The National Estuarine Research Reserve System represents a diverse collection of ecosystems among which environmental conditions differ dramatically, making it inherently difficult to determine the extent to which patterns and properties identified in one estuary are transferable to those of any other. The primary objective of our study was to develop a multivariate classification framework for comparison of these estuaries and identify the primary sources of environmental variability in each. Using a 4-year dataset from the National Estuarine Research Reserve System-Wide Monitoring Program, combined with principal components analysis, we identified distinct patterns among 21 reserves that allowed grouping based on the primary factors shaping physicochemical variability. Salinity and temperature were the primary factors shaping variability in the reserves, an observation that was corroborated by similar multivariate analysis of data from 33 published studies of non–National Estuarine Research Reserve systems representing a wide range of coastal and estuarine waters. We then investigated the effect of temperature and salinity on biological processes in these systems by using the ratio of bacterial production to chlorophyll-a as a response variable. Salinity and temperature had different but significant effects on bacterial production/chlorophyll-a ratios, suggesting in turn that these properties contribute to the balance between autotrophic and heterotrophic planktonic processes in estuarine ecosystems. Our study confirms the universal role of salinity and temperature in shaping the variability among even the most diverse systems and provides a valuable classification framework for comparison of reserves within the context of the entire National Estuarine Research Reserve System. Use of this classification approach may provide insight into the extent to which results from investigative studies in one reserve may be applicable to others, a valuable application when the effect of environmental stressors is considered.
INTRODUCTION
The National Estuarine Research Reserve System (NERRS) is composed of more than 4000 km2 of protected coastal, estuarine, and wetland areas. These reserves represent a diverse collection of aquatic ecosystems that experience a wide range of environmental conditions. Significant scientific utility of the reserves has been realized through the establishment of a System-Wide Monitoring Program (SWMP) and subsequent characterization through evaluation of long-term datasets (Kennish, 2004; Wenner et al., 2004). This, in turn, has improved our understanding of the range and variability of environmental conditions in the reserves and provides the basis for comparison to other NERRS and non-NERRS estuaries.
Estuaries collectively represent a diverse array of dynamic aquatic systems that vary substantially in their state of enrichment, salinity, morphology, and hydrodynamics (Kennish, 1992). The temporal and spatial variability of these systems is such that environmental conditions for even a single category of estuary (e.g., drowned river valleys) may vary tremendously both within and among such systems (Day et al., 1989). In addition, the primary factors regulating plankton community activity and abundance are highly variable among estuarine systems and include anthropogenic nutrient inputs (Revilla et al., 2000), light availability (Harding, Meeson, and Fisher, 1986), salinity (Pradeep Ram, Nair, and Chandramohan, 2003), season, temperature, and organic matter source and quality (Apple, del Giorgio, and Kemp, 2006; Lomas et al., 2002; Raymond and Bauer, 2000). As a result, identifying representative estuarine systems is challenging, since the primacy of any one parameter in driving the physical and biological variability is, in itself, highly variable among systems. Yet identifying the physico-chemical factors driving estuarine variability is fundamental to determining the extent to which patterns and properties identified in one location (e.g., a NERRS reserve) are transferable to the understanding of any other estuary.
In studies of estuarine ecosystems, algal biomass (i.e., chlorophyll-a concentrations) is typically used as an index of system-level productivity, trophic state, and autotrophic carbon production (Cloern, 2001), while bacterial production (BP) is a common measure of the activity of heterotrophic microbes (Kirchman, 1993) and their carbon demand (del Giorgio and Cole, 1998). The relationship between BP and chlorophyll-a (CHLA) has been used in a number of aquatic systems as an index for heterotrophic and autotrophic coupling (e.g., Cole, Findlay, and Pace, 1988; White et al., 1991). Previous studies identify a positive correlation between BP and CHLA across a wide range of systems and relate the strength of this relationship to the coherence of autotrophic and heterotrophic processes. However, these previously published studies are limited to marine and freshwater systems, and it is not altogether clear if such a relationship applies to estuaries, where inputs of allochthonous organic matter are large and often drive net ecosystem metabolism (Caffrey, 2004; Hopkinson and Smith, 2005).
The primary objective of the present study is to characterize the NERRS reserves based on the primary factors driving physicochemical variability in each and develop a multivariate classification framework to facilitate comparisons among the reserves. This is accomplished by combining principal components analysis (PCA) with water quality and nutrient data collected as part of NERRS SWMP and using this analysis to establish relationships among the reserves. We then evaluate the effect of environmental factors on biological processes in estuarine ecosystems using the BP : CHLA relationship as a response variable. Data for these analyses were collected in Monie Bay and North Inlet–Winyah Bay NERRS reserves and combined with a comprehensive review of published studies conducted in other estuarine ecosystems.
METHODS
Data Collection: NERRS SWMP
A composite dataset combining measures of water quality and nutrients was generated using SWMP data available through NERRS ( http://cdmo.baruch.sc.edu/). Water quality parameters included temperature (°C), salinity (psu), specific conductivity (mS cm−1), dissolved oxygen (DO; in % saturation and mg L−1), turbidity (NTUs), depth (m), and pH. These parameters were monitored continuously with a YSI 6600 EDS datasonde throughout the ice-free period of the year and reported at 15- to 30-minute intervals. Grab samples for nutrients typically occurred on 3- to 4-week, intervals with minimum sampled parameters including CHLA (μg L−1), dissolved orthophosphate (PO43−), ammonium (NH4−), nitrite (NO2−;), nitrate (NO3−), nitrite + nitrate (NOx), and total dissolved inorganic nitrogen (DIN = NH4+ + NOx). All dissolved nutrients were reported in mg (N or P) L−1. Measures of phaeophyton, silicate, total suspended solids, total nitrogen, total dissolved nitrogen, total phosphorus, total dissolved phosphorus, dissolved organic nitrogen, particulate organic carbon, and dissolved organic carbon were not collected at all reserves or on all sampling events and were thus omitted from our multivariate analyses. Downloaded water quality and nutrients files for all years and reserves were date–time matched and concatenated into a single file using a Visual Basic (VB.NET version 8.0, Microsoft) parsing program written by the authors, which paired discrete nutrient sampling events with the closest continuous water quality monitoring event for each station, resulting in a minimum of 15 environmental parameters for each concatenated sampling event. Data collected as part of diel time series were not included in the analyses. Analyses focused on differences across the NERRS at the reserve level, with characterization of conditions in each reserve based on those sites for which all 15 parameters were recorded and with no attempt to differentiate sites within specific reserves. The final composite dataset represented 86 sampling stations in 21 reserves and included 2820 concatenated entries of water quality and nutrients (Table 1).
