Measuring and Modeling Diel Oxygen Dynamics in a Shallow Hypereutrophic Estuary: Implications of Low Oxygen Exposure on Aquatic Life

Abstract

Hypoxia, or low dissolved oxygen (DO) is a common outcome of excess nitrogen and phosphorus delivered to coastal waterbodies. Shallow and highly productive estuaries are particularly susceptible to diel-cycling hypoxia, which can exhibit DO excursions between anoxia (DO ≤ 1 mg L⁻¹) and supersaturated concentrations within a day. Shallow estuaries exhibiting diel-cycling hypoxia are understudied relative to larger and deeper estuaries, with very few mechanistic models that can predict diel oxygen dynamics. We utilized continuous monitoring data and the Coastal Generalized Ecosystem Model (CGEM) coupled with an Environmental Fluid Dynamics Code (EFDC) hydrodynamic model to simulate diel DO dynamics in Weeks Bay, AL. Low oxygen conditions ranging from anoxia to DO ≤ 4 mg L⁻¹ were consistently observed and simulated in the lower water column for periods of minutes to >11 hours. High frequency observations and model simulations also identified significant vertical gradients in near bottom DO that varied as much as 0.8 to 3.1 mg L⁻¹ within 0.4 m from the bottom. This spatiotemporal variability presents unique challenges to adequately quantify DO dynamics and the potential exposure of aquatic life to low oxygen conditions. Our results demonstrate the need for detailed measurements to adequately quantify the complex DO dynamics in shallow estuaries. We also demonstrate that simulation models can be successfully applied to evaluate diel oxygen dynamics in complex estuarine environments when calibrated with fine time scale data and effective parameterization of water column and benthic metabolic processes.

1. Introduction

Human disturbance of coastal watersheds has led to increased nutrient loading and eutrophication in coastal ecosystems across the globe. Excess nitrogen and phosphorus delivered to coastal waterbodies often stimulates increased algal production and respiration, which can lead to hypoxia, or low dissolved oxygen (DO), especially in the lower water column where ventilation may be inhibited (Fennel and Testa, 2019). The percentage of U.S. estuaries exhibiting low DO or hypoxia has increased by ~27% between the 1980s and 2000s (Rabalais et al., 2009), while hypoxia has been identified globally in more than 500 coastal ecosystems (Breitburg et al., 2018). Shallow and highly productive estuaries and embayments are particularly susceptible to diel-cycling hypoxia, in which DO cycles between hypoxia before dawn and normoxia or higher DO levels in late afternoon. In shallow estuaries, rates of benthic production can be high and benthic production and respiration are both relatively more important in shallow water columns, providing a small volume to dilute DO exchanges with sediments (Kemp et al., 1992). Extreme DO excursions in shallow estuaries have been observed at concentrations between near-anoxia (DO ≤ 0.2 mg L⁻¹) to well above saturated levels within a single day (Tyler et al., 2009), while a somewhat narrower range is more common (Baumann and Smith, 2018).

Diel-cycling hypoxia and oxygen dynamics in shallow estuarine systems are influenced by a multitude of biotic and abiotic factors. Shallow estuaries exhibit strong benthic-pelagic coupling (Nixon, 1979) which has been characterized by high nutrient cycling (DiDonato, 2006), high oxygen demand (Wilson and DePaul, 2017), and modulation of metabolism by an active microphytobenthos (MacIntyre et al., 1996). Even in well-mixed estuaries high oxygen consumption rates can produce hypoxia despite a lack of density stratification (Verity, 2006). Shallow systems generally categorized as “well-mixed” may also experience episodic stratification during periods of high river discharge and reduced winds (Kang and Xia, 2022). Oxygen dynamics are often influenced by abiotic factors such as wind forcing, vertical and horizontal mixing, variation in freshwater inflow, cloud cover and temperature (Duvall et al., 2022). Tyler et al. (2009 found that diel-cycling hypoxia in a Delaware tidal creek was more likely to occur during periods of calm winds and reduced solar insolation, leading to reduced oxygen production via primary production.

Varying occurrence and duration of diel-cycling hypoxia presents unique challenges to aquatic life when compared to persistent hypoxia. Fish exposed to low oxygen concentrations exhibit physiological and behavioral responses (Zhu et al., 2013) that differ depending on whether hypoxia is prolonged versus transient or cyclical (Borowiec et al., 2015; Regan and Richards, 2017). Benthic communities exhibit changes in diversity, composition, and function in associations with low oxygen exposure (Diaz and Rosenberg, 2008; Scheffer and Carpenter, 2003). While 2 mg L⁻¹ has often been suggested as a threshold for hypoxia (Vaquer-Sunyer and Duarte, 2008) lethal and sub-lethal effects occur across a range of DO levels, frequencies, and durations of exposure (Stoklosa et al., 2018; USEPA, 2000).

Quantifying low oxygen exposure in shallow estuaries depends on adequate measurement of oxygen throughout the diel cycle, a challenging proposition given the rapid changes in DO conditions that can occur. Recent studies utilizing high-frequency sensors illustrate the benefits of using continuous measurement to characterize water quality condition and ecosystem function in dynamic shallow estuaries (Duvall et al., 2022; Shen et al., 2008; Tian et al., 2022). There is growing recognition that high-frequency DO measurements are useful in many coastal ecosystems. This is especially true for shallow estuaries exhibiting diel-cycling hypoxia where hypoxia is less well-characterized than in larger, typically deeper estuaries (e.g., Chesapeake Bay) and inland seas (e.g., Baltic Sea), where hypoxia has been characterized most often as seasonally-persistent. Recent improvements in sensor technology cost and maintenance requirements make including continuous DO monitoring in studies of shallow estuaries increasingly practical, while small and shallow estuaries are significant in aggregate given that these systems combine to account for a significant fraction of the many coastlines, even within larger estuaries such as Chesapeake Bay (e.g., Tian et al., 2022)

Simulation modeling has been used effectively to study oxygen dynamics in large or deeper estuaries, but a lack of high-frequency observations necessary for calibration and parameterization can impede their successful application to shallow systems. Shallow, well-mixed estuaries can be difficult to model because important physical and biological dynamics must be evaluated at diel timescales rather than the seasonal to interannual scales that are usually the focus of modeling in deeper coastal systems (e.g., Chen et al., 2015). Some models have struggled to adequately simulate the dynamic range of DO conditions in shallow ecosystems (e.g., Brady, 2014; Testa et al., 2021; Tian et al., 2022; Wan et al., 2012) and relatively few models have focused ondiel oxygen dynamics. Nonetheless, mechanistic models are useful for synthesizing our understanding of ecological processes that control the relationship between nutrients and oxygen at the diel time scale. Mechanistic models can also inform policy development related to nutrient management in shallow ecosystems, where relationships between nutrient loading and diel-cycling hypoxia are not well quantified. Further, because these models explicitly address interactions between hydrodynamic and biogeochemical processes, they can be used to predict responses to a changing environment, including responses to nutrient loads in the context of global climate change (Testa et al., 2021).

