美国河口分类

Exploring Estuarine Nutrient Susceptibility D O N A L D S C A V I A * A N D Y O N G L I U School of Natural Resources & Environment, University of Michigan, Ann Arbor, Michigan 48109 Received December 1, 2008. Revised manuscript received March 12, 2009. Accepted March 17, 2009. The susceptibility of estuaries to nutrient loading is an important issue that cuts across a range of management needs. We used a theory-driven but data-tested simple model to assist classifying estuaries according to their susceptibility to nutrients. This simple nutrient-driven phytoplankton model is based on fundamental principles of mass balance and empirical response functions for a wide variety of estuaries in the United States. Phytoplankton production was assumed to be stoichiometrically proportional to nitrogen load and an introduced “efficiency factor” intended to capture the myriad processes involved in converting nitrogen load to algal production. A Markov Chain Monte Carlo algorithm of Bayesian inference was then employed for parameter estimation. The model performed remarkably well for chlorophyll estimates, and the predicted estimates of primary production, grazing, and sinking losses are consistent with measurements reported in the literature from a wide array of systems. Analysis of the efficiency factor suggests that estuaries with the ratio of river inflow to estuarine volume (Q/V) greater than 2.0 per year are less susceptible to nutrient loads, and those with Q/V between 0.3 and 2.0 per year are moderately susceptible. This simple model analysis provides a first-order screening tool for estuarine susceptibility classification. Introduction Eutrophication is a threat to coastalwaters that ismost often a result of society-mediateddeliveryof excessnutrients (1-4). This overenrichment can lead to serious andnegative effects, such as harmful algal blooms, habitat loss, biodiversity changes, bottom oxygen depletion, and fishery loss (4, 5). Determining nutrient loading targets to ameliorate these impacts is ultimately anestuary-specificenterprise; however, there is also a growing need to understand more generally why some systems are more susceptible than others so that management guidance can be provided across systems (6). The diversity of estuaries has made classification an important anddifficult question for researchers anddecision makers since the1950s (7-9). TheNationalResearchCouncil proposed 12 factors that control estuarine responses, in- cluding physiographic setting, primary production, nutrient load, dilution, water residence time, stratification, hypsog- raphy, grazing of phytoplankton, suspended materials load and light extinction, denitrification, spatial and temporal distributions of nutrient inputs, and allochthonous organic matter inputs (4). Some recent U.S. classification efforts include a dissolved concentration potential (DCP) index (2), an Assessment of Estuarine Trophic Status (ASSETS) meth- odology (10), theCoastal andMarineEcologicalClassification Standard (CMECS) conceptual classification (11), stressor- response relationships developed over broad geographical scales (12), and a multivariate regression analysis as part of a synthesis toguidedevelopmentof estuarinenutrient criteria (13). Similar efforts have also been developed for Peninsular Malaysia (14), Portugal, the EU Water Framework Directive (15), andEnglandandWales (16). A reviewof 26 classification schemes found that past systems focusedmainly on terrest- rial and aquatic systems and for specific regions and habitat types (9, 12). Kurtz et al. (6) reviewed dozens of classification schemes and concluded that the distinctions among ap- proaches appear to be between hierarchical and nonhier- archical structures, data-driven and theory-driven, and functional vsphysical structural and that someclassifications combine two or more methods or combine classification with other tools like modeling. Our approach is nonhierarchical, theory-drivenbut data- tested, and functional. It is a modeling approach to identify key features useful for classification.We use a simplemodel, based on fundamental principles ofmass balance, empirical response functions, and an introduced estuarine efficiency term for a wide variety of estuaries to explore the basis for their susceptibility to nutrient loads, ultimately contributing to a classification scheme to guide nutrient control policies. As such, our aim is todevelop a screeningmodel for estuaries in general, not a prediction or forecastingmodel for specific estuaries. Methods Data Sources. Data for 99 estuaries are described in NEEA EstuariesDatabase (http://ian.umces.edu/neea) (3). For our analysis, we used 75 of those systems: 14 estuaries were dropped from our analysis based on extreme physical characteristics (e.g., very shallow, very deep, long residence time, or excessive loads). Ten others were dropped because early attempts with our model generated estimates of estuarine