RESEARCH ARTICLE Update on the Temperature Corrections of Global Air-Sea CO2

10.1029/2022GB007360
Flux Estimates
Key Points:
Yuanxu Dong1,2  , Dorothee C. E. Bakker1  , Thomas G. Bell2  , Boyin Huang3  ,
• T he impact of the warm bias in an
Peter Landschützer4  , Peter S. Liss1, and Mingxi Yang2 
in situ sea surface temperature data
set and the cool skin effect on air-sea
Centre for Ocean and Atmospheric Sciences, School of Environmental Sciences, University of East Anglia, Norwich, UK,
1
carbon dioxide (CO2) flux estimates
are revisited Plymouth Marine Laboratory, Prospect Place, Plymouth, UK, 3National Centers for Environmental Information, National
2

• The updated temperature corrections Oceanic and Atmospheric Administration, Ashddeeville, NC, USA, 4Max Planck Institute for Meteorology, Hamburg,
imply a smaller increase in net ocean Germany
CO2 uptake (∼35%) compared to a
previous study (∼50%)
• The revised observation-based CO2
flux agrees well with the independent
Abstract  The oceans are a major carbon sink. Sea surface temperature (SST) is a crucial variable in the
ocean carbon inventory calculation of the air-sea carbon dioxide (CO2) flux from surface observations. Any bias in the SST or any
upper ocean vertical temperature gradient (e.g., the cool skin effect) potentially generates a bias in the CO2
Supporting Information:
flux estimates. A recent study suggested a substantial increase (∼50% or ∼0.9 Pg C yr −1) in the global ocean
Supporting Information may be found in CO2 uptake due to this temperature effect. Here, we use a gold standard buoy SST data set as the reference
the online version of this article. to assess the accuracy of insitu SST used for flux calculation. A physical model is then used to estimate the
cool skin effect, which varies with latitude. The bias-corrected SST (assessed by buoy SST) coupled with the
Correspondence to: physics-based cool skin correction increases the average ocean CO2 uptake by ∼35% (0.6 Pg C yr −1) from 1982
Y. Dong and D. C. E. Bakker, to 2020, which is substantially smaller than the previous correction. After these temperature considerations, we
[email protected]; estimate an average net ocean CO2 uptake of 2.2 ± 0.4 Pg C yr −1 from 1994 to 2007 based on an ensemble of
[email protected]
surface observation-based flux estimates, in line with the independent interior ocean carbon storage estimate
corrected for the river induced natural outgassing flux (2.1 ± 0.4 Pg C yr −1).
Citation:
Dong, Y., Bakker, D. C. E., Bell, T.
G., Huang, B., Landschützer, P., Liss,
Plain Language Summary  The global oceans play a major role in taking up carbon dioxide (CO2)
P. S., & Yang, M. (2022). Update released by human activity from the atmosphere. Accurate sea surface temperature (SST) measurements and
on the temperature corrections of quantification of any upper ocean temperature gradients (e.g., cool skin effect) are critical for ocean CO2 uptake
global air-sea CO2 flux estimates.
estimates. We determine a slight warm bias in the SST data set used for CO2 flux calculation by utilizing a gold
Global Biogeochemical Cycles,
36, e2022GB007360. https://doi. standard reference buoy SST data set. We then derive a physics-based temperature correction for the ubiquitous
org/10.1029/2022GB007360 cool skin effect on the ocean surface. The temperature revised CO2 flux bridges the gap between estimates from
the surface observation-based air-sea CO2 fluxes and from the independent ocean carbon inventory.
Received 16 FEB 2022
Accepted 15 AUG 2022