In this study we collected continuous monitoring data and measured DO fluxes at the sediment water interface and in the plankton to further evaluate DO dynamics in Weeks Bay, AL, a shallow hypereutrophic estuary with a 36-year continuous DO record. We also coupled the Environmental Fluid Dynamics Code (EFDC) hydrodynamic model with the Coastal Generalized Ecosystem Model (CGEM) to simulate hourly water quality (Hamrick, 1992; Jarvis et al., 2020; Lehrter et al., 2017). We selected CGEM, which has been developed relatively recently (Lehrter et al., 2017), for this application because of its numerous formulation options and application of internal cell quota nutrient-dependent phytoplankton growth (Droop, 1973). The objectives of this paper were to (1) validate CGEM model performance for simulating production, respiration, and oxygen dynamics at diel timescales, (2) evaluate processes responsible for observed diel variability in vertical oxygen concentration gradients, and (3) quantify the frequency and duration of low oxygen exposure.

2. Methods

2.1. Site Description

Weeks Bay is a shallow (mean depth ~1.4 m) microtidal (tide range ~0.4 m) sub-estuary of Mobile Bay, a large subtropical estuary on the northern Gulf of Mexico coast of the United States (Figure 1, Schroeder et al., 1990). The dominant freshwater sources for the bay are the Fish and Magnolia Rivers, which deliver 73% and 27% of freshwater inputs (Caffrey et al., 2014). The watershed area is 516 km², about 71 times the surface area of the estuary (7.2 km²). Land-use is predominantly agriculture (44%; Lehrter, 2006). Watershed dissolved inorganic nitrogen (DIN) concentrations exceeding 140 µM (Lehrter, 2008) contribute to gross primary production (GPP) rates of approximately 825 g carbon m⁻² y⁻¹ (Caffrey et al., 2014), which is well above the threshold for hypertrophic as defined by Nixon (1995. Weeks Bay provides nursery habitat for a multitude of fish, crustaceans, and shellfish such as shrimp, bay anchovy, and blue crab that support a robust commercial fishery industry providing ~$500 million/year for Alabama (NOAA, 2022). The Weeks Bay National Estuarine Research Reserve (NERRS) has supported monitoring and research in Weeks Bay since 1986.

Figure 1.

Location of Weeks Bay and regional monitoring sites used as model forcings (top left). Weeks Bay watershed land use and monitoring sites (bottom left). Weeks Bay model grid, boundary forcings, and monitoring sites (right).

2.2. Field Sampling and Measurement

Field sampling was conducted from May to October 2015 and during 10-days of intensive monitoring beginning 8/7/17. Vertical profiles of temperature, salinity, dissolved oxygen (DO), pH, photosynthetically active radiation (PAR), chlorophyll-a fluorescence, and CDOM fluorescence were collected at 9 stations using a Sea-Bird Electronics SBE25plus CTD system (Stations EPA01–09 in Fig. 1). A Wet-Labs model WQMx Water Quality Monitor (WQM) was deployed from May to October 2015 at a mid-estuary station co-located with the Weeks Bay National Estuarine Research Reserve (NERRS) station WKBMB (Fig. 1). WQMs were deployed on a buoy with the sensor 0.1 m below the surface and measured temperature, salinity, DO, chlorophyll-a fluorescence, CDOM fluorescence, and turbidity. Additional time series of water temperature and DO were obtained for Weeks Bay using MiniDOT (Precision Measurement Engineering, Inc.) sensors (Station EPA05). Sensor calibrations were validated against air-saturated water before and after deployment and recorded data at 15-minute intervals for two weeks. A MiniDOT sensor was deployed near the sediment water interface at station EPA05, co-located with NERRS station WKBMB.

Rates of primary production and respiration were measured using bottle incubations collected monthly from May to October 2015 following Murrell et al. (2018 at stations EPA01, EPA05, and EPA08. Data collected in 2017 focused on measuring benthic production and respiration at stations EPA01, EPA05, and EPA08 to improve model parameterization. Continuous in-situ benthic flux measurements were made using MiniDOT sensors secured inside a clear open chamber that was placed over the sediment. Chambers were deployed in the mid-morning for a period of 7–9 hours before recovery. After recovery, chambers were wiped clean before being re-deployed overnight.

Temperature, salinity, and DO were also recorded at 0.5 m above the bottom using a YSI 6600 series sonde operated by the Weeks Bay NERRS as part of the NERRS System Wide Monitoring Program (Sanger et al., 2002). Data were obtained via the NERRS Centralized Data Management Office (NERRS Centralized Data Management Office, 2022), and included sites FR01 (Fish River outflow site), WKBMB (mid-bay site), and WKBWB (Week Bay inlet). Bottom DO at these sites along with other environmental variables including bottom water salinity, bottom water temperature, wind speed, solar radiation, water level (tide), rainfall, and Fish River discharge were processed at hourly time scales between April through September for multivariate statistical analysis to examine the factors controlling bottom DO. This data set included 4,392 observations for each variable. The principal component analysis (PCA) was used to determine the number of principle components with eigenvalues ≥ 1, which was then used in a factor analysis (FA) to extract the latent factors involved in the variation of the data set. Varimax orthogonal rotation with Kaiser normalization was used for FA. The PCA/FA analyses were conducted with Analyse-it Ultimate Edition (Analyse-it Software, Ltd.).

2.3. Hydrodynamic Model

We implemented a three-dimensional hydrodynamic model for Weeks Bay using the Environmental Fluid Dynamics Code (EFDC) initially developed by Hamrick (1992. EFDC has been applied in estuarine ecosystems across the U.S. and globally (Wan et al., 2012; Wool et al., 2003), including similar shallow temperate estuaries (Xia et al., 2010; Xia et al., 2011). A curvilinear grid with a general vertical coordinate (GVC) layering scheme (TetraTech, 2006) was utilized to better simulate both shallow and deeper portions of the Bay. Most of Weeks Bay was represented using 5 vertical sigma layers. Three and four sigma layers were used for shallow areas of the Magnolia and Fish Rivers, respectively. Seven layers were used in the deeper channel between Weeks Bay and Mobile Bay. The mean horizontal resolution of the model was 0.01 km², with a mean sigma layer thickness of 0.3 m across the center of the bay, which is nearly uniform in depth. Bathymetry was interpolated from using NOAA soundings (Patchen et al., 2012) where available with supplementing data from historical measurements in Mobile Bay and the tributary rivers.