Data Collection: Published Literature
A composite dataset of literature values from 33 independent studies reporting simultaneous observations of bacter-ioplankton production and CHLA concentrations in estuarine, coastal, and marine surface waters was compiled. Classification of systems as estuarine, coastal, or marine (i.e., open ocean) was generally provided by the author, with distinction between marine and coastal systems being based on proximity to land, potential for terrestrial influence, and water depth. This dataset included 779 paired independent measures of BP and CHLA (Table 2). When available, concurrent measures of temperature (n = 633), salinity (n = 530), and inorganic nutrient concentrations (i.e., PO43−, n = 450; NH4+, n = 272; NOx, n = 401; DIN, n = 314) were also recorded. Almost all studies (>90%) were conducted in temperate estuarine and coastal systems, with the exception of one tropical estuary (Bano, Moran, and Hodson, 1997) and four open ocean studies (Bode et al., 2001; Caron et al., 2000; Conan et al., 1999; Sherr and Sherr, 2001). We avoided studies reporting values for BP and CHLA that were averaged over spatial or temporal (>1 day) scales. Estimates of BP were typically reported in carbon units and converted to a standard rate of μgC L−1 h−1. Rates of leucine and thymidine incorporation were converted to carbon units using 3.1 kg C mol Leu−1 (Simon and Azam, 1989) and 1.89 kg C mol TdR−1 (Ducklow, Kirchman, and Anderson, 2002), respectively. Bacterial production reported as change in cell abundance was converted to carbon units using 20 fg C cell−1 (Lee and Fuhrman, 1987). When tabular data were not available in the published document, they were either provided by authors or extracted directly from figures using DataThief software (B. Tummers, http://www.datathief.org).
Field Sampling
Monie Bay is a tidally influenced subestuary of the Chesapeake Bay and one of three components of Maryland's Chesapeake Bay NERRS. This system is located on the eastern shore of Maryland and consists of a semienclosed bay and three tidally influenced creeks varying in their agricultural land use, salinity, and other watershed characteristics. North Inlet–Winyah Bay is located on the central South Carolina coast and consists of two components: a tidally influenced lagoonal estuary (North Inlet) and a larger coastal plain estuary (Winyah Bay) that receives inputs from five coastal rivers (Buzzelli et al., 2004). Sampling of Monie Bay occurred monthly during 2000–2003 and is described in detail by Apple, del Giorgio, and Newell (2004). North Inlet–Winyah Bay was sampled during July and September 2006 and followed the same protocol as that in Monie Bay. Briefly, stations were accessed via a small boat and measures of water column chemistry (i.e., temperature, salinity, and Secchi depth) were recorded. Approximately 20 L of subsurface (<0.5 m) water were collected, transported back to the laboratory, and subsampled for BP, CHLA, and dissolved nutrients following standard protocols (Apple, del Giorgio, and Newell, 2004).
Statistical Analyses
All statistical analyses, including standard least squares regressions, one- and two-way analyses of variance (ANOVA), means comparisons (Tukey-Kramer honestly significant difference, or HSD; α 0.05), and PCA were performed using the JMP version 5.0.1 statistical software package (SAS Institute Inc., Cary, NC). Values for BP, CHLA concentrations, and dissolved nutrients were log-transformed prior to analyses. Comparison of means and PCA were also performed individually on SWMP data assigned to one of eight relevant biogeographic regions ( http://nerrs.noaa.gov/), including the Acadian (WEL, GRB), Californian (TJR, ELK), Carolinian (NOC, NIW, ACE, SAP), Columbian (SOS, PDB), Fjord (KAC), Louisianan (APA), Virginian (NAR, WQB, MUL, DEL, CBV, CBM), and West Indian (JOB, RKB). (See Table 1 for explanation of abbreviations.)
RESULTS
Water Quality and Nutrient Parameters (SWMP Data)
Overall means for the entire SWMP dataset are reported in Table 3. Mean values were also calculated for each SWMP parameter in each of the 21 reserves (Table 4){ label needed for table-wrap[@id='i1551-5036-55-sp1-59-t402'] }{ label needed for table-wrap[@id='i1551-5036-55-sp1-59-t403'] } and ordered using significant differences as identified by Tukey-Kramer HSD. Highest mean temperatures were recorded at JOB and RKB (28°C) and lowest at KAC (7°C). Mean salinity was highest at JOB (35) and ELK (32) and lowest at CBM (0.1) and DEL (3). Salinity and pH were positively correlated in the composite dataset, with almost neutral pH recorded in DEL (6.9), WEL (7.0), and CBM (7.0) and most alkaline conditions at ELK (8.1). Percent saturation and absolute concentration (mg L−1) of DO were well correlated, with the highest means recorded at KAC (106% saturation and 10.6 mg L−1) and lowest recorded at TJR (61% saturation and 4.9 mg L−1) and NIW (57% saturation and 4.8 mg L−1). The most turbid systems were ACE, NOC, and SAP (77, 61, and 40 NTU, respectively), with the highest values consistently recorded in ACE. Least turbid were NAR and WQB (both 2.0 NTU), although mean values for these two reserves were not statistically different than the majority of the other reserves. The greatest water depth was recorded at KAC (6.4 m), followed by GTM (2.5 m) and PDB (2.2 m). The shallowest systems were TJR and CBV (0.3 m).