efficiency that were quite unrealistic (see below and Supporting Information). The remaining 75 estuaries (37 drowned rivers; 19 lagoons; 9 coastal bays; 10 fjords) still represent a diversity of depths (0.5 to 46 m), volumes (1.7 × 107 to 2.9 × 1010 m3), residence times (4 to 979 days), total nitrogen (TN) loads (1.3 × 104 to 5.3 × 107 kg/year), and summer surface chlorophyll concentrations (2.3 to 24.8 µg/L) (see Supporting Information). Freshwater discharge, salinity, and ocean boundary nitrogen concentrations were also obtained from this database; however, we found the reportedvalues foroceansalinitywere inconsistentwithother published values for some subtributaries of the Chesapeake Bay. Accordingly,we recalculatedwater residence times (see below), based on updated salinity estimates for the Chester, Choptank,Rappahannock,Tangier/Pocomoke,andYork river subestuaries from 1222, 713, 185, 1120, and 121 days to 276, 85, 108, 586, and 92 days, respectively. Growingseasonchlorophyll a concentrationswerederived from Sea-viewing Wide Field-of-view Sensor (SeaWiFS) imagery reportedmonthly for 1997 to 2004 (http://geoportal. kgs.ku.edu/estuary/) (17). We used June-August averages for each of the 7 years. Annual average total nitrogen daily loads, based on the most recent SPARROW model updates (18), were provided by the U.S. Geological Survey (R. Alexander, personal communication). Because SPARROW is not well suited for the relatively flat Florida watersheds, we used NOAA-report fluxes reported on the NEEA Web site. Model Development. While models can be useful tools for describing and predicting specific estuarine responses to* Corresponding author e-mail: scavia@umich.edu. Environ. Sci. Technol. 2009, 43, 3474–3479 3474 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 10, 2009 10.1021/es803401y CCC: $40.75  2009 American Chemical Society Published on Web 04/10/2009 Administrator 高亮 Administrator 下划线 Administrator 下划线 Administrator 下划线 Administrator 下划线 changes in nutrient loads (e.g., refs 19-24), they can also be useful inexploringmoregeneral responses toprovide insights into what controls their susceptibility to eutrophication (25, 26, 16, 27). We developed a nutrient-driven phytoplank- ton model, simplified from previous studies on lakes and estuaries (28-31), that relates summer average phytoplank- ton biomass to spring TN daily loads and estuarine physical characteristics. Rather than model detailed nitrogen dynam- ics (including phytoplankton uptake and biogeochemical cycling), wemodeled summer phytoplankton production as proportional to spring TN load, similar to earlier work simulating biological oxygen demand in the Gulf of Mexico and the Chesapeake Bay (20, 22, 24). The nitrogen loading rate was converted to phytoplankton carbon production by multiplying load by a factor encompassing the C:N ratio for nitrogen-limitedproduction, the relationshipbetweenspring averagedailyandannual averagedaily load, andan“estuarine conversion efficiency factor” intended to capture processes converting nitrogen load to algal production. This is admit- tedly a very strong simplification, but it served the purpose of relating production to load and introducing the efficiency factor that became very useful in assessing estuarine susceptibility. We discuss this in detail below; however, we used this bulk property, the estuarine conversion efficiency, to calibrate themodel and then to explore how it variedwith various estuarine properties. Phytoplankton losses aremodeled as afirst-order sinking rate and a zooplankton grazing term modeled as quadratic inphytoplanktonbiomass. This is similar to approachesused for zooplankton mortality (32-34) under the assumption that zooplankton abundance varies with phytoplankton abundance. Thus, the overall rate of change of mixed-layer phytoplankton carbon (B) is: dB dt ) In- QoutB V1 - vs1 ′ B- LB2 (1) In) TNL V1 R) TNR +TNO V1 R (2) TNO)QinNO, V1) zfV, vs1 ′ ) vs z1 ) vs zfz (3) where B is phytoplankton biomass (g C/ m3), In is phy- toplankton production (g C/ m3/day) derived from spring nutrient load (TNL, g N/day) and calibration term (R, g C/g N);Qout is the outflow to the ocean (m3/day), vs1′ is the sinking rate (1/day), vs is the sinking velocity (m/day), z1 is themixed layer depth (m), z is the estuary average depth (m), L is the grazing loss rate (m3/g C/day), TNL is the sumof TNR (spring riverine TN load, g N/day) and TNO (ocean nitrogen influx, g N/day) (ignoring atmospheric deposition sources and N fixation), NO is the ocean nitrogen concentration (mg/L); Qin is ocean inflow (m3/day), V1 is the mixed layer volume (m3), and V is the estuary volume (m3). The ratio of mixed layer depth to total depth zf is 1.0 for well-mixed estuaries and assumedtobe0.5 for stratifiedestuaries.Weassumed lagoons and all other estuaries with depth <3.0 m were well mixed and that all fjords were stratified. The water residence time (WRT, day), Qout, and Qin can be calculated from average