Introduction
Over the past decade, several syntheses have highlighted the
significant contribution of the land–ocean aquatic continuum (LOAC) to the
global atmospheric CO2 budget (Cole et al., 2007; Battin et al., 2009;
Mackenzie et al., 2012; Bauer et al., 2013; Ciais et al., 2013; Raymond et
al., 2013; Regnier et al., 2013a). In a recent review, Regnier et al. (2013a)
proposed that inland waters (streams, rivers and lakes) and estuaries outgas
1.1 and 0.25 PgC yr-1, respectively, while continental shelf seas take
up 0.2 PgC yr-1. However, CO2 data are too sparse and unevenly
distributed to provide global coverage and large uncertainties remain
associated with these estimates. The inland water outgassing could for instance
reach 2.1 PgC yr-1 with 86 % coming from streams and rivers
(Raymond et al., 2013), a value which is about twice that reported in Regnier
et al. (2013a) and in the IPCC Fifth Assessment Report (Ciais et al.,
2013). The most recent global budgets for the estuarine CO2 source and
the continental shelf CO2 sink also reveal significant discrepancies,
both falling within the 0.15–0.4 PgC yr-1 range (Laruelle et al.,
2010, 2013; Cai, 2011; Bauer et al., 2013; Dai et al., 2013). None of these
estimates, however, fully resolves the seasonality in CO2 fluxes because
temporal coverage of the global data is insufficient. Complex seasonal
dynamics of CO2 exchanges between the atmosphere and individual
components of the LOAC have been reported in previous studies which have
highlighted the potential importance of the intra-annual variability for
local and regional CO2 budgets (e.g., Kempe, 1982; Frankignoulle et al.,
1998; Jones and Mulholland, 1998; Degrandpré et al., 2002; Thomas and
Schneider, 1999; Wallin et al., 2011; Regnier et al., 2013a; Rawlins et al.,
2014). Here, we extend the analysis to the sub-continental scale, and present
the spatial and seasonal variability of CO2 fluxes at the air–water
interface (FCO2) for the entire northeast North American LOAC, from streams to
the shelf break. This region of unprecedented data coverage allows us to
produce, for the first time, empirically derived monthly maps of CO2
exchange at 0.25∘ resolution. Our results allow us to investigate the
seasonal CO2 dynamics across the inter-connected systems of the LOAC and
elucidating their response to contrasting intra-annual changes in climate
conditions.
Methods
Our study area is located along the Atlantic coast of the northern US and
southern Canada and extends from the Albemarle Sound in the South section to the
eastern tip of Nova Scotia in the North section. It corresponds to COSCAT 827 (for
Coastal Segmentation and related CATchments) in the global coastal
segmentation defined for continental land masses by Meybeck et al. (2006)
and extrapolated to continental shelf waters by Laruelle et al. (2013).
COSCATs are homogenous geographical units that divide the global coastline
into homogeneous segments according to lithological, morphological, climatic,
and hydrological properties. The area corresponding to COSCAT 827 comprises
447×103 km2 of watersheds and 357×103 km2 of coastal
waters, amongst which 15×103 km2 of estuaries. It is one of the
best monitored regions in the world with several regularly surveyed rivers
(Hudson, Susquehanna, York, Connecticut) and some of the most extensively
studied coastal waters (Degrandpré et al., 2002; Chavez et al., 2007;
Fennel et al., 2008; Fennel and Wilkin, 2009; Previdi et al., 2009; Fennel,
2010; Shadwick et al., 2010, 2011; Signorini et al., 2013). For the purpose
of this study, the area was divided into a North and a South section (Fig. 1).
