Modeling Physical and Biogeochemical Controls on Dissolved Oxygen in Chesapeake Bay: Lessons Learned from Simple and Complex Approaches

Abstract

We compared multiple modeling approaches in Chesapeake Bay to understand the processes controlling dissolved oxygen (O₂) cycling and compare the advantages and disadvantages of the different models. Three numerical models were compared, including: (1) a 23-compartment biogeochemical model coupled to a regional scale, salt- and water-balance box model, (2) a simplified, four-term model formulation of O₂ uptake and consumption coupled to a 3D-hydrodynamic model, and (3) a 23-compartment biogeochemical model coupled to a 3D-hydrodynamic model. All three models reproduced reasonable spatial and temporal patterns of dissolved O₂, leading us to conclude that the model scale and approach one chooses to apply depends on the scientific questions motivating the study. From this analysis, we conclude the following: (1) Models of varying spatial and temporal scales and process resolution have a role in the scientific process. (2) There is still much room for improvement in our ability to simulate dissolved O₂ dynamics in coastal ecosystems. (3) An ever-increasing diversity of models, three of which are presented here, will vastly improve our ability to discern physical versus biogeochemical controls on O₂ and hypoxia in coastal ecosystems.

5.1 Introduction

Depleted dissolved oxygen (O₂) conditions have been a feature of Chesapeake Bay for at least the past nine decades (Newcombe and Horne 1938) and reflect both the Bay’s high-productivity and the physical isolation of bottom waters from the atmosphere. Although Chesapeake Bay may be naturally susceptible to hypoxia development, analyses of long-term data indicate that summer hypoxic and anoxic water volumes have increased over the past several decades (Hagy et al. 2004) in response to some combination of elevated nutrient loading and large-scale climatic changes (Scully 2010a; Murphy et al. 2011). Because hypoxic conditions have many negative consequences for living resources and restrict habitat availability (Brady and Targett 2013; Buchheister et al. 2013), there is a strong emphasis in the management of this system to alleviate low-O₂ conditions in the Bay through nutrient input reductions within the watershed (Boesch et al. 2001).

Our understanding of O₂ dynamics in coastal ecosystems like Chesapeake Bay is complicated by the fact that O₂ is controlled by a diverse suite of physical and biogeochemical processes. Some of these processes covary or are linked (e.g., freshwater and nutrient inputs), and each has its own dominant time and space scales. Elevated freshwater input, for example, delivers inorganic nutrients that fuel phytoplankton production in seaward Bay regions, but also delivers inorganic particles that limit light availability in landward regions and may cause advection of phytoplankton biomass downstream, away from landward regions (e.g., Miller and Harding 2007). Freshwater input is also associated with elevated stratification during summer that reduces vertical diffusion of O₂ (Boicourt 1992), yet elevated freshwater flow also leads to higher landward advection in bottom water (Li et al. 2015), which may deliver O₂-rich water but also may transport relatively labile organic material (Kemp et al. 1997). Wind stress may lead to three-dimensional (e.g., along- and cross-channel) circulation features that physically replenish deep-water O₂ (Scully 2010b). Prevailing northward winds (up-estuary) in summer, for example, can reduce the along-channel exchange flow and the associated landward O₂ flux in the deep water, but strengthen the cross-channel circulation that exchanges O₂-poor water with overlying O₂-rich water (Scully 2010b; Li et al. 2015), and vice versa. The wind could also mix nutrient-rich water into surface layers and fuel elevated phytoplankton production (Malone et al. 1986; Li et al. 2009) that will drive additional O₂ depletion in the following months. Observing the biogeochemical response to these dynamics directly would involve near-continuous measurements of phytoplankton production and respiration, concentrations of O₂, organic carbon and key nutrients, current velocities, stratification, and other meteorological and hydrological variables (e.g., wind speed) over multiple seasons. Considering the impractical and high-cost nature of such efforts, numerical models can be used as tools to understand the response of O₂ to external forcing in a way that cannot reasonably be accomplished using observational studies alone.

Accordingly, much effort has been invested in building models to simulate O₂ dynamics on several time and space scales. Models encompassing a wide range of hydrodynamic resolution, biogeochemical complexity, and temporal scope have been applied in numerous ecosystems (Oguz et al. 2000; Justíc et al. 2007; Fennel et al. 2013; Hamidi et al. 2015), including Chesapeake Bay (Xu and Hood 2006; Liu and Scavia 2010; Scully 2010b; Brown et al. 2013; Cerco and Noel 2013; Lee et al. 2013; Feng et al. 2015) and the Great Lakes (Rucinski et al. 2010; Hamidi et al. 2015). Statistical models are often used to infer the major drivers of O₂ depletion and to guide management actions (e.g., nutrient load reductions) that alleviate low-O₂ zones, while coupled biophysical models (e.g., the three models presented in this chapter) are used to understand ecosystem interactions and feedbacks, where O₂ is one of many biogeochemically linked model variables. While multiple coupled, hydrodynamic−biogeochemical models currently exist for Chesapeake Bay (Xu and Hood 2006; Li et al. 2009; Cerco and Noel 2013), few studies have compared multiple models to emphasize their utility in answering different scientific and management questions (Irby et al. 2016).

