Introduction
Carbon fluxes along the land–ocean aquatic continuum are currently receiving
increasing attention because of their recently recognized role in the global
carbon cycle and anthropogenic CO2 budget (Bauer et al., 2013; Regnier
et al., 2013a; LeQuéré et al., 2014, 2015). Estuaries are important
reactive conduits along this continuum, which links the terrestrial and
marine global carbon cycles (Cai, 2011). Large amounts of terrestrial carbon
transit through these systems, where they mix with carbon from autochthonous,
as well as marine, sources. During estuarine transit, heterotrophic processes
degrade a fraction of the allochthonous and autochthonous organic carbon
inputs, supporting a potentially significant, yet poorly quantified CO2
evasion flux to the atmosphere. Recent estimates suggest that
0.15–0.25 PgC yr-1 is emitted from estuarine systems worldwide
(Borges and Abril, 2012; Cai, 2011; Laruelle et al., 2010; Regnier et al.,
2013a; Laruelle et al., 2013, Bauer et al., 2013). Thus, in absolute terms,
the global estuarine CO2 evasion corresponds to about 15 % of the
open-ocean CO2 uptake despite the much smaller total surface area.
Currently, estimates of regional and global estuarine CO2 emissions are
mainly derived on the basis of data-driven approaches that rely on the
extrapolation of a small number of local measurements (Cai, 2011; Chen et
al., 2013; Laruelle et al., 2013). These approaches fail to capture the
spatial and temporal heterogeneity of the estuarine environment (Bauer et
al., 2013) and are biased towards anthropogenically influenced estuarine
systems located in industrialized countries (Regnier et al., 2013a). Even in
the best surveyed regions of the world (e.g., Australia, western Europe, North
America, or China) observations are merely available for a small number of
estuarine systems. In addition, if available, data sets are generally of low
spatial and temporal resolution. As a consequence, data-driven approaches can
only provide first-order estimates of regional and global estuarine CO2
emissions.
Integrated model–data approaches can help here, as models provide the means
to extrapolate over temporal and spatial scales and allow the disentanglement of the
complex and very dynamic network of physical and biogeochemical processes
that control estuarine CO2 emissions. Over the past decades,
increasingly complex process-based models have been applied, in combination
with local data, to elucidate the coupled carbon–nutrient cycles on the
scale of individual estuaries (e.g., O'Kane, 1980; Soetaert and Herman,
1995; Vanderborght et al., 2002; Lin et al., 2007; Arndt et al., 2009; Cerco
et al., 2010; Baklouti et al., 2011). However, the application of such model
approaches remains limited to the local scale due to their high data
requirements for calibration and validation (e.g., bathymetric and geometric
information and boundary conditions), as well as the high computational
demand associated with resolving the complex interplay of physical and
biogeochemical processes on the relevant temporal and spatial scales
(Regnier et al., 2013b). Complex process-based models are thus not suitable
for application on a regional or global scale and, as a consequence, the
estuarine carbon filter is, despite its increasingly recognized role in
regional and global carbon cycling (e.g., Bauer et al., 2013), typically not
taken into account in model-derived regional or global carbon budgets (Bauer
et al., 2013). The lack of regional and global model approaches that could
be used as stand-alone applications or that could be coupled to regional
terrestrial river network models (e.g., Global NEWS, Seitzinger et al., 2005;
Mayorga et al., 2010; SPARROW, Schwarz et al., 2006) and continental shelf
models (e.g., Hofmann et al., 2011) is thus critical.
The Carbon-Generic Estuary Model (C-GEM (v1.0); Volta et al., 2014) has been
developed with the aim of providing such a regional and/or global modeling tool
that can help improve existing, observationally derived first-order
estimates of estuarine CO2 emissions. C-GEM (v1.0) has been
specifically designed to reduce data requirements and computational demand
and, thus, tackles the main impediments to the application of estuarine
models on a regional or global scale. The approach takes advantage of the
mutual dependency between estuarine geometry and hydrodynamics in alluvial
estuaries and uses an idealized representation of the estuarine geometry to
support the hydrodynamic calculations. It thus allows running steady-state
or fully transient annual to multi-decadal simulations for a large number of
estuarine systems, using geometric information readily available through
maps or remote sensing images. Although the development of such a
regional–global tool inevitably requires simplification, careful model
evaluations have shown that, despite the geometric simplification, C-GEM
provides an accurate description of the hydrodynamics, transport, and
biogeochemistry in tidal estuaries (Volta et al., 2014). In addition, the
model approach was successfully used to quantify the contribution of
different biogeochemical processes to CO2 air–water fluxes in an
idealized, funnel-shaped estuary forced by typical summer conditions
characterizing a temperate western European climate (Regnier et al., 2013b).
Volta et al. (2016b) further investigated the effect of estuarine geometry
on the CO2 outgassing using three idealized systems and subsequently
established the first regional carbon budget for estuaries surrounding the
North Sea by explicitly simulating the six largest systems of the area
(Volta et al., 2016a), including the Scheldt and the Elbe, for which detailed
validation was performed.
Limits of the 0.5∘ resolution watersheds corresponding to
tidal estuaries of the US East Coast. Three subregions are delimited with
colors, and orange stars represent the location of previous studies.
Here, we extend the domain of application of C-GEM (v1.0) to quantify
CO2 exchange fluxes, as well as the overall organic and inorganic carbon
budgets for the full suite of estuarine systems located along the entire US East Coast, one of the most intensively monitored regions in
the world. A unique set of regional data, including partial pressure of
CO2 in riverine and continental shelf waters (pCO2; Signorini et
al., 2013; Laruelle et al., 2015), riverine biogeochemical characteristics
(Lauerwald et al., 2013), estuarine eutrophication status (Bricker et al.,
2007), and estuarine morphology (NOAA, 1985) is available. These
comprehensive data sets are complemented by local observations of carbon
cycling and CO2 fluxes in selected, individual estuarine systems (see
Laruelle et al., 2013, for a review), making the US East Coast an ideal region for a first, fully explicit regional evaluation of
CO2 evasion resolving every major tidal estuary along the selected
coastal segment. The scale addressed in the present study is unprecedented so
far (> 3000 km of coastline) and covers a wide range of
estuarine morphological features, climatic conditions, land-use and land
cover types, and urbanization levels. The presented study will not
only allow a further evaluation of C-GEM (v1.0) but will also provide the
first regional-scale assessment of estuarine CO2 evasion along the US East Coast (25–45∘ N). Furthermore, it will help explore general
relationships between carbon cycling and CO2 evasion, and readily
available estuarine geometrical parameters.
After a description of the model itself and of the data set used to set up
the simulations, a local validation is presented, which includes salinity,
pCO2, and pH longitudinal profiles for two well-monitored systems (the
Delaware Bay and the Altamaha River estuary). The averaged
rates of
CO2 exchange at the air–water interface for each year simulated by the model for 13
individual estuaries are also compared with observed values reported in the
literature. Next, regional-scale simulations for 42 tidal estuaries of the US
East Coast provide seasonal and yearly integrated estimates of the net
ecosystem metabolism (NEM), CO2 evasion, and carbon-filtering capacity
(CFilt). Model results are then used to elucidate the estuarine
biogeochemical behavior along the latitudinal transect encompassed by the
present study (30–45∘ N). Finally, our results are used to derive
general relationships between carbon cycling and CO2 evasion, and
readily available estuarine geometrical parameters.
Regional description and model approach
Observation-based carbon budget for the US East Coast
The study area covers the Atlantic coast of the United States (Fig. 1), from
the southern tip of Florida (25∘ N) to Cobscook Bay (45∘ N)
at the United Sates–Canada boundary. This area encompasses distinct climatic zones and
land cover types and exhibits a variety of morphologic features (Fig. 1). The
region can be subdivided into several subregions following a latitudinal
gradient (Signorini et al., 2013). In this study, we define three subregions
following the boundaries suggested by the COSCAT segmentation (Meybeck et
al., 2006; Laruelle et al., 2013) and the further subdivision described in
Laruelle et al. (2015). From north to south, the regions are called the North
Atlantic (NAR), Mid-Atlantic (MAR), and South Atlantic (SAR) regions (Fig. 1). Total carbon
inputs from watersheds to US East Coast estuaries (Table 1) have been
estimated to range from 4.0 to 10.7 Tg C yr-1 (Mayorga et al., 2010;
Shih et al., 2010; Stets and Strieg, 2012; Tian et al., 2010, 2012),
consisting of dissolved organic carbon (DOC; ∼ 50 %), dissolved
inorganic carbon (DIC; ∼ 40 %), and particulate organic carbon (POC;
∼ 10 %). In addition, a statistical approach has been applied to
estuaries of the region to quantify organic carbon budgets and net ecosystem
productivity (NEP) using empirical models (Herrmann et al., 2015).
Estimates of total annual riverine input from watersheds to
estuaries (Tg C yr-1). The ranges are based on Stets and
Striegl (2012), Global NEWS (Mayorga et al., 2010), Hartmann et al. (2009),
SPARROW (Shih et al., 2010), and DLEM (Tian et al., 2010, 2012). Modified from
Najjar et al. (2012).
DIC
DOC
POC
Total
NAR
0.2–0.8
0.3–2.1
0.1–0.2
0.6–3.1
MAR
1.4–1.8
0.5–2.3
0.1–0.3
2.0–4.4
SAR
0.4–1.4
0.9–1.6
0.1–0.2
1.4–3.2
Total
2.0–4.0
1.7–6.0
0.3–0.7
4.0–10.7
Recent studies estimated that, along the US East Coast,
rivers emit 11.4 Tg C yr-1 of CO2 to the atmosphere (Raymond et
al., 2013), while continental shelf waters absorb between 3.4 and
5.4 Tg C yr-1 of CO2 from the atmosphere (Signorini et al.,
2013). A total of 13 local, annual mean estuarine CO2 flux
estimates across the air–water interface based on measurements are also
reported in the literature and are grouped along a latitudinal gradient
(Table 2). Four of these estimates are located in the SAR: Sapelo Sound, Doboy Sound, Altamaha Sound (Jiang et al., 2008), and
the Satilla River estuary (Cai and Wang, 1998). Three studies investigate
CO2 fluxes in the MAR: the York River estuary
(Raymond et al., 2000) and the Hudson River (Raymond et al., 1997). There is
also a comprehensive CO2 flux study for the Delaware Estuary that will be published
after the completion of this work (Joeseof et al., 2015). Six systems are
located in the NAR: the Great Bay, the Little Bay,
the Oyster estuary, the Bellamy estuary, the Cocheco estuary (Hunt et al.,
2010, 2011), and the Parker River estuary (Raymond and Hopkinson, 2003). The
mean annual flux per unit area from these local studies is
11.7 ± 13.1 mol C m-2 yr-1 and its extrapolation to the
total estuarine surface leads to a regional CO2 evasion estimate of
3.8 Tg C yr-1. This estimate is in line with that of Laruelle et
al. (2013), which proposes an average CO2 emission
rate of 10.8 mol C m-2 yr-1, for the same region. Thus, CO2 outgassing could
remove 35 to 95 % of the riverine carbon loads during estuarine transit.
