Estimating daily primary production and nighttime respiration in estuaries by an in situ carbon method

A	  Dawn	  Dusk	  Dawn	  Carbon	  method	  was	  developed	  to	  estimate	  daily primary	  productivity	  and	  respiration	  in	  Narragansett	  Bay	  at	  9	  Narragansett	  Bay Fixed	  Site	  Monitoring	  Network	  stations.	  The	  method	  utilizes	  YSI	  temperature, salinity,	  and	  pH	  measurements	  and	  measured	  alkalinity	  values.	  The	  method	  was compared	  to	  a	  previously	  verified	  Dawn	  Dusk	  Dawn	  Oxygen	  method	  for Narragansett	  Bay	  developed	  by	  Smith	  (2011).	  The	  methods	  compared	  well	  with correlations	  coefficients	  between	  0.69	  –	  0.96	  for	  all	  four	  categories	  (surface production,	  surface	  respiration,	  bottom	  production,	  and	  bottom	  respiration)	  and both	  summers.	  In	  all	  categories,	  2014	  comparisons	  were	  more	  highly	  correlated than	  2013. Metabolic	  rate	  sensitivity	  to	  pH	  and	  alkalinity	  analyses,	  pH	  stability,	  and accuracy,	  were	  conducted	  to	  quantify	  error.	  The	  YSI	  pH	  sensors	  were	  stable	  over the	  dawn-­‐dusk-­‐dawn	  time	  period	  (i.e.	  24	  hours),	  with	  an	  average	  change between	  15	  minute	  readings	  of	  0.01	  units.	  Based	  on	  the	  manufacturers	  stated accuracy	  for	  the	  YSI	  sensor	  of	  0.2	  pH	  units,	  the	  surface	  metabolic	  rate	  estimates could	  be	  under	  or	  over	  estimated	  by	  -­‐11	  to	  23%	  if	  the	  pH	  sensor	  was	  reading	  low or	  high	  by	  a	  systematic	  0.2	  unit	  offset.	  The	  bottom	  estimates	  could	  be	  over estimated	  by	  8	  to	  11%,	  based	  on	  the	  same	  offset.	  The	  metabolic	  rate	  estimates are	  not	  significantly	  affected	  by	  a	  change	  in	  alkalinity,	  with	  ANOVA	  p	  values	  >0.9 for	  all	  categories	  and	  two	  stations.	  Four	  comparisons	  between	  a	  Satlantic	  SeaFET pH	  sensor	  and	  YSI	  pH	  sensors	  were	  conducted	  with	  four	  variable	  results	  due	  to


LIST OF TABLES
This may be artificially lowering the estimates of surface production at the North Prudence site.