Model boundary conditions included flow measurements from USGS stations 02378500 (Fish River) and 02378300 (Magnolia River). Because the upstream reach of the Fish River in the EFDC model is downstream of the USGS gauge station, the magnitude of flows was uniformly increased by a factor of 1.56 to account for sub-watershed contributions below the gauge station. Additional sub-watershed contributions that were not included in the Fish and Magnolia River discharges were estimated using the area-weighted average of discharge per unit of watershed area from the two rivers. Freshwater boundary temperatures for the Fish and Magnolia Rivers were quantified using measurements from the Alabama Department of Environmental Management (ADEM) station MGNB-101, which is co-located with the Magnolia River boundary in our model.

Open boundary conditions in Mobile Bay included water surface elevations measured at NOAA station 8735180 (Dauphin Island, AL) adjusted to match water elevations in Weeks Bay at NOAA station 8732828 as part of the model calibration (Fig. 1). Temperature and salinity at the open boundary were quantified using measurements from NOAA station 8735180 (Dauphin Island, AL) and 1435881600 BSCA1, respectively (Fig. 1). Meteorological data, including rainfall, air temperature, barometric pressure, relative humidity, cloud cover, solar radiation, dew point temperature, wind speed and direction, and potential evapotranspiration were derived from measurements at the NOAA NERRS station WKXA1 (NDBC, 2021), located at Weeks Bay, and NOAA National Climatic Data Center SA station WBAN #13894 at the Mobile Bay Regional Airport (NCDC, 2021).

2.4. Ecosystem Model

Water quality was simulated using the Coastal Generalized Ecosystem Model (CGEM), a three-dimensional mechanistic simulation model first implemented in the northern Gulf of Mexico (Jarvis et al., 2021; Jarvis et al., 2020; Lehrter et al., 2017). CGEM is a biogeochemical model that simulates nutrients (nitrogen, phosphorus, silica), carbon, oxygen, phytoplankton, and zooplankton in the physical environment provided by the hydrodynamic model. A detailed description of the modeled state variables is provided in appendices in Lehrter et al. (2017. We coupled CGEM to EFDC using the binary hydrodynamic file linkage created by EFDC, which was originally created to couple with EPA’s Water Quality Analysis Simulation Program (WASP). Hydrodynamic output from EFDC was written to a netcdf format using a utility program to provide the physical transport and mixing environment to the CGEM water quality model. This implementation of CGEM utilized a sediment instant remineralization parameterization, which assumes all organic matter reaching the bottom is instantly remineralized (Fennel et al., 2006). This formulation has the benefit of directly linking oxygen consumption rates to the amount of organic matter deposited to the sediment, helping to more closely couple benthic and pelagic processes in shallow ecosystems (Supplemental Text). Oxygen production by microphytobenthos (MPB) was simulated using an exponential oxygen production-irradiance relationship based on data from Lehrter et al. (2014

$$\text{MPB}=0.33\times \text{PA}{\text{R}}_{\text{se}{\text{d}}^{2.93}}$$ (1)

where PAR_sed is the irradiance (mol photons m⁻² d⁻¹) calculated at the sediment-water interface. Additional descriptions of simulated benthic production and sediment remineralization are available in the Supplemental Text.

Supplemental Tables S1 through S5 provide a comprehensive list of CGEM parameter inputs. Freshwater boundary conditions for CGEM were defined using the USGS Load Estimator (LOADEST; Runkel et al., 2004) using data from ADEM stations FI-1 and MGNB-101 for the Fish and Magnolia Rivers, respectively (Supplemental Figure S1). Open boundary conditions were applied using mean concentrations at station EPA09 during 2015 (Supplemental Table S6).

2.5. Model Calibration

EFDC and CGEM were calibrated to 2015 data for state variables of the water column and biological process rates of the water column and sediment. Hydrodynamic calibration focused on matching time series from continuous water surface elevation, temperature, and salinity and vertical structure from survey based vertical CTD profiles. CGEM calibration focused on DO concentrations and rates of primary production and respiration.

To quantify model performance, field data were used with model output to examine the coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE), normalized RMSE (NRMSE), and Index of Agreement (IA) as:

$${\text{R}}^2=\frac{\left(\text{n}{\sum }_{\text{i}=1}^{\text{n}}\left({\text{M}}_{\text{i}}\times {\text{O}}_{\text{i}}\right)\right)-\left({\sum }_{\text{i}=1}^{\text{n}}{\text{O}}_{\text{i}}\times {\sum }_{\text{i}=1}^{\text{n}}{\text{M}}_{\text{i}}\right)}{\sqrt{\left[\text{n}{\sum }_{\text{i}=1}^{\text{n}}\left({\text{M}}_{\text{i}}^2\right)-{\sum }_{\text{i}=1}^{\text{n}}{\left({\text{M}}_{\text{i}}\right)}^2\right]\times \left[\text{n}{\sum }_{\text{i}=1}^{\text{n}}\left({\text{O}}_{\text{i}}^2\right)-{\sum }_{\text{i}=1}^{\text{n}}{\left({\text{O}}_{\text{i}}\right)}^2\right]}}$$ (2)

$$\text{MAE}=\frac{{\sum }_{\text{i}=1}^{\text{n}}\left|{\text{M}}_{\text{i}}-{\text{O}}_{\text{i}}\right|}{\text{n}}$$ (3)

$$\text{RMSE}=\sqrt{\frac{\sum {\left({\text{O}}_{\text{i}}-{\text{M}}_{\text{i}}\right)}^2}{\text{n}}}$$ (4)

$$\text{NRMSE}=\frac{\text{RMSE}}{\overline{\text{O}}}$$ (5)

$$IA=1.0-\frac{{\sum }_{i=1}^n\left|M_i-O_i\right|}{{\sum }_{i=1}^n\left[\left|M_i-\overline{O}\right|+\left|O_i-\overline{O}\right|\right]}$$ (6)

where $O_i$ is an observed value, $M_i$ the modeled value for n observations, and $\overline{O}$ is the mean of $n$ observations.

