clmatc Data Center, Ashevlle, North Carolna) for Flamngo, Royal Palm, Tamam Tral, and Homestead (Fgure 1). These relatons, whch have r 2 generally greater than 0.5, provded the means of extrapolatng the recent record of ranfall n the bay back n tme over the perod 1965 through We calculated a bay-wde average ranfall from the extrapolated records, usng area weghts based on Thessen polygons, to estmate annual and monthly, bay-wde average ranfall (Fgure 6b) Evaporaton. Evaporaton has not been drectly measured n Florda Bay, and as yet lttle effort has been made to evaluate the long-term evaporaton rate or ts seasonal or regonal varatons. Several years (1965 through 1970) of pan evaporaton observatons are avalable from a Natonal Weather Servce cooperatve observng staton (data avalable from the Natonal Clmatc Data Center) at Flamngo on the southwest Florda manland (Fgure 1). The annual average of these data (approxmately 210 cm yr 1 ) appeared to be too hgh to be accepted as drect estmates of evaporaton n Florda Bay. By comparson, Morton [1986] estmates annual evaporaton from Lake Okeechobee s 162 cm based on a calculaton of ts water budget. Recently, Pratt and Smth [1999] estmated an annual evaporaton rate of about 73 cm by usng a Dalton law formula and data collected at three stes n the bay. However, snce some data needed to apply ths formula were obtaned from a fourth staton outsde of the bay, t s not known what magntude of error mght have been ntroduced nto ther estmate. Because of the lack of relable evaporaton estmates we derved our own estmate of the long-term, baywde, annual average evaporaton rate, E B, from our calbraton of the steady state box model usng the annual averages of the observed salnty data (see secton 3.4.1). Then, we estmated monthly values of evaporaton, Q E, by multplyng the long-term, annual rate by monthly weghts, w k, derved from the seasonal pattern n the pan data from Flamngo: where Q E E B w k, (6) 12 w k 1. k 1 Fgure 6. (a) Monthly runoff nto Florda Bay was estmated from measured dscharges n Taylor Sough (1965 to 1995) and n the C111 canal (1970 to 1995). The volumetrc fluxes were dvded by the surface area of Florda Bay (200 km 2 ) to yeld an equvalent bay-wde depth for runoff. The dashed lne s the total runoff for each calendar year. (b) Monthly ranfall onto Florda Bay for 1965 to 1995 was estmated from relatonshps between long-term ranfall data n the Everglades and more recent (short term) ranfall data n Florda Bay. The dashed lne s the total ranfall for each calendar year Salnty Models Steady state box model. The steady state box model served two purposes. It allowed us (1) to estmate the unknown evaporaton flux and (2) to nvestgate what effect year-to-year varatons n net freshwater supply has had on bay salnty. We formulated the box model followng the approach of Mller and McPherson [1991] at Charlotte Harbor. In ths approach the net effects of resdual crculaton and hydrodynamc mxng were accounted for by a (unknown) net exchange flux, Q T, for each regon of the bay that represented the cumulatve nflux of seawater flowng nto that regon. These exchange fluxes, expressed n cm yr 1, were assumed to be constant wth respect to season and year. Each regon also receved a net supply of freshwater, Q F, as a result of runoff, drect ranfall, and evaporaton. Invokng mass conservaton for both water and salt led to an expresson for the steady state salnty n each regon. On an annual average bass a flux of water, Q T, wth salnty, S ocn, entered each basn from the ocean and a flux of water, Q T plus Q F, returned to the ocean wth the salnty n the basn. Equatng the nflow and outflow, advected fluxes of salt led to an expresson for the annual average, steady state salnty n the efflux (S ann,.e., an estmate of the annual average salnty n a gven regon of the bay): Q T S ann S ocn, (7) Q T Q F where Q F Q P Q R Q E and the ranfall, runoff, and evaporaton fluxes are annual average values. We appled the box model separately to each of the four regons n the bay (Fgure 3). Annual runoff and ranfall volumes (Fgure 6) were unformly dstrbuted over the entre area of the bay. In the case of runoff we assumed that mxng wthn the bay was vgorous enough to redstrbute runoff throughout the bay from ts localzed ponts of dscharge n less than a year. In the case of ranfall the avalable records from ponts wthn the bay were nsuffcent to characterze any spatal dstrbuton that was not unform. We thus appled the same ranfall, Q P, and runoff, Q R, to each regon, and we