In this chapter, we describe three modeling packages that include both simplified and complex biogeochemical and hydrodynamic configurations. Our objectives are to compare and contrast these models to illustrate the different types of simulations that investigators can use to answer key scientific questions. We also aim to assess quantitatively how well the applied models reproduce observations of dissolved O₂ in Chesapeake Bay, discuss their respective limitations, and suggest where and how the models might be utilized in future studies based on their spatio-temporal dimensions and levels of biogeochemical and hydrodynamic complexity.

5.2 Methods and Approach

High-resolution, coupled hydrodynamic–biogeochemical ocean models are valued for their ability to accurately represent physical and biogeochemical process in coastal ecosystems, especially with increasingly affordable computing power. Less spatially resolved models (regional scale) are still useful for research needs, as they can be executed quickly (minutes), represent biogeochemistry at the scales where it has been predominantly measured (regionally and seasonally), and can be used to compute simple and unambiguous budgets of key variables. In this chapter, we describe and analyze three different numerical models, including: (1) a course-scale mass-balance transport model (“box model”) coupled to a water column and sediment model with detailed simulations of biogeochemical processes (Row-Column Aesop, or RCA); (2) a 3D, hydrodynamic model using the Regional Ocean Modeling System (ROMS) coupled to a simple, four-term O₂ biogeochemistry model; and (3) the ROMS coupled to a multi-compartment, water column and sediment biogeochemical process model (RCA).

5.2.1 Box Model with Biogeochemistry (BM-RCA)

A coarse spatial-scale model was developed for Chesapeake Bay, using a water column and sediment biogeochemical model and physical transport calculated using a salt- and water-balance model (box model, or BM). The spatial domain of this model includes 17 control volumes of water (9 surface-layer “boxes”, 8 bottom-layer “boxes”; Figs. 5.1 and 5.2), whose vertical separation is based on mean-pycnocline depths for each region (Hagy 2002). The box model solves a series of linear algebraic equations to compute advective and non-advective exchanges between control volumes based upon freshwater inputs and salt distributions, and given the small number of regions (17), the computation requires minimal computing resources. The biogeochemical model (Row-Column Aesop, or RCA) was developed by HydroQual, Inc. and has been applied in many coastal ecosystems (e.g., Long Island Sound, Massachusetts Bay). RCA is the most recent extension of the family of water quality models that originated from the Water Quality Analysis Simulation Program (WASP) used by the United States Environmental Protection Agency (Di Toro et al. 1983). The model allows for up to three phytoplankton groups, as well as water column state variables representing the following; (1) particulate and dissolved organic carbon, nitrogen and phosphorus, (2) dissolved inorganic nitrogen, phosphorus, and silica, (3) biogenic particulate silica, and (4) O₂ (Fig. 5.3). RCA also includes a state variable that represents O₂ equivalents associated with sulfide and methane released at the sediment-water interface. Nitrification and denitrification are modeled in both the water column and sediments, where RCA includes a sediment biogeochemical model, which has two layers that represent the near-surface aerobic and underlying anaerobic environments and simulates the cycling of carbon, O₂, nitrogen, phosphorus, silica, and sulfur. The sediment model predicts sediment-water fluxes of dissolved O₂, nitrate, ammonium, phosphate, dissolved methane, and sulfide, where the latter two constituents contribute to the water column state variable called “oxygen equivalents” that acts as a reservoir for non-nitrogen reduced solutes that contribute to O₂ demand. Sulfide is produced via sulfate reduction in the sediment model, where it is subsequently stored temporarily in particulate form as iron monosulfide (Cornwell and Sampou 1995), oxidized in the sediment (contributing to sediment O₂ demand), or released to the overlying water. A more extensive description of the RCA modeling package is found elsewhere (Testa et al. 2013, 2014; Xue et al. 2014). Model simulations were run on a 6-h time step over the years 1986–2006.

Fig. 5.1

Maps of Chesapeake Bay, including (left panel) bathymetry with red area representing the O₂ budget region (see text), (middle panel) the ROMS model grid with wet cells in red, and (right panel) locations of water column monitoring stations (blue circles), sediment-water flux observation stations (red squares), the location of PAR data used for simulations (yellow star), and water column respiration measurements (green hexagons). For water column monitoring stations, select stations are labeled with abbreviated station names relative to the official monitoring station names (e.g., CB22 = CB2.2 and CB33–CB43 = CB3.3C–CB4.3C). Middle and right panels reprinted from Testa et al. (2014) with permission from Elsevier

Fig. 5.2

Diagram of salt- and water-balance (box) model for Chesapeake Bay, including relevant freshwater inputs, transport coefficients (Q, Q’, E), box identifiers, and key tributary estuaries. In Box 5–9, Δs represents salt exchanges between the boxes and adjacent, connected tributaries. Diagram adapted from Hagy (2002). An aerial view of this model is included in Fig. 5.1 (right panel)

Fig. 5.3

(left panel) Simple O₂ model, based on empirical relationships, used to investigate physical controls on hypoxia. (right panel) Schematic diagram of the major state variables and transformation processes in RCA, reprinted from Testa et al. (2014) with permission from Elsevier