INTRODUCTION
Coastal waters around the world have experienced high nutrient loading from urban centers and agricultural land for over a century (Diaz andRosenberg 2008, Smith 2003). The negative impacts associated with high nutrient loading, eutrophication and particularly eutrophication induced hypoxia (Cloern 2001, Diaz 2001, Diaz and Rosenberg 2008, Gooday et al. 2009, Howarth et al. 2011, Kemp et al. 2005, were not recognized as marine water quality issues until 1969 when discussed in "Eutrophication: Causes, Consequences, Correctives" published by the U.S. National Academy of Sciences (Nixon 2009).
Eutrophication can alter the characteristics and function of ecosystems (Boesch and Rabalais 1991, Elmgren 1989, Pearson and Rosenberg 1978, Rosenberg et al. 1990) including a shift from macro benthos primary production to pelagic phytoplankton as the dominant producers and reduced light penetration (Bonsdorff et al. 1997). The decomposition of excess phytoplankton can lead to hypoxia, which is known to be a stressor and lethal to many benthic and pelagic species (Diaz et al. 2004, Gray and Ying 2002, Pihl et al. 1991, Pihl et al. 1992).
Narragansett Bay is no exception to these issues. The Bay has experienced eutrophication and it's negative consequences for decades (Bergondo et al. 2005, Bonsdorff et al. 1997, Carpenter et al. 1998a, Codiga et al. 2009, Corrales and Maclean 1995, D'Avanzo et al. 1996, Deacutis 2008, Melrose et al. 2007, Nixon 1995. In response to concerns that hypoxia would continue to expand throughout Narragansett Bay, as well as a large fish kill in Greenwich Bay, the Rhode Island Department of Environmental Management has taken steps to reduce nitrogen concentrations of point source (i.e. Wastewater Treatment Facility effluent) flowing into the Bay over the last decade. Point source nutrient reductions as well as implementation of more efficient fertilization techniques have become common practice in developed countries as awareness of marine eutrophication has expanded.
Estuaries are dynamic systems that do not always respond to nutrient reductions in the same way (Conley et al. 2009, Duarte et al. 2009, Kemp et al. 2009). The effects of climate change are evident in Narragansett Bay with an increase in water temperature by more than 1°C in the last 60 years (Smith et al. 2010). This increase in temperature has led to a reduction, and in many years, the elimination of the traditional winter--spring phytoplankton bloom in Narragansett Bay possibly due to increased grazing pressure (Oviatt et al. 2002, Smith et al. 2010). Nixon (2009) stressed a further reduction in nutrients to Narragansett Bay could reduce the winter--spring bloom to the point where minimal benthic-pelagic coupling occurred in the spring, reducing the regeneration of nutrients later in the season. This storage of nutrients in the benthos has traditionally been a source to producers later in the summer when the water column nutrients are depleted, and helps support the growth of secondary producers (Nixon et al. 2009). Given these complex relationships between the physical, chemical, and biological components of estuaries, understanding how primary production of Narragansett Bay responds to nitrogen reductions is critical to understanding the whole ecosystem, and to making informed decisions on the extent of nitrogen reductions needed to reduce hypoxia without negatively effecting the growth of many commercial species.
The first response expected after nutrient reduction is a change in the production and respiration in a water body. Here we introduce a new carbon based method to estimate integrated daily metabolic rates. A method has been developed to estimate net daily primary productivity between dawn and dusk, and nighttime respiration rates between dusk and dawn. The technique utilizes in situ temperature, salinity and pH sensors throughout Narragansett Bay as well as measured alkalinity of water samples collected at the sensor sites.
The estimation of primary production and respiration in marine waters has been occurring for almost 100 years, with Gaarder and Gran (1927) performing the first oxygen 'light and dark bottle' incubations in the Oslo Fjord in 1916. This method has been employed by many researchers since then (Bender et al. 1987, Nixon and Oviatt 1972, Oviatt et al. 1981, Oviatt et al. 1986b, Smith 2011 and provided the first insight into how important phytoplankton and other marine primary producers are to an ecosystem. The oxygen light and dark bottle incubations are suitable in highly productive systems and can be used to estimate net and gross productivity as well as respiration.
While the oxygen light and dark bottle method works well for highly productive areas, where the productivity is very low, such as the open ocean, a change in oxygen during the incubation time is sometimes undetectable. As an alternative, a radioactive carbon ( 14 C) method was developed (Steeman Nielsen 1952) for use in the oligotrophic open ocean based on the uptake of 14 C to quantify the phytoplankton primary production during incubation. The 14 C method can be used to detect smaller changes in primary productivity and has been used in hundreds of studies worldwide including estuarine studies (Kelly et al. 1985, Oviatt et al. 1986b, Oviatt 2008, Peterson 1980, Sampou and Oviatt 1991. In the 14 C method the water is filtered to remove large grazers prior to incubation. The resulting 14 C derived productivity measurement is an intermediate estimate between gross and net primary production (Bender et al. 1999, Ostrom et al. 2005 As an alternative to incubations, sampling open water over the dawn dusk dawn time period for consecutive days and measuring the oxygen concentration of the water with the Winkler titration method was employed to estimate primary productivity and respiration (Caffrey 2004, D'Avanzo et al. 1996, Nixon et al. 1976, Odum and Hoskin 1958, Oviatt et al. 1993, Oviatt et al. 1986a, Oviatt et al. 1986b, Sampou and Oviatt 1991, Vaudrey 2007. These studies were among the first to capture the effects of in situ processes on metabolic rates and would later become the framework for future in situ dawn dusk dawn studies (Middleton andReeder 2003, Smith 2011). Oxygen respiration through grazing 8 and vertical mixing of cells throughout the water column were incorporated with the diel oxygen curve method and mesocosm experiments (Oviatt et al. 1986a, Oviatt et al. 1986b, Oviatt et al. 1987 removed the effects of advection on oxygen that are typically difficult to account for in open water in situ sampling regimes. In addition to oxygen metabolic rates, chlorophyll a is often used to estimate net primary production in estuaries. Chlorophyll a fluorescence is measured either by extraction, in situ sensors, or satellite measurements and is often used as a proxy for biomass, although fluorescence of chlorophyll a per cell depends on size, species, and environmental conditions (Falkowski and Kiefer 1985). BZpIo models were designed to estimate primary production using the relationship between chlorophyll a biomass (B), euphotic depth (Zp) and irradiance (Io), and the production estimated from 14 C incubations (Keller 1988).
Once the relationship between the parameters is established, the productivity of the system can be estimated from chlorophyll a measurements and light intensity. BZpIo models are often utilized in studies of eutrophic estuaries since these models perform best when the water column is light limited, where the euphotic depth is less than the overall depth (Brush and Brawley 2009, Brush 2002, Canion et al. 2013, Goebel et al. 2006, Smith 2011. This approach allows for water column integration to achieve a production in gC m --2 day --1 , but the relationships between the parameters are system specific and must be determined for each different ecosystem studied. With rapidly changing technology, in situ sensors have become a frequent choice for measurement of physical, chemical, and biological parameters at high temporal resolution. As part of the plan to reduce the flux of nitrogen into Oxygen method (DDD--O2) that calculated net primary production as the change between dawn and dusk in situ oxygen, and nighttime respiration as the change between dusk and the following dawn in situ oxygen. The DDD--O2 method was verified using concurrent estimates of production from 14 C incubations, oxygen light and dark bottle incubations and a comparison with an in situ integrated 15-minute change in oxygen method (Smith 2011). The 15--minute method summed the change between each 15--minute oxygen reading between dawn and dusk for each day, and for dusk and the following dawn to estimate net production and respiration. A comparison between the DDD--O2 method and a 15--minute integrated change in oxygen method indicated there was no statistical difference between the daily primary production estimated by either method, suggesting that advection played a minor role in the metabolic rate estimates from the DDD--O2 method (Smith 2011).
Oxygen in the mixed layer equilibrates with the atmosphere on daily time scales and thus a wind dependent air--sea gas exchange model for Narragansett Bay was used to correct for oxygen exchange in the DDD--O2 method (Smith 2011  (2007) to standardize to a 10 m wind height. Smith (2011) quantified the effect of diffusion on metabolic rates by comparing the metabolic surface rates estimated with and without an air--sea gas exchange parameter. On average, the metabolic rates estimated with a diffusion parameter were 3.09% and 3.23% higher for Mt. View and Bullocks Reach (located 3 km northwest of Conimicut Point, in Providence River) than the metabolic rates estimated without an air--sea gas correction. There was no significant difference between the two sets of metabolic rates at the 5% level, on average.
Biomass is reported in units of carbon per volume, and thus a measurement of metabolic rates in oxygen must be converted to carbon using a photosynthetic quotient (PQ) and respiratory quotient (RQ). Smith et al. (2012) derived a PQ for Narragansett Bay taking into account species composition, distribution and nutrient composition changes in recent years.
Anaerobic respiration has been observed in Narragansett Bay bottom waters for decades (Doering et al. 1987, Nowicki 1994, Sampou and Oviatt 1991, Seitzinger et al. 1980. More recently, focus on denitrification in estuaries has increased (Ehrlich 2014, Fulweiler et al. 2010, Fulweiler et al. 2007, Fulweiler et al. 2013, Herbert 1999. Denitrification utilizes nitrate (NO3 --) as the terminal electron acceptor when oxygen is not present and is observed in hypoxic and anoxic waters. The DDD--O2 oxygen method does not capture remineralization of organic matter to carbon dioxide through denitrification since the process has no effect on dissolved oxygen concentrations. If anaerobic respiration were a significant contributor to remineralization, the DDD--C method would include this fraction as part of the total respiration rate.
We compared the DDD--O2 method to a Dawn Dusk Dawn Carbon (DDD--C) method. The advantages of the DDD--C method were the elimination of the need for an air--sea gas exchange coefficient to estimate diffusion, and the estimate of a PQ and RQ to convert units from oxygen to carbon. As with the oxygen method, the in situ sensors provide un--paralleled temporal coverage and suitable spatial coverage of Narragansett Bay. The comparison of the DDD--C method and the DDD--O2 method enabled estimations of time and site specific PQ and RQ values for Narragansett Bay.
Several questions addressed in this manuscript include, are metabolic rate estimates from the DDD--C method comparable with the estimates from the DDD--O2 method? Can the DDD--C method be used to detect a change over time in metabolic rates in Narragansett Bay due to nitrogen reduction? Are the YSI pH sensors adequately precise and accurate for use in the DDD--C method? Is bi-weekly alkalinity sampling sufficient to capture variation in alkalinity within Narragansett Bay and what effect does a change in alkalinity have on estimates of metabolic rates? How do the calculated photosynthetic and respiratory quotients compare with the ones estimated by Smith (2012) for Narragansett Bay? Is advection altering our estimates of metabolic rates for Narragansett Bay?