3. Results

The 2015 simulation period represents an average year for freshwater discharge into Weeks Bay. Peak flow occurred in mid-April near 40 m³ s⁻¹ for both the Fish and Magnolia Rivers (Fig. 2). Mid-bay salinity observed at NERRS station WKBMB was between 0 and 10 ppt for much of the spring and summer and increased to a mean of 12 ppt in August-September. Water temperature for the entire spring-summer period averaged 29.4°C, with daily maximum summer (June-September) temperatures between June-September between 31–33°C (Fig. 2).

Figure 2.

Hydrograph of Fish and Magnolia River discharge (panel A) during the model simulation period (1/1/15–10/1/15). Timeseries of dissolved oxygen (panel B), salinity (panel C), and temperature (panel D) from the National Estuarine Research Reserve (NERRS) monitoring site WKBMB. Salinity and dissolved oxygen data between 3/30–4/29 were removed due to sensor problems. Triangles represent survey sampling dates. Grey shading depicts the ~7 day deployment of MiniDOT sensors depicted in Fig. 6.

3.1. Hydrodynamic Calibration

Simulated water surface elevations compared to data from NOAA station 8732828 (Fig. 2), had an R² of 0.91, RMSE of 0.08, and IA of 0.97 (Supplemental Table S7). The hydrodynamic model successfully simulated seasonal and sub-daily temperature and salinity dynamics at the mid-bay NERRS WKBMB station (Fig. 3), but simulated bottom salinity varied 7 ppt on average less than observed variability at tidal time scales. Simulated vertical salinity and temperature structure was similar to observations from stations EPA01-EPA08, with R² values from 0.78 to 0.92 and 0.81 to 0.94 for temperature and salinity, respectively (Supplemental Table S7).

Figure 3.

Simulated (black) versus observed (red) water surface elevation at NOAA 8732828 (A,D), temperature at NERRS WKBMB (B,E) and salinity at NERRS WKBMB (C,F). Data are displayed for the entire simulation period (A-C) and for a select 9-day period highlighted in grey (D-F) to illustrate diel variability and model performance.

3.2. Water Column and Sediment Rate Processes

Measurements of water column integrated Primary Production (PPI) based on bottle incubations ranged from 5 to 16 g O₂ m⁻² d⁻¹, with a 27% decrease in mean production along the estuarine gradient from near the mouth of the Fish River to the pass near Mobile Bay (Fig. 4). Simulated PPI varied from 2 to 24 g O₂ m⁻² d⁻¹, a range approximately 2-fold greater than field observations. Peak rates occurred between June and August (Supplemental Figures S2 and S4). Measurements of water column integrated respiration (WRI) based on bottle incubations were 2 to 6 g O₂ m⁻² d⁻¹, with a mean of 4 g O₂ m⁻² d⁻¹, which is about 40% of measured PPI. Simulated WRI averaged ~1 g O₂ m⁻² d⁻¹ across the Bay, which is 25% of the mean of measured WRI (Fig. 4).

Figure 4.

Measured and simulated boxplots of water column (A) and benthic (B) rate processes. Positive and negative values indicate primary production and respiration, respectively. Light grey shading in panel A and B represent the average summer water column integrated primary production rates and benthic respiration measured by Mortazavi et al. (2012), respectively. The dark grey area in panel B represents measured benthic respiration rates from EPA (2011).

Measurements of benthic primary production were 1 to 5 g O₂ m⁻² d⁻¹ and increased from the Fish River outlet to the pass. Simulated benthic production rates were also between 1 and 5 g O₂ m⁻² d⁻¹. Measured benthic respiration rates were less than simulated rates, averaging ~1–3 g O₂ m⁻² d⁻¹ compared to 1–8 g O₂ m⁻² d⁻¹ in simulations (Fig. 4). Estimates of benthic respiration from EPA (2011 and Mortazavi et al. (2012 ranged from 1 to 3.9 g O₂ m⁻² d⁻¹, which encompasses both the observed and simulated rates from this study. Simulated benthic contribution to total integrated primary production averaged 21% (Supplemental Figure S4). Simulated surface chlorophyll concentrations were 0–120 mg m⁻³ and were within the range of measured surface chlorophyll (Supplemental Figure S5). Simulated and measured surface nitrate was 0–45 mmol m⁻³ (Supplemental Figure S6). Measured phosphate was 0–1.6 mmol m⁻³, whereas simulated values were 0–0.7 mmol m⁻³ (Supplemental Figure S6).

3.3. Observed Diel Oxygen Dynamics

Continuous DO observations from the NERRS monitoring stations varied from zero to >150% of the saturation concentration. sometimes within a single day (Fig. 2). DO at the NERRS mid-bay station averaged 6 mg L⁻¹ with overnight DO decreasing to below 4 mg L⁻¹ and daytime DO exceeding 9 mg L⁻¹ from June to September. DO at summer average salinity (12 ppt) and temperature (29°C) is 7.2 mg L⁻¹. DO ≤ 2 mg L⁻¹ was also observed in the upper bay near the Fish River outflow on 18% of days in August and September (Fig. 5). Surface oxygen concentrations averaged between 8 and 10 mg L⁻¹ at station EPA05, with daily minimum DO ≤4 mg L⁻¹ (Fig. 5). Vertical profiles collected between 11:00 AM and 1:00 PM throughout summer showed that DO was nearly homogeneous by mid-day across much of the Bay although during a mid-summer survey (7/28/15) DO approached hypoxia near the bottom (Supplemental Figure S7).

Figure 5.

Simulated (red) versus observed (black) daily dissolved oxygen box plots at NERRS continuous monitoring sites FR01 (A & D) and WKBMB (B & E) and EPA surface buoy at EPA05 (C & F). Data are displayed for the summer (A-C) and for a period of one month (D-F) to illustrate diel variability and model performance.

In the continuous DO time series from 9/14/2015 to 9/22/2015, DO was near anoxia for at least 2 hours on 75% of nights (Fig. 6). Near-bottom observations at night were less than mid-depth observations at the NERRS sensors 94% of the time. Although DO at mid-depth during the same period was less than 2.5 mg L⁻¹ on 38% of nights, DO never fell below 2 mg L⁻¹. Minimum bottom water DO occurred between 2:00 and 8:00 AM, after which DO increased rapidly to between 10 and 11 mg L⁻¹ by 1:00 to 3:00 pm. During 9/12 to 9/22, 20% of DO measurements overall were less than 2 mg L⁻¹, while no DO measurements at 0.5 m above the bottom were less than 2 mg L⁻¹ (Fig. 8). DO below 4 mg L⁻¹ accounted for 36% of observations near the bottom compared to 19% of observations at the NERRS sensor. Thus, the duration of exposure to DO less than 2 and 4 mg L⁻¹ was 17% higher near the bottom compared to 0.5 m above the bottom.