The key advantages of the BM-RCA approach are that it can be executed rapidly on a personal computer (20 years in ~5 min for the 17 regions) and that it captures regional patterns of O₂, carbon, and nutrient dynamics, which can be validated using available monitoring data and rate-process measurements. Its short run times make it amenable to sensitivity analysis and scenario experiments and allow it to be implemented by a wide range of users (e.g., resource managers) who may not have access to advanced computational resources. Because of its well-defined boundaries and relatively few transport terms, the model output is easily post-processed to provide regional and seasonal mass-balance budgets of key variables. The disadvantages of this modeling approach include its inability to capture lateral and vertical variability in physical and biogeochemical processes, as well as the patchy spatial patterns in plankton productivity and biomass. In particular, short temporal- and spatial-scale physical dynamics in response to wind and tidal forcing cannot be adequately represented in this model, and these processes are known to be important in Chesapeake Bay and other coastal ecosystems (Li and Zhong 2009; Scully 2010a).

5.2.2 Hydrodynamic 3D Model with Simple Oxygen (ROMS-SDO)

The second model in this analysis (ROMS-sDO) is a simple, empirical O₂ model implemented in a relatively high-resolution hydrodynamic model (the Regional Ocean Modeling System; ROMS). This model was motivated by the need to capture the key seasonal patterns in biogeochemical O₂ uptake so that experiments related to the effects of altered hydrodynamic variability due to wind stress, tidal mixing, and freshwater inflow can be achieved. This approach is designed to take into account the strong seasonal variation of both uptake and production in the water column (Fig. 5.3). Based on observed measurements of sediment O₂ demand (SOD), water column respiration (WCR), and phytoplankton community production (PhP), the approach empirically relates these processes to the state variables (e.g., O₂ concentration, ambient water temperature T), and/or environment forcing (e.g., photosynthetically available radiation PAR), allowing for O₂ consumption/production to vary with space (horizontal and vertical) and time to the first order (see details in Li et al. 2015).

$$ \begin{aligned} SOD & = 9.90 \times 1.7845^{{T/10^{o} {\text{C}}}} \times \left( { \, \frac{{O_{2} }}{{Ox + 59\,{\text{mmol}}\,{\text{O}}_{ 2} {\text{ m}}^{ - 3} }}} \right) \\ PhP & = 31.25 \times (1.0101 + 0.0314PAR + 0.1966T) \\ WCR & = 3.3 \times e^{0.0715T} \\ \end{aligned} $$

ROMS has been implemented in Chesapeake Bay and validated against a wide range of observational data and has demonstrated considerable capability in reproducing estuarine dynamics at tidal, synoptic, and seasonal time-scales (Li et al. 2005). We use an application of this model with a 160 × 240 grid in the horizontal direction (about 500 m grid size) for ROMS-sDO and a grid 80 × 120 grid in the horizontal direction (about ~1 km grid size) for ROMS-RCA (Sect. 6.2.3). Both ROMS-sDO and ROMS-RCA include 20 layers in the vertical dimension (Fig. 5.1). The open ocean boundary consists of 10 constituents (M2, S2, N2, K2, K1, O1, P1, Q1, Mf, and Mm) for tidal forcing, de-tided observations for non-tidal water elevations, and monthly climatology for salinity and temperature. In order to provide the hindcasts of hydrodynamic fields, the model is forced by daily river discharge along with zero salinity and seasonal water temperature, and by winds, net shortwave and downward longwave radiation along with air temperature, relative humidity, and pressure. Surface water temperature was nudged to the observed SST field. Further details of the application of ROMS in Chesapeake Bay are described elsewhere (e.g., Li et al. 2005).

The ROMS-sDO approach is advantageous because it uses reasonable, but computationally meager representations of biogeochemical O₂ production and uptake to allow high temporal- and spatial-scale simulations of O₂ dynamics in coastal ecosystems. The disadvantages of this approach include the fact that the empirical O₂ uptake formulations generally cannot capture spatial and interannual variability related to nutrient loading and availability (i.e., eutrophication) and do not capture biogeochemical feedbacks, which limit the model’s ability to be run for realistic multi-year experiments. Thus, we display the viability of this approach in reproducing seasonal O₂ dynamics, focusing ROMS-sDO simulations to a single year (1989) when many of the observations used to build the empirical models were made.

5.2.3 Hydrodynamic 3D Model with Biogeochemistry (ROMS-RCA)

The most complex model formulation we present in this chapter is a “soft-coupling” of ROMS to the multi-compartment, dynamic biogeochemical model RCA (as described above). The term “soft-coupling” represents the fact that ROMS hydrodynamic model simulations were performed first, where the output was saved and subsequently used to provide hydrodynamic fields (e.g., current velocity) to drive RCA simulations. This model includes high spatial and temporal resolution simulations of a wide range of hydrodynamic and biogeochemical processes. Hourly averages of ROMS temperature, salinity, and transport terms are used to force RCA along with external loads of nutrients and organic carbon based on daily freshwater inputs and monthly fortnightly nutrient concentrations (Fig. 5.1). RCA is simulated on a 10-min time step, and we utilized a multitude of available monitoring data to characterize boundary and initial conditions, as well as external flows and loadings (Testa et al. 2014; Li et al. 2016).