Mean CHLA was highest in DEL, CBM, and SAP (10.0, 9.0, and 7.6 μg L−1, respectively) and lowest in KAC, TJR, and NAR (all <1.2 μg L−1). Highest mean PO43− occurred at TJR (0.44 mg L−1), SAP (0.12 mg L−1), and ACE (0.09 mg L−1), while the most PO43− deplete conditions were at MUL (0.01 mg L−1) and APA (0.008 mg L−1). As with PO43−, the highest mean NH4+was observed at TJR (1.25 mg L−1), followed by JOB (0.16 mg L−1) and NOC (0.13 mg L−1). Lowest means were observed for KAC, WQB, and NAR (all <0.03 mg L−1. Means in NOx did not follow patterns in PO43− or NH4+. Highest mean NOx was recorded in CBM (0.65 mg L −1). DEL (0.75 mg L−1) and lowest in JOB, RKB, and NIW (all <0.01 mg L−1).
There was a highly significant effect of biogeographic region on all parameters in the composite SWMP dataset (ANOVA, p <0.0001), and significant differences existed between means within all biogeographic regions (Tukey-Kramer HSD, α = 0.05). Data from the West Indian region exhibited the highest mean temperature (27°C) and salinity (31), while lowest means for these temperature and salinity were recorded in the Fjord and Virginian regions (7°C and 13, respectively). Highest DO was recorded in the Fjord region (106% saturation and 10.6 mg L−1), as was the greatest mean water depth. Reserves from the Carolinian region had the lowest DO (70% saturation and 5.1 mg L−1), were the most turbid (44 NTU), and had the highest CHLA (5.2 μg L−1). Lowest mean turbidity and CHLA were recorded in the Fjord region (3 NTU and 0.9 μg L−1, respectively). Highest mean PO43− (0.08 mg L−1) and NH4− (0.1 mg L−1) occurred in the Californian region and lowest in the Louisianan (0.004 mg L−1) and Fjord (0.01 mg L−1) regions. The most NOx-enriched region was Columbian (0.1 mg L−1), and the most deplete region was West Indian (0.01 mg L−1).
Principal Components Analysis
Principal components analysis of paired water quality and nutrients measurements derived from SWMP data identified two composite variables (hereafter PC1 and PC2) that explained 48% of the variability within the dataset (n = 1973), with 28% and 20% attributed to PC1 and PC2, respectively (Figure 1, Table 5). Ten parameters were used in PCA (temperature, salinity, pH, water depth, percent saturation and concentration of DO, turbidity, and dissolved concentrations of PO43−, NH4+, and NOx). Two parameters available in the composite dataset were omitted from our analyses: specific conductivity due to redundancy with salinity and DIN because it is derived from NOx and NH4+ and well correlated with these parameters (r2 = 0.96, p <0.0001). CHLA was initially omitted from PCA because this parameter was not recorded at JOB and including CHLA would have eliminated this reserve from our comparative analyses. An additional PCA was performed including CHLA, which produced loading patterns similar to those of the first analysis and with CHLA grouping closely with NH4+, PO43−, and turbidity. In both analyses, PC1 had high positive loadings for salinity and pH and was negatively correlated with NOx, whereas PC2 was positively correlated with temperature and negatively correlated with both DO percent saturation and concentration (Figure 1, Table 5). Coordinate vectors for each parameter included in PCA (solid arrows), and mean factor loadings for each reserve (solid circles) allow visualization of the controlling environmental factors most important for each NERRS (Figure 1). Horizontal distribution of reserves (i.e., along PC1) generally represented a gradient of salinity and NOx concentrations, while vertical distribution represented one of temperature and DO. This resulted in warmer and more saline systems being located predominantly in the upper right quadrant, cooler saline systems in the lower right quadrant, and freshwater-influenced reserves to the left of the origin. Groupings of reserves generally coincided with the biogeographic region in which they were located (shaded regions in Figure 1). Mean annual temperature was well correlated with mean factor loadings from each reserve on PC2 (Figure 2; r2 = 0.76, p <0.0001, n = 21), although two reserves (KAC and TJR) deviated markedly from this general trend. Removal of these outliers from the regression analysis greatly improved the relationship (i.e., r2 = 0.87, p <0.0001, n = 19).
SWMP data was divided into subsets representing each biogeographic region and evaluated using PCA. Cumulative percent variability explained by PCA for each biogeographic region was equal to or greater than that of the composite SWMP dataset and ranged from 48% (West Indian region) to 66% (Columbian region). For data from the Columbian region, all 10 water quality parameters were highly and positively correlated with either PC1 or PC2 (i.e., eigenvectors ≥ 0.7). In general, analysis of data from each biogeographic region generated patterns similar that of the entire dataset, with temperature and salinity driving most of the variability and exhibiting negative correlations with DO and dissolved nutrients, respectively. Exceptions to this pattern included weaker temperature loading in Acadian, Californian, and West Indian regions (i.e., eigenvectors of 0.46, 0.28, and 0.38, respectively); weak salinity loading in the Fjord region (0.46); and general absence of nutrient influence in the Carolinian region.
PCA of literature data produced loading patterns that were strikingly similar to SWMP (Table 5), with a negative correlation between heavily loaded salinity (−0.64) and dissolved nutrients (>0.77) on PC1 and a strong positive temperature loading on PC2 (0.94). PCA of literature data, including BP, also revealed a strong positive correlation between temperature and BP and heavy loading on PC2. Analysis of literature data produced negative loadings of NH4+ and NOx on PC1 that were much stronger than those of SWMP data. Although loading of CHLA on PC1 and PC2 was similar to that of SWMP data, it was not as strongly correlated with dissolved nutrients. Because of limited overlap in parameters reported by published studies, literature dataset PCA included only five parameters (temperature, salinity, PO43−, NH4+, and NOx), and the dataset was much smaller (n = 207). This analysis explained 75% of the variability within the literature data, with 56% and 19% attributed to PC1 and PC2, respectively.