The boundary is set on land to distinguish the regions subject to seasonal
ice freeze and snowfalls from those that are not (Armstrong and Brodzik,
2001). This delineation attributes 96 % of the estuarine surface area to
the South section due, for the most part, to the contribution of Chesapeake
Bay which accounts for about two thirds of the estuarine area. The
delineation extends further into the coastal waters in such a way that the
Scotian Shelf and the Gulf of Maine correspond to the North section and the
Mid-Atlantic Bight and Georges Bank to the South section. The riverine data
are calculated from pH and alkalinity measurements extracted from the GLObal
RIver CHemistry Database (GLORICH (Hartmann et al., 2014), previously used in Lauerwald et al.,
2013), while continental shelf values are calculated from the Surface Ocean
CO2 Atlas (SOCAT v2.0) database which contains quality controlled
direct pCO2 measurements (http://www.socat.info/, Bakker et al., 2014).
Geographic limits of the study area with the location of the
riverine (GLORICH database, in green; Lauerwald et al., 2013) and continental
shelf waters data used for our calculations (SOCAT 2.0 database, in red;
Bakker et al., 2014). The location of the estuarine studies used is indicated
by purple squares.
Rivers
CO2 evasion from rivers (FCO2) was calculated monthly per 15 s
grid cell (resolution of the hydrological routing scheme HydroSHEDS 15 s,
Lehner et al., 2008) from estimates of the effective stream/river surface
area Aeff [m2], gas exchange velocity k [m d-1], and
water–atmosphere CO2 concentration gradient Δ[CO2]
[µmol l-1]:
FCO2=Aeff×k×Δ[CO2].
The calculation of Aeff first requires estimation of the total
stream/river surface area, A. The latter was calculated from the linear
stream network derived from the HydroSHEDS 15 s routing scheme using a
minimum threshold on the catchment area of 10 km2, and estimates of
stream width derived from the annual mean discharge Qann using
the equations of Raymond et al. (2012, 2013) (Eqs. 2, 3). Values of A were not
calculated for each individual month, as the discharge–stream width
relationship
only hold true for Qann (Raymond et al., 2013). Qann
was obtained using HydroSHEDS 15 s to route the gridded data of average
annual runoff from the UNH/GRDC composites (Fekete et al., 2002).
ln(B[m])=2.56+0.423×ln(Qann[m3s-1])
(Eq. 2 after Raymond et al., 2012),
ln(B[m])=1.86+0.51×ln(Qann[m3s-1]
(Eq. 3 after Raymond et al., 2013),
where B is stream width [m] and
Qann is annual average discharge
[m3 s-1].
For each 15 s raster cell covered by lake and reservoir
areas as represented in the global lake and wetland database of Lehner and
Döll (2004), A was set to 0 km2. Aeff was then
derived from A to account for seasonal stream drying and ice cover
inhibiting FCO2. Seasonal stream drying was assumed for each 15 s cell
and month when the monthly average discharge Qmonth is
0 m3 s-1. Values of Qmonth were calculated similarly
to that of Qann using the gridded data of average monthly runoff
from the UNH/GRDC composites (Fekete et al., 2002). Ice cover was assumed for
each 15 s cell and month when the mean air temperature (Tair),
derived from the WorldClim data set of Hijmans et al. (2005), is below
-4.8∘ C (Lauerwald et al., 2015). In case of ice cover
and/or stream drying, Aeff is set to 0 m2. Otherwise
Aeff equals A.
Values of k were first calculated as standardized values for CO2 at a
water temperature (Twater) of 20∘ C (k600), from stream
channel slope CS and estimates of flowing velocity V (Eq. 4). Using the
Strahler order (Strahler, 1952) to perform the segmentation of
the stream network, CS was calculated for each segment by dividing the
change in its altitude by its length. Information on altitude was derived
from the HydroSHEDS elevation model. V was calculated from Qann based
on the equations of Raymond et al. (2012, 2013) (Eqs. 5, 6). Similarly to the
stream width, the V-Q relationship only holds true for Qann (Raymond et
al., 2013), and this is why only annually average values for V and k600
could be calculated. The k value for each month was calculated from
k600, an estimate of the average monthly water temperature Twater (Lauerwald et al., 2015; Raymond et al., 2012).
k600[md-1]=V[ms-1]×CS[1]×2841+2.02
(Eq. 4 after Raymond et al.,
2012),
ln(V[ms-1])=-1.64+0.285×ln(Qann[m3s-1])
(Eq. 5 after Raymond et al.,
2012),
ln(V[ms-1])=-1.06+0.12×ln(Qann[m3s-1])
(Eq. 6 after Raymond et al.,
2013),
where k600 is the standardized gas exchange velocity for CO2 at
20∘ C water temperature [m d-1], Qann is annual average discharge [m3 s-1], V stream flow velocity [m s-1], and CS
channel slope [dimensionless].