The key advantages of this model are that it simulates high-resolution spatial and temporal dynamics of biogeochemical and physical processes, such that even with changing nutrient loading and climatic conditions, the model can be run to reproduce reasonable estuarine dynamics over many years. This allows for experiments with altered nutrient loading, freshwater input, and wind stress to be conducted that result in improved understanding of myriad estuarine processes over multiple time and space scales. The disadvantages of this approach include high computational costs and the simulation of physical and biogeochemical processes at scales (e.g., daily, ~1 km) that are difficult to validate given limited spatial and temporal scales of observational data.

5.2.4 Calibration and Validation Datasets

All models require validation to ensure their ability to accurately reproduce the processes and state variables that are intended to be understood and simulated. Although many models are validated against state variables, including concentrations of nutrients, O₂, carbon, and phytoplankton biomass (chlorophyll-a), it is equally important to validate these models against rate processes, such as plankton community photosynthesis and respiration rates, organic matter decay rates, and nutrient transformation and sediment-water exchange rates. Validation of both “state” and “rate” observations allow one to determine that the model is predicting concentrations based on appropriate transformation rates, as opposed to situations where the model accurately simulates “state” concentrations generated by compensating but inaccurate rate processes (e.g., offsetting errors). For Chesapeake Bay, we were able to calibrate and validate model behavior for diverse variables and process for which a wide range of data exist across time and space scales. Validations of the water column state variables (e.g., chlorophyll-a or chl-a, dissolved and particulate nutrients, dissolved and particulate organic carbon, etc.) were performed using fortnightly monthly observations of these variables at several depths and stations within Chesapeake Bay (http://www.chesapeakebay.net/data). In addition, rates of water column respiration were compared to measurements of O₂-uptake in dark bottles at several stations (Sampou and Kemp 1994; Smith and Kemp 1995), while rates of photic-layer net primary production were validated with observations based on O₂ incubations and empirical model computations based on ¹⁴C uptake measurements (Smith and Kemp 1995; Harding et al. 2002). Observed sediment-water fluxes of dissolved O₂ were estimated from time-course changes in solute concentrations during incubations of intact acrylic sediment cores collected at key stations in Chesapeake Bay (Fig. 5.1; Cowan and Boynton 1996).

5.3 Insights Gained from Model Simulations

In this section, we review the performance of each of the three models and highlight selected results from the simulations. We also emphasize the types of questions that can and cannot be answered with a given modeling approach and how these approaches lead to improved understanding of the dynamics of O₂ and hypoxia in Chesapeake Bay.

5.3.1 Comparison of Model Performance

All three models reproduced seasonal cycles of bottom-water dissolved O₂ along the central axis of Chesapeake Bay with reasonable accuracy (Fig. 5.4). Taylor diagrams (Fig. 5.4) graphically illustrate several metrics of model-data agreement (Taylor 2001), and each model was compared in such a diagram for the time period during which simulations were made, including 1996–2005 for ROMS-RCA, 1985–2006 for BM-RCA, and 1989 for ROMS-sDO. O₂ simulations from each model were highly correlated to observed values, where r-values exceeded 0.7 in all cases except the upper-Bay regions in ROMS-sDO (Fig. 5.4). These two stations (CB2.2 and CB3.1) for ROMS-sDO also tended to have the highest root-mean-squared difference values (RMSD; Fig. 5.4). Although this model-data mismatch was only based upon a single year, it reveals the potential inability for an empirical O₂ model to capture variability in respiration in the upper Bay, where observed O₂ variability is high and driven by interannual variations in the accumulation of phytoplankton biomass in bottom-waters (Testa and Kemp 2014). Somewhat over-predicted O₂ at these upper-Bay stations may also explain why ROMS-sDO tends to under-predict hypoxic volume in the early summer (see Sect. 6.3.3), as hypoxia tends to initiate in the upper Bay. In general, the ROMS-based models with high spatial resolution tended to capture variability in bottom-water O₂ better than the regionally based BM-RCA, yet BM-RCA was able to capture seasonal patterns of bottom-water O₂, as well as interannual variability associated with changes in freshwater and nutrient inputs (Fig. 5.4). That said, ROMS-RCA predicted short-term variations (daily weekly) in bottom-water O₂ that BM-RCA did not. Although the high-frequency observations necessary to validate these model simulations do not currently exist along the main channel of Chesapeake Bay, high-frequency variability in O₂ is expected to occur given tidal mixing and the passage of storm fronts.