1. Introduction
Author Contributions:
Conceptualization: Yuanxu Dong, Since the Industrial Revolution, humans have emitted large amounts of carbon dioxide (CO2) to the atmosphere,
Dorothee C. E. Bakker, Mingxi Yang which is the main reason for observed global warming. The oceans are a major CO2 sink, accounting for ∼25%
Formal analysis: Yuanxu Dong (∼2.8 Pg C yr −1 for the last decade) of the annual anthropogenic CO2 emissions (Friedlingstein et al., 2022) and
Investigation: Yuanxu Dong, Dorothee
C. E. Bakker, Peter Landschützer, Peter S. ∼40% of all anthropogenic CO2 released since industrialization (Gruber et al., 2019; Sabine et al., 2004).
Liss, Mingxi Yang
Methodology: Yuanxu Dong, Boyin The global air-sea CO2 flux is often estimated by the bulk method, combining in situ fCO2w (fugacity of CO2
Huang, Peter Landschützer in seawater) measurements (e.g., from the surface ocean CO2 Atlas, SOCAT; Bakker et al., 2016) with a wind
Software: Yuanxu Dong, Boyin Huang, speed-dependent gas transfer velocity (e.g., Wanninkhof,  2014; see Methods). Thanks to the SOCAT (http://
Peter Landschützer
Supervision: Dorothee C. E. Bakker, www.socat.info/) community, a key data set of fCO2w has been available since 2011 (Pfeil et al., 2013; Sabine
Thomas G. Bell, Peter S. Liss, et al., 2013). The latest SOCAT version, SOCAT v2021, contains 30.6 million quality-controlled fCO2w observa-
Mingxi Yang tions from 1957 to 2020 with an accuracy better than 5 μatm (Bakker et al., 2016, 2021).

Sea surface temperature (SST) is key for bulk air-sea CO2 flux estimates. Takahashi et al. (2009) reported a 13%
increase in ocean CO2 uptake by correcting for a 0.08 K warm bias in SST. CO2 is a water-side controlled gas
© 2022. The Authors.
This is an open access article under (Liss & Slater, 1974), and thus air-sea CO2 exchange is mainly limited by transfer within the ∼20–200 μm mass
the terms of the Creative Commons boundary layer (MBL, Figure 1; Jähne, 2009). The MBL temperature should be used for the CO2 flux calculation,
Attribution License, which permits use, but it is impractical to measure in situ SST within the very thin MBL. The bulk seawater temperature (TBulk) meas-
distribution and reproduction in any
medium, provided the original work is ured concurrently with fCO2w (typically at ∼5 m depth by ship) in SOCAT is often used for the bulk air-sea CO2
properly cited. flux calculation by assuming a well-mixed upper ocean (top ∼10 m) without any vertical temperature gradients.

DONG ET AL. 1 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Validation: Dorothee C. E. Bakker, However, two temperature issues might generate bias in the CO2 flux estimates by using the SOCAT SST. The
Thomas G. Bell
first issue is the ship's intake depth (∼5 m instead of micrometers) and the other is the location of the SST sensor
Visualization: Yuanxu Dong
Writing – original draft: Yuanxu Dong (within the warm hull of the ship instead of in the unperturbed seawater).
Writing – review & editing: Yuanxu
Dong, Dorothee C. E. Bakker, Thomas G. First, the SOCAT SST represents the bulk seawater temperature, which might not be equal to the temperature
Bell, Boyin Huang, Peter Landschützer, at the MBL because many processes can generate vertical temperature gradients in the upper ocean. There is a
Peter S. Liss, Mingxi Yang
temperature gradient (red line in Figure 1) in the thermal boundary layer (TBL and gray shaded area) relating
to air-sea heat exchange. Infrared radiometer measurements indicate that the skin temperature at ∼10 μm depth
(TSkin) is on average ∼0.17 K (Donlon et al., 2002) lower than the subskin temperature (TSubskin, at ∼0.1–1 m depth)
because the ocean surface generally loses heat through longwave radiation and latent and sensible heat fluxes
(the so-called cool skin effect; e.g., Donlon et al., 2002, 2007; Minnett et al., 2011; Robertson & Watson, 1992;
Zhang et al., 2020). Another process that might create an upper ocean temperature gradient is the diurnal warm
layer effect. Water close to the surface (e.g., at 0.5 m depth) is sometimes warmer than deeper water (e.g., at 5 m
depth) due to daytime solar insolation, especially under conditions of clear sky and low wind speed (Gentemann
& Minnett, 2008; Prytherch et al., 2013; Ward et al., 2004). The warming leads to stabilization of the surface layer
and thus helps maintain a layered upper ocean structure. The diurnal warm layer effect is not as ubiquitous as the
cool skin effect (Fairall et al., 1996), and the warm layer is complex to characterize. In the absence of the warm
layer effect, the bulk seawater temperature (TBulk) is approximately equal to TSubskin, and TThermal (temperature at
the base of the TBL) because the water below the TBL is well-mixed by turbulence.