Figure 6.

Timeseries of measured (A) and simulated (B) vertical dissolved oxygen gradients at station EPA05. Black dotted lines represent sensor height in panel A and the center of the CGEM sigma layer in panel B.

Figure 8.

Cumulative distributions of dissolved oxygen concentrations for the near bottom (black line) and 0.5 m above bottom (blue line). The left panel includes observations taken using the near-bottom MiniDOT oxygen sensor (black line) and the NERRS SONDE (blue line). The right panel includes the CGEM bottom sigma layer (layer 5; black line), the CGEM above bottom layer (layer 4, blue line), and the NERRS SONDE (dashed blue line). Percent differences in probability of hypoxia (2 mg L⁻¹; vertical dashed line) and low oxygen exposure at 4 mg L⁻¹ are depicted.

The PCA analysis of hourly bottom DO data measured at FR01, WKBMB, and WKBWB together with other environmental variables identified five principal components with eigenvalues ≥ 1, explaining 74% of the total variance of the water quality data set. Five factors were extracted from FA with orthogonal rotation (Table 2). In general, the absolute value of loading score > 0.5 indicates the factor strongly influences the variable while the absolute value of loading score close to 0 suggests a weak influence on the variable (Liu et al., 2003; Ouyang, 2005; Wan et al., 2014). Given this large data set, we consider a medium loading score of 0.2~0.5 to still present some significant influence on the variable. The result shows that a significant proportion of variability is explained by the factors for bottom DO at the Weeks Bay inlet (WKBWB) and mid-bay (WKBMB) but not at Fish River (FR01). Factor 1 receives strong loading from bottom DO at Weeks inlet (−) and bottom salinity (+), along with medium loading from bottom DO at mid-bay (−), tidal water level (+), and Fish River discharge (−). Factor 2 is driven by tidal water level (+) with some influence on bottom DO at Fish River (−). Factor 3 is dominated by bottom DO at mid-bay (−) with medium loading from bottom DO at Weeks Bay Inlet (−) and Fish River (−) as well as tidal water level (+) and solar radiation (−). Factor 4 receives strong loadings from wind (+) and solar radiation (+) while Factor 5 is dominated by Fish River discharge. Both factors have weak influence on bottom DO. Note that the sign of + or – denotes the direction of change. For example, DO measured at all three locations changes in the same direction with solar radiation in Factor 3.

Table 2.

Factor loadings matrix of factor analysis (Orthogonal rotation [varimax] with Kaiser normalization) for environmental variables affecting DO at three NERRS monitoring stations. Strong factor loadings (absolute value >0.5) are in bold type and medium loadings (0.2–0.5) in italic.

Variables	Factor 1	Factor 2	Factor 3	Factor 4	Factor 5	Communality
Bottom temperature	−0.08	0.19	0.19	−0.26	0.44	0.35
Bottom salinity	0.65	−0.10	0.05	0.04	0.25	0.50
Wind speed	0.04	−0.01	0.02	0.75	−0.08	0.58
Rainfall	0.04	−0.01	0.05	0.12	0.00	0.02
Solar radiation	−0.14	−0.06	−0.21	0.62	0.01	0.45
Fish River discharge	−0.26	−0.04	0.04	−0.06	-0.53	0.35
Water level - tide	0.24	-0.87	0.25	0.27	−0.19	1.00
Fish River DO (FR01)	−0.04	−0.21	−0.20	−0.07	−0.02	0.09
Mid-Bay DO (WKBMB)	−0.22	0.12	-0.90	0.07	−0.06	0.89
Weeks Inlet DO (WKBWB)	-0.96	0.04	−0.28	−0.02	0.02	1.00
Cumulative variance explained with PCA (%)	22.6	41.0	53.2	64.3	73.9

3.4. CGEM Simulated Oxygen Dynamics

CGEM successfully simulated both seasonal and daily oxygen dynamics at different monitoring locations and depths throughout the bay (Fig. 5). Near-river DO concentrations averaged ~6 mg L⁻¹ between June and September, varying between 2 and 10 mg L⁻¹ (Fig. 5A, 5D). Simulated DO at this site was within the daily range of DO observations, however simulated diel DO patterns were less dynamic than observations in late summer (August-September). Simulated DO at the NERRS mid-bay site similarly matched seasonal and diel patterns, with greater diel variability than the near-river site (Fig. 5B, 5E). Simulated DO at this site also averaged ~6 mg L⁻¹, with consistent mid-summer diel variability between 2 and 10 mg L⁻¹. Simulated surface DO remained near saturation in the summer, averaging ~8 mg L⁻¹ and varying daily between 6 and 10 mg L⁻¹ (Fig. 5C, 5F). Simulated surface DO was generally 1 to 2 mg L⁻¹ less than observations.

CGEM also reproduced the fine scale vertical DO gradients recorded by continuous monitoring sensors at the mid-bay station (Fig. 6). Simulated DO was near zero in bottom water in the overnight-early morning hours on 88% of days, consistent with observations (75%), and returned to 100% saturation by mid-day. Simulated oxygen concentrations at mid-depths were 1 to 2 mg L⁻¹ higher than surface DO when density stratification was present due to high primary production in the mid- and bottom-layer. Simulated DO was less than observed DO overnight on some days, as was the case on 9/14 (Fig. 6). Although observed mid-depth DO was low, simulated DO at mid and upper depths was less than observations. Between April to September simulated bottom layer DO was lowest in the early morning hours between 4 to 6 AM (Fig. 7A), similar in both concentration and timing to low morning DO observed during the two-week MiniDOT deployment (Fig. 6). Primary production between the hours of 6 AM through 6 PM caused a net increase in bottom layer DO, with the highest production rates and net increase in DO occurring in the early afternoon between 1 and 4 PM (Fig. 7). Simulated total respiration rates were sustained throughout the day and night and were generally lower than daytime primary production. Biological DO production and respiration rates were between 4–15 times greater than net advective and diffusive oxygen exchange in the bottom layer between adjacent grid cells from April to September (Fig. 7C).

Figure 7.