Fig. 5.4

(left panel) Time series of bottom-water O₂ concentrations at three stations in Chesapeake Bay, including observed concentrations (open circles) and those predicted by the BM-RCA (red line) and ROMS-RCA model (blue lines). (right panel) Taylor diagrams for bottom-water O₂ concentrations from the three models at several stations in Chesapeake Bay

As mentioned previously, we consider it important to validate model process rates in addition to concentration measurements. We thus compared rates of water column respiration (as O₂ uptake) and SOD measured at several stations to those rates predicted by ROMS-RCA (Fig. 5.5). Water column respiration rates, which were available from multiple studies and several stations in the main stem of Chesapeake Bay (Sampou and Kemp 1994; Smith and Kemp 1995), compared favorably with ROMS-RCA-simulated respiration over several seasons (Fig. 5.5a, c and e). Although modeled respiration rates were slightly higher than observations during March to May at three stations (10.00–14.69 (model) versus 7.39–11.17 (observed) mmol m⁻³ d⁻¹), summer (June–August) and November rates were comparable. Where measurements from multiple years were available for comparison (spring and summer at CB6.1), model estimates fell within the range of observations. At station CB4.3, anoxia during mid-summer prevented the measurement of respiration with oxygen-based techniques, but respiration was predicted by the model because anoxia was not always predicted in this region by ROMS-RCA (Figs. 5.4 and 5.5). ROMS-RCA also captured seasonal variability in sediment oxygen demand (hereafter SOD) at three stations, but tended to over-predict SOD in the middle Bay (Fig. 5.5). As with water column respiration, ROMS-RCA did not predict the true anoxic conditions that were observed at this station and SOD was allowed to persist because O₂ was available for uptake from the overlying water (Fig. 5.5). Despite this, overlying O₂ was sufficiently low and sediment pore water sulfide concentrations were high-enough for the sediment model to generate fluxes of sulfide to the water column, which would subsequently consume water column O₂. The sulfide fluxes correspond to seasonal minima in both bottom-water O₂ concentrations and the aerobic layer depth within the sediment model, which limits the storage of sulfide (Cornwell and Sampou 1995). ROMS-sDO-simulated respiration and SOD rates are plotted for comparison, to illustrate how they were similar across stations with identical seasonal variation, which contrasts with ROMS-RCA predictions that varied spatially and temporally (Fig. 5.5).

Fig. 5.5

(left panel; a, c, e) Comparisons of water column respiration rates modeled by ROMS-RCA (1996–2005 mean is solid black line, minima and maxima are dotted lines) and ROMS-sDO (green lines) with observed values indicated by red circles. (right panel; b, d, f) Comparisons of sediment oxygen demand (SOD) modeled by ROMS-RCA (1996–2005 mean is solid black line, minima and maxima are dotted lines) sediment-water sulfide (H₂S) flux modeled by ROMS-RCA (blue lines) and SOD modeled by ROMS-sDO (green lines) with observations indicated by red circles (±1 SD). Both left and right panels include comparisons at an upper Bay (a, b; CB2.2), middle Bay (c, d; CB4.3C), and lower-Bay (e, f; CB6.1) station

5.3.2 Insights Gained from BM-RCA

Despite the simplified physical transport model used to drive BM-RCA, it reproduces seasonal, regional, and interannual variability in bottom-water dissolved O₂ (Fig. 5.4). Given the coarse resolution in this model, these validations are restricted to the deepest stations along the main channel of central Chesapeake Bay (shallow stations flanking the channel are excluded). Thus, although BM-RCA is a useful tool to understand the effects of interannual variations of river flow and nutrient load on deep-water O₂ at regional scales, it cannot resolve episodic or spatially resolved dynamics (e.g., vertical profiles) in the Bay. However, the straightforward division of the Bay into a limited number of segments in BM-RCA allows for the computation of regional budgets of O₂, carbon, and related nutrients.

For example, several questions remain related to the timing and location of the source of organic matter fueling hypoxia in Chesapeake Bay (Kemp et al. 1997). Budget calculations for particulate organic carbon (POC) in each box during two seasons (Spring = March–April, Summer = June–July) averaged over the 1986–2006 period, suggest the key role of both vertical sinking and landward longitudinal transport as mechanisms for POC delivery to bottom waters (Fig. 5.6). Landward net POC imports (the potential fuel for O₂ depletion) were greatest in lower-Bay regions, but muted in upper-Bay regions (Fig. 5.6). This is consistent with the suggestion that landward, bottom-water transport of organic carbon resulting from net surface-layer carbon production in seaward Bay regions is a key aspect of the Bay carbon budget supporting O₂ depletion (e.g., Kemp et al. 1997). Thus, although BM-RCA cannot capture small-scale variability in O₂ (which is often important) it reveals nutrient loading controls on deep-water O₂ and seasonal and regional transport of organic carbon.

Fig. 5.6

Mean particulate organic carbon (POC) budgets for middle-Bay regions as computed from BM-RCA model computations in Chesapeake Bay for the March to April (top) and June–July (bottom) periods. Net Import is equivalent to the balance between the transport terms

5.3.3 Insights Gained from ROMS-SDO

Recently, state-of-the-art hydrodynamic models have been coupled to relatively simple formulations for biogeochemical O₂-uptake (Hetland and DiMarco 2008; Scully 2010a; Li et al. 2015). The motivation for these efforts has been to quantify biogeochemical effects on O₂ in a simple and computationally meager way to allow an emphasis on variations in physical controls. Such an approach for Chesapeake Bay has proven useful to understand the effects of wind speed and direction on O₂ dynamics, but these simulations also provide an opportunity to understand the balance between biogeochemical O₂ uptake and physical replenishment.