METHODS
This study evaluated an in situ carbon method to estimate daily metabolic rates in Narragansett Bay, Rhode Island. Metabolic rates were determined by measurements of rates of dissolved carbon dioxides changes based on temperature, salinity, pH and alkalinity. A change in the carbon dioxide concentration in water was reflected in the pH of the water. Since carbon dioxide was removed from water during photosynthesis and released during respiration, changes in pH can be used to estimate changes in fixed carbon. The method estimated daily net productivity, or system apparent production, using the difference in carbon dioxide between dusk and dawn and system night respiration as the difference between dawn and the previous dusk, both estimates were converted to grams of carbon, to provide an integrated estimate of daily system metabolic rates per unit volume.
Two sets of measurements were gathered during this study to assess daily rates of carbon change: dawn and dusk data for temperature, salinity, and pH from nine of the Narragansett Bay Fixed Site Monitoring Network (NBFSMN) stations and bi--weekly sampling of alkalinity at the NBFSMN sites ( Figure 1).

Monitoring site procedures
The NBFSMN sites were equipped with two Yellow Spring Incorporated (YSI) 6600 series data loggers, one at 1m below surface and the other 0.5m above the bottom. Each surface data logger recorded temperature, salinity, dissolved oxygen (DO), pH, Chlorophyll a fluorescence and depth every 15 minutes. The bottom data logger recorded temperature, salinity, dissolved oxygen (DO), pH, and depth every 15 minutes. GSO Dock site had only one data logger at 1 -2 m below the surface depending on tide. The manufacturer published accuracy for the pH sensor was ± 0.2 units and the resolution was 0.01, however the sensors were calibrated and post calibrated in the lab and corrected for drift over the two--week deployment. The accuracy of the temperature sensors was ± 0.15°C with a resolution of 0.01°C. The accuracy of the specific conductivity sensors was ± 0.5% of reading + 0.001 mSiemens cm --1 and the resolution was 0.01 mSiemens cm --1 . Fifty mSiemens cm --1 is roughly equal to 32.8 ppt at 25°C (the instrument performs an internal calibration to calculate exact salinity).
Each station was serviced every two weeks by swapping the existing data logger with a newly calibrated data logger. Instruments were calibrated and maintained with quality control measures in the laboratory, field, and post deployment. The pH sensor was calibrated using two pH buffers, pH 7 and pH 10.
Salinity was calibrated using a specific conductivity solution of 50 µSiemens cm --1 .
The sensors were post calibrated using the same solutions to quantify drift over the two--week period.
The data were reviewed, corrected, and documented before distribution in

Alkalinity samples
Water samples were collected every two weeks at the same time as the instrument swap from the 9 stations in the study from July 2013 -September 2013 and June 2014 -September 2014. Water was collected using a Niskin bottle from 1 m below the surface and 0.5 m above the bottom (same depths as data loggers). A 250 ml dark Nalgene bottle was triple rinsed using site water and then filled from the bottom using tubing and allowed to overflow by 1 volume.
Samples were kept in a cooler on ice until returned to laboratory and kept in a refrigerator at 1.5°C for up to 24 hours until analysis.
In the laboratory, alkalinity was determined by potentiometric titration The average error of all of the CRMs titrations was ± 113 µmol kg --1 or approximately 5% of the reading. The acid used for titration was 0.1M HCl that had been standardized using TRIS to determine the exact concentration of the acid. The density is also calculated and used to calculate the total alkalinity. All alkalinity calculations and analysis performed using R 3.0.1 (R 2013) using the 'AT' function in the SeaCarb Package. The package was written by Andrew Dickson and uses the Non--linear least squared method described in Dickson et al. .