Values of Δ(CO2) were derived from monitoring data with
calculated pCO2river (12 300 water samples, from 161
locations, Lauerwald et al., 2013), and an assumed pCO2atmosphere of 390 µatm. Lauerwald et al. (2013) calculated
pCO2river values from pH, alkalinity, water temperature, and,
where available, major ion concentrations, using the hydrochemical modeling
software PhreeqC v2 (Parkhurst and Appelo, 1999). The pCO2 values were
converted into concentrations, [CO2], using Henry's constant (Henry,
1803) for each sample at its observed temperature Twater using
the equation of Telmer and Veizer (1999). In order to minimize the influence
of extreme values, the results were aggregated to median values per sampling
location and month for which at least three values were available. These
median values per sampling location and month were then used to calculate
maps of Δ[CO2] at a 15 s resolution. To this end, an inverse
distance-weighted interpolation was applied. This method allows us to predict a
value for each grid cell from observed values at the four closest sampling
locations, using the inverse of the squared distance between the position on
the grid and each sampling locations as weighting factors. To account for
downstream decreases in pCO2river, which are often reported in
the literature (Finlay, 2003; Teodoru et al., 2009; Butman and Raymond,
2011), the interpolation was applied separately to three different classes of
streams and rivers defined by Qann, for which sufficiently large
subsets of sampling locations could be retained: (1)
Qann< 10 m3 s-1 (n= 76), (2)
10 m3 s-1≤ Qann< 100 m3 s-1(n= 47), and (3) Qann≥ 100 m3 s-1 (n= 38). The
three maps of Δ[CO2] per month were then recombined according to
the spatial distribution of Qann values. The FCO2 values
were first calculated using Eq. (1) at the high spatial resolution of
15 s for each month. The results were then aggregated to a 0.25∘
resolution and 3-month period and reported as area-specific values
referring to the total surface area of the grid cell. At the outer
boundaries, only the proportions of the cell covered by our study area are
taken into account. The difference between the FCO2 values calculated using
the equations of Raymond et al. (2012) and Raymond et al. (2013) was used as an estimate of the
uncertainty of the mean yearly FCO2. The aforementioned method is consistent with the
approach of Raymond et al. (2013), which used two distinct sets of equations
for k and A to estimate the uncertainty in these parameters and their
combined effect on the estimated FCO2.
Summary of the data used for the FCO2 calculations in
compartment of the LOAC.