ROMS-sDO was run for Chesapeake Bay in the year 1989, where observations of primary production, water column respiration, and sediment O₂ demand were available to develop empirical formulations for these processes (Li et al. 2015). ROMS-sDO captured seasonal patterns of deep-water O₂ across many stations in Chesapeake Bay, and thus reasonably reproduced the annual cycle of hypoxic volume (Figs. 5.4 and 5.7). To understand the balance between O₂ uptake and physical replenishment, a budget was computed for O₂ by integrating model O₂-equation over a selected control volume for Chesapeake Bay, which encompasses all deep waters below 10-m depth (from the mean sea level) in the main stem to the north of York River mouth, and the northern boundary intercepts the shoaling bathymetry in the upper Bay (see Fig. 5.1 for location). The budget revealed five types of changes for deep-water O₂, including the water column respiration over the entire control volume, the sediment respiration across the seafloor, the along-channel advection of oceanic high-O₂ water across the lower-Bay section, and the vertical advection and diffusion of O₂ across the 10-m interface (Li et al. 2015). Over the April to July period, biogeochemical O₂ uptake (water column + sediment respiration) exceeded inputs via horizontal and vertical advection and diffusion, resulting in the drawdown of bottom-water O₂ and the development of hypoxia (Fig. 5.7). Over the course of spring and early summer (as in all months) this biogeochemical O₂ uptake was dominated by water column respiration (Fig. 5.7), which is consistent with cross-system analyses that suggest sediment O₂ uptake is a small fraction of total water column uptake in systems deeper than 5–8 m (Kemp et al. 1992). Interestingly, during the spring period when O₂ is drawn down, advection of O₂ is a large fraction of total input to this region due to stronger circulation resulting from buoyancy-induced along-estuary density gradient and favorable prevailing wind directions (Li et al. 2015). Although vertical diffusion is the dominant term for the physical components of the O₂ budget during mid-summer, advection is once again important during later-summer and fall, when physical imports exceed biogeochemical uptake, leading to replenishment of bottom-water O₂ (Fig. 5.4) and the decline in hypoxic volume (Fig. 5.7).

Fig. 5.7

Time series of ROMS-sDO-simulated (top panel) hypoxic volume (O₂ < 62.5 μM) in the main stem of Chesapeake Bay, (middle panel) monthly averages of the vertical diffusive and sum of the advective fluxes and the volume-integrated water column respiration (WCR) and sediment O₂ demand (SOD), and (bottom panel) monthly net O₂ flux (physical transport + biogeochemical uptake) in the control volume, (see Fig. 5.1, left panel) where positive value represent net replenishment and negative values represent net uptake

Simulations using ROMS-sDO thus make an important contribution to our understanding of O₂ dynamics in Chesapeake Bay. It became clear that even relatively simple models are useful in quantifying the seasonal and regional balance between O₂ uptake and replenishment, and discerning which processes (advection versus diffusion, water column versus sediment respiration) contribute most to variability at a given time of year. Although such models can therefore be used to investigate interannual variations in physical input and be subject to experiments that quantify the effects of freshwater input and altered wind patterns, they cannot be directly used to understand interannual variations in O₂ resulting from altered nutrient loading and other biological considerations.

5.3.4 Insights Gained from ROMS-RCA: Interannual Variation

The most complex and well-resolved model presented in this chapter is ROMS-RCA, which includes a dynamic water column and sediment biogeochemical model coupled to a relatively high-resolution hydrodynamic ocean model. This model permits investigations into short temporal- and spatial-scale dynamics as in ROMS-sDO and the analysis of individual terms in the O₂ budget, but it adds an additional value in that specific biogeochemical mechanisms can be examined, as well as interannual variability of dissolved O₂ associated with altered nutrient loading. These model simulations spanned a 10-year period in Chesapeake Bay using realistic climatic and freshwater forcing, and were also executed for a single year (2000) to examine responses specific to altered nitrogen and phosphorus loading scenarios.

A dissolved O₂ budget analysis similar to that presented for ROMS-sDO was performed on the 10-year simulation of ROMS-RCA, with the addition of separate terms for horizontal advection and diffusion, as well as vertical advection and diffusion (Fig. 5.8). O₂ budgets were calculated for two periods (March–May and June–August), to capture periods when hypoxia initiates (March–May) and when hypoxic volume is fully developed at its seasonal maximum (June–August). In both periods, biogeochemical uptake exceeds physical replenishment, but interannual variability is much higher for the spring period (March–May) than during the summer period (June–August; Fig. 5.8). Perhaps, more importantly, physical replenishment of O₂ tends to be proportional to spring O₂ uptake, where interannual variation in spring O₂ uptake covaries with physical O₂ inputs, both of which are correlated to winter–spring nutrient loading. In addition, physical replenishment of O₂ is highest during summer, which is the period when biogeochemical O₂ uptake is at seasonal maxima (Fig. 5.8). This reveals that vertical and horizontal gradients in O₂ that are setup by biogeochemical uptake influence the physical replenishment fluxes, regardless of season. Pie charts representing the relative contribution to the total O₂ budget of the various physical and biogeochemical terms reinforce the results of ROMS-sDO, where water column respiration is the dominant uptake term and advective inputs of O₂ are comparable in magnitude to diffusive inputs in multiple seasons (Fig. 5.8).