Dawn dusk dawn metabolic rate estimation by a carbon method
From the 15 minutes buoy measurements, temperature, salinity, and pH, closest to the sunrise and sunset times are identified each day. The dissolved inorganic carbon (DIC) concentration at dawn and dusk were calculated using in situ temperature, salinity, pH, and measured alkalinity and the dissolved carbonate system disassociation constant equations in Pilson (2013) and the carbon dioxide equation from Oviatt (1986b) ( Table 1). The estimation of changes in fixed carbon were made (gC m --3 day --1 or gC m --3 night --1 ), using the change in DIC converted to moles of carbon. See Appendix A for the method.
The rate that carbon dioxide equilibrates with the atmosphere is much slower than oxygen (Williams and Follows 2011), thus over the time period being sampled (dawn dusk dawn), the flux into or out of the mixed layer is negligible.
The timescales of air--sea equilibrium were estimated using the equations from Williams and Follows (2011). For a non -reactive gas (dissolved oxygen), the timescale to equilibrium is estimated by: where is the time in seconds it takes to reach equilibrium, h is the depth of the mixed layer, and Kg is the gas transfer velocity as a function of wind speed.
For a reactive gas, such as carbon dioxide: where DIC is the concentration of all carbonate species (CO2 aq, HCO3 --, CO3 2--), [CO2*] is the concentration of CO3 2--, and B is the Revelle buffer factor, defined as: The average DIC, ! * , and B parameters were determined using all pH and alkalinity data used in the study and the CO2SYS.m function in Matlab v. R2013a 1) 15--minute temperature, salinity and pH data for each of the 9 stations, from which the dawn and dusk readings were selected. Sites were run individually.
3) The alkalinity data were from samples measured every two weeks, surface and bottom. The measurement was applied to a week prior to sampling and a week after sampling.

Method comparison
The DDD--O2 method for estimating metabolic rates by in situ changes in oxygen (taking into account air sea gas exchange) (Appendix B, C) has been has been compared to 14 C measurements of primary production, oxygen light and dark bottom estimates of net primary production and respiration rates, and a 15-minute integrated change in oxygen method (Smith 2011). The DDD--O2 method (Appendix B) will be used to compare the metabolic rate estimates from the DDD--C method.
The metabolic rates calculated using the DDD--O2 method were converted from gO2 m --3 day --1 to gC m --3 day --1 by a photosynthetic quotient (PQ) and respiratory quotient (RQ). The PQ and RQ were calculated using the average of each site for a given category (surface production, surface respiration, bottom production, bottom respiration) and year. The PQ is defined as the moles of oxygen produced per mole of carbon uptake.
The RQ is the opposite, the moles of oxygen consumed per mole of remineralized The average PQ and RQ for all sites combine, differentiated by year and category, were used to convert the oxygen metabolic rates to carbon metabolic rates. Alkalinity sample collection began late in summer 2013 (July 24 th , 2013).
The first sampling occurred on July 11 th , however the methods changed between this first round and the July 24 th sampling, thus the first set of data were discarded. The

Error estimation -pH measurements
A pH stability analysis was conducted to ensure that the 15--minute pH readings from the buoy sensors were stable. The mean absolute differences between 15--minute readings were calculated for all sites and both summers. 20 The manufacturer stated accuracy of the pH sensor was 0.  There is a duplicate set to allow for a seamless swap of instruments, leading to at least 34 different pH sensors used throughout a summer. The comparison between the two sensors will indicate whether there was an offset between the SeaFET and the YSI pH sensors.

Error estimation-alkalinity measurements
In order to determine whether bi--weekly alkalinity sampling provided sufficient resolution of the alkalinity in Narragansett Bay, an alkalinity frequency test was performed. From January 6  bi--weekly, weekly, and twice weekly. For the alkalinity sensitivity analysis, the percent change between all three (pairwise comparisons) estimates of metabolic rates was calculated. The normality of the data was testing using a Shapiro--Wilk test and both a one--way Analysis of Variance (ANOVA) and a Kruskal--Wallis test were conducted using R with alkalinity sampling frequency as the grouping factor and metabolic rates as the dependent variable.

Hydrodynamics at North Prudence
An Acoustic Doppler Current Profiler (ADCP) was deployed near the North currents from this dataset, the impact of advection was examined by estimating if a water mass at one site could reach another site's sensors, taking into account magnitude and direction, in one tidal cycle. Due to the location of the ADCP, interactions between North Prudence and the three closest sites were considered (Conimicut Point, Poppasquash Point, Mt. View).

Alkalinity relationship with salinity in Narragansett Bay
A preliminary equation for total alkalinity calculated from salinity was investigated for summer 2013 and summer 2014. Future work includes a more detailed examination of the data and the mechanisms that drive change in alkalinity within the Bay.

Euphotic Depth
Light profiles were taken at every station at the same time as the sonde swap and alkalinity sample collection. Profiles were conducted using a Li--Cor light meter, with a hand held (model LI--250A), deck sensor (model LI--190R), and a spherical underwater sensor . Light was taken every meter at sites over 3.0 m deep, and every 0.5 m at sites with depths less than 3.0 m (Greenwich Bay, GSO Dock). Light readings were recorded both on the down cast and on the up cast, the light extinction and euphotic depth, defined as the depth at which 1% of the surface light reaches, was determined using the cast with the most consistent light (fewest passing clouds).

Method comparison
The DDD--C method estimates of metabolic rates were highly correlated to the estimates from the DDD--O2 method ( (Table 3).