About 75 % of the air-water exchange occurs in tidal estuaries
(2.8 Tg C yr-1), while lagoons and small deltas contribute the
remaining 25 %. Although these simple extrapolations from limited
observational data are associated with large uncertainties, they highlight
the potentially significant contribution of estuaries to the CO2
outgassing in the region. However, process-based quantifications of regional
organic and inorganic C budgets, including air–water CO2 fluxes for the
estuarine systems along the US East Coast, are not available.
Published local annually averaged estimates of FCO‾2 in mol C m-2 yr-1 for estuaries along the US East Coast.
Name
Long (∘)
Lat (∘)
FCO‾2
Reference
Observed
Modeled
Altamaha Sound
-81.3
31.3
32.4
72.7
Jiang et al. (2008)
Bellamy
-70.9
43.2
3.6
3.9
Hunt et al. (2010)
Cocheco
-70.9
43.2
3.1
3.9
Hunt et al. (2010)
Doboy Sound
-81.3
31.4
13.9
25.7
Jiang et al. (2008)
Great Bay
-70.9
43.1
3.6
3.9
Hunt et al. (2011)
Little Bay
-70.9
43.1
2.4
3.9
Hunt et al. (2011)
Oyster Bay
-70.9
43.1
4
3.9
Hunt et al. (2011)
Parker River estuary
-70.8
42.8
1.1
3.9
Raymond and Hopkinson (2003)
Sapelo Sound
-81.3
31.6
13.5
20.6
Jiang et al. (2008)
Satilla River
-81.5
31
42.5
25.7
Cai and Wang (1998)
York River
-76.4
37.2
6.2
8.1
Raymond et al. (2000)
Hudson River
-74
40.6
13.5
15.5
Raymond et al. (1997)
Selection of estuaries
The National Estuarine Eutrophication Assessment (NEEA) survey (Bricker et
al., 2007), which uses geospatial data from the NOAA Coastal Assessment Framework (CAF) (NOAA,
1985), was used to identify and characterize 58 estuarine systems discharging
along the Atlantic coast of the United States. From this set, 42 tidal
estuaries, defined as a river stretch of water that is tidally influenced
(Dürr et al., 2011), were retained (Fig. 1) to be simulated by the C-GEM
model, which is designed to represent such systems. Using outputs from
terrestrial models (Hartmann et al., 2009; Mayorga et al., 2010), the
cumulated riverine carbon loads for all the nontidal estuaries that are
excluded from the present study amount to 0.9 Tg C yr-1 , which
represents less than 15 % of the total riverine carbon loads of the
region. These 16 systems are located in the SAR (Eq. 10) and in the MAR
(Eq. 6).
The northeastern part of the domain (NAR; Fig. 1; Table 1) includes 11
estuaries along the Gulf of Maine and the Scotian Shelf, covering a
cumulative surface area of 558 km2. It includes drowned valleys, rocky
shores, and a few tidal marshes. The climate is relatively cold (annual
mean = 8 ∘C) and the human influence is relatively limited
because of low population density and low freshwater input. The mean
estuarine water depth is 12.9 m and the mean tidal range is 2.8 m.
The central zone (MAR) includes 18 tidal estuaries accounting for a total
surface area of 9298 km2. The Chesapeake Bay and the Delaware estuaries
alone contribute more than 60 % to the surface area of the region. In
this region, estuaries are drowned valleys with comparatively high river
discharge and intense exchange with the ocean. Several coastal lagoons,
characterized by a limited exchange with the ocean, are located here, but are
not included in our analysis. The MAR is characterized
by a mean annual temperature of 13 ∘C and is strongly impacted by
human activities due to the presence of several large cities (e.g., New York,
Washington, Philadelphia, Baltimore) and intense agriculture. The mean water
depth is about 4.7 m and the tidal range is 0.8 m.
The SAR includes 13 tidal estuaries covering a
total surface area of 959 km2. These systems are generally dendritic
and surrounded by extensive salt marshes. The climate is subtropical with an
average annual temperature of 19 ∘C. Land use includes agriculture
and industry, but the population density is generally low. Estuarine systems
in the SAR are characterized by a shallow mean water depth of 2.9 m and a
tidal range of 1.2 m.
Model setup
The generic 1-D reactive-transport model (RTM) C-GEM (Volta et al., 2014) is
used to quantify the estuarine carbon cycling in the 42 systems considered in
this study. The approach is based on idealized geometries (Savenije, 2005;
Volta et al., 2014) and is designed for regional- and global-scale
applications (Regnier et al., 2013b; Volta et al., 2014, 2016a). The model
approach builds on the premise that hydrodynamics exert a first-order
control on estuarine biogeochemistry (Arndt et al., 2007; Friedrichs and
Hofmann, 2001) and CO2 fluxes (Regnier et al., 2013a). The method takes
advantage of the mutual dependence between geometry and hydrodynamics in
tidal estuaries (Savenije, 1992) and the fact that, as a consequence,
transport and mixing can be easily quantified from readily available
geometric data (Regnier et al., 2013a; Savenije, 2005; Volta et al., 2016b).
Idealized estuarine geometry and main parameters. Parameters
indicated by green arrows are measured; b is calculated. See Sect. 2.3.1
for further details.
Estuarine surface area (a) and mean annual freshwater discharge
(b) for each tidal estuary of the US East Coast. Estuarine surface
area is expressed as a percentage of the entire surface area of the region
(19 830 km2).
Description of idealized geometries for tidally averaged
conditions
Although tidal estuaries display a wide variety of shapes, they nevertheless
share common geometric characteristics that are compatible with an idealized
representation (Fig. 2; Savenije, 1986, 2005). For tidally averaged
conditions, their width B (or cross-sectional area A) can be described by
an exponential decrease as a function of distance, x, from the mouth
(Savenije, 1986, 2005).
B=B0×exp-xb
B (m) is the tidally averaged width, B0 (m) the width at the mouth,
x (m) the distance from the mouth (x=0), and b (m) the width convergence
length (Fig. 2). The width convergence length, b, is defined as the
distance between the mouth and the point at which the width is reduced to B0e-1. It is directly related to the dominant hydrodynamic forcing. A high
river discharge typically results in a prismatic channel with long
convergence length (river-dominated estuary), while a large tidal range
results in a funnel-shaped estuary with short convergence length (marine-dominated estuary). At the upstream boundary, the estuarine width is given
by the following equation.
BL=B0×exp-Lb
L denotes the total estuarine length (m) along the estuarine
longitudinal axis.
The total estuarine surface S (m2) can be estimated by integrating
Eq. (1) over the estuarine length.
S=∫0LBdx=b×B0×1-exp-Lb
The width convergence length is then calculated from B0, BL,
L, and the real estuarine surface area (SR) by inserting Eq. (2) in
Eq. (3).
b=SRB0-BL
SR is calculated for each system using the SRTM water body data (Fig. 3a), a
geographical data set encoding high-resolution worldwide coastal outlines in a
vector format (NASA/NGA, 2003). While such a database exists for a well-monitored region such as the US East Coast, resorting to using the
idealized estuarine surface area (S) is necessary in many other regions.
The longitudinal mean, tidally averaged depth h (m) is obtained from the
NEEA database (Bricker et al., 2007).
Using this idealized representation, the estuarine geometry can be defined by
a limited number of parameters: the width at the mouth (B0), the
estuarine length (L), the estuarine width at the upstream limit
(BL), and the mean depth h. These parameters can be easily
determined from local maps or Google Earth using geographic information
systems (GISs) or they can be obtained from databases (NASA/NGA, 2003).
Hydrodynamics, transport, and biogeochemistry
Estuarine hydrodynamics are described by the one-dimensional barotropic,
cross-sectionally integrated mass and momentum conservation equations for a
channel with arbitrary geometry (Nihoul and Ronday, 1976; Regnier et al.,
1998; Regnier and Steefel, 1999).
rs∂A∂t+∂Q∂x=0∂U∂t+U∂U∂x=-g∂ζ∂x-gUUCz2H
t is the time (s), x the distance along the longitudinal axis
(m), A the cross-sectional area A= H×B (m2), Q the
cross-sectional discharge Q=A×U (m3 s-1), U the flow
velocity Q/A (m s-1), rs the storage ratio
rs=Bs/B, Bs the storage width (m), g the gravitational acceleration (m s-2), ξ the elevation (m), H the total water depth H=h+ξ(x,t) (m), and Cz the Chézy
coefficient (m1/2 s-1). The coupled partial differential equations
(Eqs. 5 and 6) are solved by specifying the elevation ξ0(t) at the
estuarine mouth and the river discharge Qr(t) at the upstream
limit of the model domain.
The one-dimensional, tidally resolved, advection–dispersion equation for a
constituent of concentration C(x,t) in an estuary can be written as follows (e.g.,
Pritchard, 1958).
∂C∂t+QA∂C∂x=1A∂∂xAD∂C∂x+P
Q(x,t) and A(x,t) denote the cross-sectional discharge and area,
respectively, and are provided by the hydrodynamic model (Eqs. 5 and 6).
P(x,t) is the sum of all production and consumption process rates affecting
the concentration of the constituent. The effective dispersion coefficient
D (m2 s-1) implicitly accounts for dispersion mechanisms
associated with sub-grid scale processes (Fischer, 1976; Regnier et al., 1998).
In general, D is maximal near the sea, decreases upstream, and becomes
virtually zero near the tail of the salt intrusion curve (Preddy, 1954; Kent,
1958; Ippen and Harleman, 1961; Stigter and Siemons, 1967). The effective
dispersion at the estuarine mouth can be quantified by the following relation
(Savenije, 1986).
D0=26×h01.5×N×g0.5
h0 (m) is the tidally averaged water depth at the estuarine mouth
and N is the dimensionless Canter-Cremers estuary number defined as the
ratio of the freshwater entering the estuary during a tidal cycle to the
volume of salt water entering the estuary over a tidal cycle (Simmons,
1955).