Fig. 5.8

Time series of ROMS-RCA-simulated total diffusive + advective O₂ inputs to the hypoxic region (red bars) and the volume-integrated water column respiration (WCR) and sediment O₂ demand (SOD) in the same region (green bars) for the March–May (top) and June to August (bottom) periods as computed from ROMS-RCA. Pie charts indicate the fraction of the total O₂ budget contributed by each diffusive, advective, or biogeochemical term for the given period

5.3.5 Insights Gained from ROMS-RCA: Response to Nutrient Loading

The second set of analyses presented in this section includes a sensitivity test of Chesapeake Bay O₂ dynamics to altered loadings of nitrogen (N) and phosphorus (P). Whereas the ROMS-sDO simulations predict the same biogeochemical O₂ uptake in a given year and place to understand interannual variability in physical effects, these experiments use the same physical forcing (from the year 2000 in ROMS-RCA) to isolate variations in nutrient loading effects. These experiments reveal that hypoxic volume days (HVD) were consistently higher under elevated nutrient loads, but the response was stronger for N relative to P. This resulted from widespread N limitation in seaward Bay regions during summer months (Malone et al. 1996) that led to higher net primary production (NPP) and phytoplankton biomass under elevated N loads or much lower NPP and biomass under reduced loads (Fig. 5.9). Relatively lower sensitivity to P loads results from the fact that P is limiting in winter–spring in the upper and middle Bay and that phytoplankton growth during this season appears to be a less important driver of summer hypoxic volumes (e.g., Newell et al. 2007). The fact that HVD was more sensitive to combined NP load changes reveals the potential for alternating nutrient limitation if the load is dominated by either N or P, especially during transitional periods in Chesapeake Bay where P is limiting in spring and N is limiting in summer (Malone et al. 1996). These results highlight previously underemphasized seasonal dynamics associated with hypoxia-nutrient load relationships, as well as the interacting role of N and P loads in controlling hypoxic volume, which have been highlighted in other large coastal ecosystems (Conley et al. 2009; Greene et al. 2009; Laurent and Fennel 2014).

Fig. 5.9

(top panel) Total ROMS-RCA-simulated water column respiration in three regions (R3-upper Bay, R5-middle Bay, R7-lower Bay) of Chesapeake Bay (see Fig. 5.1 for regional boundaries) computed for each of 10 N load experiments during summer (June–August). (middle panel) Comparison of modeled hypoxic volume days (HVD) for several different nutrient loading scenarios (0.25x, 1x, 2x) for nitrogen only (orange bars at left), phosphorus only (green bars in middle), and nitrogen + phosphorus (cyan bars at right). (bottom panel) Relationship between January to May total nitrogen loading and HVD from model simulations (green circles), where the vertical dashed line indicates the TN load observed for the year 2000. Middle and bottom panels reprinted from Testa et al. (2014) with permission from Elsevier

HVD responded nonlinearly to January to May total nitrogen (TN) loads varying over 2 orders of magnitude (e.g., Murphy et al. 2011). These simulations suggest that HVD would saturate (600 km³-d) at loads approaching twice that of conditions in 2000. From a management perspective, this indicates that current nutrient reduction goals should be expected to result in observable reductions in hypoxic volume. Simple mechanistic models that simulate Chesapeake Bay hypoxia also predict similar, nonlinear relationships between TN load and hypoxic volume (Liu and Scavia 2010), although volumes at the low end of the loading range have not been observed. If HVD is plotted against N load for each year from the 1996–2005 simulation, the HVD tends to fall below the logistic curve (Fig. 5.9) for years with below-average Susquehanna Flow (lower HVD/load), while HVD/load is higher in above-average flow years. This suggests that HVD is sensitive to physical circulation or additional nutrient inputs under high-flow conditions.

Perhaps, the clearest conclusion of the nutrient load simulations for Chesapeake Bay is the importance of summer NPP and respiration in driving the Bay’s response to N loading. This is consistent with recent historical data analyses, which have suggested that declines in late summer (July–August) hypoxic volume are associated with modest declines in January to May Susquehanna River N loads (Murphy et al. 2011). Statistical analyses do not, however, provide the specific mechanisms connecting reduced winter–spring N loads to July–August hypoxic volumes. As this and other studies (Malone et al. 1996) have shown, N limitation is the primary control on phytoplankton growth during summer throughout much of Chesapeake Bay. Model simulations clearly display that NPP, phytoplankton biomass, and respiration during the summer period are more sensitive to N additions than during spring (Fig. 5.9), but spring (March to May) NPP and water-column respiration rates were also enhanced by elevated nutrient loads (as in Fig. 5.8). In a related ROMS-RCA model simulation, where summer phytoplankton was not allowed to grow (Fig. 5.10), bottom-water O₂ was replenished to non-hypoxic levels beginning in mid-June. The implication of this result is that summer phytoplankton growth and subsequent respiration are necessary to maintain hypoxia throughout summer. Additionally, N-loading enhancement of lower-Bay water column respiration was a primary driver of interannual variations in hypoxic volume in the 10-year simulation in Chesapeake Bay. These simulations clearly identify N load as a major driver of mid- to late summer hypoxic volume in Chesapeake Bay, and they provide mechanisms to link N load to hypoxia.