8 1812 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY Fgure 7. Average monthly freshwater fluxes to Florda Bay for 1970 to The net freshwater supply (fne lne) fluctuates between defct and surplus because peaks n monthly patterns of ranfall and runoff (bold and dotted lnes) lag the peak n monthly evaporaton (dashed lne) by about 4 months. calbrated the unknown annual exchange fluxes, Q T, and annual evaporaton fluxes, Q E, for each regon by ndvdually fttng the model to the observed salnty data (usng least squares mnmzaton of the resduals). Followng calbraton for each regon, we calculated the annual, bay-wde evaporaton rate, E B, as the average of the four regonal values of Q E (see (6)) Dynamc, spatally explct model. We used a dynamc, spatally explct mass balance model to nvestgate the combned nfluence of seasonal and nterannual varatons n net freshwater supply for two reasons. We needed a dynamc model because resdence tmes n the bay exceed 1 month and we could not assume a steady state salnty response on a seasonal or monthly tmescale. We needed a spatally explct model to examne whether the nfluence of runoff on a monthly scale would be confned to basns near the nflow along the Everglades coast. We developed a model for ths purpose that mantans a runnng account of the water and salt budgets n each of 44 well-mxed basns wthn the bay (Fgure 3). The boundares of these basns follow the system of the anastomosng banks that dssect the bay. Ths geometry was chosen because the banks are the prmary controls on fluxes wthn the bay and the basns offer a natural framework for mass balance accountng. Ths approach traces ts roots to the Keulegan s [1967] model for the response of a coastal lagoon to forcng by ocean tdes actng through an nlet. Tdal exchange through the nlet s modeled usng Mannng s equaton and the head dfference between zero-velocty water bodes at ether end of the nlet channel. We adapted ths approach to condtons n Florda Bay where exchange s governed by the constrcton of shallow banks, not narrow nlets, and we extended t to a network of basns nterconnected by flows over banks. The dynamc mass balance model (Flux Accountng and Tdal Hydrology at the Ocean Margn (FATHOM)) calculates exchange wth a coastal ocean and mxng among basns n a bay as the results of tdally drven water fluxes across shallow banks. At each hourly tme step the model solves for unform, hydraulc flow across each bank based on the depth, wdth, and frctonal roughness of the bank and water levels n upstream and downstream basns. Mannng s equaton for frcton flow n channels [see Henderson, 1966] s used to calculate water velocty as a functon of depth wth a vertcal resoluton of 0.3 m. These veloctes are used wth cross-sectonal areas of banks to calculate water fluxes. Salt fluxes are then calculated from water fluxes and the salnty of an upstream basn. Detals of the banks representaton and the hydraulc-equaton solutons are gven by Cosby et al. [1999]. Bascally, FATHOM smulates mxng of salt between adjacent basns as tdally drven flows over a seres of wers. In addton to the clmate data needed for the box model, FATHOM requres tde data to set the open-water boundary condtons for the bay. Hourly tde stages along the Gulf of Mexco and Atlantc boundares of Florda Bay were nterpolated from NOAA tde tables for locatons along the southern Florda coast and along the Florda Keys. We used semdurnal tdes and appled the same annual pattern for all years from 1965 to The effects of wnd tdes and wnd mxng were not ncluded n ths applcaton of FATHOM. We assumed the Gulf of Mexco and Atlantc salnty to be constant at 35. The seasonal estmates of evaporaton derved from the box model (equaton (6)) were appled usng the same total evaporaton and seasonal pattern for all basns and all years smulated. Our estmated