N=Qb×TP
In this equation, Qb is the bank-full discharge
(m3 s-1), T is the tidal period (s), and P is the tidal prism
(m3). For each estuary, N can thus be calculated directly from the
hydrodynamic model. The variation in D along the estuarine gradient can be
described by Van der Burgh's equation (Savenije, 1986).
∂D∂x=-KQrA
K is the dimensionless Van der Burgh's coefficient and the minus sign
indicates that D increases in the downstream direction (Savenije, 2012). The Van
der Burgh's coefficient is a shape factor that has values between 0 and 1
(Savenije, 2012), and it is a function of estuarine geometry for tidally
average conditions. Therefore, each estuarine system has its own
characteristic K value, which correlates with geometric and hydraulic scales
(Savenije, 2005). Based on a regression analysis covering a set of 15 estuaries, it has been proposed to constrain K from the estuarine geometry
(Savenije, 1992).
K=4.32×h00.36B00.21×b0.14with0<K<1
Reaction processes P considered in C-GEM comprise aerobic degradation,
denitrification, nitrification, primary production, phytoplankton mortality,
and air–water gas exchange for O2 and CO2 (Fig. 4 and Table 3).
These processes and their mathematical formulation are described in detail in
Volta et al. (2014) and Volta et al. (2016a).
Conceptual scheme of the biogeochemical module of C-GEM used in
this study. State variables and processes are represented by boxes and ovals, respectively. Modified from Volta et al. (2014).
State variables and processes explicitly implemented in CGEM.
State variables
Name
Symbol
Unit
Suspended particulate matter
SPM
gL-1
Total organic carbon
TOC
µM C
Nitrate
NO3
µM N
Ammonium
NH4
µM N
Phosphate
DIP
µM P
Dissolved oxygen
DO
µM O2
Phytoplankton
Phy
µM C
Dissolved silica
DSi
µM Si
Dissolved inorganic carbon
DIC
µM C
Biogeochemical reactions
Gross primary production
GPP
µM C s-1
Net primary production
NPP
µM C s-1
Phytoplankton mortality
M
µM C s-1
Aerobic degradation
R
µM C s-1
Denitrification
D
µM C s-1
Nitrification
N
µM N s-1
O2 exchange with the atmosphere
FO2
µM O2 s-1
CO2 exchange with the atmosphere
FCO2
µM C s-1
SPM erosion
ESPM
gL-1 s-1
SPM deposition
DSPM
gL-1 s-1
The nonlinear partial differential equations for the hydrodynamics are
solved by a finite difference scheme following the approach of Regnier et
al. (1997), Regnier and Steefel (1999), and Vanderborght et al. (2002). The
time step Δt is 150 s and the grid size Δx is constant along
the longitudinal axis of the estuary. The grid size default value is 2000 m,
but can be smaller for short-length estuaries to guarantee a minimum of
20 grid points within the computational domain. Transport and reaction terms
are solved in sequence within a single time step using an operator splitting
approach (Regnier et al., 1997). The advection term in the transport equation
is integrated using a third-order-accurate total-variation-diminishing (TVD)
algorithm with flux limiters (Regnier et al., 1998), ensuring monotonicity
(Leonard, 1984), while a semi-implicit Crank–Nicholson algorithm is used for
the dispersion term (Press et al., 1992). These schemes have been extensively
tested using the CONTRASTE estuarine model (e.g., Regnier et al., 1998;
Regnier and Steefel, 1999; Vanderborght et al., 2002) and guarantee mass
conservation to within < 1 %. The reaction network (including
erosion-deposition terms when the constituent is a solid species) is
numerically integrated using the Euler method (Press et al., 1992). The
primary production dynamics, which take into account the combined effects of
nutrient limitation and light attenuation in the water column induced by its
background turbidity and suspended particle matter (SPM) concentration, requires vertical resolution of
the photic depth. The latter is calculated according to the method described
in Vanderborght et al. (2007). This method assumes an exponential decrease in
the light in the water column (Platt et al., 1980), which is solved using a
Gamma function.
Boundary and forcing conditions
Boundary and forcing conditions are extracted from global databases and
global model outputs that are available at 0.5∘ resolution.
Therefore, C-GEM simulations are performed at the same resolution according
to the following procedure. First, 42 coastal cells corresponding to tidal
estuaries are identified in the studied area (Fig. 1). If the mouth of an
estuary is spread over several 0.5∘ grid cells, those cells are
regrouped in order to represent a single estuary (e.g., Delaware estuary), and
subsequently, a single idealized geometry is defined as described above. The
model outputs (Hartmann et al., 2009; Mayorga et al., 2010) and databases
(Antonov et al., 2010; Garcia et al., 2010a, b) used to constrain our
boundary conditions are representative of the year 2000.
For each resulting cell, boundary and forcing conditions are calculated for
the following periods: January–March, April–June, July–September, and
October–December. This allows for an explicit representation of the seasonal
variability in the simulations.
External forcings
Transient physical forcings are calculated for each season and grid cell
using monthly mean values of water temperature (World Ocean Atlas; Antonov et
al., 2010; Locarini et al., 2010) and seasonal averaged values for wind speed
(Cross-Calibrated Multi-Platform (CCMP) Ocean Surface Wind Vector Analyses
project; Atlas et al., 2011). Mean daily solar radiation and photoperiods
(corrected for cloud coverage using the ISCCP Cloud Data Products; Rossow and
Schiffer, 1999) are calculated depending on latitude and day of the year
using a simple model (Brock, 1981).
Annual river carbon loads of TOC (a), annual DOC fluxes from
wetlands (b), annual river carbon loads of DIC (c), and annual TC fluxes (d).
All fluxes are indicated per watershed.
Riverine discharge, concentrations, and fluxes
River discharges are extracted from the UNH/GRDC runoff data set (Fekete et
al., 2002). These discharges represent long-term averages (1960–1990) of
monthly and annual runoff at 0.5∘ resolution. The data set is a
composite of long-term gauging data, which provide average runoff for the
largest river basins, and a climate-driven water balance model (Fekete et
al., 2002). Total runoff values are then aggregated for each watershed at the
coarser 0.5∘ resolution (Fig. 3b). Next, seasonal mean values (in
m2 s-1) are derived in order to account for the intra-annual
variability in water fluxes. Based on annual carbon and nutrient inputs from
the watersheds (Mg yr-1), mean annual concentrations (mmol m-3)
are estimated for each watershed using the UNH/GRDC annual runoff
(km2 yr-1). Mean seasonal concentrations are then calculated from
the seasonally resolved river water fluxes of a given subregion.
Annual inputs of DOC, POC, and inorganic nutrients are derived from the
global NEWS2 model (Mayorga et al., 2010). Global NEWS is a spatially
explicit, multielement (N, P, Si, C), and multi-form global model of nutrient
exports from rivers. In a nutshell, DOC exports are a function of runoff,
wetland area, and consumptive water use (Harrison et al., 2005). No
distinction is made between agricultural and natural landscapes since they
appear to have similar DOC export coefficients (Harrison et al., 2005).
Sewage inputs of organic carbon (OC) are ignored in Global NEWS
because their inclusion did not improve model fit to data (Harrison et al.,
2005). POC exports from watersheds are estimated using an empirical
relationship with SPM (Ludwig et al., 1996). Inorganic nitrogen (DIN) and
phosphorus (DIP) fluxes calculated by Global NEWS depend on agriculture and
tropical forest coverage, fertilizer application, animal grazing, sewage
input, atmospheric N deposition, and biological N fixation (Mayorga et al.,
2010). The inputs of dissolved silica (DSi) are controlled by soil bulk
density, precipitation, slope, and presence of volcanic lithology (Beusen et
al., 2009).
The DIN speciation is not provided by the Global NEWS2 model. The NH4 and
NO3 concentrations are therefore determined independently on the basis
of an empirical relationship between ammonium fraction (NH4 / DIN ratio) and
DIN loads (Meybeck, 1982). Dissolved oxygen (DO) concentrations are extracted
from the water quality criteria recommendations published by the US Environmental Protection Agency (EPA, 2009). The same source is used
for phytoplankton concentrations, using a ratio of chlorophyll a to phytoplankton
carbon of 50 g C (gChla)-1 (Riemann et al., 1989) to convert the
EPA values to carbon units used in the present study.
Inputs of DIC and total alkalinity (ALK) are
calculated from values reported in the GLORICH database (Hartmann et al.,
2009). For each watershed, seasonal mean values of DIC and ALK concentrations
are estimated from measurements performed at the sampling locations that are
closest to the river–estuary boundary. The spatial distribution of annual
inputs of TOC = DOC + POC, DIC, and TC = TOC + DIC from
continental watersheds to estuaries are reported in Fig. 5a, c, and d,
respectively. The contribution of tidal wetlands to the total organic carbon (TOC) inputs is also
shown (Fig. 5b). Overall, the total carbon (TC) input over the entire model domain is
estimated at 4.6 Tg C yr-1, which falls in the lower end of previous
reported estimations (Najjar et al., 2012).
Inputs from tidal wetlands
The DOC input of estuarine wetlands (Fig. 5b) scales to their fraction, W,
of the total estuarine and is calculated using the Global NEWS
parameterization.
Y_DOC=E_Cwet×W+E_Cdry×1-W×Ra×QactQnatY_DOCwetY_DOC=E_Cwet×WE_Cwet∗W+E_Cdry×(1-W)
Y_DOC is the DOC yield (kg C km-2 yr-1) calculated for the
entire watershed, Y_DOCwet is the estimated DOC yield from
wetland areas (kg C km-2 yr-1), Qact/Qnat
is the ratio between the measured discharge after dam construction and before
dam construction, E_Cwet and E_Cdry
(kg C km-2 yr-1) are the export coefficients of DOC from wetland
and non-wetland soils, respectively. W is the percentage of the land area
within a watershed that is covered by wetlands, R is the runoff
(m yr-1), and a is a unit-less calibration coefficient defining how
non-point source DOC export responds to runoff. The value of a is set to
0.95, consistent with the original Global NEWS-DOC model of Harrison et
al. (2005). The carbon load Y_DOCwet is then exported as a
diffuse source along the relevant portions of estuary. The estuarine segments
receiving carbon inputs from tidal wetlands are identified using the National
Wetlands Inventory of the US Fish and Wildlife Service (US Fish and Wildlife
Service, 2014). The inputs from those systems are then allocated to the
appropriate grid cell of the model domain using a GIS. The flux calculated is
an annual average that is subsequently partitioned between the four seasons
as a function of the mean seasonal temperature, assumed to be the main
control of the wetland–estuarine exchange. This procedure reflects the
observation that in spring and early summer, DOC export is low as a result of
its accumulation in the salt marshes induced by high productivity (Dai and
Wiegert, 1996), (Jiang et al., 2008). In late summer and fall, the higher
water temperature and greater availability of labile DOC contribute to higher
bacterial remineralization rates in the intertidal marshes (Cai et al., 1999;
Middelburg et al., 1996; Wang and Cai, 2004), which induce an important
export. This marsh production–recycle–export pattern is consistent with the
observed excess DIC signal in the offshore water (Jiang et al., 2013). DIC
export from tidal wetlands is neglected here because it is assumed that OC is
not degraded before reaching the estuarine realm. Although this assumption
may lead to an overestimation of OC export from marshes and respiration in
estuarine water, it will not significantly affect the water pCO2 and
degassing in the estuarine waters because mixing is faster than respiration.