Compartment
Parameter
Description
Source
Reference
Rivers
pCO2
CO2
GLORICH
Hartmann et al. (2014)
partial pressure
Lauerwald et al. (2013)
-
River network, digital
HydroSHEDS 15 s
Lehner et al. (2008)
elevation model (DEM)
-
Runoff
UNH/GRDC
Fekete et al. (2002)
T
Air temperature
-
Hijmans et al. (2005)
-
Lake surface
Global Lake and
Lehner and Döll (2004)
area
Wetland Database
Estuaries
As
Surface Area
SRTM water body data set
NASA/NGA (2003)
-
CO2 exchange rate
Average of
Raymond et al. (1997)
local estimates
Raymond et al. (2000)
Raymond and Hopkinson (2003)
Hunt et al. (2010)
Shelves
As
Surface area
COSCAT/MARCATS
Laruelle et al. (2013)
Segmentation
ΔpCO2
pCO2 gradient at the
SOCAT database
Bakker et al. (2014)
air–water interface
k
Calculated using
CCMP database
Altas et al. (2011)
wind Speed
K0′
Solubility, calculated using
SOCAT database
Bakker et al. (2014)
salinity, water temperature
Estuaries
The yearly averaged CO2 exchange at the air–water interface was obtained
from local estimations of emission rates in seven estuaries located within
the study area (see Table 1). The limited number of observations does not
allow us to resolve the seasonality in CO2 emissions. The yearly average
local CO2 emission rates range from 1.1 molC m-2 yr-1 in
the Parker River to 9.6 molC m-2 yr-1 in the Hudson River
estuary, for a mean value of 4.2 molC m-2 yr-1 for the seven
systems. This value was then multiplied by the estuarine surface areas
extracted from the SRTM water body data set (NASA/NGA, 2003), to estimate the
bulk outgassing for the North and South sections of COSCAT 827. It should be
noted that the methods used to estimate the CO2 emission rates differ
from one study to the other (i.e., different relationships relating wind speed
to the gas transfer coefficient). However, in the absence of a consistent and
substantial estuarine pCO2 database for the region, we believe that
our method is the only one which allows one to derive a regional data driven
estimate for the CO2 outgassing from estuaries which would otherwise require the use of reactive transport models (Regnier et al., 2013b). Similar approaches have been used in the past to produce global estuarine
CO2 budgets (Borges et al., 2005; Laruelle et al., 2010, 2013; Cai,
2011; Chen et al., 2013). The standard deviation calculated for the emission
rates of all local studies was used as an estimate of the uncertainty of the
regional estuarine FCO2.
Continental shelf waters
Monthly CO2 exchange rates at the air–water interface were calculated
in continental shelf waters using 274 291 pCO2 measurements extracted
from the SOCAT 2.0 database (Bakker et al., 2014). For each measurement, an
instantaneous local CO2 exchange rate with the atmosphere was calculated
using Wanninkhof's equation (Wanninkhof, 1992), which is a function of a
transfer coefficient (k), dependent on the square of the wind speed above
sea surface, the apparent solubility of CO2 in water (K0′)
[moles m-3 atm-1], which depends on surface water temperature and
salinity, and the gradient of pCO2 at the air–water interface
(ΔpCO2) [µatm].
FCO2=As×k×K0′×ΔpCO2
The parameterization used for k is that of Wanninkhof et al. (2013), and
all the data necessary for the calculations are available in SOCAT 2.0 except
for wind speed, which was extracted from the CCMP database (Atlas et al.,
2011). The resulting CO2 exchange rates were then averaged per month for
each 0.25∘ cell in which data were available. Average monthly
CO2 exchange rates were calculated for the North and South sections
using the water surface area and weighted rate for each cell, and those
averages were then extrapolated to the entire surface area As of
the corresponding section to produce FCO2. In effect, this corresponds
to applying the average exchange rate of the section to the cells devoid of
data. To refine further the budget, a similar procedure was also applied to 5
depth segments (S1 to S5) corresponding to 0–20, 20–50, 50–80, 80–120 and
120–150 m, respectively, and their respective surfaces areas were extracted
from high resolution bathymetric files (Laruelle et al., 2013). The choice of
slightly different methodologies for FCO2 calculations in rivers and
continental shelf waters stems from the better data coverage in the
continental shelf, which allows the capturing of spatial heterogeneity within
the region without using interpolation techniques. The standard deviation
calculated for all the grid cells of the integration domain was used as the
uncertainty of the yearly estimates of FCO2. A more detailed
description of the methodology applied to continental shelf waters at the
global scale is available in Laruelle et al. (2014).
Spatial distribution of the CO2 exchange with the atmosphere in
rivers and continental shelf waters aggregated by seasons. The fluxes are net
FCO2 rates averaged over the surface area of each 0.25∘ cell
and a period of 3 months. Positive values correspond to fluxes towards the
atmosphere. Winter is defined as January, February, and March, Spring as
April, May, and June, and so forth.