monthly ranfall for the bay (Fgure 6b) was evenly dstrbuted over the 44 basns, and monthly runoff (Fgure 6a) was added at fve nflow ponts along the north shore of the East Regon. Runoff dstrbuton among the nflow ponts was determned by the proportons of the total runoff contrbuted by measured flows n Taylor Slough and the C111 canal. Generally, the nfluence of the C111 canal has been to redstrbute runoff to the easternmost parts of the northern boundary [Lorenz, 1999] relatve to hstorcal condtons. We derved the length/depth dstrbuton of each bank and the volume/depth dstrbuton of each basn from Geographc Informaton System (GIS) data that had a resoluton of 20 m. On the bass of ths GIS data we assgned to each bank one of four wdths (300, 1000, 3000, or 4000 m). We appled a value of 0.1 for Mannng s n, the frcton coeffcent, for all banks (based on the lterature for sedments and substrates smlar to those on the banks n Florda Bay [e.g., Henderson, 1966]). FATHOM calculates hourly values of water level and mean salnty for each basn; monthly average salnty was calculated for each basn based on these hourly values. Because of the smplfyng assumptons nherent n the representaton of tdal exchange n the model, the varaton of salnty s not correctly represented at tmescales less than that represented n the varaton of monthly average salnty. For all but the East Regon (whch conssts of a sngle basn), smulated monthly salnty values from two adjacent basns were averaged to provde regonal salnty estmates that corresponded to the observed data (Fgure 3). We dd not calbrate FATHOM n any formal sense (e.g., optmzaton by least squares to ft the salnty data). The nputs descrbed above were appled to the model, and the smulated salnty values were used wthout further adjustment. 4. Results and Dscusson 4.1. Water Balance for Florda Bay Usng the annual averages of ranfall and runoff for 1970 to 1995 estmated from our data for these nputs (Fgure 6) and the annual evaporaton estmated from the applcaton of the box model to the salnty data for 1987 through 1995 (see secton 4.1.2), we derved an annual water balance for Florda Bay for ranfall of 98 cm yr 1, for runoff of 9 cm yr 1, for evaporaton of 110 cm yr 1, and for net freshwater supply to the bay of 3 cmyr 1. We averaged the freshwater fluxes for each month from 1970 through 1995 to derve an average seasonal cycle of the water balance for Florda Bay (Fgure 7).

9 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY 1813 Monthly ranfall vared from about 4 cm month 1 durng the dry season to greater than 15 cm month 1 at the peak of the rany season. Estmated evaporaton was lowest n wnter (about 6 cm month 1 ) and reached a peak n early summer (13 cm month 1 ). Monthly runoff under the management practces n place durng the perod was unformly low (less than 2 cm month 1 for all months), but there appeared to be a tendency toward slghtly hgher runoff n late summer (Fgure 7) Importance of ranfall. Under water-management practces from 1970 through 1995 the average annual volume of runoff nto Florda Bay was less than one tenth of the average annual volume of drect ranfall onto the bay. Ths dstngushes Florda Bay from other nearby estuarne areas where the rato of runoff to drect ranfall was 1 to 2 orders of magntude greater (Table 2). Wthn the bay the effects of runoff can be locally more mportant. For nstance, n the East Regon where almost all drect runoff actually entered the bay, the rato of runoff to ranfall was larger (approxmately 0.5). Comparng salnty condtons n the East Regon wth those n the Central Regon provded an ndcaton of the spatal varaton n the magntude of the effect of runoff to ranfall ratos on salnty wthn the bay. Both regons were relatvely solated from exchange wth the ocean, but the Central Regon receved lttle freshwater nflow from runoff. Salnty varatons n the two regons were smlar (and hgh compared to other areas of the bay), but the mean salnty n the Central Regon (no runoff) was 10 hgher than that n the East Regon (Table 1) Estmated evaporaton. The calbraton of the box model for each regon n the bay provded an estmate of annual average evaporaton n each regon for 1987 