The second issue is the potential warm bias in the SOCAT SST. The SST community has identified a warm bias in
shipboard SST measurements in the ICOADS (International Comprehensive Ocean-Atmosphere Data Set; Huang
et al., 2021; Kennedy et al., 2011, 2019; Reynolds & Chelton, 2010). This might be because ship SST measure-
ments are affected by engine room warming because the SST sensor is often located in the engine room or some-
where in the ship interior (Kennedy et al., 2019). The SSTs in SOCAT were almost exclusively measured by ship-
board systems (98%), meaning that a warm bias also likely exists in the SOCAT SST data set. It is worth noting
that the percentage of the SST data measured by research vessels in SOCAT is likely higher than in the ICOADS
shipboard SST data set. The SST measured by research ships (typically external to the ship's hull) is expected to
have a higher accuracy than the SST measured by commercial ships (often in the ship's interior/within the engine
room), so the warm bias in SOCAT SST may well be different with the warm bias in ICOADS ship SST.

Satellite observation of SST represents a consistent estimate of subskin temperature and avoids the diurnal warm
layer effect and any potential warm bias issue. Satellite SST thus has been proposed as an alternative to calcu-
late the bulk air-sea CO2 flux (Goddijn-Murphy et al., 2015; Shutler et al., 2019; Watson et al., 2020; Woolf
et al., 2016). Results based on a satellite SST data set suggest a ∼25% increase (i.e., warm bias correction; the
cool skin correction results in another ∼25% increase) in ocean CO2 uptake compared to the flux estimate based
on the SOCAT SST (Watson et  al.,  2020). However, the satellite SST is not measured concurrently with the
fCO2w. Colocating the 1° × 1°, monthly gridded satellite SSTs with individual fCO2w in SOCAT might introduce
extra uncertainties. In addition, various issues in satellite SSTs (e.g., cloud masking, impact of aerosol, diurnal
variability, uncertainty estimation, and validation) have not been fully resolved, especially at high latitudes and
in coastal and highly dynamic regions (O’Carroll et al., 2019). A comparison of eight global gap-free satellite/
blended SST products showed that their global mean ranged from 20.02°C to 20.17°C for the period 2003–2018
(Yang et al., 2021). Therefore, the current accuracy of satellite SST means that it probably does not allow an
optimal estimate of the global air-sea CO2 flux.

SST observations from drifting buoys are unaffected by engine room warming, and are expected to provide the
best quality reference temperature to assess bias in the ship SST, and satellite SST retrievals (Huang et al., 2021;
Kennedy et al., 2011, 2019; Kent et al., 2017; Merchant et al., 2019; Reynolds & Chelton, 2010). This work
utilizes drifting buoy SST as the reference temperature to determine the accuracy of the SOCAT SST and to
correct for any bias in the SOCAT SST data set.

Subskin temperature with a cool skin correction represents the skin temperature, which can be used to calculate
air-sea CO2 flux. Watson et al. (2020) reported a ∼25% increase in ocean CO2 uptake by considering a constant
cool skin effect (−0.17 K, Donlon et al., 2002) from 1982 to 2020. In this study, the cool skin effect estimated by
a physical model (Fairall et al., 1996) and by an empirical model (Donlon et al., 2002) are compared at a global
scale. The updated temperature corrections are then used to estimate their impact on the global air-sea CO2

DONG ET AL. 2 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

flux. The revised global air-sea CO2 flux based on an ensemble of CO2 flux
products (Fay et al., 2021) is then compared with the ocean carbon inventory
(Gruber et al., 2019).