Photosynthetic and respiratory quotient
The photosynthetic quotients calculated from the comparison of the two methods were within an acceptable range of values, 1.  (Table 4), but all RQs were less than 1, the value often used in estuarine studies (Caffrey 2004, Collins et al. 2013, Oviatt et al. 1986b.

pH stability, accuracy and sensitivity analyses
The pH sensors took a reading every 15 minutes and although only the dawn and dusk measurements are used for this method, the 15--minute readings were analyzed to determine if the pH sensor had high precision. The pH stability analysis indicated that the YSI pH sensor readings over the dawn dusk dawn time period were stable between 15--minute intervals. The mean absolute value difference between 15--minutes readings of the YSI pH sensors ranged between 0.01 -0.02 pH units for the mean of all stations except Greenwich bay (Table 5).
For both years, Greenwich Bay had higher variability between readings than the other sites with means ranging between 0.03 -0.04 pH units, although the overall metabolic rates at the Greenwich Bay site were on average twice that of the other stations. Most importantly, the change in pH between 15--minute readings was in the direction consistent with productivity during the daytime and respiration during nighttime (pH increases during productivity and decreases during respiration), indicating that any change between dawn and dusk was likely a true change in pH and not an error in the sensor. For each category (surface production, surface respiration, bottom production and bottom respiration) and each of the 5 sites used in the analysis, metabolic rates were estimated with the measured pH, and with systematic offsets of ± 0.2 pH units, to create three separate sets of metabolic rate estimates.
The pH sensitivity analysis conducted on five sites of data showed that the average percent change in calculated metabolic rates for all sites when the surface pH was increased by 0.2 pH units was 23% and 20% for production and respiration, respectively. The average percent change in metabolic rates when the surface pH was decreased by 0.2 pH units for all sites was a decrease in metabolic rates by 9% to 11% for production and respiration, respectively ( Table   6). The bottom estimates of metabolic rates were increased by 10--11% on average for production and 8% for bottom respiration ( Table 6). The timeline graphs of Conimicut Point for all four categories indicated that the higher magnitude metabolic rate estimates (more positive for production and more negative for respiration) were accentuated by the change in pH (Figure 5). The other 4 sites show a similar trend (Appendix D).
A one--way ANOVA for each category/site combination indicated that there was no significant difference in the means of metabolic rate estimates between the three datasets at the 5% level, on average. The only exception was Conimicut Point surface respiration, where the mean of the estimates from the pH -0.2 dataset was significantly different from the estimates from the pH + 0.2 units dataset, however neither of the means from the offset pH datasets were significantly different from the estimates from the measured pH metabolic rates, determined using Tukey's Honest Significant Difference test (Table 7).
Production and respiration rates estimated with the DDD--C method were not constant throughout the Bay (Table 8). Conimicut point was the most variable and productive site out of the five sites used in the pH sensitivity analysis, thus the increase or decrease in pH at this site resulted in a greater change in total fixed carbon rates at this site compared to the others. The mean summer daily surface productivity at Conimicut Point for 2013 is 0.33 gC m --3 day --1 , with a possible range based on ± 0.2 of 0.29 -0.41 gC m --3 day --1 . In contrast, the mean summer daily surface productivity at North Prudence is 0.11 gC m --3 day --1 with a possible range of 0.10 -0.13 gC m --3 day --1 ( Table 9). The YSI pH sensor accuracy was compared with a Satlantic SeaFET pH sensor with an accuracy of 0.02 units. The comparison between the SeaFET pH sensor and the YSI pH sensor during a 4--day tank deployment showed that there was a consistent offset of 0.03 pH units with YSI reading higher, when the pH ranged between 8.14 -8.24 units ( fouling. The post--calibration of the YSI sensor indicated that over the 3--week deployment in June--July, the sensor had drifted 0.05 units.

Alkalinity sensitivity
To test the calculated metabolic rates sensitivity to alkalinity, three different alkalinity datasets were used. From the original dataset collected in January -February 2015, 3 values were chosen that were sampled two weeks apart each to comprise the biweekly alkalinity dataset. Alkalinity samples that were sampled a week apart (5 total) comprised the weekly dataset and the twice weekly alkalinity dataset contained all 9 values sampled during that time period.
The increase in sampling frequency, and increased range of alkalinity values, had almost no effect on the resulting metabolic rates at either Conimicut Point or Quonset Point. For all 4 comparison categories, and 3 sets of metabolic rate estimates, the metabolic rates on average only changed by 2 -8% (Table 10) (Table 11).

Alkalinity and salinity relationship
Using an ordinary least squares regression, there was no relationship between measured alkalinity and salinity for the all combine data. However, when separated by year, there was a significant relationship (p = 0.01, at 5% level) between the two for 2013, but not for 2014 (Figure 9). The preliminary equation for 2013 was: y = 39.73x + 645.7, r 2 = 0.21, r = 0.46 The equation for 2014: y = --11.11x + 2318.52, r 2 = 0.01, r = --0.09 where x was salinity and y was measured alkalinity for both equations.