through 1995 (Table 3). We averaged the evaporaton rates for each regon to estmate the annual evaporaton rate for Florda Bay as a whole. Ths average annual, bay-wde evaporaton rate was approxmately 110 cm yr 1, sgnfcantly lower than the estmates derved from the pan data at Flamngo (210 cm yr 1 ) and the water budget for Lake Okeechobee (162 cm yr 1 ). Wthn the bay the estmates of annual evaporaton vared spatally (Table 3). The estmated rates were almost dentcal n the Central, South, and West Regons (approxmately 130 cm yr 1 ), but the estmated rate for the East Regon was more than 30% lower. A spatal pattern n evaporaton over the bay (related to water depth, bottom cover, etc.) was expected, and, theoretcally, the calbraton of the box model could recover some of that pattern (to the degree to whch the pattern s reflected n the annual average salntes used to calbrate the model). Table 2. Comparson of Annual Freshwater Input Fluxes for Florda Bay Wth Other Florda Estuares Estuary Area, km 2 Runoff, a cm Ranfall, a cm Inflow, a cm Runoff/ Ranfall Florda Bay b c Charlotte Harbor d Indan Rver Lagoon e a Annual volume s dvded by area of estuary. b Runoff and ranfall are annual averages for 1970 through c Ths s the sum of gauged flows n Taylor Slough and n the C111 canal at S18C. d Source s Mller and McPherson [1991]. e Source s Smth [1993]. Table 3. Summary of the Box Model Calbraton for Basn Tde, a cm Q T, b cm yr 1 Q E, b cm yr 1 r 2 East South Central West a Source s N. P. Smth and P. A. Ptts (Harbor Branch Oceanographc Insttuton, unpublshed report, 1996). b Values of Q T and Q E were estmated durng calbraton by nonlnear regresson usng observed annual average salnty. The box model explaned a much smaller proporton of the varaton n the annual average salnty values n the East Regon than elsewhere n the bay (based on the r 2 values, Table 3). The lower estmated evaporaton and the lower explaned varance n the east may reflect an underestmaton of the freshwater fluxes ether from runoff or drect ranfall nto ths regon. Ths underestmaton probably resulted, at least n part, from our decsons (1) to apply annual runoff unformly to each regon under the assumpton that mxng of runoff throughout the bay was complete wthn a year and (2) to gnore the effect of ranfall contrbutons from the Taylor Slough area below the dscharge gauge. If most of the runoff (and some addtonal ranfall) had been added to the smulatons for the East Regon, the estmated evaporaton (and perhaps the explaned varance) would have been hgher n that regon. Addng less runoff to the other regons would have resulted n lower estmated evaporaton rates. Lackng observatons on dstrbuton and mxng of runoff n the bay (and the necessary resoluton n the steady state, spatally aggregated box model structure), we could not evaluate these potental bases n the regonal evaporaton estmates. We therefore averaged the annual evaporaton rates from all four regons to provde our best, unbased estmate of the bay-wde annual evaporaton rate Influence of Net Freshwater Supply on Salnty Sources of nterannual varaton. The average annual net freshwater supply to Florda Bay s essentally zero. However, there have been large fluctuatons n both the annual ranfall and annual runoff to the bay (Fgure 6). From 1965 to 1995, annual drect ranfall onto the bay vared from about 75 cm yr 1 to about 140 cm yr 1, a range of nterannual varaton of 65 cm yr 1 (range s 65% of mean value). For the same perod the range of nterannual varaton of runoff nto the bay was only 23 cm yr 1 (from 0 to 23 cm yr 1 wth a range of 250% of mean value). Gven the absolute magntudes and ranges of nterannual varatons of ranfall and runoff, t seems lkely that annual varatons of salnty n Florda Bay for the last 3 decades have been prmarly affected by varatons n annual ranfall and only to a lesser extent by changes n annual runoff even though the percentage of changes n runoff have been greater Results of the steady state model. A comparson of the annual average salnty values smulated by the steady state box model wth the observed monthly salnty data supported our concluson that nterannual varatons n ranfall and runoff explaned much of the varaton of salnty n all regons of the bay (Fgure 8). The box model, whch was appled to each