2. Methods
2.1.  Global Air-Sea CO2 Flux Estimates

The bulk air-sea CO2 flux equation is:

(1)
𝐹𝐹 = 𝐾𝐾660 (𝑆𝑆𝑆𝑆∕660)−0.5 (𝛼𝛼𝑤𝑤 𝑓𝑓 CO2𝑤𝑤 − 𝛼𝛼𝑖𝑖 𝑓𝑓 CO2𝑎𝑎 )

where F (mmol m −2 day −1) is the air-sea CO2 flux and K660 (cm h −1) is the
gas transfer velocity (e.g., Wanninkhof, 2014) normalized to a Sc (Schmidt
number) of 660. The Sc is defined as the ratio of the kinematic viscosity of
water (m 2 s −1) and the molecular diffusivity of CO2 (m 2 s −1). The CO2 solu-
bility (mol L −1 atm −1) at the base of the MBL and at the air-sea interface is
Figure 1.  A schematic of the upper ocean (0–10 m depth) using an example
represented by αw and αi, respectively (Figure 1). Sc and α are calculated from
where temperature is influenced by a positive (ocean heat loss) sensible heat
flux and carbon dioxide (CO2) is being taken up by the ocean. The gray shaded seawater temperature and salinity (Wanninkhof et al., 2009; Weiss, 1974). Sc
area represents the thermal boundary layer (TBL), and the red line represents is equal to 660 for CO2 at 20°C and 35 psu seawater. The CO2 fugacity (μatm)
the temperature gradient in the TBL. The mass (in this case, CO2) boundary at the base of the MBL and just above the air-sea interface is represented by
layer (MBL) is embedded within the TBL. The blue line corresponds to the fCO2w and fCO2a, respectively.
CO2 concentration gradient within the MBL. The TBL is characteristically
10 times thicker than the MBL because heat is transferred about an order To calculate the global air-sea CO2 flux, fCO2w measured at the equilibrator
of magnitude quicker than CO2 (Jähne, 2009). Sea surface temperature is
temperature is first corrected to the in situ bulk temperature (SOCAT SST).
a general term for all temperatures mentioned in the figure. TInterface: the
temperature at the air-sea interface; TSkin: the skin temperature at ∼10 μm Seawater at ∼5  m depth (ranging from 1 to 10  m depth depending on the
depth measured by an infrared radiometer; TMass: the temperature at the base of ship or sampling platform) is sampled from the ship's underway water intake
the MBL (20–200 μm depth); TThermal: the temperature at the base of the TBL and pumped through an equilibrator. The equilibrated CO2 mole fraction in
(0.1–2 mm depth); TSubskin: the temperature of seawater below the TBL at a the air of the headspace (χCO2w) is measured in a gas analyzer. χCO2w is
depth of ∼0.1–1 m such as measured by drifting buoys; TBulk: the temperature
then converted to equilibrator fugacity (fCO2w_equ) (Text S1 in Supporting
at 1–10 m depth as measured at the typical depth of a ship's seawater intake.
TInterface, TMass, and TThermal are conceptual (black text), whereas TSkin, TSubskin, Information S1). fCO2w_equ is further corrected by the chemical temperature
and TBulk are from actual measurements (practical, blue text). Figure developed normalization (Takahashi et al., 1993) to obtain fCO2w in the bulk seawater:
from Donlon et al. (2007).
(2)
𝑓𝑓 CO2w = 𝑓𝑓 CO2w equ 𝑒𝑒0.0423(𝑇𝑇𝑤𝑤 bulk −𝑇𝑇equ )

where Tw_bulk is the seawater temperature measured concurrently with fCO2w

at the ship's water intake at typically 5 m depth. Seawater fCO2w measure-
ments are then interpolated to obtain a global gap-free fCO2w product (at 1° × 1°, monthly resolution, e.g., Land-
schützer et al., 2013). A global gap-free SST data set is generally one of the independent input variables for the
fCO2w interpolation process. Other variables in Equation 1 are calculated using a global gap-free SST product and
related data sets (e.g., mole fraction of atmospheric CO2 for the calculation of fCO2a). Finally, globally mapped
fCO2w, fCO2a, Sc, αw, αi, and gas transfer velocity (K660, estimated using a global gap-free wind speed data set) are
used for the CO2 flux calculation via Equation 1.