Results and discussion
Figure 2 shows the spatial distribution of FCO2 along the LOAC
integrated per season. Throughout the year, river waters are a strong source
of CO2 for the atmosphere. Significant differences in the intensity of
the CO2 exchange at the air–water interface can nevertheless be
observed between the North and South sections, both in time and space. During
winter, there is nearly no CO2 evasion from rivers in the North section
due to ice coverage and stream drying. Over the same period, the CO2
emissions from the South section range from 0 to
5 gC m-2 season-1. During spring, the pattern is reversed and
rivers in the North exhibit higher outgassing rates than in the South section
with maximum emissions rates of
> 10 gC m-2 season-1. This trend is maintained
throughout summer while during fall, the entire COSCAT displays similar
emission rates without a clear latitudinal signal. Continental shelf waters
display a very different spatial and seasonal pattern than that of rivers.
During winter, the North section is predominantly a mild CO2 sink, with
rates between +2 and -5 gC m-2 season-1, which intensifies
significantly in the South section (-2 to
> -10 gC m-2 season-1). During spring, an opposite
trend is observed, with a quasi-neutral CO2 uptake in the South section
and a strong uptake in the North section, especially on the Scotian Shelf.
The entire COSCAT becomes a net CO2 source in summer with emission rates
as high as 5 gC m-2 season-1 in the Mid-Atlantic Bight. During
fall, the Gulf of Maine and Georges Bank remain CO2 sources, while the
Scotian Shelf and the Mid-Atlantic Bight become again regions of net CO2
uptake.
The monthly integrated FCO2 for the North and South sections provides
further evidence of the contrasting seasonal dynamics for the two areas
(Fig. 3a and b). In the North section, CO2 evasion from rivers is almost
zero in January and February, rises to a maximum value of 0.26±0.05 TgC month-1 in May and then progressively decreases until the
end of the year. These low winter values are explained by the ice cover
inhibiting the gas exchange with the atmosphere. The steep increase and
FCO2 maximum in spring could be related to the flushing of water from
the thawing top soils, which are rich in dissolved organic carbon (DOC) and
CO2. Additionally, the temperature rise also induces an increase in
respiration rates within the water columns(Jones and Mulholland, 1998;
Striegl et al., 2012). Rivers and the continental shelf in the North section
present synchronized opposite behaviors from winter through spring. In the
shelf, a mild carbon uptake takes place in January and February (-0.04±0.25 TgC month-1), followed by a maximum uptake rate in April (-0.50±0.20 TgC month-1). This CO2 uptake in spring has been
attributed to photosynthesis associated with the seasonal phytoplankton bloom
(Shadwick et al., 2010). Continental shelf waters behave quasi-neutral during
summer (<0.05±0.09 TgC month-1) and emit CO2 at a
high rate in November and December (>0.15±0.21 TgC month-1). Overall, the rivers of the North section emit
1.31±0.24 TgC yr-1, while the continental shelf waters take up
0.47±0.17 TgC yr-1. The very limited estuarine surface area (0.5
103 km2) only yields an annual outgassing of 0.03±0.02 TgC yr-1. The shelf sink calculated for the region differs from
that of Shadwick et al. (2011) which reports a source for the Scotian
Shelves, in contrast to the current estimate. Our seasonally resolved budget
is however in line with the -0.6 TgC yr-1 sink calculated by
Signorini et al. (2013) using an 8-year data set as well as with the
simulations of Fennel and Wilkin (2009), which also predict sinks of -0.7
and -0.6 TgC yr-1 for 2004 and 2005, respectively. No similar
analysis was so far performed for inland waters.
Areal-integrated monthly air–water CO2 flux for rivers and the
continental shelf waters in the North section (a), South section
(b), and entire study area (c). Positive values correspond
to fluxes towards the atmosphere. The boxes inside each panel correspond to
the annual carbon budgets for the region including the lateral carbon fluxes
at the river–estuary interface, as inorganic (IC) and organic carbon (OC).
The values in grey represent the uncertainties of the annual fluxes.
Surface areas, CO2 exchange rate with the atmosphere, and
surface integrated FCO2 for the North and South sections of COSCAT 827,
subdivided by river discharge classes and continental shelf water depth
intervals.