Covariation of BP and CHLA
The literature dataset revealed a highly significant positive relationship between BP and CHLA across a range of marine, coastal, and estuarine systems (Figure 3). Approximately 24% of the variability in BP was explained by CHLA, with a positive y-intercept of 0.37 μgC L−1 h−1. Superimposed on this figure, but not included in regression statistics, are data from field sampling of Monie Bay and North Inlet conducted as part of the present study. The relationship between BP and CHLA reported in other studies was corrected for discrepancies in rate units and converted to common units of μgC L−1 h−1 prior to comparison, revealing similar slope and y-intercept when Cole, Findlay, and Pace (1988) and White et al. (1991) were considered but steeper slope and lower y-intercept than those reported by Murrell (2003). Within our dataset, open ocean systems were characterized by lower BP and CHLA, with data from studies in the Atlantic Ocean (Bode et al., 2001; Caron et al., 2000) exhibiting higher BP per unit CHLA relative to those from the Pacific (Sherr and Sherr, 2001). The majority of coastal and estuarine systems had higher rates of BP and higher concentrations of CHLA relative to open ocean data, with the highest values observed in estuarine systems. Paired measures of BP and CHLA from Monie Bay had a similar distribution to that of other estuarine systems but no significant relationship (p ≤ 0.6, n = 159). Four systems emerged from the literature and field data that exhibited highly significant and relatively strong covariation of BP and CHLA (Figure 4), including Indus River Delta (Bano et al., 1997), Urdaibai estuary (Revilla et al., 2000), Menai Strait (Blight et al., 1995), and North Inlet (present study).
Temperature was positively correlated with BP : CHLA and explained approximately one-third of the variability in the BP vs. CHLA relationship (Figure 5A). The relationship between BP : CHLA and salinity was parabolic across the entire salinity range (Figure 5B), with a second-order polynomial providing a highly significant fit (i.e., p <.0001) and explaining 25% of the variability. The salinity associated with this minimum value (i.e., first derivative) was 17, reflecting the relatively low values of BP per unit CHLA in mesohaline waters. We did not observe a significant relationship between BP : CHLA and ambient nutrient concentrations (i.e., PO43−, NH4+, NOx, and DIN).
DISCUSSION
Principal Components Analysis
The overview of mean water quality and nutrient values presented in this study (e.g., Tables 3 and 4{ label needed for table-wrap[@id='i1551-5036-55-sp1-59-t402'] }{ label needed for table-wrap[@id='i1551-5036-55-sp1-59-t403'] }) and others (e.g., Wenner et al., 2004) is useful for placing environmental conditions at any one reserve within the context of all others. However, comparisons based on a single environmental parameter cannot represent all the factors that contribute to the physicochemical character of a reserve and are thus limited in the ability to reveal the extent to which reserves are similar or may differ. In this regard, the use of PCA—and more specifically, the distribution of reserves along PC1 and PC2 axes—is a meaningful and integrative means of classifying reserves based on similarities in environmental conditions. For example, reserves located in the upper right quadrant of Figure 1 represent those systems that are more saline and located in warmer climates (e.g., JOB in Puerto Rico, RKB and GTM in Florida, and SAP in Georgia), while saline yet much colder systems appear in the lower left (e.g., KAC in Alaska). Many reserves that share geographic proximity but differ substantially with respect to distribution of sites, hydrography, and estuary type still group closely together on the PCA biplot (e.g., NAR with WQB, CBV with MUL, and DEL with CBM), suggesting that landscape-scale characteristics or regional climatological factors may often have a significant impact relative to local environmental conditions.
Salinity and Nitrogen Loading
Salinity had the greatest impact on variability in environmental conditions among the reserves, as evidenced by the strong positive loading on PC1. However, loading of other parameters (e.g., NOx and pH) on PC1 suggests that it is not salinity alone but rather the delivery of nitrogen associated with freshwater inputs that drives variability in SWMP data. This allows a more accurate characterization of physicochemical processes of these estuarine systems. The apparent influence of salinity may therefore actually be a proxy for more complex processes related to nutrient and organic matter loading in these systems. Reserves that appear to be most influenced by this phenomenon include DEL and CBM, as evidenced by their strong negative correlation with PC1. Also located along the axis of PC1 are three other mid-Atlantic reserves, including MUL in New Jersey, CBV, and NOC, suggesting that the influence of salinity and freshwater delivery of nitrogen may be a phenomena associated with climatic conditions and/or regional-scale watershed characteristics. A similar relationship between salinity and dissolved nutrient concentrations has been observed previously in multivariate analyses of other NERRS reserve data (i.e., Monie Bay; Apple, del Giorgio, and Newell, 2004) and may be a characteristic of estuarine systems with tributaries that drain agriculturally developed watersheds. Further multivariate analysis of SWMP data incorporating available information regarding watershed characteristics would provide valuable insight into the role of landscape-scale characteristics on the variability in the reserves.