estuarine salinity (Sal1), ocean boundary salinity (Sal0), and river discharge (Q), all of which are in the NEEA Estuaries Database, and water and salt balances, as: QinSal0)QoutSal1 Q+Qin)Qout w{Qout)Q Sal0Sal0- Sal1Qin)Q Sal1Sal0- Sal1 (4) WRT is defined as: WRT) V Qout ) V Q × Sal0- Sal1 Sal0 (5) To explore the model’s ability to reproduce summer phytoplankton concentrations,we solved eq1 at steady state under the assumption that this will provide analytical power and adequate distinctions among estuaries (35). While phytoplanktonbiomass certainly varies over shorter periods and for most estuaries those differences are generally attenuated at annual scales (36), there remains sufficient discrimination amongestuaries for this analysis. The steady- state solution, obtained by setting (dB)/(dt) ) 0, is: B) -(Qout+V1vs1 ′ )+ √(Qout+V1vs1′ )2+ 4InLV 12 2LV1 (6) Parameter Estimation. Bayesian analysis has been in- creasingly applied in ecology (37, 38) because of its ability to handle uncertainty, incorporate prior information such as data andmodeling experience, and develop probabilistic assessments to support decision making (39). Compared to traditionalmodeling approaches, Bayesian-estimatedmodel parameters are embodied in theposteriorprobability density functions,whichprovidecredible intervals forbothparameter values and predicted states under various probability levels (40). See Supporting Information for details. We used WinBUGS (version 1.4.3) (41), called from R (version 2.6.0; R2WinBUGS (version 2.1-8)) (42). With WinBUGS, we estimated a single value for parameters L,vs, and C:CHL across all estuaries and individual values of R for each estuary. Our previous modeling analysis revealed that the four parameters are correlated (see Supporting Informa- tion), so we used the following informative priors, based on literature information and experience (39), to provide relia- ble parameter estimates: L ∼ N(0.80,0.25)I(0,); C:CHL ∼ N(50,20)I(0,); vs ∼ N(0.3, 0.10)I(0,). The numbers in the brackets represent themean and the standard deviation and I(0,) denotes censoring to eliminate negative values. We conducted a sensitivity analysis on the forms of the informa- tivepriors (see Supporting Information) and found theywere relatively insensitive to the variance changes. We used a noninformative prior for R, assuming a normal distribution with unspecified mean and common variance, since we did not have credible prior information for this derived property (see Supporting Information) and we wanted to allow the algorithm maximum flexibility in its estimation. Four goodness-of-fit measures were used to test model results between predicted and observed values: correlation coefficient, slope of the regression, coefficient of determi- nation, R2, and the root mean squared error (RMSE) (see Supporting Information). Results C:CHL and Chlorophyll Estimates. The model performed remarkably well, with a correlation between predicted and observed chlorophyll of 0.99 (Supporting Information). The slope of the regression fit is 0.96 with an intercept of 0.17 which is not significantly different from zero. R2 is 0.99 with a RMSE of 0.50 and scale-independent RMSE of 0.051. We used a potential scale reduction factor, Rhat, to determine model convergence. Resulting Rhat values are all close to 1.0, indicating themodel converges well (42). Themean and standarddeviationof theposteriordistribution for thecarbon to chlorophyll ratio was 56 ( 10.6, well within the range reported in the literature (43-47). While these estimates are satisfying, it is important to also compare our calculated production and loss rates to observations because even simplemodels are capable ofmatching state variables based on erroneous, yet compensating, rate processes (e.g., ref 28). VOL. 43, NO. 10, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 3475 Administrator 高亮 Administrator 下划线 Administrator 高亮 Administrator 高亮 Administrator 高亮 Administrator 高亮 Administrator 高亮 So we made those comparisons to ensure that this is not simply sophisticated curve-fitting. PhytoplanktonPrimaryProduction.Model estimates for growing-season phytoplankton primary production ranged between 0.05 and 8.0 g C/m2/day, with first, second, and third quartiles of 0.24, 0.45, and 0.78 g C/m2/day. These estimates represent the central tendency of production for eachestuarywithdistributionsassociatedwithR (eq2). Thus, the overall distribution of the model production estimate is a mixture distribution. For comparison, we compiled sum- mariesofphytoplanktonprimaryproduction for112estuaries andcoastal systems (48-51) andcompared theirdistributions to our model. For cases where production estimates were reported as annual average daily rates, we assumed that 70% of the annual productionoccurs during the 7month growing season.Comparing thenotchedboxplots (Figure1) illustrates that the distribution of our predictions are indistinguishable from those empirical estimates. Grazing