10 1814 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY Fgure 8. The smulated salnty values from the box model (bold lne) when compared to the observed salnty data (fne lne) for 1987 to 1995 revealed the magntude of salnty varatons assocated wth nterannual varatons n ranfall and runoff for each aggregated basn n Florda Bay. The box model was drven only by annual ranfall and runoff and smulated the salnty of all four basns over the 9 years wth an effcency of 37%. regon usng annual fluxes, attaned an effcency of 37% for 1987 through 1995 (Table 4). (For each regon the smulated average salnty for a gven year was used for all months wthn that year when calculatng the effcency by (2).) By contrast, the effcency score for the locaton model, the long-term mean salnty for each regon, was just 16% (Table 4). (For each regon the mean salnty was used for all months n all years when calculatng the effcency by (2).) The dfference n effcency can be attrbuted to the nfluence on salnty of nterannual fluctuatons n ranfall and runoff. That s, approxmately 21% of the varance of salnty n Florda Bay resulted from nterannual varatons of freshwater fluxes. The values of the exchange flux, Q T, n the box model can be nterpreted as the water renewal rate, a functon of tdal exchange wth the Gulf of Mexco and the Atlantc. We compared the magntudes of the calbrated exchange fluxes wth observed tdal ampltudes n each regon of the bay (Table 3) and found a strong correlaton, provdng partal, qualtatve corroboraton of the model calbraton. The calbrated values of Q T (Table 3) were used to estmate resdence tmes n each regon of the bay. Assumng that the average water depth n each regon s 100 cm, resdence tmes n years were defned as 100/Q T. These estmates of resdence tmes ranged from 0.3 to 0.6 years and ndcated that water n the East and Central Regons would requre over a year to be completely replaced by exchange flux. The results of the resdence tme analyss and the fact that observed fluctuatons n the salnty data appear to lag the smulated salnty (Fgure 8) for all regons suggested that annual average salnty was not n steady state wth annual varatons n the net freshwater supply anywhere wthn the bay Seasonal effects. On a monthly bass the average net freshwater supply fluctuated consderably between negatve and postve values (Fgure 7). The average net supply of freshwater was postve durng the rany season (from June through October) and was negatve n the wnter and sprng. Generally, salnty values durng wnter and sprng exceeded the salnty of the adjacent ocean; durng the rany season, salnty values dropped to below ocean salnty (Fgure 5). Although annual runoff was small compared to annual ranfall, the seasonal varaton n runoff was an mportant component of the seasonal varaton n net freshwater supply. For example, total net freshwater supply durng the rany season was about 22 cm, of whch more than 30% was contrbuted by runoff from Taylor Slough and the C111 canal. Ths mples that changes n the amount or the tmng of the seasonal components of runoff may have greater mpact than changes n annual totals alone Results from the dynamc model. FATHOM attaned an effcency score of 51% when compared to salnty observatons for 1987 through 1995, compared to 37% attaned by the steady state box model (Table 4). Ths suggested that approxmately 14% of the varance n observed salnty n Florda Bay was related to seasonal varatons n net freshwater supply. The ncreased effcency largely resulted because FATHOM very successfully smulated both the seasonal fluctuatons and nterannual trends n salnty observatons n the East and Central Regons (Fgure 9). There was a spatal Table 4. Types of Varablty n Salnty Derved From Comparson of Model Results Types of Varablty a Model Effcency (Eff), % Spatal basn means Spatal and nterannual box model Locaton, nterannual and seasonal FATHOM FATHOM s Flux Accountng and Tdal Hydrology at the Ocean Margn model. a Varablty categorzed as spatal (among basns, nterannual), based on annual averages, and seasonal, based on monthly averages.