(A) Hourly DO for the simulated bottom layer and measured DO at Mid-Bay station WKBMB between April-September. Mean surface irradiance includes the 25^th and 75^th percentiles (shaded orange). (B) Hourly change in simulated bottom layer and measured DO. (C) Hourly DO budget for simulated bottom layer primary production, total respiration (water column and sediments), and net advective and diffusive flux (horizontal and vertical) at WKBMB.

Observations indicate a 17% difference in the duration of exposure (defined here as percent of observations) to hypoxia between the bottom layer and lower water column (Fig. 8). Simulations produced a similar difference (16%) between sigma layers 4 and 5, with hypoxia present nearly 25% of the time during summer (Fig. 8). Simulated oxygen exposure below 4 mg L⁻¹ occurred ~55% of the time in sigma layer 5 versus 29% in sigma layer 4. To evaluate bottom layer exposure at different oxygen concentrations we calculated the cumulative hours below oxygen thresholds of 1, 2, 3, and 4 mg L⁻¹ for the period from peak spring rainfall (April) through the end of summer (October) with the model simulations. Low DO exposure varied substantially along the estuarine gradient between the near river site (EPA01) and mouth of Weeks Bay (EPA08), with the greatest exposure occurring in the middle reach of the Bay (Fig. 9). At mid-bay station EPA05 there were 390 hours near anoxia (≤1 mg L⁻¹) and 800 hours less than 2 mg L⁻¹. Anoxia and DO < 2 mg L⁻¹ were less common at EPA01 (about 50 and 100 hours, respectively) and EPA08 (about 100 and 300 hours, respectively (Fig. 9). DO exposure below 4 mg L⁻¹ was similar between EPA05 and EPA08, with both sites exhibiting greater than 1,600 hours of low DO exposure during the 6-month period between spring and summer.

Figure 9.

Simulated timeseries of bottom layer cumulative exposure to low oxygen conditions between peak spring rainfall (April) and the end of summer (October). Simulation results are shown at EPA stations EPA01, EPA05, and EPA08.

4. Discussion

High-frequency observations in this study demonstrate the need for detailed measurements to adequately quantify the complex DO dynamics in shallow estuaries. This is especially true for eutrophic shallow estuaries, where high nutrient loads may sustain high rates of production and respiration, driving high diel DO variability. The measurements needed to quantify these dynamics can be made using a layered strategy whereby different sampling approaches are utilized to target different temporal and spatial scales. Our results also demonstrate that simulation models can be successfully applied to evaluate oxygen dynamics in complex estuarine environments at diel timescales when paired with high-frequency observations.

4.1. Vertical gradients of diel hypoxia

Continuous monitoring observations in this study reveal a precipitous DO gradient between the near bottom and lower water column driven by enhanced water column and benthic metabolism. When nightly hypoxia was observed, DO concentrations near the bottom (~0.1 m) were between 0.8 and 3.1 mg L⁻¹ lower than DO recorded at a height of 0.4 m above the bottom. This steep vertical DO gradient can result in underestimation of low oxygen conditions and hypoxia when fixed sensors are located outside the area of maximum oxygen demand. For example, DO observations from the mid-bay NERRS sensor show only episodic nightly excursions below 2 mg L⁻¹ (Fig. 5b), with less than 2% of summer (June-September) observations exhibiting hypoxia and ~11% of observations of DO ≤ 4 mg L⁻¹ (Fig. 8). Given high primary productivity and heterotrophic metabolism of Weeks Bay such scarce observations of hypoxia at this site are lower than expected, and indeed represent an ~17% underestimate of hypoxic observations when compared to targeted near bottom measurements and simulation results.

It is difficult to adequately measure DO dynamics in shallow estuaries due to the dynamic physical and biological processes that affect important rate processes. Diel DO and primary production rates can vary due to a number of abiotic environmental forcings, including temperature, solar insolation, ebb tide, wind mixing, and daily streamflow (Tyler et al., 2009). These forcings can induce spatial heterogeneity in low oxygen conditions due to localized shifts between heterotrophy and autotrophy (Mortazavi et al., 2012), proximity to freshwater inflow (Novoveska and MacIntyre, 2019), and variable wind and tidal mixing (Duvall et al., 2022; Wenner et al., 2009). In Weeks Bay low DO at three NERRS monitoring sites varied spatially in response to their proximity to external forcings outside the Bay (Table 2). Low DO near the inlet (WKBWB) to Mobile Bay was correlated with higher tides and lower bottom water salinity. While the mid-bay site was to a lesser extent impacted by tidal influence, reduced solar insolation was an important factor in low oxygen conditions, likely due to dampened primary production and DO saturation (Tyler et al., 2009).

Strategic monitoring that combines different sampling methods to separately quantify spatial and temporal patterns of variability can improve DO characterization and help elucidate important environmental controls. This is especially important in shallow systems where benthic-pelagic coupling exerts a greater influence on water column DO. Total integrated respiration rates (water column plus benthic rates) measured in this study were capable of inducing hypoxia from saturation (8 mg L⁻¹) in the lower third of the mid-bay water column in less than 5 hours (± 2.1 hours). Measured benthic production rates in this study were between 13–30% of integrated water column production, consistent with estimates of 21% and 27% from Schreiber and Pennock (1995 and Caffrey et al. (2014, respectively. High benthic production is not surprising given the shallow depth and near bottom light penetration in Weeks Bay (Amacker, 2013; Caffrey et al., 2014), with benthic algae estimated to comprise ~25% of total microalgal biomass (Schreiber and Pennock, 1995). High primary production rates in this study were capable of re-establishing saturation from hypoxia in 1.9 hours (± 0.8 hours) based on water column and sediment rate measurements. In-situ bottom water observations similarly identified a > 4 mg L⁻¹ increase in DO within ~5 hours of the first morning light (Fig. 7A).

In Weeks Bay high primary production rates nearly twice those of nearby Gulf coast estuaries (Caffrey et al., 2014) are reflected in its multi-decadal record of hypoxia and heterotrophic metabolism (Caffrey et al., 2014; Mortazavi et al., 2012). While ongoing continuous monitoring has characterized the system as experiencing summertime diel hypoxia, the layered monitoring strategy applied in this study suggests that diel DO gradients and overnight low oxygen conditions are more extreme than previously documented. In contrast with deeper systems where seasonally persistent hypoxia constitutes a greater portion of the sub-pycnocline water mass, adequately characterizing DO in shallow estuaries may require a targeted monitoring strategy that addresses fine scale temporal and spatial variability between the benthos and upper water column. The need for these measurements are further supported by steep vertical oxygen gradients observed in similar shallow estuarine environments at water depths of ~2 m (Liu et al., 2022) and with hypoxic thickness as small as ~0.5 m (Huang et al., 2019).