andSinking LossRates.Themean and standard deviation of grazing and sinking parameters were 0.69 ( 0.27 m3/g C/day for L and 0.21 ( 0.07 m/day for vs, well within expected ranges for grazing (33, 16) and sinking (46, 47). In addition, grazing loss as a percent of primary production was 66 ( 18%, compared to 24 ( 15% for sedimentation, suggesting that grazing is often themain loss term. Estimates for the Strait of Georgia (52), Halifax Harbor (53), ChesapeakeBay (54),MobileBay (55), andApalachicola Bay (56) all suggest that grazing was the primary factor controlling phytoplankton biomass. The consistency of the general patterns of model output and these observations is demonstrated by comparing the frequency distributions of model output to these field measurements across a wide array of systems (Figure 2 and Supporting Information). While some of our grazing estimates seem to be a bit higher than those reported in the literature, the overall comparison is quite good. EstuarineEfficiency.Theabove comparisonsofmodeled andmeasuredproduction, sinking, andgrazingdemonstrate that the model not only fits the observed phytoplankton chlorophyll concentration across this diverse set of estuaries but also fits key rate processes well. This lends credence to using the model to explore relative estuarine sensitivity through our estimates estuarine efficiency. The Bayesian estimated mean value of R that best fit chlorophyll observa- tions ranged between 0.52 g C/g N and 159.5 g C/g N, with theuncertainty around individual values relatively constant. Mean and standard deviation of the coefficients of variation were 17 ( 1%. This calibration term,R, is composed of three factors: the nitrogen-limited C:N ratio for production, a factor relating average spring daily nutrient loads to annual average daily loads, and the estuarine efficiency factor. Because we want to explore the efficiency term,weneed to factor out the other two; although it is important to note that the scaling factors influence the absolute value but not the patterns of R across estuaries. The Redfield C:Nmolar ratio is often used for these types of estimates; however, recent evidence suggests that under nitrogen-limited conditions, carbon overconsumption (57) drives the C:N ratio higher. For our analysis, we used 12.7 (10.9mass ratio), basedonanaverageof14estimates reported in the literature (58-64). Inmost estuarine systems, average daily spring loads are considerably higher than the annual average.Forouranalysis,weassumedtheaveragedaily spring load was 2.0 times the annual average daily load. Thus, to estimate estuarine efficiency we divided R by 21.8 (10.9× 2), producing efficiency terms between 0.02 and 7.34 (inter- quartile range: 0.34-2.28). Estuaries with efficiency terms greater than 1.0 can be considered “recyclers”; those below 1.0 can be considered “N sink” systems or highly flushed systems. This is discussed further below. Discussion Themodel reproduced summer chlorophyll concentrations as a function of total nitrogen load and the physical characteristics of the estuary for a wide range of estuarine types and conditions (Supporting Information). This was based on several simplifying assumptions, the most useful ofwhichwas the introductionof anestuarine efficiency term, representing the fraction of the spring nitrogen load con- verted to algal biomass. There are, of course,manyprocesses thatmodulate that conversionand reduceoverall conversion efficiency, including denitrification, delivery of unavailable nutrient forms, sedimentburial, and rapidflushingcompared to algal production. There are also processes that enable recycling of nitrogen and increase the conversion efficiency. Our analysis does not distinguish among those processes but rather explores their net effect. We explored how predictedestuarineefficiency, ε)R/21.8, variedwithdifferent FIGURE 1. Frequency distribution of net primary production estimates for our study (75 estuaries) and that summarized in Montes-Hugol et al. (15 estuaries) (51), Smith and Hollibaugh (22 estuaries) (50), Boynton et al. (45 estuaries) (48), and Underwood and Kromkamp (30 estuaries) (49). 3476 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 10, 2009 estuarine properties and found themost useful relationship with the ratio of river discharge to estuarine volume (Q/V) (Figure 3, Supporting Information). Note that Q is the river discharge, not the sum of that discharge and ocean inflow, which is convenient because the latter is more difficult to estimate. In this analysis, efficiency appeared to decrease roughly with the inverse square root of Q/V: ε ) 0.908(Q/V)-0.47 (R2) 0.53), where ε represents mean values arising from the 75 estimated normal distributions. This is logical because load generally increases with inflow (Q) and, for a given estuarine volume, one would expect the system to be less efficient in processing that load and, in fact, be overloaded for high values of Q. Conve