Temperature and Dissolved Oxygen
Temperature was another important factor explaining variability in SWMP data, as evidenced by the positive loading of temperature on PC2 (Table 5) and highly significant correlation between mean reserve temperature and factor loadings on PC2 (Figure 2). In addition to revealing the universal effect of temperature on SWMP data, PCA provides insight into the role of temperature-dependent heterotrophic processes in regulating ambient DO concentrations. The strong negative correlation of DO (mg L−1) and temperature revealed by both PCA (Table 5) and direct comparisons (r = −0.69, p <0.0001) is not surprising, as temperature strongly dictates the physical solubility of oxygen in water. However, when expressed as percent saturation, which removes the physical effects of temperature and salinity, the strong negative correlation remains. This pattern was also observed by Caffrey (2004) in an earlier analysis of SWMP data and probably represents a general characteristic among estuarine ecosystems. Although the drivers of this phenomenon are many, it likely results from the positive effect of temperature on planktonic (Kemp et al., 1992; Sampou and Kemp, 1994) and benthic (Caffrey, 2004; Cowan and Boynton, 1996) respiration, as well as the disproportionate effect of temperature on heterotrophic relative to autotrophic processes (Lomas et al., 2002). Indeed, PCA of literature data revealed a strong correlation between temperature and BP but none between temperature and CHLA. The strong temperature dependence of DO—specifically, percent saturation—emphasizes the important contribution of heterotrophic processes in estuaries. Not surprisingly, these processes are subsidized by significant inputs of allochthonous organic matter that are characteristic of most estuaries and that contribute to the net-heterotrophic conditions typically encountered (e.g., Hopkinson and Smith, 2005).
The temperature dependence of DO that we have observed is not absolute, and deviation of reserves from the otherwise strong positive relationship between temperature and factor loadings on PC2 (Figure 2) reveals those systems in which DO variability deviates substantially from what might typically be expected. For example, based on the relationship observed for most reserves, loading on PC2 was lower than expected (i.e., DO was higher) for KAC. This system experiences some of the lowest temperatures and turbidities (and thus highest light availability) recorded in SWMP data. Higher salinities characteristic of this reserve suggest, in turn, that the role of allochthonous organic matter is minimal as a subsidy for heterotrophic activities. Thus, uncharacteristically high DO concentrations in KAC may be attributed in part to higher solubility associated with colder waters but more importantly to elevated primary production relative to temperature-dependent heterotrophic activity (Billen and Becquevort, 1990).
Tijuana River (TJR) also deviated from the otherwise strong relationship between temperature and PC2 factor loading, with higher-than-expected loadings indicating uncharacteristically low DO concentrations. In the case of TJR, the temperature dependence of DO may be more strongly influenced by factors known to drive hypoxic conditions in estuaries, such as inorganic nutrients (Kemp et al., 2005; Paerl et al., 1998), inputs of allochthonous dissolved organic matter, and subsequent degradation by heterotrophic microbial processes (Bano, Moran, and Hodson, 1997; Carlsson, Graneli, and Segatto, 1999). TJR is one of the most phosphate-replete reserves, having 8 of the top 10 measurements for PO43−;from the entire dataset, 4 of which were in excess of 1.0 mg L−1. A similar pattern was observed for ammonium in this reserve. This enrichment was reflected in DO, with over half of the 40 sampling events within the SWMP dataset that recorded extremely low DO concentrations (i.e.,<1 mg L−1) occurring in TJR. Low DO may function as a positive feedback mechanism in this system, with reduced, hypoxic conditions causing the release of sediment-bound phosphate (Scudlark and Church, 1989) and accumulation of water column ammonium (Kemp et al., 1990). Such a relationship between hypoxia and ambient NH4+;and PO43− concentrations is often attributed to nutrient-driven increases in primary production and subsequent consumption of algal dissolved organic matter by heterotrophic microbes (Cloern, 2001). However, TJR experiences vigorous tidal flushing that prevents the development of persistent algal blooms (Fry, Gace, and McClelland, 2003), and most of the organic matter and nutrients delivered to this system come from raw sewage (Gersberg et al., 2006). The absence of elevated concentrations of organic matter in surface waters discharged from TJR (Zeng, Khan, and Tran, 1997) indicates that most of the inputs of organic matter are removed, leading to the conclusion that low DO conditions in this system are driven by heterotrophic consumption of allochthonous organic matter that is present in the water column or that has been deposited in estuarine sediments. In this regard, TJR represents an extreme example of how excessive inputs of allochthonous organic matter and heterotrophic processes can combine to produce netheterotrophic conditions and further degrade water quality in estuarine ecosystems.
Covariation of BP and CHLA
The scope and resolution of SWMP data are both extensive and comprehensive, yet the lack of measures of water column respiration, or another index of bacterioplankton community activity in the SWMP data, prevents us from directly addressing the magnitude of heterotrophic processes and their covariation with planktonic autotrophs within the NERRS. Instead, we investigated one aspect of this coherence in estuarine ecosystems using a synthesis of published studies reporting simultaneous measures of BP and CHLA and then compared this with patterns seen in two specific NERRS sites. Studies of the relationship between BP and CHLA have been conducted in a subtropical estuary (e.g., Pensacola Bay; Murrell, 2003) and in reviews of published literature focusing on marine and freshwater systems (Cole, Findlay, and Pace, 1988; White et al., 1991). These have generally reported a positive covariation, with the coherence of BP and CHLA serving as an index of ecosystem status and resource supply. Our analysis of studies conducted in estuarine ecosystems also identified a highly significant positive relationship between BP and CHLA that was comparable to that reported by others (Figure 3), with one notable discrepancy being a lower y-intercept (i.e., lower BP per unit CHLA) and greater slope when compared to those of Murrell (2003), suggesting the influence of warmer water temperatures and more direct coupling between BP and CHLA in Pensacola Bay.