Spatial trends in metabolic rates
There was a north south gradient in metabolic rates from Conimicut Point The Greenwich Bay site had the largest system apparent production and nighttime respiration, and the highest variability, of all nine study sites for both summers. Conimicut Point was the next most productive, higher than Sally Rock, which is located within Greenwich Bay. Conimicut Point, Greenwich Bay, and Sally Rock regularly have the highest number of hypoxic days of the nine sites (Table 12). For all sites, the metabolic rates were variable, and do not show blooms occurring throughout the summer.
The bottom respiration rates do not followed the same gradient trends, with both Conimicut Point and North Prudence exhibiting the second and third lowest bottom respiration rates for both summers. In contrast, Greenwich Bay has the highest metabolic rates for surface and bottom, both years. The bottom production and respiration at Greenwich Bay both years, is 3 -5 times higher than any other site (

Method comparison
While the two models for oxygen and carbon production and respiration were highly correlated in both summers, it is interesting that in each category, they were more highly correlated in summer 2014 than summer 2013 (Table 2, Figures 2,3). One possibility as to why this occurred is a difference in the water quality between the two summers. Summer 2013 was more hypoxic than average for Narragansett Bay, while 2014 was very below average for number of hypoxic days (Table 12), with the exception of Greenwich Bay and Sally Rock which exceeded or equaled the average number of hypoxic days. When the water is hypoxic, anaerobic respiration may be the primary remineralization process. In this case, the respiration would be reflected in the DDD--C method but not the DDD--O2 method. However, for seven out of the nine sites, the oxygen converted to carbon respiration rates were higher in 2013 than the carbon respiration rates, although in most cases the difference was 0.01 gCm --3 night --1 (Table 8, 13). There is not a large difference in the bottom respiration between the two methods that could be attributable to anaerobic respiration, and this is likely not the cause for higher correlations between the metabolic rate estimates in summer 2014.
One--way ANOVAs indicated that for 2013, the estimates of metabolic rates from the DDD--O2 and DDD--C methods were significantly different at the 5% level from each other for all categories of surface production, surface respiration, bottom production and bottom respiration (Table 3). In 2014, only bottom respiration means were significantly different from each other. Interestingly, the bottom respiration had the highest correlation between the two methods out of any of the categories or summers. These differences are likely to be a result of the PQ and RQ used to convert the oxygen metabolic rates to carbon metabolic rates.
In 2013, the average quotient for each of the 4 categories had a larger standard deviation than the average quotient calculated for each category in 2014 (Table     32 4). The average PQ and average RQ of all sites was used to convert all of the oxygen production and respiration rates to carbon metabolic rates. In 2013, the between site quotient variability was higher than in 2014, thus the average quotient was less representative of all of the sites in 2013 (Table 4), leading to a difference in metabolic rate means between the two methods for 2013, and not 2014.
The 2014 bottom respiration difference in means may also be an artifact of the respiratory quotient used. The RQ used to convert oxygen respiration rates to carbon respiration rates was 0.72 ± 0.24 with one outlier (Poppasquash Point) removed (Table 4). The average RQ without Poppasquash Point removed was 0.58, which led to an even greater difference between the means of the two methods when used to convert from oxygen to carbon. Conimicut Point and North Prudence both had low Bottom RQs for 2014, 0.51 and 0.53, respectively (Table 4). These were not removed from the dataset since they did not fall outside of two standard deviations from the averaged data, however, they are much lower than the other sites, and may have artificially lowered the average bottom RQ for 2014, leading to statistically different means between the two methods.
Although the means in 2013 and bottom respiration in 2014 were statistically different, they were not ecologically different (Table 3).

Photosynthetic and respiratory quotient
The photosynthetic quotient for 2013 surface production, 1.4 ± 0.23, was the same as the PQ derived for Narragansett Bay by Smith et al. (2012). For the 33 2014 surface production, the PQ was estimated to be 1.22 (Table 4), close to the PQ estimated for Narragansett Bay by Oviatt et al. (1986aOviatt et al. ( , 1986b  indicating that production and respiration rates were roughly equal throughout the Bay (Table 3).

pH stability, accuracy and sensitivity analyses
A large concern of this study was whether the manufacturer stated accuracy of the YSI sensor (0.2 units) was acceptable for use in carbon metabolic rate method for Narragansett Bay. The pH stability analysis indicated that the YSI sensor was stable over the dawn dusk dawn period, it was not changing erratically (Table 5) and was therefore acceptable for use in this method. The Greenwich Bay site had the highest average surface and bottom daily estimates of production and respiration rates for both summers (  Figure 5). This is due to the carbonate species buffering system, at a higher baseline pH more metabolism must occur to change the buffered pH by 0.2 units than would have to occur at lower pH values. The percent change varied by site, with surface values at all sites being more affected by varied pH than the bottom metabolic rate estimates.
The sensitivity analysis showed that, at worst, the surface production estimates could be off by --10 -23%, with all other estimates of metabolic rates affected by --11 -11% (Table 6). At higher productivity sites such as Conimicut Point and Greenwich Bay this translates into a potential difference in metabolic rates of up to 0.08 gC m 3 day --1 (Table 9). Despite these relatively high percent changes between the metabolic rates estimated from the measured pH and the offset pH, the one--way ANOVAs computed for the pH sensitivity analysis indicated that for surface production, surface respiration, bottom production, and bottom respiration, for all sites, the metabolic rates estimated with pH ± 0.2 units was not significantly different than the metabolic rates estimated from the measured pH, at the 5% level. Although the means are not significantly different from one another, a difference in surface production based on a possible error in the pH measurements should not be over  (Table 15), well above the offsets observed in the pH sensor comparisons. Since the sensitivity analysis indicated that an offset of ± 0.2 units did not significantly change the metabolism estimates, the offsets observed would have minimal effect on the metabolic rate estimates; the YSI pH sensor is appropriate for use in estimating metabolic rates in Narragansett Bay. If nutrient reductions decrease metabolic rates in Narragansett Bay leading to a more oligotrophic system, the YSI sensors may no longer be suitable for use at the lower bay stations where production is already reduced compared to the upper bay stations since the true variance in pH may be within the error of the sensor.