Concentrations at the estuarine mouth
For each estuary, the downstream boundary is located 20 km beyond the mouth
to minimize the bias introduced by the choice of a fixed concentration
boundary condition to characterize the ocean water masses (e.g., Regnier et
al., 1998). This approach also reduces the influence of marine boundary
conditions on the simulated estuarine dynamics, especially for all organic
carbon species whose concentrations are fixed at zero at the marine boundary.
This assumption ignores the intrusion of marine organic carbon into the
estuary during the tidal cycle but allows a focus on the fate of terrigenous
material and its transit through the estuarine filter. DIC concentrations are
extracted from the GLODAP data set (Key et al., 2004), from which ALK and pH
are calculated assuming CO2 equilibrium between coastal waters and the
atmosphere. The equilibrium value is computed using temperature (WOA2009;
Locarnini et al., 2010) and salinity (WOA2009; Antonov et al., 2010) data,
which vary both spatially and temporally. The equilibrium approach is a
reasonable assumption because differences in partial pressure ΔpCO2 between coastal waters and the atmosphere are generally much
smaller (0–250 µatm; Signorini et al., 2013) than those reported
for estuaries (ΔpCO2 in the range 0–10 000 µatm;
Borges and Abril, 2012). Salinity, DO, NO3, DIP, and DSi concentrations
are derived from the World Ocean Atlas (Antonov et al., 2010; Garcia et al.,
2010a, b). NH4 concentrations are set to zero in marine waters. For all
variables, seasonal means are calculated for each grid cell of the boundary.
Modeled (lines) and measured (crosses) salinities in the Delaware
Bay estuary for January (a), February (b), May (c), and June (d). The two lines
correspond to high and low tides.
Biogeochemical indicators
The model outputs (longitudinal profiles of concentration and reaction rates)
are integrated in time over the entire volume or surface of each estuary to
produce the following indicators of the estuarine biogeochemical functioning
(Regnier et al., 2013b): the mean annual NEM,
the air–water CO2 flux (FCO2), the carbon and nitrogen filtering
capacities (CFilt and NFilt), and their corresponding element budgets. The
NEM (molC yr-1) (Caffrey, 2004; Odum, 1956) is defined as the
difference between net primary production (NPP) and total heterotrophic
respiration (HR) on the system scale.
NEM=∫0365∫0LNPPx,t-Raerx,t-Rdenx,t×Bx×Hx,tdxdt
NPP is in mol C m-3 yr-1),
Raer is the aerobic degradation of organic matter
(in mol C m-3 yr-1), and Rden is the denitrification
(in mol C m-3 yr-1) (see Volta et al., 2014, for detailed
formulations). NEM is thus controlled by the production and decomposition
of autochthonous organic matter, by the amount and degradability of organic
carbon delivered by rivers and tidal wetlands, and by the export of
terrestrial and in-situ-produced organic matter to the adjacent coastal zone.
Following the definition of NEM, the trophic status of estuaries can be net
heterotrophic (NEM < 0), when HR exceeds NPP, or net autotrophic
(NEM > 0), when NPP is larger than HR because the burial and
export of autochthonous organic matter exceeds the decomposition of
river-borne material.
The FCO2 (mol C yr-1) is defined as the following.
FCO2=∫0365∫0LRCO2x,t∗BxdxdtRCO2(xt)=-vp(x,t)CO2aq(x,t)-K0x,t×PCO2(x,t)
RCO2 (molC m-2 yr-1) is the rate of exchange in
CO2 at the air–water interface per unit surface area. vp is
the piston velocity (m yr-1) and is calculated according to Regnier et
al. (2002) to account for the effect of current velocity and wind speed,
[CO2(aq)] is the concentration of CO2 in the estuary
(mol m-3), K0 is Henry's constant of CO2 in seawater
(mol m-3 atm-1), and PCO2 is the atmospheric partial
pressure in CO2 (atm).
The carbon-filtering capacity (as a percentage) corresponds to the fraction of the
river-borne supply that is lost to the atmosphere and is defined here as the
ratio of the net outgassing flux of CO2 and the total inputs of C, e.g.,
total carbon expressed as the sum of inorganic and organic carbon species,
both in the dissolved and particulate phases.
CFilt=FCO2∫0365Q×TCrivdt×100
[TC]riv denotes the total concentration of C in each riverine
input.
Fluxes per unit area for FCO2 and NEM, denoted as
FCO2‾ and NEM‾, respectively,
are defined in mol C m-2 yr-1 and are calculated by dividing the
integrated values calculated above by the (idealized) estuarine surface S.
NEM‾=NEMS×1000FCO‾2=FCO2S×1000
Seasonal values for the biogeochemical indicators are calculated using the
same formula as above, but calculating the integral over a seasonal rather
than annual timescale (i.e., 3 months).
Model–data comparison
C-GEM has been specifically designed for an application on a global and/or regional
scale, requiring the representation of a large number of individual and often
data-poor systems. Maximum model transferability and minimum validation
requirements were thus central to the model design process and the ability of
the underlying approach to reproduce observed dynamics with minimal
calibration effort has been extensively tested. The performance C-GEM's
one-dimensional hydrodynamic and transport models using idealized geometries
have been evaluated for a number of estuarine systems exhibiting a wide
variety of shapes (Savenije, 2012). In particular, it has been shown that the
estuarine salt intrusion can be successfully reproduced using the proposed
modeling approach (Savenije 2005; Volta et al., 2014, 2016b). In addition,
C-GEM's biogeochemistry has also been carefully validated for geometrically
contrasting estuarine systems in temperate climate zones. Simulations for the
Scheldt Estuary (Belgium and the Netherlands), a typical funnel-shaped
estuary, were validated through model–data and model–model comparisons (Volta
et al., 2014, 2016a). Furthermore, simulations for the Elbe estuary
(Germany), a typical prismatic shape estuary that drains carbonate terrains
and thus exhibits very high pH, was validated against field data (Volta et
al., 2016a). In addition, carbon budgets calculated using C-GEM for six European estuaries discharging in the North
Sea have been compared with budgets derived from observations (Volta et al., 2016a). Although C-GEM has been specifically designed and
tested for the type of regional application presented here, its
transferability from North Sea to US East Coast estuaries was further
evaluated by assessing its performance in two US East Coast estuaries. First,
the hydrodynamic and transport model was tested for the Delaware Bay (MAR).
The model was forced with the monthly, minimal, and maximal observed
discharges
at Trenton over the period between 1912 and 1985 (UNH/GRDC database; GRDC,
2014). Simulated salinity profiles are compared with salinity observations
(the months with the highest number of
data entries), which were extracted from the UNH/GRDC database from January, February, May, and June. Figure 6
shows that the model captures both the salinity intrusion length and the
overall shape of the salinity profile well. In addition, the performance of
the biogeochemical model and specifically its ability to reproduce pH and
pCO2 profiles was evaluated by a model–data comparison for both the
Delaware Bay (MAR) in July 2003 and the Altamaha River estuary (SAR) in
October 1995. Similar to Volta et al. (2016a), the test systems were chosen
due to their contrasting geometries. The Delaware Bay is a marine-dominated
system characterized by a pronounced funnel shape, while the Altamaha River
has a prismatic estuary characteristic of river-dominated systems (Jiang et
al., 2008). Monthly upstream boundary conditions for nutrients, as well as
observed pH data and calculated pCO2 are extracted from data sets
described in Sharp (2010) and Sharp et al. (2009) for the Delaware and in
Cai and Wang (1998), Jiang et al. (2008), and Cai et al. (1998) for the
Altamaha River estuary. The additional forcings and boundary conditions are
set similarly to the simulation for 2000 (see Tables S2, S3, S4, S5, and S6 in
the Supplement). Figure 7 shows that measured and simulated pH values are in good
agreement with observed pH and observation-derived calculations of
pCO2. In the Delaware Bay, a pH minimum is located around 140 km and
is mainly caused by intense nitrification sustained by large inputs of
NH4 from the Philadelphia urban area, coupled to an intense
heterotrophic activity. Both processes lead to a well-developed pCO2
increase in this area (Fig. 7c). Overall, the longitudinal pCO2
profile of the Delaware estuary is characterized by values close to
equilibrium with the atmosphere in the widest section of the Delaware Bay
(near the estuarine mouth and throughout the first 40 km of the
system), with values above 1200 µatm at 150 km and beyond, where
characteristic salinities are below 5. Although the profile presented here is
simulated using boundary conditions representative of July 2003 and no
pCO2 data were available for validation for this period, a recent
study by Joesoef et al. (2015) reports a similar longitudinal pCO2
profile in July 2013. For the Altamaha River estuary, pH steadily increases
from typical river to typical coastal ocean values (Fig. 7b). In addition,
both observations and model results reveal that outgassing is very intense in
the low-salinity region, with more than a 5-fold decrease in pCO2
between a salinity of 0 and 5 (Fig. 7d).
Longitudinal profiles of pH (top) and pCO2 (bottom) for the
Delaware Bay (left) and Altamaha River estuary (right).
Spatial distribution of spatially averaged value (a) and
integrated value (b) of mean annual FCO2 (red) and -NEM
(blue) along the US East Coast. In panel (a), the notation with over bars
(FCO‾2 and -NEM‾) represents
rates per unit surface. For the sake of the comparison with
FCO‾2, Fig. 8 displays -NEM‾
because the model predicts that all estuaries in this region are net
heterotrophic.