Effects of Temperature and Salinity
Based on the primacy of temperature and salinity in driving the variability of environmental conditions in SWMP data and the important role of these parameters in regulating the composition, abundance, and activity of planktonic microbes (Apple, del Giorgio, and Kemp, 2006; Cottrell and Kirchman, 2003; Lomas et al., 2002; Malone et al., 1988; Smith and Kemp, 1995), we explored their effect on biological processes using BP : CHLA ratios as a response variable. Our analyses of literature data showed that both temperature and salinity are important factors influencing the covariation of BP and algal biomass, with no evidence that dissolved nutrient concentrations influenced this relationship (analyses not shown). Temperature had a highly significant positive relationship with BP : CHLA ratios (Figure 5A), resulting in a log-linear increase in BP per unit CHLA with increasing temperature. The decreasing exponential relationship (i.e., β <1) suggests a decrease in BP relative to CHLA at higher temperatures. Asymptotic temperature responses of BP have recently been reported in estuaries (Apple, del Giorgio, and Kemp, 2006; Murrell, 2003) and lakes (Felip, Pace, and Cole, 1996), which may drive the relationship between temperature and BP : CHLA values. The positive temperature response of BP : CHLA is also in general agreement with the strong temperature dependence of heterotrophic relative to autotrophic microbial processes in a range of estuarine ecosystems (Hoch and Kirchman, 1993; Pomeroy et al., 2000; Sampou and Kemp, 1994; Shiah and Ducklow, 1997) and the weakened coupling of BP and CHLA (i.e., lower BP per unit CHLA) observed at lower temperatures (Hoch and Kirchman, 1993; Shiah and Ducklow, 1994a).
Salinity was also a significant factor explaining variability in the relationship between BP and CHLA (Figure 5B). The response to salinity was more complex, with a weak (i.e., r2 = 0.24) but significant concave parabolic relationship between salinity and BP : CHLA ratios and lower BP per unit CHLA at intermediate salinities. Based on the first derivative of the second-order polynomial describing this relationship, we estimate this minimum salinity to be approximately 17. The occurrence of lower BP : CHLA ratios at intermediate salinities may be a function of conditions in the middle to lower estuary, where phytoplankton are liberated from light limitation and respond to elevated nutrient concentrations (Harding, 1994), resulting in an increase in CHLA relative to BP. This pattern may also be driven by lower growth efficiencies at intermediate salinities that result from elevated rates of bacterial respiration and reduced BP and are related to shifts in bacterioplankton phylogeny and/or metabolism associated with changing salinity (Bouvier and del Giorgio, 2002; Crump, Ambrust, and Baross, 1999; del Giorgio and Bouvier, 2002; Yokokawa et al., 2004). In Chesapeake Bay, the apparent negative effect of salinity on the coherence of BP and CHLA may be attributed to the well-documented shift from net heterotrophy in the upper bay to higher phytoplankton biomass and net autotrophy in the lower bay (Kemp et al., 1997). Shifts of this nature in the relationship between BP and CHLA appear to represent a recurring property along salinity gradients across a wide range of estuarine ecosystems, although the mechanisms are not well understood and may differ among estuaries.
System-Specific Patterns in the Covariation of BP and CHLA
Four estuaries emerged where the relationship between BP and CHLA was relatively strong (Figure 4): Indus River Delta (Bano et al., 1997), Menai Strait (Blight et al., 1995), Urdaibai estuary (Revilla et al., 2000), and North Inlet–Winyah Bay NERRS (present study). These systems were generally more saline (i.e., 22–37) and represent a broad geographic range and estuary types. In general, the covariation of BP and algal biomass was weaker in mesohaline systems, where alternate sources of allochthonous organic matter may undermine the reliance of bacterial metabolism on algal dissolved organic matter, and was stronger in the more saline systems, which have higher ambient water temperatures. Stronger covariation in warmer systems may reflect the important role of temperature in modulating physiological processes and may be a characteristic of the relationship between BP and CHLA in warmer, subtropical regions (Murrell, 2003). Comparison of the four systems in Figure 4 provides compelling evidence supporting the effect of temperature on the relationship between BP and CHLA. We found a strong and highly significant correlation (i.e., r = 0.83, p <0.0001) between study latitude and y-intercept, with the highest BP per unit CHLA in the subtropical Indus River Delta (∼24°10′N) and lowest in Menai Strait (∼53°10′N). Despite the apparent temperature-dependent rank order among these four systems, y-intercepts from these regressions (including that of Indus River) remained significantly lower than those reported by Murrell (2003), suggesting that factors other than temperature may be important in regulating the coherence of BP and CHLA in these systems.
Despite lower estimates of BP per unit CHLA, data from Menai Strait exhibited the strongest covariation of BP and CHLA (r2 = 0.54). This marine-dominated system in North Wales experiences frequent phytoplankton blooms and limited inputs of allochthonous organic matter (Blight et al., 1995). Murrell (2003) also observed strong coherence of BP and CHLA in Pensacola Bay, attributing this to the critical role of phytoplankton-derived substrates in supporting bacterioplankton metabolism. Indeed, the slope of the relationship between log-transformed data from this system indicates a steep increase in BP relative to CHLA at relatively low concentrations and thereby suggests a tighter coupling of BP to CHLA in this system. Similarly, we speculate that the coherence of BP and CHLA in Menai Strait represents a more direct trophic coupling of BP to algal dissolved organic matter, with the caveat that comparatively low estimates of BP per unit CHLA were observed in Menai Strait resulting from the modulating effect of lower water temperatures on the metabolic response of bacterioplankton to substrate supply (Apple, del Giorgio, and Kemp, 2006; Shiah and Ducklow, 1994a). Such direct coupling did not appear to drive the covariation of BP and CHLA in other systems, where the two parameters exhibited strong coherence. Urdaibai estuary is a turbid, nutrient-enriched environment in northern Spain where bacterioplankton carbon demand is subsidized by anthropogenic inputs of organic wastes (Revilla et al., 2000). Covariation of BP and CHLA in this system most likely results from the absence of nutrient limitation for algal and bacterial growth, combined with subsidy of the latter by inputs of allochthonous organic matter. Similar conditions may exist in the eutrophic Indus River Delta in Pakistan, where over 80% of bacterioplankton carbon demand is derived from allochthonous sources (Bano et al., 1997).