Alkalinity sensitivity
One assumption of the study was that bi--weekly alkalinity sampling would be sufficient to capture the variability of the alkalinity in Narragansett Bay. An alkalinity sensitivity analysis revealed that the variance in alkalinity throughout the bay has little effect on the metabolic rate estimates. Despite the much higher range captured in the twice--weekly alkalinity sampling, the three sets of metabolic rate estimates only varied by 2 -8 % on average over the 40 day test period (Figure 8). The p values for the ANOVA on both sites were higher than 0.99 for all four categories (Table 11) at the 95% confidence level. This indicates that while total alkalinity is necessary for calculating total carbon dioxide, the total alkalinity value does not significantly change the estimates of metabolic rates; pH is the main driver of change for the metabolic rate estimates. We conclude that measuring alkalinity every two weeks is adequate for in situ estimation of carbon apparent production and night respiration.

Alkalinity and salinity relationship
Given the lack of alkalinity effect on metabolic rate estimates, a preliminary investigation into deriving an equation to calculate total alkalinity from salinity was undertaken for Narragansett Bay. In order to make this a valuable equation, much more work will have to be put towards analyzing the alkalinity trends and fluxes, taking into account the influence from different species of nutrients, freshwater mixing, and average river flows between years.
Our preliminary investigation shows that there is a significant relationship (p = 0.01, at 5% level) between alkalinity and salinity in 2013 when river flow was high ( Figure 9a). However, in 2014, a particularly dry year, there was no relationship between alkalinity and salinity in Narragansett Bay (Figure 9b).

Spatial trends in metabolic rates
Metabolic rates varied at the different stations around the Bay (Table 8, were evident in the oxygen metabolic rates (Figure 11). The surface metabolic rates were highest at Greenwich Bay, Conimicut Point, and Sally Rock, the three sites with the highest average number of hypoxic days (Table 12), indicating that metabolic rates are a driver of hypoxia. The system apparent production and night respiration were higher at many sites in 2014 than in 2013 (Table 8), possibly due to the increased light penetration at most sites (Table 14). Although metabolic rates were higher, the hypoxia was drastically lower in summer 2014, than summer 2013 (Table 12). Codiga et al.
(2009) showed a strong relationship between June rainfall and hypoxia within Narragansett Bay. June 2013 experienced >25 cm of rainfall to the Narragansett Bay watershed, compared to <6 cm in June 2014, indicating that meteorological variability is a stronger driver on the severity of hypoxia within Narragansett Bay than production rates.

Hydrodynamics at North Prudence
To investigate the anomalously low metabolic rates at North Prudence, acoustic Doppler current profiler (ADCP) data from Rogers (2008) was examined (Figure 12