While such local validations allow the assessment of the performance of the model for
a specific set of conditions, the purpose of this study is to capture the
average biogeochemical behavior of the estuaries of the US East Coast. Therefore, in addition to the system-specific validation, published
annually averaged FCO2 estimates for 12 tidal systems located within the
study area collected over the 1994–2006 period are compared to simulated
FCO2 for conditions representative of the year 2000. Overall, simulated
FCO2 values are comparable to values reported in the literature (Table 2).
Although significant discrepancies are observed at the level of individual
systems, the model captures the overall behaviors of
estuaries along the US East Coast in terms of intensity of CO2
evasion rate remarkably well. The model simulates low CO2 efflux (< 5 mol
C m-2 yr-1) for the six systems where such conditions have been
observed, while the five systems for which the CO2 evasion exceeds 10 mol
C m-2 yr-1 are the same in the observations and in the model
runs. The discrepancies at the individual system level likely result from a
combination of factors, including the choice of model processes and their
parametrization, the uncertainties in constraining boundary conditions, and
the limited representability of instantaneous and local observations.
Results and discussion
Spatial variability of estuarine carbon dynamics
Figure 8 presents the spatial distribution of simulated mean annual
FCO‾2 and -NEM‾ (Fig. 8a), as
well as FCO2 and -NEM (Fig. 8b). In general, mean annual
FCO‾2 values are about 30 % larger than mean annual
NEM‾ values, with the exception of six estuaries situated in
the north of the coastal segment. Overall, the NEM‾ is
characterized by smaller system-to-system variability compared to the
FCO‾2 in all regions. In addition, Fig. 8 reveals
distinct differences across the three coastal segments and highlights the
important influence of the estuarine geometry and residence time, as well as
the latitudinal temperature gradient on estuarine carbon cycling.
Overall, FCO‾2 values are the lowest in the NAR
(mean flux = 17.3 ± 16.4 mol C m-2 yr-1; surface-weighted average = 23.1 mol C m-2 yr-1), consistent with
previously reported very low values for small estuaries surrounding the Gulf
of Maine (Hunt et al., 2010, 2011; Table 2). In contrast,
NEM‾ reveals a regional minimum in the NAR
(-51.2 ± 16.6 mol C m-2 yr-1; surface-weighted
average = -52.8 mol C m-2 yr-1). The MAR is characterized
by intermediate values for FCO‾2, with a mean flux
of 26.3 ± 34.6 mol C m-2 yr-1 (surface-weighted
average = 11.1 mol C m-2 yr-1) and the lowest values for
NEM‾ (-15.1 ± 14.2 mol C m-2 yr-1;
surface-weighted average =-7.4 mol C m-2 yr-1). This region
also shows the largest variability in CO2 outgassing compared to the NAR
and SAR, with the standard deviation exceeding the mean
FCO‾2, and individual estimates ranging from 3.9 to
150.8 mol C m-2 yr-1. This variability is mainly the result of
largely variable estuarine surface areas and volumes. Some of the largest
US East Coast estuaries (e.g., Chesapeake and Delaware bays), as well as some of
smallest estuaries (e.g., York River and Hudson River estuaries; Raymond et
al., 1997, 2000), are located in this region (Tables 2 and 4). The maximum
value of 150.8 mol C m-2 yr-1 simulated in the MAR is similar
to the highest FCO2 value reported in the literature
(132.3 mol C m-2 yr-1 for the Tapti estuary in India; Sarma et
al., 2012). The SAR is characterized by the highest mean
FCO‾2
(46.7 ± 33.0 mol C m-2 yr-1; surface-weighted
average = 40.0 mol C m-2 yr-1) and intermediate
NEM‾ (-36.8 ± 24.7 mol C m-2 yr-1;
surface-weighted average = -31.2 mol C m-2 yr-1).
The NAR is characterized by a regional minimum in
FCO‾2 and only contributes 4.6 % to the total
FCO2 of the US East Coast, owing to the small cumulative
surface area available for gas exchange in its 10 estuarine systems. In
contrast, the 18 MAR estuaries, with their large relative contribution to the
total regional estuarine surface area, account for as much as 70.1 % of
the total outgassing. Because of their smaller cumulated surface area
compared to those of the MAR, the 14 SAR estuaries account for merely
25.3 % of the total outgassing despite their regional maximal
FCO‾2. A similar, yet slightly less pronounced
pattern emerges for the NEM‾. The NAR, MAR, and SAR
respectively contribute 13.7, 60.7, and 25.6 % to the total regional net
ecosystem metabolism. The comparatively larger relative contribution of the
NAR to the total NEM as compared to the total FCO2 can be explained
by the importance of the specific aspect ratio for NEM. A larger ratio of
estuarine width B0 and convergence length b corresponds to a more funnel-shaped estuary, while a low ratio corresponds to a more prismatic geometry
(Savenije, 2005; Volta et al., 2014). In the NAR, estuaries are generally
characterized by relatively narrow widths and deepwater depths, thus
limiting the potential surface area for gas exchange with the atmosphere.
However, the relative contribution of each region to the total regional NEM
and FCO2 is largely controlled by estuarine surface area. Figure 9
illustrates the cumulative FCO2 (a) and NEM (b) as a function of the
cumulative estuarine surface areas. The disproportionate contribution of
large estuaries from the MAR translates into a handful of systems (Chesapeake
and Delaware bays and the main tributaries of the former, in particular)
contributing to roughly half of the regional NEM and FCO2, in spite
of relatively low individual rates per unit surface area. However, the
smallest systems (mostly located in the NAR and SAR) nevertheless
contribute a significant fraction to the total regional NEM and
FCO2. The 27 smallest systems merely account for less than 10 % of
the total regional estuarine surface area, yet contribute 38 and 29 % to
the total regional NEM and FCO2, respectively (Fig. 9). This
disproportional contribution can be mainly attributed to their high
individual FCO‾2 and NEM‾ values.
This is illustrated by the average simulated FCO‾2
for all 27 smallest systems (calculated as the sum of each estuarine CO2
outgassing per unit surface area divided by the total number of estuarine
systems), which is significantly higher (30.2 mol C m-2 yr-1) than
its surface-weighted average (14 mol C m-2 yr-1). This therefore
accounts for the disproportionate contribution of very large systems
(calculated as the sum of each estuarine CO2 outgassing divided by the
total estuarine surface area across the region).
The cumulative FCO2 (a) and NEM (b) as functions of the cumulative
estuarine surface area. Systems are sorted by increasing surface area.
Contribution of NEM, nitrification, and supersaturated riverine waters
to the mean annual FCO2 (a). Spatial distribution
of mean annual carbon filtration capacities (CFilt) and export (CExport) along the US East Coast (b).
Following the approach used in Regnier et al. (2013b), the contribution of
each biogeochemical process to FCO2 is assessed by evaluating their
individual contribution to DIC and ALK changes, taking into account the local
buffering capacity of an ionic solution when TA and DIC are changing due to
internal processes, but ignoring advection and mixing (Zeebe and Wolf-Gladrow
2001). In the present study, we quantify the effect of the NEM on the
CO2 balance, which is almost exclusively controlled by aerobic
degradation rates because the contributions of denitrification and NPP to the
net ecosystem balance are small. Nitrification, a process triggered by the
transport and/or production of NH4 in oxygenated waters, favors
outgassing through its effect on pH, which shifts the acid–base equilibrium
of carbonate species and increases the CO2 concentration. The
contribution of supersaturated riverine waters to the overall estuarine
CO2 dynamics is calculated as the difference between all the other processes
creating or consuming CO2. Figure 10a presents the contribution of the
annually integrated NEM, nitrification, and evasion of supersaturated, DIC-enriched riverine waters to the total outgassing for each system, as well as
for individual regions of the domain. The calculation of these annual values
is based on the sum of the seasonal fluxes. Model results reveal that,
regionally, the NEM supports about 50 % of the estuarine CO2
outgassing, while nitrification and riverine DIC inputs sustain about 17 and
33 % of the CO2 emissions, respectively. The relative significance
of the three processes described above shows important spatial variability.
In the NAR, oversaturated riverine waters and NEM respectively sustain 50 and
44 % of the outgassing within the subregion, while nitrification is of
minor importance (6 %). In the MAR, the contribution of riverine DIC
inputs is significantly lower (∼ 30 %) and the main contribution to
the outgassing is NEM (∼ 50 %); nitrification accounts for
slightly less than 20 % of the outgassing. In the SAR, the riverine
contribution is even lower (∼ 20 %), and the outgassing is mainly
attributed to the NEM (∼ 55 %) and nitrification
(∼ 25 %). Therefore, although the model results reveal significant
variability across individual systems, a clear latitudinal trend in the
contribution to the total FCO2 emerges from the analysis. The
importance of oversaturated riverine water decreases from north to south,
while NEM and nitrification increase along the same latitudinal gradient.
The increasing relative importance of estuarine biogeochemical processes over
riverine DIC inputs as drivers of FCO2 along the north–south gradient
is largely driven by increasing temperatures from north to south, especially
in the SAR region (Table S1).
Yearly averaged surface area (S), freshwater discharge (Q),
residence time (Rt), FCO2, and NEM of all simulated estuaries.