North
South
Total
Surface Area
Rate
FCO2
Surface Area
Rate
FCO2
Surface Area
Rate
FCO2
103 km2
gCm-2 yr-1
109 gC yr-1
103 km2
gCm-2 yr-1
109 gC yr-1
103 km2
gCm-2 yr-1
109 gC yr-1
Rivers
Q1 (Q < 1 m s-1)
0.14
2893±521
391±70
0.27
1961±353
532±96
0.41
2271±409
924±166
Q2 (1 m s-1 <Q <10 m s-1 )
0.21
2538±457
525±95
0.32
1570±283
506±91
0.53
1948±351
1032±186
Q3 (10 m s-1 <Q <100 m s-1 )
0.16
1476±267
237±43
0.30
1307±235
392±71
0.46
1366±246
629±113
Q4 (100 m s-1 <Q)
0.17
891±160
152±27
0.36
729±131
261±47
0.52
781±141
412±74
Sub-total
0.67
1939±349
1305±235
1.25
1351±243
1692±305
1.92
1557±280
2997±539
Estuaries
0.53
50±31
27±19
14.51
50±31
731±453
15.04
50±31
758±469
Shelf
S1 (depth <20m)
11.21
5±1
53±19
24.28
-3±1
-79±11
35.49
-1±1
-27±5
S2 (20 m <depth < 50 m)
26.25
-1±1
-35±12
63.88
-8±1
-521±70
90.13
-6±1
-556±108
S3 (50 m <depth < 80 m)
39.28
-3±1
-128±45
48.63
-7±1
-359±126
87.91
-6±1
-488±95
S4 (80 m <depth < 120 m)
60.69
-3±1
-209±73
25.18
-8±1
-199±27
85.87
-5±1
-409±80
S5 (120 m <depth < 150 m)
34.73
-4±1
-151±18
7.63
-12±1
-91±12
42.36
-6±1
-242±47
Sub-total
172.17
-3±1
-472±166
169.59
-7±1
-1250±169
341.77
-5±1
-1722±335
In the South section of the COSCAT, the warmer winter temperature leads to
the absence of ice cover (Armstrong and Brodzik, 2001). Our calculations
predict that the riverine surface area remains stable over time, favoring a
relatively constant outgassing between 0.1 and 0.2 TgC month-1 throughout the year, adding up to a yearly source of 1.69±0.31 TgC yr-1. Estuaries emit 0.73±0.45 TgC yr-1,
because of their comparatively large surface area (14.5×103 km2),
about 1 order of magnitude larger than that of rivers (1.2×103 km2, Table 2). It should be noted that our estimate of the estuarine
outgassing is derived from a limited number of local studies, none of which
were performed in the two largest systems of COSCAT827, which are the
Chesapeake and Delaware bays (> 80 % of the total estuarine
surface area in COSCAT 827). These estuaries are highly eutrophic (Cai,
2011), which suggests that they might be characterized by lower pCO2
values and subsequent CO2 exchange than the other systems in the
region. On the other hand, our regional outgassing of 50 gC m-2 yr-1 is already well below the global average of 218 gC m-2 yr-1 calculated using the same approach by Laruelle et al. (2013) for
tidal estuaries. The continental shelf CO2 sink is strongest in January
(-0.47±0.30 TgC month-1) and decreases until June, when a
period of moderate CO2 emission begins (max of 0.13±0.08 TgC month-1 in August) and lasts until October. Finally, November and
December are characterized by mild CO2 sinks. Such seasonal signal,
following that of water temperature, is consistent with the hypothesis of a
CO2 exchange in the South section regulated by variations in gas
solubility, as suggested by Degrandpré et al. (2002) for the
Mid-Atlantic Bight.