Comparison of Monie Bay and North Inlet
North Inlet and Monie Bay are temperate, tidally influenced coastal plain estuaries and thus share many characteristics, including an extensive network of tidal creeks, Spartina-dominated fringing marshes, and seasonal fluctuations in temperature characteristics of temperate climates. In this regard, the striking difference in the strength of the BP vs. CHLA relationship between these two systems was surprising. However, there are important contrasts between North Inlet and Monie Bay that may help explain this difference. Connectivity with coastal ocean waters is more direct in North Inlet, as indicated by its relatively high mean annual salinity (24) compared to that of Monie Bay (7.4), where water is exchanged with the mesohaline portion of Chesapeake Bay. The different water masses adjacent to Monie Bay and North Inlet lead to differences in concentration and composition of dissolved organic matter as well, with annual mean dissolved organic carbon content of floodwaters in Monie Bay being twice that of North Inlet (i.e., 756 vs. 366 μM, respectively). Furthermore, Monie Bay drains a much more extensive and agriculturally developed upland, with dissolved organic matter absorption coefficients—an index of terrestrial organic matter (McKnight et al., 2001)—being an order of magnitude higher in Monie Bay relative to North Inlet (i.e., 16.1 vs. 1.9 m−1). Monie Bay is thus a more nutrient- and organic-rich system than North Inlet, with more diverse sources of organic matter input. As a result, and despite higher annual mean CHLA in Monie Bay (10.4 μg L−1) relative to North Inlet (7.9 μg L−1), bacterioplankton in this resource-rich system are not metabolically linked to CHLA as tightly as in systems where autochthonous dissolved organic matter comprises a greater fraction of bacterioplankton carbon demand.
North Inlet also experiences relatively high mean tidal amplitude (1.5 m) and short residence time (∼15 h), with as much as 40% of the total water volume leaving the estuary with each ebb tide (Gardner, Kjerfve, and Petrecca, 2006). In contrast, Monie Bay has half the mean tidal amplitude (0.7 m) of North Inlet and much longer water residence times (Apple, del Giorgio, and Newell, 2004). Longer residence times in Monie Bay ensure that the inputs of dissolved organic matter are also retained in the estuary longer. Organic matter bioavailability at the ecosystem scale is a function of its inherent decomposition rate relative its residence time in that ecosystem. Increased residence time thus extends the temporal scales beyond which BP and phytoplankton production can be linked, which may not be reflected in static observations based on CHLA. Thus, not only are the quantity and diversity of organic matter inputs greater in Monie Bay than in North Inlet but the timescale of its availability to bacterial consumers, for both autochthonous and allochthonous inputs, is also longer. Collectively, these differences in substrate availability and composition conspire to undermine the coherence of BP and CHLA in Monie Bay.
Temperature is another important difference between North Inlet and Monie Bay; as discussed earlier, it also influences the BP vs. CHLA relationship in these systems. A number of studies across a range of aquatic systems have reported a shift in the regulation of bacterioplankton metabolism from temperature dependence to resource dependence with increasing temperature, with the influence of resource supply being more important than the constraints of temperature above approximately 20°C (Apple, del Giorgio, and Kemp, 2006; Felip, Pace, and Cole, 1996; Hall and Cotner, 2007; López-Urrutia and Morán, 2007). This leads to the influence of organic matter quality and degradability becoming more important at warmer temperatures. In this regard, the absolute difference in mean annual temperature between Monie Bay and North Inlet may not be as relevant as the difference in the amount of time bacteria are exposed to temperatures above this threshold. Water temperatures in North Inlet were above 20°C for 49% of the year, compared to only 27% of the year in Monie Bay, resulting in a discrepancy of 84 days. Thus, bacterioplankton occurred at water temperatures where they were able to respond quickly to algal dissolved organic matter. Water temperatures were well above 20°C during the time the present data were collected for North Inlet, although continuing measurements (E.M. Smith, unpublished data) indicate the significant relationship holds across seasons as well. However, despite 2 years of biweekly sampling of Monie Bay during summer months, strong coherence of BP and CHLA was never identified, suggesting that removal of the metabolic constraints imposed by lower temperatures is not enough to induce coupling and further implicates the influence of allochthonous organic matter sources on the covariation of BP and CHLA.
Our evaluation of systems in which BP and CHLA are more coherent (Figure 4) led to the conclusion that the BP vs. CHLA relationship is stronger in warmer, more saline waters. The comparison of Monie Bay and North Inlet corroborates this pattern, as mean annual water temperature (i.e., 20°C vs. 17°C) and salinity are higher in North Inlet than in Monie Bay. However, this may not necessarily result from the direct effect of salinity and temperature alone on the coherence of BP and CHLA but, rather, from the changing importance of resource supply vs. temperature in regulating bacterioplankton carbon metabolism at different temperature regimes and differences in sources of organic matter substrates between these systems.
Covariation of BP and CHLA: Additional Influencing Factors
It is generally assumed that strong covariation of BP and CHLA results from direct subsidy of bacterioplankton carbon demand by production of algal dissolved organic matter (e.g., Baines and Pace, 1991), with the strength of the BP vs. CHLA relationship reflecting the extent of this coupling. However, covariation of BP and CHLA may also occur if both bacterioplankton metabolism and phytoplankton biomass are simultaneously yet independently regulated by an external factor, such as inorganic nutrient availability (e.g., Coveney and Wetzel, 1995; Smith and Kemp, 2003; Vrede et al., 1999). Under such circumstances, nutrients subsidize algal production while bacterioplankton combine nutrients with allochthonous (i.e., nonalgal) sources of organic matter to fuel growth, evidence of which was observed in the studies discussed earlier (i.e., Bano et al., 1997; Revilla et al., 2000). The independent but proportional effect of nutrients on BP and CHLA may explain why we failed to observe a relationship between dissolved nutrient concentrations and BP : CHLA ratios, because if BP and CHLA increase proportionally along a eutrophication or nutrient gradient, BP : CHLA will remain relatively unchanged. In this regard, we observed a weak but highly significant correlation between PO43− and both BP and CHLA (i.e., r = 0.27 and 0.22, respectively; n = 450), yet no relationship between PO43−and BP : CHLA ratios. PCA revealed grouping of CHLA with NH4+, PO43−, and turbidity, with NH4+ and PO43+ likely driving increases in CHLA concentrations and turbidity increasing as a result. The absence of a correlation between BP and dissolved nutrients in PCA of literature data attests to the overwhelming effect of temperature in driving heterotrophic relative to autotrophic processes.