Long
Lat
S
Q
Rt
FCO‾2
NEM‾
FCO2
NEM
∘
∘
km2
m3 s-1
days
mol C m-2 yr-1
mol C m-2 yr-1
106 mol C yr-1
106 mol C yr-1
NAR
-67.25
44.75
7
38.5
15
3.7
-37.4
27
-270
-67.25
45.25
12
73.6
15
6.0
-56.7
71
-666
-67.25
45.25
12
73.6
15
13.8
-56.6
162
-666
-67.75
44.75
3
68.5
4
6.7
-63.5
23
-221
-68.25
44.75
14
69.5
19
4.1
-56.2
58
-791
-68.75
44.75
89
309.9
23
27.4
-58.2
2431
-5163
-69.75
44.25
50
626.6
5
32.3
-74.4
1607
-3703
-70.25
43.75
3
25.8
10
2.1
-21.0
7
-71
-70.75
41.75
288
103.6
958
5.0
-4.0
1428
-1146
-70.75
42.25
63
210.7
40
16.2
-32.9
1025
-2081
-70.75
42.75
17
105.8
3
56.3
-69.0
943
-1155
MAR
-70.75
43.25
31
29.9
11
21.6
-37.4
662
-1146
-71.25
41.75
257
28.2
808
3.9
-2.5
997
-650
-71.75
41.25
21
112.4
4
35.2
-32.6
726
-672
-72.75
40.75
20
25.4
62
30.7
-21.1
623
-430
-72.75
41.25
10
142.5
2
150.8
-36.9
1578
-386
-72.75
41.75
55
476.6
3
55.9
-45.7
3088
-2523
-73.25
40.75
19
26.8
56
31.4
-28.4
608
-550
-74.25
40.75
1192
608.2
126
15.5
-11.8
18432
-14047
-75.25
38.25
399
80.5
172
13.9
-5.0
5558
-2016
-75.25
38.75
354
31.8
357
7.5
-3.0
2659
-1076
-75.25
39.75
1716
499.0
221
10.0
-7.8
17072
-13439
-75.75
39.25
224
18.3
434
7.5
-2.9
1685
-640
-76.25
39.25
3427
717.1
352
8.1
-5.1
27646
-17352
-76.75
37.25
586
272.3
74
15.0
-10.4
8810
-6084
-76.75
37.75
154
36.3
163
10.7
-6.6
1654
-1023
-76.75
39.25
59
71.2
29
48.6
-34.6
2862
-2038
-77.25
38.25
206
30.2
268
6.1
-3.3
1265
-676
-77.25
38.75
568
259.2
118
16.7
-10.8
9488
-6134
SAR
-78.25
34.25
48
167.4
7
122.5
-62.4
5916
-3015
-79.25
33.25
47
56.3
42
43.4
-36.5
2056
-1728
-79.25
33.75
45
291.4
8
85.1
-78.7
3843
-3551
-79.75
33.25
25
33.8
15
37.9
-32.8
956
-828
-80.25
32.75
25
31.0
50
48.8
-42.5
1214
-1057
-80.25
33.25
92
75.5
61
62.7
-61.2
5769
-5625
-80.75
32.25
71
21.1
182
12.9
-7.0
918
-501
-80.75
32.75
164
63.1
95
20.6
-11.5
3372
-1879
-81.25
31.75
92
71.7
45
25.7
-20.9
2361
-1926
-81.25
32.25
130
379.8
11
51.7
-39.2
6732
-5097
-81.75
30.75
34
18.7
61
17.5
-14.7
602
-505
-81.75
31.25
130
17.7
294
5.5
-4.0
713
-523
-81.75
31.75
56
350.5
4
72.7
-67.4
4068
-3770
Contrasting patterns across the three regions can also be observed with respect
to carbon-filtering capacities, CFilt (Fig. 10b). In the NAR, over 90 %
of the riverine carbon flux is exported to the coastal ocean. However, in the
MAR, the high efficiency of the largest systems in processing organic carbon
results in a regional CFilt that exceeds 50 %. This contrast between
the NAR and the MAR and its potential implication for the carbon dynamics of
the adjacent continental shelf waters has already been discussed by Laruelle
et al. (2015). In the NAR, short estuarine residence results in a much lower
removal of riverine carbon by degassing compared to the MAR. Laruelle et
al. (2015) suggested that this process could contribute to the weaker
continental shelf carbon sink adjacent to the NAR, compared to the MAR. In
the SAR, most estuaries remove between 40 and 65 % of the carbon inputs.
The high temperatures observed and the resulting accelerated biogeochemical
process rates in this region favor the degradation of organic matter and
contribute to the increase in the estuarine capacity for filtering carbon. However,
in the SAR, a large fraction of the OC loads is derived from adjacent salt
marshes located along the estuarine salinity gradients, thereby reducing the
overall residence time of OC within the systems. The filtering capacity of
the riverine OC alone, which transits through the entire estuary, would thus
be higher than the one calculated here. As a consequence, the highest C retention
rates are expected in warm tidal estuaries devoid of salt marshes or
mangroves (Cai, 2011).
Seasonal variability in estuarine carbon dynamics
Carbon dynamics in estuaries of the US East Coast not only show a marked
spatial variability but also vary on the seasonal timescale. Table 5
presents the seasonal distribution of NEM and FCO2 for each
subregion. In the NAR, a strong seasonality is simulated for the NEM and
the summer period contributes more than a third to the annually integrated
value. The outgassing reveals a lower seasonal variability and is only
slightly higher than summer outgassing during fall and lower during spring.
In the MAR, summer contributes more to the NEM (> 28 % of
the yearly total) than any other season, but seasonality is less pronounced
than in the NAR. Here, FCO2 values are highest in winter and particularly low
during summer. In the SAR, summer accounts for 30 % of the NEM, while
spring contributes 21 %. FCO2 is relatively constant throughout
the year, suggesting that seasonal variations in carbon processing decrease
towards the lower latitudes in the SAR. This is partly related to the low
variability in river discharge throughout the year in lower latitudes
(Table S1). In riverine-dominated systems with low residence times, such as,
for instance, the Altamaha River estuary, the CO2 exchange at the
air–water interface is mainly controlled by the river discharge because the
time required to degrade the entire riverine organic matter flux exceeds the
transit time of OC through the estuary. Therefore, the riverine-sustained
outgassing is highest during the spring peak discharge periods. In contrast,
the seasonal variability in FCO2 in long-residence, marine-dominated
systems with large marsh areas (e.g., Sapelo and Doboy Sound) is essentially
controlled by seasonal temperature variations. Its maximum is reached during
summer when marsh plants are dying and decomposing, as opposed to spring when
marshes are in their productive stage (Jiang et al., 2008). These contrasting
seasonal trends have already been reported for different estuarine systems in
Georgia, such as the Altamaha Sound, the Sapelo Sound, and the Doboy Sound
(Cai, 2011). On the scale of the entire US East Coast, the seasonal
trends in NEM reveal a clear maximum in summer and minimal values during
autumn and winter. The seasonality of FCO2 is much less pronounced
because the outgassing of oversaturated riverine waters throughout the year
contributes to a large fraction of the FCO2 and dampens the effect of
the temperature-dependent processes (NEM and denitrification). In our
simulations, the competition between temperature and river discharge is the
main driver of the seasonal estuarine carbon dynamics. When discharge
increases, the carbon loads increase proportionally and the residence time
within the system decreases, consequently limiting an efficient degradation
of organic carbon input fluxes. In warm regions like the SAR, the temperature
is sufficiently high all year round to sustain high C processing rates and
this explains the reduced seasonal variability in NEM.
Seasonal contribution to FCO2 and NEM in each subregion. The
seasons displaying the highest percentages are indicated in bold. Winter is
defined as January, February, and March; spring is defined as April, May, and June, and so
on.
Region
S
NEM
Winter
Spring
Summer
Fall
FCO2
Winter
Spring
Summer
Fall
km2
mol C yr-1
%
%
%
%
mol C yr-1
%
%
%
%
NAR
558
-16.3 109
14.7
21.2
37.0
27.2
7.2 109
26.3
18.9
26.5
28.3
MAR
9298
-72.2 109
21.9
25.9
28.8
23.4
108.3 109
29.8
23.3
20.7
26.2
SAR
959
-30.5 109
24.6
20.9
30.3
24.2
39.2 109
26
23.4
27
23.6
Regional carbon budget: a comparative analysis
The annual carbon budget for the entire US East Coast is summarized in
Fig. 11a. The total carbon input to estuaries along the US East Coast
is 4.6 Tg C yr-1, of which 42 % arrives in organic form and
58 % in inorganic form. Of this total input, salt marshes contribute
0.6 Tg C yr-1, which corresponds to about 14 % of the total carbon
loads and 32 % of the organic loads in the region. The relative
contribution of the salt marshes to the total carbon input increases towards
low latitudes and is as high as 60 % in the SAR region. Model results
suggest that 2.7 Tg C yr-1 is exported to the continental shelf
(25 % as TOC and 75 % as DIC), while 1.9 Tg C yr-1 is emitted
to the atmosphere. The overall carbon-filtering capacity of the region thus
equals 41 % of the total carbon entering the 42 estuarine systems
(river plus salt marshes). Because of the current lack of a benthic module in
C-GEM, the water column carbon removal occurs entirely in the form of
CO2 outgassing and does not account for the potential contribution of
carbon burial in sediments. The estimated estuarine carbon retention
presented here is thus likely a lower-bound estimate. Reported for the modeled
surface area of the region, the total FCO2 value of 1.9 Tg C yr-1
translates into a mean air–water CO2 flux of about
14 mol C m-2 yr-1. This value is slightly higher than the
estimate of 10.8 mol C m-2 yr-1 calculated by Laruelle et
al. (2013) on the basis of local FCO‾2 estimates
assumed to be representative of yearly averaged conditions (see Sect. 2.1).
The latter was calculated as the average of 13 annual
FCO‾2 values reported in the literature (Table 2),
irrespective of the size of the systems. This approach is useful and widely
used to derive regional and global carbon budgets (Borges et al., 2005;
Laruelle et al., 2010; Chen et al., 2013). However, this approach may lead to
potentially significant errors (Volta et al., 2016a) due to the uncertainty
introduced by the spatial interpolation of local measurements to large
regional surface areas, while useful and widely used for deriving regional and
global carbon budgets.
Regional C budgets are sparse. To our knowledge, the only other published
regional assessment of estuarine carbon and CO2 dynamics comes from
a relatively well-studied region: the estuaries flowing into the North Sea in
western Europe (Fig. 11b). This budget was calculated using a similar
approach (Volta, 2016a) and thus provides an ideal opportunity for a
comparative assessment of C cycling in these regions. However, it is
important to note that there are also important differences in the applied
model approaches and those differences should be taken into account when
comparing the derived budgets. In particular, the northwestern European study is based
on a simulation of only the six largest systems (Elbe, Scheldt, Thames, Ems,
Humber, and Weser), accounting for about 40 % for the riverine carbon
loads of the region. It assumes that the intensity of carbon processing and
evasion in all other smaller estuaries discharging into the North Sea
(16 % of the carbon loads) can be represented by the average of the six
largest system simulation results. In addition, the Rhine–Meuse system, which
alone accounts for 44 % of the carbon riverine inputs of the region, was
treated as a passive conduit with respect to carbon due to its very short
freshwater residence time (Abril et al., 2002). The contribution of
salt marshes to the regional carbon budget was also ignored because their
total surface area is much smaller than along the US East Coast (Regnier et
al., 2013b). Another important difference is the inclusion of seasonality in
the present study, while the budget calculated for the North Sea is derived
from yearly average conditions (Volta et al., 2016a).
Overall, although both regions receive similar amounts of C from rivers (4.6
and 5.9 Tg C yr-1 for the US East Coast and the North Sea,
respectively), they reveal significantly different carbon-filtering capacities.