The analysis of the intensity of the river CO2 outgassing reveals that
the smallest streams (Q<1 m3 s-1, Q1 in Table 2)
display the highest emission rates per unit surface area, with values ranging
from 1961 gC m-2 yr-1 in the South section to
2893 gC m-2 yr-1 in the North section. These values gradually
decrease with increasing river discharge to 729 gC m-2 yr-1 in
the South section and 891 gC m-2 yr-1 in the North section for
Q> 100 m3 s-1 (Q4, Table 2). The emission rates for
this latter class of rivers are consistent with the median emission rate of
720 gC m-2 yr-1 proposed by Aufdenkampe et al. (2011) for
temperate rivers with widths larger than 60–100 m. Aufdenkampe et
al. (2011) also report a median emission rate of
2600 gC m-2 yr-1 for the smaller streams and rivers, which falls
on the high end of the range calculated for Q1 in the present study. The
surface area of the river network is relatively evenly distributed amongst
the four discharge classes of rivers (Table 2). Yet, river sections for which
Q < 10 m3 s-1 (Q1+Q2) contribute to 65 % of the
total CO2 outgassing although they only represent 51 % of the
surface area. This result therefore highlights that streams and small rivers
are characterized by the highest surface-area-specific emission rates. The
higher outgassing rates in the North section are a consequence of higher
ΔCO2 values since average k values are similar in both
sections. In rivers with Qann<10 m3 s-1, the
ΔCO2 is about twice as high in the North than in the South
section from April to August (Table 2). The calculation of pCO2 from
alkalinity and pH presumes however that all alkalinity originates from
bicarbonate and carbonate ions and thus tends to overestimate pCO2
because non-carbonate contributions to alkalinity, in particular organic
acids, are ignored in this approach. The rivers in Maine and New Brunswick,
which drain most of the northern part of COSCAT 827, are characterized by
relatively low mineralized, low pH waters, rich in organic matter. In these
rivers, the overestimation in pCO2 calculated from the alkalinity
attributed to the carbonate system only was reported to be in the range of
13–66 % (Hunt et al., 2011). Considering that rivers in the southern
part of COSCAT 827 have lower DOC concentrations and higher dissolved
inorganic carbon (DIC) concentrations, the higher FCO2 rates per
surface water area reported in the northern part could partly be due to an
overestimation of their pCO2 values. However, a direct comparison of
average pCO2 values does not confirm this hypothesis. For the two
Maine rivers (Kennebec and Androscoggin rivers), Hunt et al. (2014) report an
average pCO2 calculated from pH and DIC of 3064 µatm. In our
data set, three sampling stations are also located in these rivers and
present lower median pCO2 values of 2409, 901 and 1703 µatm
for Kennebec River at Bingham and North Sidney and for Androscoggin River at
Brunswick, respectively. A probable reason for the discrepancy could be that
we report median values per month while Hunt et al. (2014) report arithmetic
means, which are typically higher.
On the continental shelf, the shallowest depth interval is a CO2 source
in the North section while all other depth intervals are CO2 sinks
(Table 2). The magnitude of the air–sea exchange for each segment is
between the values calculated for estuaries (50 gC m-2 yr-1) and the nearby open ocean (∼20 gC m-2 yr-1, according to Takahashi et al., 2009). This trend along a depth
transect, suggesting a more pronounced continental influence on nearshore
waters and a strengthening of the CO2 shelf sink away from the coast
was already discussed in the regional analysis of Chavez et al. (2007) and
by Jiang et al. (2013), specifically for the South Atlantic Bight. Modeling
studies over a larger domain including the upper slope of the continental
shelf also suggest that the coastal waters of the Northeast US are not a
more intense CO2 sink than the neighboring open ocean (Fennel and
Wilkin, 2009; Fennel, 2010). Our analysis further suggests that the
continental influence is more pronounced in the North section. Here, the
shallowest waters (S1) are strong net sources of CO2 while the
intensity of the CO2 sink for the other depth intervals gradually
decreases, but only to a maximum value of -4 gC m-2 yr-1 for S5.
This value is about 3 times smaller than in the South section (-12 gC m-2 yr-1).