Fig. 5.10

Seasonal cycle of ROMS-RCA-simulated bottom-layer dissolved O₂ at four stations in Chesapeake Bay under two conditions: (1) the “Base Run,” or simulation under normal conditions (blue lines), and (2) a simulation where summer phytoplankton growth is prevented (“No Summer Production,” red lines)

5.4 Summary and Synthesis

5.4.1 Lessons Learned from Different Models

Simulation studies for each of the three numerical models presented in this chapter provide a unique contribution to our current understanding of O₂ cycling and hypoxia in Chesapeake Bay. The modeling tool chosen to answer a particular research question should be based on a need to balance model complexity with sufficient resolution of the spatio-temporal scales and process description needed to investigate a particular suite of research questions.

BM-RCA was able to accurately reproduce seasonal and regional O₂ dynamics in Chesapeake Bay, as well as to quantify interannual variability in chlorophyll-a and O₂ concentrations over a two-decade-long period. Regional budgets of particulate organic carbon derived from the model revealed the importance of landward longitudinal advection in delivering labile carbon to the seasonally hypoxic region of Chesapeake Bay. The low computational cost of this model and its generic physical transport calculations make it highly portable to other coastal systems. BM-RCA, however, cannot resolve lateral patterns in O₂ uptake and transport or capture fine-scale patterns of O₂, thus preventing accurate computations of hypoxic area and volume. For example, to investigate recently recognized channel-shoal interactions associated with lateral circulation as a key replenishment process of O₂ (Scully 2010a) and associated phytoplankton responses (Malone et al. 1986), a finer-scale 3D modeling approach is required (e.g., ROMS-sDO or ROMS-RCA). ROMS-sDO can also be used to understand the nature of other hydrodynamic processes on dissolved O₂ such as the effects of tidal/wind mixing and freshwater input. ROMS-sDO, however, cannot be used to simulate the interannual variability of dissolved O₂ and hypoxic and anoxic volume in the Bay because it does not capture interannual changes in water column and sediment O₂ uptake associated with changes in nutrient load. Thus, to understand interannual variations, we need to examine variation in nutrient loading from year to year, along with the associated spatially-resolved patterns of nutrient and organic matter transport and cycling. In this case, ROMS-RCA is chosen over simplified models (BM-RCA or ROMS-sDO) because of its sufficient resolution and complexity in both physics and biogeochemistry.

5.4.2 Considerations for the Future

Despite the recent advances in modeling dissolved O₂ dynamics and hypoxia in Chesapeake Bay and other coastal waters worldwide, there is vast room for additional analysis and model improvement. This chapter was designed to help the coastal system modeling community by illustrating what can be learned about the effects of seasonal and interannual variability in physical forcing and nutrient loading on hypoxia in a particular system (Chesapeake Bay). The lessons learned can likely be applied elsewhere. Despite what was learned from 10 years of ROMS-RCA model simulations for the Bay, an extension of the simulations beyond a decade may be necessary to fully test the model’s ability to reproduce interannual variability in hypoxia and capture trends. Biogeochemical model simulations, despite their limitations (e.g., unconstrained parameters, missing processes), are only as good as the hydrodynamic simulations used to drive them. Several applications of ROMS in Chesapeake Bay have successfully reproduced current velocities and longitudinal salinity and temperature structure (e.g., Li et al. 2005; Irby et al. 2016), but have been unable to generate accurate gradients of vertical salinity, often under-predicting these gradients in some times and places (Irby et al. 2016). To improve simulations of dissolved O₂ concentrations (vertical gradients of which tend to follow those of salinity), hydrodynamic simulations must improve this aspect of estuarine dynamics. New efforts to compare the accuracy and utility of multiple coupled hydrodynamic–biogeochemical ocean models may be a way forward in enhancing existing modeling tools.

5.4.3 Summary

In this chapter, we summarize and compare three modeling systems for simulating dissolved O₂ and hypoxia dynamics in Chesapeake Bay. We conclude that each of these modeling approaches has its advantages and disadvantages, and the choice of which to apply depends on the scientific questions that are to be addressed. For example, if one seeks a tool to do sensitivity tests or examine regional patterns in biogeochemistry, a model like BM-RCA may be sufficient. On the other hand, if the driving questions are related to climatic effects on O₂ dynamics, a model similar to ROMS-sDO may be adequate. However, if one is interested in examining interannual variability in biogeochemical processes, biophysical interactions and feedbacks, or small-scale processes, a model like ROMS-RCA is necessary. Although the potential levels of complexity and resolution accessible with modern coastal biophysical models continue to increase, relatively simple models still have a role for addressing broader and more general research questions to understand coastal systems dynamics.