While the estuaries of the US East Coast filter 41 % of the
riverine TC loads, those from the North Sea only remove 8 % of the
terrestrially derived material. This is partly due to the large amounts of
carbon transiting through the passive Rhine–Meuse system. The regional
filtering capacity is higher (15 %) when this system is excluded from the
analysis. However, even when neglecting this system, significant differences
in filtering efficiencies between both regions remain. The FCO2 value from the
North Sea estuaries (0.5 Tg C yr-1) is significantly lower than the
1.9 Tg C yr-1 computed for the US East Coast. The reason for
the lower evasion rate in northwestern European estuaries is essentially twofold.
First, the total cumulative surface area available for gas exchange is
significantly lower along the North Sea, in spite of comparable flux
densities calculated using the entire estuarine surface areas of both regions
(14 and 23 mol C m-2 yr-1 for the US East Coast and the
North Sea, respectively). Second, although the overall riverine carbon loads
are comparable in both regions (Fig. 11), the ratio of organic to inorganic
matter input is much lower in the North Sea area because the regional
lithology is dominated by carbonate rocks and mixed sediments that contain
carbonates (Dürr et al., 2005; Hartmann et al., 2012). As a consequence,
TOC represents less than 20 % of the riverine loads and only 10 % of
the carbon exported to the North Sea. In both regions, however, the increase
in the ratio of inorganic to organic carbon between input and output is
sustained by a negative NEM (Fig. 11). Although the ratios themselves may
significantly vary from one region of the world to another, as evidenced by
these two studies, a NEM-driven increase in the inorganic fraction within
carbon load along the estuarine axis is consistent with the global estuarine
carbon budget proposed by Bauer et al. (2013). On the US East Coast,
the respiration of riverine OC within the estuarine filter is partly
compensated for by OC inputs from marshes and mangroves in such a way that the
input and export IC/OC ratios are closer than in the North Sea region.
Annual carbon budget of the estuaries of the US East Coast
(a) and of the coast of the North Sea (b; modified from Volta et al.,
2016a).
Scope of applicability and model limitations
Complex multidimensional models are now increasingly applied to
quantitatively explore carbon and nutrient dynamics along the land–ocean
transition zone on seasonal and even annual timescales (Garnier et al.,
2001; Arndt et al., 2007, 2009; Arndt and Regnier, 2007; Mateus et al.,
2012). However, the application of such complex models remains limited to
individual, well-constrained systems due their high data requirements and
computational demand resulting from the need to resolve important physical,
biogeochemical, and geological processes on relevant temporal and spatial
scales. The one-dimensional, computationally efficient model C-GEM has been
specifically designed to reduce data requirements and computational demand
and to enable regional and/or global scale applications (Volta et al., 2014,
2016a). However, such a low data demand and computational efficiency
inevitably requires simplification. The following paragraphs critically
discuss these simplifications and their implications.
Spatial resolution
Here, C-GEM is used with a 0.5∘ spatial resolution. While this
resolution captures the features of large systems, it is still very coarse
for relatively small watersheds, such as those of the St. Francis River,
Piscataqua River, May River, or Sapelo River. For instance, the five
estuaries reported by Hunt et al. (2010, 2011, see sect. 2.6) are all small
systems covered by the same watershed at a 0.5∘ resolution. Only
watersheds whose areas span several grid cells can be properly identified and
represented (i.e., Merrimack or Penobscot with six and nine cells, respectively).
Hydrodynamic and transport model
C-GEM is based on a theoretical framework that uses idealized geometries and
significantly reduces data requirements. These idealized geometries are fully
described by three easily obtainable geometrical parameters (B, B0, and H). The
model thus approximates the variability in estuarine width and cross section
along the longitudinal axis through a set of exponential functions. A
comprehensive sensitivity study (Volta et al., 2014) has shown that
integrated process rates are generally sensitive to changes in these
geometrical parameters because of their control on estuarine residence times.
For instance, Volta et al. (2014) demonstrated that the NEM is particularly
sensitive to the convergence length. Similarly, the use of constant depth
profile may lead to variations of about 10 % in NEM (Volta et al., 2014).
Nevertheless, geometrical parameters are generally easy to constrain,
especially in well-monitored regions such as the US East Coast. Here, all
geometrical parameters are constrained on the basis of observed estuarine
surface areas and average water depths. In addition, the model also accounts
for the slope of the estuarine channel. This approach ensures that simulated
estuarine surface areas, volumes, and thus residence times are in good
agreement with those of the real systems and it minimizes uncertainties
associated with the physical setup.
In addition, the one-dimensional representation of the idealized estuarine
systems does not resolve two- or three-dimensional circulation features
induced by complex topography and density-driven circulation. While C-GEM
performs well in representing the dominant longitudinal gradients, its
applicability to branched systems or those with aspect ratios for which a
dominant axis is difficult to identify (e.g., Blackwater estuary, UK; Pearl
River estuary, China; Tagus estuary, Portugal; Bay of Brest, France) is
limited.
Biogeochemical model
Although the reaction network of C-GEM accounts for all processes that
control estuarine FCO2 (Borges and Abril, 2012; Cai, 2011), several,
potentially important processes, such as benthic–pelagic exchange processes,
phosphorous sorption–desorption and mineral precipitation, a more complex
representation of the local phytoplankton community, grazing by higher
trophic levels, or multiple reactive organic carbon pools, are not included.
Although these processes are difficult to constrain and their importance for
FCO2 is uncertain, the lack of their explicit representations induces
uncertainties in CFilt. In particular, the exclusion of benthic processes
such as organic matter degradation and burial in estuarine sediments could
result in an underestimation of CFilt. However, because very little is known
on the long-term fate of organic carbon in estuarine sediments, setting up
and calibrating a benthic module proves a difficult task. Furthermore, to a
certain degree, model parameters (such as organic matter degradation and
denitrification rate constant) implicitly account for benthic dynamics. We
nonetheless acknowledge that, by ignoring benthic processes and burial in
particular, our estimates for the estuarine carbon filtering may be
underestimated, particularly in the shallow systems of the SAR.
Biogeochemical model parameters for regional and global applications are
notoriously difficult to constrain (Volta et al., 2016b). Model parameters
implicitly account for processes that are not explicitly resolved and their
transferability between systems is thus limited. In addition, published
parameter values are generally biased towards temperate regions in
industrialized countries (Volta et al., 2016b). A first-order estimation of
the parameter uncertainty associated with the estuarine carbon removal
efficiency (CFilt) can be extrapolated from the extensive parameter
sensitivity analyses carried out by Volta et al. (2014, 2016b). These
comprehensive sensitivity studies on end-member systems have shown that the
relative variation in CFilt when a number of key biogeochemical parameters
are varied by 2 orders of magnitude varies by ±15 % in prismatic
(short residence time on the order of days) to ±25 % in funnel-shaped
(long residence time) systems. Thus, assuming that uncertainty increases
linearly between those bounds as a function of residence time, an uncertainty
estimate can be obtained for each of our modeled estuaries. With this simple
method, the simulated regional CFilt value of 1.9 Tg C yr-1 would be associated
with an uncertainty range comprised between 1.5 and 2.2 Tg C yr-1. Our
regional estuarine CO2 evasion estimate is thus reported with moderate
confidence. Furthermore, in the future, this uncertainty range could be
further constrained using statistical methods such as Monte Carlo simulations
(e.g., Lauerwald et al., 2015).
Boundary conditions and forcings
In addition, simulations are only performed for climatological means over the
period 1990–2010 without resolving interannual and secular variability.
Boundary conditions and forcings are critical as they place the modeled
system in its environmental context and drive transient dynamics. However,
for regional applications, temporally resolved boundary conditions and
forcings are difficult to constrain. C-GEM places the lower boundary
condition 20 km from the estuarine mouth into the coastal ocean and the
influence of this boundary condition on simulated biogeochemical dynamics is
thus limited. At the lower boundary condition, direct observations for
nutrients and oxygen are extracted from databases such as the World Ocean
Atlas (Antonov et al., 2010). However, lower boundary conditions for OC and
pCO2 (zero concentration for OC and assumption of pCO2
equilibrium at the seaside) are simplified. This approach does not allow
the address of the additional complexity introduced by biogeochemical dynamics in
the estuarine plume (see Arndt et al., 2011). However, these dynamics only play a
secondary role in the present study, which focuses on the role of the
estuarine transition zone in processing terrestrially derived carbon.
Constraining upper boundary conditions and forcings is thus more critical.
Here, C-GEM is forced by seasonally averaged conditions for Q, T, and
radiation. To date, Global NEWS only provides yearly averaged conditions for a
number of upper boundary conditions (Seitzinger et al., 2005; Mayorga et al.,
2010), representative of the year 2000. Simulations are thus only partly
transient (induced by seasonality in Q, T, and radiation) and do not
resolve short-lived events such as storms or extreme drought conditions. In
addition, direct observations of upper boundary conditions are rarely
available, in particular over seasonal or annual timescales. For the US East Coast estuaries, direct observations are only available for O2,
chlorophyll a, DIC, and ALK. For DIC and alkalinity, boundary conditions are
constrained by calculating the average concentration over a period of about
3 decades. In addition, observational data are extracted at the station
closest to the model's upper boundary, which might still be located several
kilometers upstream or downstream of the model boundary. Upper boundary
conditions of POC, DOC, DIN, DIP, and DSi are extracted from Global NEWS and are thus
model-derived. As a consequence, our results are thus intimately dependent on
the robustness of the Global NEWS predictions. These values are usually only
considered as robust estimates for watersheds larger than ∼ 10 cells
(Beusen et al., 2005), which only correspond to 13 of the 42 estuaries
modeled in this study.
Model–data comparison
The generic nature of the applied model approach renders a direct validation
of model results on the basis of local and instantaneous observational data
(e.g., longitudinal profiles) difficult. In particular the applications of
seasonally and/or annually averaged or model-deduced boundary conditions, which are
not likely representative of these long-term average conditions, do not lend
themselves well to comparison with punctual measurements. Therefore, model
performance is evaluated on the basis of spatially aggregated estimates
(e.g., regional FCO2 estimates based on local measurements) rather than
system-to-system comparisons with longitudinal profiles from specific days.
However, note that the performance of C-GEM has been intensively tested by
specific model–data comparisons for a number of different systems (e.g.,
Volta et al., 2014, 2016a) and we are thus confident of its predictive
capabilities.
Despite the numerous simplifying assumptions inevitably required for such a
regional assessment of carbon fluxes along the land–ocean continuum, the
presented approach does nevertheless provide an important step forward in
evaluating the role of land–ocean transition systems in the global carbon
cycle. It provides a first robust estimate of carbon dynamics based on a
theoretically well-founded and carefully tested, spatially and temporally
resolved model approach. This approach provides novel insights that go
beyond those gained through traditionally applied zero-salinity methods or
box model approaches. In addition, it also highlights critical variables and
data gaps and thus helps guide efficient monitoring strategies.