The relationship observed between BP and CHLA was significant yet surrounded by considerable scatter and thus relatively weak (r2 = 0.24). Although salinity and temperature explain some of this variability, there are additional factors that were not investigated as part of our study or were not included in the composite dataset that might further explain this variability. The covariation of BP to algal biomass can be influenced by temporal lags between primary production and availability of algal substrates for consumption by bacterioplankton (Billen, 1990), reliance of BP on nonalgal substrates (Baines and Pace, 1991; Bano, Moran, and Hodson, 1997), differences in extracellular organic carbon release per unit primary production (Baines and Pace, 1991), and direct (and disproportional relative to phytoplankton) effect of nutrients on bacterioplankton production (Vrede et al., 1999). Coherence between BP and CHLA may also be undermined by variability in bacterial growth efficiency (del Giorgio and Cole, 1998), high turbidity (Pomeroy et al., 2000), allochthonous inputs of organic matter (Hoch and Kirchman, 1993; Revilla et al., 2000), and disproportionate effect of temperature on bacteria relative to phytoplankton (Shiah and Ducklow, 1994a). The extent to which measurements of CHLA accurately represent in situ algal biomass and production may also be compromised by variability in the carbon-to-chlorophyll ratio of phytoplankton (Cloern, 1995), contribution of accessory pigments to primary production (Marra, Trees, and O'Reilly, 2007), and variability in the strength of autotrophic–heterotrophic coupling that appears to be a function of the average cell size and/or composition of the phytoplankton community (Serret et al., 2001; Smith and Kemp, 2001). Although an investigation of these factors is beyond the scope of this study, it is important to recognize the limitations of using BP : CHLA ratios as an index of the coupling between heterotrophic and autotrophic plankton communities.
CONCLUSIONS
Our analyses identified salinity and temperature as important factors shaping the variability in both SWMP data and BP : CHLA ratios, yet the mechanisms underlying their role in shaping water quality parameters vs. the relationship between BP and CHLA are distinct. For example, the negative covariation of salinity and NOx suggests that it is not salinity alone but also nitrogen loading associated with freshwater inputs that drive variability in SWMP data. In contrast, the influence of salinity on BP : CHLA ratios most likely reflects spatial patterns in the balance of heterotrophic and autotrophic processes along the estuarine gradient. The chemical and biological dependence of water column DO on temperature explains the primacy of this parameter as a driver of variability in SWMP data, whereas it is the disproportionate effect of temperature on heterotrophic relative to autotrophic processes that drives the temperature dependence of BP : CHLA. Inputs of allochthonous organic matter and dissolved nutrients are closely linked to salinity and temperature and subsequent changes in water quality, as these serve as tracers for inputs and drivers of metabolism, respectively. In this regard, the ability to use SWMP to track and predict aspects of water quality would be greatly enhanced by including measures of dissolved and particulate carbon, as well as total nutrients. Much greater ecological understanding would, of course, be gained through a systematic inclusion of ecologically relevant process measurements in SWMP. Given the netheterotrophic status of most NERRS sites, integrative measures of bacterial community activity would be one logical place to begin.
Collectively, the NERRS represents a diverse array of estuaries that differ significantly with respect to the magnitude and range of in situ environmental conditions. It is thus difficult to categorize these systems in a manner that allows straightforward yet accurate comparisons among other reserves and other non-NERRS estuaries. Analysis of SWMP data using PCA identifies the primary sources of variability in each reserve and as such creates an integrative comparison of environmental conditions and valuable classification framework. Comparison of reserves using PCA-related analyses provides insight into the extent to which conclusions drawn from research conducted in one reserve might be applicable to another and into predicting the response of reserves to a particular stressor (Kurtz et al., 2006). This framework for comparison also helps isolate outlier reserves or those exhibiting unique physicochemical properties, as with the relationship between temperature and DO in Kachemak Bay and Tijuana River. Analyses of this nature can provide a mechanism for moving from basic monitoring of the reserves into hypotheses-driven studies regarding the factors that regulate physical, chemical, and biological processes in these and other estuarine systems.
Acknowledgments
This research was funded in part by grants from the NERRS Graduate Research Fellowship Program, Cooperative Institute for Coastal and Estuarine Environmental Technology, and Baruch Marine Research Institute Visiting Scientist Award. We thank D. Miller, M. Kemp, and R. Newell for revisions during the development of this manuscript and the panel of anonymous reviewers for their time, effort, and valuable contributions. We also acknowledge the contribution of additional data from D. Kirchman, P. del Giorgio, T. Bouvier, E. Koepfler, and D. Vaqué (Hivern project; Grant No. MAR98-0932 to M. Estrada).
LITERATURE CITED
Table 1.
Reserves, biogeographic regions, and stations included in analysis of SWMP composite dataset with years for which data were available and total observations
Table 2.
Summary of studies from which data were collected as part of our review of published literature
Table 3.
Mean values, standard error, sample size, and range of parameters included in the SWMP composite dataset and review of published studies conducted in coastal and estuarine systems
Table 4A. Temperature, salinity, pH, and depth at each reserve (mean ± SE [n]); values are derived from analysis of the SWMP composite dataset
Table 4B. Dissolved oxygen, chlorophyll-a concentrations, and turbidity at each reserve (mean ± SE [n]). Values are derived from analysis of the SWMP composite dataset
Table 4C. Dissolved nutrient concentrations at each reserve (mean ± SE [n]). Values are derived from analysis of the SWMP composite dataset
Table 5.
Results from PCA of composite SWMP (n = 1973) and literature review (n = 207) data, including eigenvalues, percent variability, and rotated eigenvectors