Annually, river and estuarine waters of the entire COSCAT 827 outgas 3.0±0.5 and 0.8±0.5 TgC yr-1, respectively, while continental shelf
waters take up 1.7±0.3 TgC yr-1 (Fig. 3c). The total riverine
carbon load exported from rivers to estuaries for the same area has been
estimated to be 4.65 TgC yr-1, 45 % as dissolved and particulate
organic carbon (2.10 TgC yr-1, Mayorga et al., 2010) and 55 % as
dissolved inorganic carbon (2.55 TgC yr-1, Hartmann et al., 2009). The
ratio of organic to inorganic carbon in the river loads is about 1 in the
North section and 1.4 in the South section. This difference stems mainly from
a combination of different lithogenic characteristics in both sections and
the comparatively higher occurrence of organic soils in the North section
(Hunt et al., 2013; Hossler and Bauer, 2013). Estimates of the total amount
of terrestrial carbon transferred to the riverine network are not available,
but the sum of the river export and the outgassing, which ignores the
contribution of carbon burial and lateral exchange with wetlands, provides a
lower bound estimate of 7.65 TgC yr-1. Under this hypothesis,
∼ 40 % of the terrestrial carbon exported to rivers is emitted to
the atmosphere before reaching estuaries. In spite of higher emission rates
per unit surface area in the North section (Table 2), the overall efficiency
of the riverine carbon filter is essentially the same in the two sections (40
and 38 % outgassing for the North and the South sections, respectively).
On the shelf, however, the South section exhibits a significantly more
intense CO2 sink (-1.25±0.2 TgC yr-1) than in the North
section (-0.47±0.2 TgC yr-1). A possible reason for this
difference can be found in the contribution of the estuarine carbon filter.
In the South section, where 96 % of the estuarine surface area is
located, these systems contribute to an outgassing of 0.73 TgC yr-1
while in the North section, their influence is negligible. Cole and
Caraco (2001) estimated that 28 % of the DOC entering the relatively
short Hudson River estuary is respired in situ before reaching the
continental shelf and it is thus likely that the estuarine outgassing in the
South section is fueled by the respiration of the organic carbon loads from
rivers. In contrast, the absence of estuaries in the North section favors the
direct export of terrestrial organic carbon onto continental shelf waters
where it can be buried and decomposed. The respiration of terrestrial organic
carbon could therefore explain why the strength of the shelf CO2 sink is
weaker in this portion of the domain. Such filtering of a significant
fraction of the terrestrial carbon inputs by estuaries has been evidenced in
other systems (Amann et al., 2010; 2015). This view is further substantiated
by the similar cumulated estuarine and continental shelf FCO2 fluxes
in both sections (Fig. 3a and b). Naturally, other environmental and physical
factors also influence the carbon dynamics in shelf waters and contribute to
the difference in CO2 uptake intensity between both sections. For
instance, in the North section, the Gulf of Maine is a semi-enclosed basin
characterized by specific hydrological features and circulation patterns
(Salisbury et al., 2008; Wang et al., 2013) which could result in longer
water residence times promoting the degradation of shelf-derived organic
carbon. Other potential factors include the plume of the Saint Lawrence
Estuary, which has also been shown to transiently expend over the Scotian
Shelf (Kang et al., 2013), the strong temperature gradient, and the
heterogeneous nutrient availability along the region which may result in
different phytoplankton responses (Vandemark et al., 2011; Shadwick et al.,
2011). Additionally, modeling studies evidenced the potential influence of
sediment denitrification on water pCO2 through the removal of fixed
nitrogen in the water column and consequent inhibition of primary production
(Fennel et al., 2008; Fennel, 2010). This removal was estimated to be of
similar magnitude as the lateral nitrogen loads, except for estuaries of the
Mid-Atlantic Bight (MAB) region (Fennel, 2010). It can nonetheless be
suggested that the estuarine carbon filter in the South section of COSCAT 827
is an important control factor of the CO2 sink in the Mid-Atlantic
Bight, which is stronger than in any other area along the entire Atlantic
coast of the US (Signorini et al., 2013).