System-scale integrated biogeochemical
indicators expressed as functions of the depth-normalized residence time
expressed as the ratio of the estuarine surface S and the river discharge Q
for all seasons. Panels (b), (d), and (e) represent -NEM, FCO2, and CFilt,
respectively. Panels (a) and (c) represent -NEM and FCO2 normalized by a
temperature Q10 function. Black lines are the best-fitted linear
regressions obtained using all the points. Grey lines are best fit using only
the estuaries from the MAR and SAR regions.
Towards predictors of the estuarine carbon processing
The mutual dependence of geometry and transport on each other in tidal estuaries and,
ultimately, their biogeochemical functioning (Savenije, 1992; Volta et al.,
2014), allows the relation of easily extractable parameters linked to their shape or
hydraulic properties to biogeochemical indicators. In this section, we
explore the relationships between such simple physical parameters and
indicators of the estuarine carbon processing NEM‾,FCO‾2, and CFilt. In order to account for the effect
of temperature on C dynamics, -NEM‾ and
FCO‾2 are also normalized to the same temperature
(arbitrarily chosen to be 0 ∘C). These normalized values are obtained by
dividing -NEM‾ and FCO‾2 by a
Q10 function f(T) (see Volta et al., 2014). This procedure allows
the consideration of the exponential increase in the rate of several temperature-dependent processes contributing to the NEM (photosynthesis, organic
carbon degradation, etc.). Applying the same normalization to
-NEM‾ and FCO‾2 is a way of
testing how intimately linked NEM and FCO2 are in estuarine systems.
Indeed, linear relationships relating one to the other have been reported
(Mayer and Eyre, 2012). The three indicators are then investigated as a
function of the ratio between the estuarine surface S and the seasonal
river discharge Q. The surface area is calculated from the estuarine width
and length, as described by Eq. (2), in order to use a parameter that is
potentially applicable to other regions for which direct estimates of the
real estuarine surface area are not available. Since the freshwater residence
time of a system is obtained by dividing volume by river discharge, the S/Q
ratio is also intimately linked to residence time. Here, we choose to exclude
the estuarine depth from the analysis because this variable cannot be easily
quantified from maps or remote sensing images and would thus compromise the
applicability of a predictive relationship on the global scale. However, from
dimensional analysis, S/Q can be viewed as a water residence time
normalized to meter depth of water. As shown by Eq. (3), S only requires
constraining B0 and width convergence length b, two parameters that can
readily be extracted from the Google Earth engine. Global databases of river
discharges, as for instance RivDIS (Vörösmarty et al., 1996), are also
available in such a way that the S/Q ratio can potentially be extracted for
all estuaries around the globe.
Figure 12a reveals that small values of S/Q are associated with the most
negative NEM‾/f(T). The magnitude of the
NEM‾ then exponentially decreases with increasing
values of S/Q. Estuaries characterized by small values of S/Q are mainly
located in the NAR subregion and correspond to small surface area, and thus
short-residence-time systems. It is possible to quantitatively relate
-NEM‾/f(T) and S/Q through a power-law function
(y = 25.85 x-0.64 with r2= 0.82). The coefficient of
determination remains the same when excluding estuaries from the NAR region
and the equation itself is not significantly different, although those
estuaries on their own do not display any statistically significant trend
(Table 6). The decrease in the intensity of the net ecosystem metabolism in
larger estuaries (Fig. 8), characterized by high S/Q ratios, can be related
to the extensive consumption of the organic matter pool during its transit
through the estuarine filter. However, when reported for the entire surface
area of the estuary, larger systems (with high values of S/Q) still reveal
the most negative surface-integrated NEM (Fig. 12b). It can also be noted
that some estuaries from the NAR region display very low values of -NEM.
These data points correspond to fall and winter simulations for which the
temperature was relatively cold (< 5 ∘C) and biogeochemical
processing was very low.
Regressions and associated coefficient of determination between the
depth-normalized residence time (S/Q) and -NEM‾/f(T),
FCO‾2/f(T), and CFilt.
Region
-NEM‾/f(T)
FCO‾2/f(T)
CFilt
NAR
y = 27.84 x-0.17 r2= 0.11
y = 6.07 x0.00 r2= 0.00
y = 15.08 log10(x)+ 4.86 r2= 0.40
MAR
y = 26.03 x-0.63 r2= 0.86
y = 34.36 x-0.58 r2= 0.68
y = 40.46 log10(x)+ 9.60 r2= 0.70
SAR
y = 28.36 x-0.71 r2= 0.76
y = 32.82 x-0.66 r2= 0.80
y = 23.19 log10(x)+ 43.71 r2= 0.46
MAR + SAR
y = 25.85 x-0.64 r2= 0.82
y = 31.64 x-0.58 r2= 0.70
y = 33.30 log10(x)+ 24.88 r2= 0.57
NAR + MAR + SAR
y = 28.98 x-0.66 r2= 0.82
y = 12.98 x-0.33 r2= 0.30
y = 40.64 log10(x)+ 11.84 r2= 0.70
The overall response of FCO‾2/f(T) to S/Q is
comparable to that of -NEM‾/f(T) (Fig. 12c), with lower
values of FCO‾2 observed for high values of S/Q.
However, for S/Q<3 days m-1, the
FCO‾2 values are very heterogeneous and contain
many, low-FCO‾2 outliers from the NAR region. These
data points generally correspond to low-water-temperature conditions, which
keep pCO2 low, even if the system internally generates enough CO2
via NEM. Thus, the well-documented correlation between
NEM‾ and FCO‾2 (Maher and
Eyre, 2012) does not seem to hold for systems with very short residence
times. For systems with S/Q > 3 days m-1, we obtain a
regression FCO2= -0.64 × NEM + 5.96 with r2 = 0.46, which compares well with the relation
FCO2= -0.42 × NEM + 12 proposed by Maher and
Eyre (2012), who used 24 seasonal estimates from small Australian estuaries.
However, our results suggest that this relationship cannot be extrapolated to
small systems such as those located in the NAR. Figure 12d, which reports
non-normalized FCO2, reveals a monotonous increase in FCO2 with
S/Q. This suggests that, unlike the NEM for which the normalization by a
temperature function allow the explanation of most of the variability, FCO2
is mostly controlled by the water residence time within the system. Discharge
is the main FCO2 driver in riverine-dominated systems, while
interactions with marshes drive the outgassing in marine-dominated
systems surrounded by marshes. Net aquatic biological production (NEM being
negative or near 0) in large estuaries (with large S/Q) is another
important reason for low FCO2 in such systems. For example, despite the
higher CO2 degassing flux in the upper estuary of the Delaware, strong
biological CO2 uptake in the mid-bay and near zero NEM in the lower bay
result in a much lower FCO2 for the entire estuary (Joesoef et al.,
2015). In systems with S/Q<3 days m-1, the short-residence-time prevents the excess CO2 of oversaturated water from being entirely
exchanged with the atmosphere and simulations reveal that the estuarine
waters are still oversaturated in CO2 at the estuarine mouth. Thus, the
inorganic carbon, produced by the decomposition of organic matter, is not
outgassed within the estuary but exported to the adjacent continental shelf
waters. This result is consistent with the observation-based hypothesis of
Laruelle et al. (2015) for the NAR estuaries. As a consequence of the
distinct behavior of short residence time systems, the coefficient of
determination of the best-fitted power-law function relating
FCO‾2 and S/Q is only significant if NAR systems
are excluded (y = 31.64 x-0.58 with r2= 0.70). This thus
suggests that such relationships (as well as that proposed by Maher and Eyre,
2012) cannot be applied to any system but only those for which
S/Q>3 day m-1.
Finally, Fig. 12e reports the simulated mean seasonal carbon-filtering
capacities as functions of the depth-normalized residence time. Not
surprisingly, and in overall agreement with previous studies on nutrient
dynamics in estuaries (Nixon et al., 1996), the carbon-filtering capacity
increases with S/Q. The best statistical relation between CFilt and S/Q
is obtained when including all three regions, resulting in r2= 0.70
(y = 40.64 log10(x)+ 11.84). Very little C removal occurs in
systems with S/Q<1 day m-1. For systems characterized by
longer depth-normalized residence times, CFilt increases regularly and
reaches 100 % for S/Q>100 day m-1. Such high values
are only observed for very large estuaries from the MAR region (Delaware and
Chesapeake bays); the majority of our systems had an S/Q range between 1
and 100 day m-1. The quantitative assessment of estuarine filtering
capacities is further complicated by the complex interplay of estuarine and
coastal processes. Episodically, marked spatial variability in concentration
gradients near the estuarine mouth may lead to a reversal of net material
fluxes from coastal waters into the estuary (Regnier at al., 1998; Arndt et
al., 2011). Our results show that this feature is particularly significant for
estuaries with a large width at the mouth and short convergence length
(funnel-shaped or bay-type systems). These coastal nutrient and carbon
inputs influence the internal estuarine C dynamics and lead to filtering
capacities that can exceed 100 %. This feature is particularly
significant in summer, when riverine inputs are low and the marine material
is intensively processed inside the estuary.
Previous work investigated the relationship between freshwater residence
time and nutrient retention (Nixon et al., 1996; Arndt et al., 2011;
Laruelle, 2009). These studies, however, were constrained by the scarcity of
data. For instance, the pioneering work of Nixon et al. (1996) only relied on
a very limited number (< 10) of quite heterogeneous coastal systems,
all located along the North Atlantic. Here, our modeling approach allows us
to generate 168 (42 × 4) data points, each representing a
system-scale biogeochemical behavior. Together, this database spans the
entire spectrum of estuarine settings and climatic conditions found along the
US East Coast. In addition, the ratio S/Q used as a master variable
for predicting temperature-normalized -NEM‾,
FCO‾2, and CFilt only requires a few easily
accessible geometric parameters (B0, b, and L) and an estimate of the
river discharge. While it is difficult to accurately predict
FCO‾2 for small systems such as those located in the
NAR region, the relationships found are quite robust for systems in which
S/Q > 3 days m-1. Most interestingly, CFilt values
reveal a significant correlation with S/Q and could be used in combination
with global riverine carbon delivery estimates such as Global NEWS 2 (Mayorga
et al., 2010) to constrain the estuarine CO2 evasion and the carbon
export to the coastal ocean on the continental and global scales.