4.2. Simulation of diel oxygen dynamics

The absence of mechanistic models applied to diel DO dynamics in shallow estuaries limits our understanding of critical processes linking nutrients to low oxygen. This scarcity of models further inhibits informed policy development aimed at protecting water quality. While regression models have been applied to describe abiotic factors controlling diel-cycling hypoxia (Tyler et al., 2009), few mechanistic models have been implemented that attempt to simulate sub-daily oxygen dynamics (Shen et al., 2008; Testa et al., 2021; Tian et al., 2022). A common challenge among mechanistic models is underestimation of daily production and respiration rates, resulting in well calibrated seasonal dynamics that considerably underestimate nightly oxygen demand and hypoxia (Brady, 2014; Testa et al., 2021). Improving simulation of rate processes requires a coordinated monitoring and modeling effort that ensures measurements match simulation needs in space and time (Ganju et al., 2016). Our ability to effectively calibrate CGEM in Weeks Bay was vastly improved by targeted water column and sediment flux rate measurements necessary to constrain model parameters that impact benthic-pelagic coupling, including sediment oxygen demand, benthic primary production, and water column production and respiration.

Obtaining realistic simulation results at diel time scales also requires highly spatially resolved models that effectively simulate small scale hydrodynamics and allow for sufficient carbon production and respiration to consume oxygen (Ganju et al., 2016; Testa et al., 2021). CGEM in Weeks Bay was implemented at high spatial (mean horizontal: 0.01 km²; mean vertical: 0.3 m) and temporal (10 seconds) resolutions to adequately simulate the physical and biological dynamics controlling organic matter production and respiration. Simulations captured diel DO variability driven by carbon production and respiration rates that were well within our observations and previous measurements in Weeks Bay (Fig. 4). This is an important consideration when evaluating the link between nutrients and metabolic processes that induce hypoxia, as mechanistic models can produce widely variable phytoplankton production rates based on relatively small changes in highly sensitive parameter sets (Golosov et al., 2021).

Diel DO gradients in our simulations were principally biologically mediated with minimal contribution from advective and diffusive oxygen fluxes (Fig. 7C). This contrasts with simulations by Tian et al. (2022 which attribute 43 to 59% of oxygen exchange to advection and between 35 to 48% to photosynthesis and respiration in the Corsica River (Chesapeake Bay). The differences between models may result from a lack of horizontal spatial heterogeneity of DO in Weeks Bay compared to the Corsica River. Morphology and flow patterns of each system play an important role in mediating the physical and biological factors affecting DO, particularly in systems like the Corsica River where primary production varies spatially (Tian et al., 2022). DO in Weeks Bay is principally controlled by irradiance-driven primary production and respiration (Table 2), similar to other shallow estuarine systems (D’Avanzo and Kremer, 1994; Sawabini et al., 2015).

The highly dynamic nature and close coupling of benthic and water column processes in shallow estuaries requires effective parameterization of sediment metabolism. In this study CGEM utilized an instant remineralization approach to link carbon production with respiration and hypoxia over hourly time scales. This is a common sediment parameterization applied in mechanistic models of larger coastal ecosystems (Fennel et al., 2013; Laurent and Fennel, 2014; Pauer et al., 2020), and to our knowledge has not been successfully validated in shallow estuaries. Simulation results indicate that an instant remineralization approach can be used to simulate sediment oxygen demand at depths less than 2 m and under high rates of organic matter deposition. While this approach is reasonable given the short vertical distance in which settling occurs in Weeks Bay, instant remineralization may limit resuspension of organic material that may otherwise accumulate in bottom sediments. Sediment resuspension can be driven by wind mixing, tropic tides, and enhanced freshwater discharge (Ha and Park, 2012), and may be an important factor in stimulating water column algae production in shallow, eutrophic, microtidal estuaries (McGill, 2019). Application of a more complex sediment diagenesis model, such as the two-layer model used by Testa et al. (2021 in the Chester River estuary, likely provides more realistic approximations of burial, mixing, and aerobic/anaerobic reactions necessary to evaluate changes in benthic function over longer timescales. Improving the detail and breadth of validation of simulation models to include variables other than oxygen concentration and relevant rates would be useful for scenario-based analysis, including scenarios that seek to predict the system response to long term changes in nutrient inputs.

4.3. Low DO exposure and implications for aquatic life

Sufficient DO concentrations are essential for the survival of aerobic aquatic organisms. While most water quality studies apply a hypoxic threshold of 2 mg L⁻¹ to describe ecosystem health as it relates to DO, biological studies have identified a much broader range of impacts at higher oxygen concentrations (Vaquer-Sunyer and Duarte, 2008). Defining thresholds for the effects of low oxygen exposure are further complicated in systems that experience diel-cycling oxygen patterns due to a range of lethal (Keppel et al., 2015) and sublethal impacts (Donelan et al., 2021; Morrell and Gobler, 2020). Even relatively small changes in DO concentration (<0.5 mg L⁻¹) and duration of exposure (<4 hours) can have measurable and ecologically important impacts (Morrell and Gobler, 2020). Estimating the aquatic life impact from low DO is thus critically dependent adequate measurement of DO and may be further improved by effective use of models to address knowledge gaps in spatiotemporal variability.

Observations in the mid-water column of Weeks Bay by the NERRS sensor indicate an average daily hypoxia exposure over the summer of less than 1 hour, a greater than 6-hour underestimate when compared to bottom layer exposure simulated by CGEM (Fig. 10). Although not consequential for all species, a difference of nearly 8 hours of exposure to DO less than 2 mg L⁻¹ can have lethal consequences for many organisms (Vaquer-Sunyer and Duarte, 2008) and behavioral avoidance consequences for motile species. While not observed by the NERRS sensor, short-term near bottom DO deployments identified bottom layer anoxia that simulations suggest could last 0.5 to 2.6 hours each night in summer (Fig. 10B). Higher DO exposure at concentrations less than 4 mg L⁻¹ were also of sufficient duration to present additional risk of sub-lethal effects due to prolonged exposure at low DO concentrations above the 2 mg L⁻¹ hypoxia threshold (Hrycik et al., 2016; USEPA, 2000).

Figure 10.