CONCLUSION
The DDD--C method was reliable for estimating metabolic rates in Narragansett Bay. Each method of metabolism estimation has advantages and disadvantages, as said best by Oviatt et al. (1986b), "Every measure of primary production has its own complexities". The in situ carbon method approach presented here has high temporal resolution, captures the effects of biological vertical mixing and grazing, accounts for anaerobic respiration and eliminates a need for an air--sea gas exchange coefficient. However, it is not without limitations. As with the DDD--O2 method, the DDD--C method does not account for advection, an advantage of incubation studies, and it does suffer from a possible error introduced by the pH sensors.
This method may be particularly useful in estuaries where metabolic rates are high, estuaries where air--sea gas exchange is elevated, and possibly estuaries with high anaerobic respiration rates. Nonetheless, the development of this method provides another option for estimating metabolic rates, and when used in conjunction with the DDD--O2 method, can be used to estimate photosynthetic and respiratory quotients for a system.    Table 3. The average metabolic rate for each category for both summers is listed (gCm --3 day --1 or night --1 ). The p values are from t tests comparing the DDD Oxygen method metabolic rates to the DDD Carbon method metabolic rates. All categories for 2013 are significantly different from each other, where as in 2014, only bottom respiration is significantly different. Despite being significantly different, these differences would not lead to a difference in classification of the water body as eutrophic or oligotrophic.  Table 4. The photosynthetic and respiratory quotients for all stations and summers. The fields with asterisks were not included in the averages because they were over 2 standard deviations of the mean of the rest of the data points in that category. The averages for each year and category were used to convert gO2 m --3 day --1 to gCm --3 day --1 .  Table 6. The percent change between the original metabolic rate estimates, and the estimates from increasing or decreasing the pH by 0.2 units. On average, if the pH is increased by 0.2 units in the surface waters, the metabolic rate estimate will increase by 23% and 21% for production and respiration, respectively. If the surface pH is decreased by 0.2 units, the metabolic rate estimates will decrease by 9--11% for production and respiration. The bottom metabolic rate estimates increase by 8--11% regardless of whether the pH is increased or decreased.  Table 7. Carbon metabolic rates were estimated with the measured pH, increased pH by 0.2 units and decreased pH by 0.2 units. The results for each category were compared using a one--way ANOVA and indicated that only Conimicut Point surface respiration was significantly different. A Tukey's HSD test indicated that there was only significant difference between the metabolic rates estimated from the increased and decreased pH sets, there was no difference between the metabolic rates from the measured pH and either of the increased or decreased pH metabolic rates datasets.  Table 8. Average daily carbon metabolic rates for each summer and site (gC m --3 d --1 or n --1 ). In 2014, bottom production and respiration were both statistically higher than in 2013, surface values were not significant.  Table 9. Mean and possible metabolic rates (gCm --3 day --1 or gCm --3 night --1 ) if the YSI pH sensor is off by 0.2 units in either direction. Surface values for Conimicut Point, the most productive site of these five, could be off by 0.08 gCm --3 day --1 if the YSI pH sensor is reading 0.2 pH units high of the true pH. Bottom values are not highly affected.  Table 10. Average percent change in metabolic rate estimates when the sampling frequency for alkalinity was changed from bi--weekly (BW) to weekly (W) and twice weekly (TW). A change in alkalinity sampling frequency only resulted in changes in metabolic rates of 1--8% on average.   Table 13. Average daily oxygen method metabolic rates for each summer and site (gC m --3 d --1 or n --1 ). Data were converted to carbon using the average PQ or RQ for each category and summer. North Prudence was anomalously low compared to surround sites, as seen in the carbon metabolic rates as well. The bottom production rates, particularly in 2013 when euphotic depth was shallower, may be reflecting oxygen mixed into the bottom layer, in addition to algal production of oxygen in the bottom, increasing overall rates of bottom production.  Figure 2. The two methods compared well in 2013. The black line is the 1 to 1 line, and the red and gray lines are the regression and 95% confidence intervals respectively. Correlations ranged from 0.69 for bottom production to 0.81 for surface respiration. Metabolic rate estimates from the oxygen method have been concerted to carbon using the average PQ and RQ from the comparison of the two methods for each year and category.  Figure 3. The two methods for estimating metabolic rates were compared for summer 2014. The black line is the 1 to 1 line, and the red and gray lines are the regression and 95% confidence intervals respectively. Metabolic rate estimates from the oxygen method have been concerted to carbon using the average PQ and RQ from the comparison of the two methods for each year and category. The correlations for 2014 were much higher than in 2013. In summer 2014, the bottom respiration had the highest correlation between the two Dawn Dusk Dawn methods. The carbon method captures anaerobic respiration through the change in carbon dioxide, where as the oxygen method could not detect this respiration. This difference may be responsible for the large discrepancy in correlations for bottom respiration between the two years.  . Metabolic rate estimates from the oxygen method have been converted to carbon using the average PQ and RQ from the comparison of the two methods for each year and category. Summer 2013 had more variability between the two methods, with Greenwich Bay exhibiting the most variability. The Greenwich Bay site has the highest production and respiration of the nine study sites, followed by Conimicut Point. The black points represent the Greenwich Bay site in these plots and the blue points are Conimicut Point.   Figure 7. The second deployment at Conimicut Point started with very high agreement between the two sensors, however on June 21 st , a bloom started and fouling occurred on the SeaFET sensor, leading to drift over the rest of the deployment. The trends are similar between the two sensors, but there is an offset present for the majority of the deployment. Post calibration data for the YSI sensor indicated that the sensor drifted 0.05 units over the three week deployment.  Figure 9. A preliminary investigation into a relationship between salinity and alkalinity in Narragansett Bay showed that there is a moderate relationship between the two for 2013, a very wet year with high river flow, but no relationship for 2014, which was drier than average, indicating that this relationship may be heavily flow dependent. Figure 10. The surface carbon metabolic rate spatial patterns changed between the two summers. A gradient down the West Passage from Conimicut Point (CP) to GSO Dock is persistent both years with North Prudence exhibiting suppressed metabolic rates compared to the surrounding sites both years. Overall, metabolic rates were higher in 2014 for all categories, but the difference between surface means were not significant at the 95% confidence level. Note: scales are different between years. Figure -5)./(temp_RiseS_aH + temp_RiseS_Kb)).*(temp_RiseS_aH.^2./(temp_RiseS_K1.*(temp_RiseS_aH ... + 2.*temp_RiseS_K2)) + (temp_RiseS_aH + temp_RiseS_K2)./(temp_RiseS_aH + 2.*temp_RiseS_K2)).*(12./10.^3)); %The conversion at the end converts TCO2 from umol/L to g m-2 --> % 1 mole/10^6 umol, 1 mol C/1 mol CO2, 12g C/ 1 mol C, 10^3L/m^3 eval (['RiseS_TCO2',num2str(cruise_idx) (-5)./(temp_SetS_aH + temp_SetS_Kb)).*(temp_SetS_aH.^2./(temp_SetS_K1.*(temp_SetS_aH ... + 2.*temp_SetS_K2)) + (temp_SetS_aH + temp_SetS_K2)./(temp_SetS_aH + 2.*temp_SetS_K2)).*(12./10.^3)); %The conversion at the end converts TCO2 from umol/L to g m-2 --> % 1 mole/10^6 umol, 1 mol C/1 mol CO2, 12g C/ 1 mol C, 10^3L/m^3 eval (['SetS_TCO2',num2str(cruise_idx) -5)./(temp_RiseB_aH + temp_RiseB_Kb)).*(temp_RiseB_aH.^2./(temp_RiseB_K1.*(temp_RiseB_aH ... + 2.*temp_RiseB_K2)) + (temp_RiseB_aH + temp_RiseB_K2)./(temp_RiseB_aH + 2.*temp_RiseB_K2)).*(12./10.^3)); %The conversion at the end converts TCO2 from umol/L to g m-2 --> % 1 mole/10^6 umol, 1 mol C/1 mol CO2, 12g C/ 1 mol C, 10^3L/m^3 eval (['RiseB_TCO2',num2str(cruise_idx) D.2 --Impact of increasing or decreasing pH on metabolic rates for North Prudence, Mt. View, Quonset Point, and GSO Dock, 2013 Figure D.2--1 Surface Production at North Prudence is visually the most impacted by a change in the pH, whereas the bottom production and respiration are not effected greatly due to a change in pH. The percent change due to pH is constant across all sites, but the metabolic rate estimates vary.  Figure D.2--2 Mt. View has higher metabolic rates than North Prudence, and the impact on the metabolic rates was greater for Mt. View, though the percent change is the same. Surface estimates are more impacted than bottom estimates from an error in pH.  Figure D.4 --1 Both summers, Greenwich Bay has the highest production and respiration in the bottom waters. This site is the shallowest site, allowing light to reach the bottom at times. In 2014, water clarity was increased compared to 2013, and metabolic rates were increased for most stations. Note, the scales on the y--axis are different between years.