11 NUTTLE ET AL.: INFLUENCE OF NET FRESHWATER SUPPLY ON SALINITY 1815 Fgure 9. The smulated salnty values from the Flux Accountng and Tdal Hydrology at the Ocean Margn (FATHOM) model (bold lne) when compared to the observed salnty data (fne lne) for 1987 to 1995 revealed the magntude of salnty varatons assocated wth both nterannual and seasonal varatons n ranfall and runoff for each aggregated basn n Florda Bay. FATHOM was drven by monthly ranfall and runoff and smulated the salnty for all four basns over the 9 years wth an effcency of 51%. component to runoff n FATHOM; all runoff was appled to the small bays along the northeastern margn of the bay that border the East Regon (Fgure 3), and no runoff was appled drectly to the Central Regon. Ths agreed wth the locaton of the sources of runoff (Fgure 1) and contrasted to the way runoff was appled n the box model (unformly to all regons). Because of the ncreased spatal resoluton and seasonal nature of the nputs, FATHOM smulated salnty n the East and Central Regons much more successfully than dd the box model. However, FATHOM dd not smulate the salnty varatons observed n the West Regon wth the same success. The West Regon s adjacent to the boundary wth the Gulf of Mexco where salnty n the model was assumed to be constant. However, salnty does vary n the gulf adjacent to the bay, and the lack of ths source of varaton n the model contrbuted to the dscrepancy. Taken together, these seasonal results and the results from the annual analyss of net freshwater supply suggested that varatons n the net freshwater supply nfluenced salnty n Florda Bay at both the seasonal and nterannual tmescales. Three factors, (1) locaton wthn the bay, (2) nterannual varaton of ranfall and runoff, and (3) seasonal varaton of runoff and precptaton, accounted for approxmately 51% of the observed salnty varaton n the bay, each component contrbutng approxmately equally (16%, 21%, and 14%, respectvely). Other mportant sources of varaton not ncluded n these analyses, but whch mght have explaned much of the remanng 49% of salnty varaton, were temporal and spatal patterns of evaporaton, wnd-drven mxng and exchange wth the coastal ocean, spatal patterns of ranfall over the bay, and varatons of salnty at the Gulf of Mexco and Atlantc boundares of the bay Model Reconstructons of Long-Term Salnty Varatons ( ) We compared smulaton results from each model wth salnty observatons from the complete 31-year ( ) database to assess the predctve ablty of the models. The 31- year record contans the temporally dense data used to calbrate the models (9 years from 1987 to 1995 comprsng approxmately 50% of the observatons) and a sparser record that contans approxmately the same number of observatons over a longer perod (22 years). Our purpose was not so much to formally test the models (that would have requred that we evaluate only that data not used n calbraton) as t was to extend the models to dentfy crtcal areas n whch mprovements could be made to both models and the supportng data. The locaton model, based on regonal means of the evaluaton perod (1987 through 1995), attaned an effcency of 20% when appled to the complete 31-year record (Table 4). Ths was not much dfferent from the 16% effcency acheved for the evaluaton perod and suggested that the effects of locaton have not changed much over the 3 decades under consderaton. Lkewse, effcency for the steady state box model appled to all of the data was not sgnfcantly dfferent from the box model effcency acheved on the evaluaton perod (Table 4). We nferred from ths that patterns of nterannual varatons n freshwater fluxes and the patterns of response n annual average salnty were relatvely unform over the perod. However, the effcency score for FATHOM decreased from 51% when appled to the shorter evaluaton perod to 28% for the complete 31-year perod probably because of the data used to drve the models and the salnty data tself. Snce the effcency for FATHOM declned even though the effcences of the other models dd not, the qualty and quantty of the data for the earler perod specfc to FATHOM must have dffered from that n the later evaluaton perod. These knds of data nclude spatally explct patterns of ranfall and runoff, monthly patterns of freshwater fluxes, and temporal and spatal varatons n salnty values along the Gulf of Mexco and Atlantc Ocean boundares. Gven that FATHOM was the most spatally and temporally complex of the models, t should