Mean daily exposure (hours) to various dissolved oxygen thresholds at station EPA05/WKBMB between April and September. The top panel depicts hourly exposure simulated in CGEM’s bottom layer (sigma layer 5). The middle panel depicts hourly exposure 0.5 m above the bottom measured by the NERRS SONDE. The bottom panel depicts the difference in hours of exposure between CGEM bottom layer and NERRS SONDE 0.5 m above bottom. April data is excluded from the NERRS dataset (middle panel) due to a sensor error.

While diel-cycling DO involves transient low DO exposure overnight, prolonged exposure to low diel DO can also have cumulative impacts over the course of days to weeks. For example, laboratory experiments demonstrated significant growth reduction of juvenile summer flounder under extreme (1 to 11 mg O₂ L⁻¹) diel cycling for 20 days, ultimately resulting in mortality after two to three weeks of exposure (Davidson, 2015). In Weeks Bay exposure to recurring diel low DO is a threat to aquatic species that may experience near-bottom hypoxia and anoxia between ~25% and ~10% of the summer, respectively (Fig. 8). In comparison, DO conditions are considerably better in the mid-water column where little to no anoxia is present and hypoxia occurs less frequently (~9% of summer). These differences in vertical DO conditions again highlight the risk of underestimating aquatic life exposure to seasonal and prolonged diel low DO conditions when measurements from a single monitoring sensor are considered.

Temporal estimates of DO condition address expected time scales of exposure without considering avoidance behavior and may therefore represent the maximum duration of exposure. This is particularly important in small and shallow systems like Weeks Bay where DO gradients are extreme and avoidance behavior may significantly affect actual exposure. Few studies have tried to address how combined spatial and temporal variability impact low DO exposure and effects in three-dimensions due to challenges in adequately measuring and modeling dynamic DO conditions and translating exposures to aquatic life effects (LaBone et al., 2021; Miller Neilan and Rose, 2013). Highly productive shallow estuaries can produce striking spatiotemporal DO variability that challenge collection of spatially representative data, demanding targeted monitoring strategies that incorporate continuous monitoring throughout the water column to adequately assess potential DO exposure. Mechanistic models are also valuable tools in the assessment of ecosystem condition and as forcings for spatiotemporal exposure assessments (LaBone et al., 2019). Models may be further applied to evaluate critical processes controlling diel DO dynamics and to predict ecosystem response and aquatic life benefits following management actions.

Realistically estimating low DO exposure is important given the potential for significant consequences to a number of commercially important species that utilize Weeks Bay and similar shallow estuaries for spawning and nursery habitat, including blue crab, bay anchovy, brown shrimp, oysters, flounder, and gulf menhaden (Miller-Way et al., 1996). Fish kills of gulf menhaden in Weeks Bay have been documented by the popular press (SEDAR, 2018), and hypoxia driven fish kills of blue crab, shrimp, flounder, and red drum have occurred in the larger Mobile Bay estuary (May, 1973). Larvae and juvenile bay anchovy are particularly susceptible to episodic low oxygen exposure, and may experience 50% egg and larvae mortality at DO concentrations ranging between 1.6–2.8 mg L⁻¹ for durations less than 12 hours (Chesney, 1989). Blue crab, a commercial species valued at $63.7 million annually across the Gulf of Mexico (Posadas, 2017), can experience significant larval and juvenile mortality at concentrations below 4 mg L⁻¹ (Tomasetti et al., 2021; Tomasetti et al., 2018). The combined environmental, economic, and social impacts of diel low DO warrant continued efforts to improve monitoring strategies and develop better models capable of informing more effective management and policy actions to improve water quality.

5.0. Conclusions and future work

Our integrated measurement and modeling of diel oxygen dynamics in Weeks Bay provide unique insights into the highly dynamic nature of biotic and abiotic controls on diel oxygen in shallow coastal ecosystems, with further implications in low DO exposure for sensitive biological species. Both observations and simulations identified significant vertical gradients in near bottom DO that varied as much as 3.1 mg L⁻¹ within 0.4 m of the bottom, attributable largely to high primary productivity and benthic processes in the estuary. Low DO ranging from anoxia to ≤4 mg L⁻¹ were consistently observed and simulated in the lower water column for periods of minutes to >11 hours. Given the range of species response to varying durations and concentrations of low DO, we believe it is important to move beyond the standard 2 mg L⁻¹ threshold for hypoxia when characterizing DO status of coastal ecosystems. While many large estuaries experience episodic low oxygen and hypoxia that can last for weeks to months, shallow estuaries exhibiting diel-cycling DO present a greater challenge to characterizing the DO regime and overall habitat quality. Improving estimates of DO exposure in both space and time using layered monitoring strategies and simulation models will help better inform water quality criteria aimed at protecting aquatic life in these dynamic shallow ecosystems.

Critical to this evaluation is the implementation of CGEM, an important milestone for application of mechanistic models to shallow diel-cycling estuaries. CGEM successfully reproduced diel DO dynamics and important rate processes, providing a unique perspective on spatiotemporal DO dynamics that could not be achieved through observations alone. Future model implementation in Weeks Bay and other shallow estuaries should be applied to inform nutrient management efforts and ecosystem response to global climate change, the impacts of which are poorly understood in shallow estuaries (Brady, 2014; Testa et al., 2021). To address prediction of future conditions more effectively, forthcoming modeling efforts should assess application of more advanced sediment diagenesis models as well as a more explicit benthic algae parameterization to improve simulation of benthic-pelagic coupling over extended simulation periods.

Table 1.

Hydrodynamic model performance statistics for mid-bay continuous monitoring stations

Station	Measured				Simulated				R2	Mean Abs Error	RMSE	Norm RMSE	IA
	Mean	Median	5th %tile	95th %tile	Mean	Median	5th %tile	95th %tile
Water Surface Elevation (m)
NOAA 8732828	0.07	0.10	−0.30	0.41	0.08	0.11	−0.33	0.49	0.91	0.06	0.08	0.33	0.97
Temperature (°C)
NERRS WKBMB	24.00	26.50	11.00	32.00	23.50	26.50	9.70	32.70	0.95	−0.58	2.09	0.09	0.91
EPA05 Surface	29.02	29.30	24.25	32.74	23.01	24.52	9.21	32.07	0.70	0.22	1.70	0.06	0.58
Salinity (ppt)
NERRS WKBMB	7.00	6.50	0.00	14.80	8.90	7.80	2.70	17.90	0.74	1.96	3.30	0.47	0.55
EPA05 Surface	7.27	5.07	1.17	16.34	6.92	5.20	1.64	15.51	0.90	−0.47	1.64	0.23	0.90