Table 1 summarizes the SST types that should be used to calculate variables in Equation 1. Sc should be calcu-
lated from the temperature utilized to derive K660 (e.g., TBulk for the K660 derived from the dual-tracer method;
e.g., Ho et al., 2006; Nightingale et al., 2000). The air-sea interface temperature (TInterface) should be used for the
calculation of fCO2a and αi, while the temperature at the base of the MBL (TMass) should be employed to calculate
fCO2w (via Equation 2) and αw. However, Woolf et al. (2016) suggested that TThermal might be a better temper-
ature for calculating fCO2w and αw. The seawater carbonate system creates a unique situation for air-sea CO2
exchange, which does not exist for other gases. Seawater temperature changes cause chemical repartitioning of
the carbonate species (CO2, carbonic acid, bicarbonate, and carbonate; Zeebe & Wolf-Gladrow, 2001). We find
that the timescale of this repartitioning equilibration (e-folding time >10 s for typical seawater; Johnson, 1982;
Zeebe & Wolf-Gladrow, 2001) is much longer than the timescale (∼1 s) of water mixing below the MBL but
within the TBL, where viscous dissipation dominates the water mixing (Jähne, 2009; Jähne et al., 1987; Woolf
et al., 2016). The explanation of the timescales is detailed in Text S2 in Supporting Information S1. Although

DONG ET AL. 3 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Table 1
Variables and Relevant Sea Surface Temperature (SST) Types for Global Air-Sea Carbon Dioxide Flux Estimates and Their Relative Importance for the Flux
Estimate (After Woolf et al., 2016)

Variable (x) Conceptual SST Practical SST product 𝐴𝐴 𝜕𝜕𝐥𝐥𝐥𝐥(𝒙𝒙)

𝜕𝜕𝑻𝑻
  𝐴𝐴 𝜕𝜕𝐥𝐥𝐥𝐥(flux)
𝜕𝜕𝑻𝑻
 
Sc −0.5 TBulk Global gap-free TSubskin 2.5% K −1 2.5% K −1
αi TInterface TSkin (Global gap-free TSubskin with a cool skin correction) −2.5% K  −1
100% K −1
fCO2a TInterface TSkin (Global gap-free TSubskin with a cool skin correction) −0.2% K −1 10% K −1
αw TThermal Global gap-free TSubskin −2.5% K  −1
−100% K −1
Individual fCO2w TThermal Individual TSubskin (In situ TBulk with any bias correction) 4.23% K −1 160% K −1
Mapped fCO2w TThermal Global gap-free TSubskin <4.23% K  −1 a
<160% K −1 a
Note. The back-of-the-envelope calculation in the last column is for fCO2w of ∼380 μatm, fCO2a of ∼390 μatm, and ΔfCO2 of −10 μatm, values typical for the last
decade (Landschützer et al., 2020).
 aThe interpolation method (e.g., MPI-SOMFFN neural network technique; Landschützer et al., 2013) can largely dampen the effect of SST on mapped fCO2w.

there is a temperature gradient in the TBL due to the cool skin effect, the carbonate species are not expected to
have time to thermally adjust, which suggests that TThermal is the optimal temperature for calculating fCO2w and αw.

TThermal, TMass, and TInterface are conceptual temperatures, which can be approximated by practical temperatures
(Figure 1). Satellite SST, which represents the subskin temperature, is a good approximation for TThermal (Shutler
et al., 2019; Watson et al., 2020; Woolf et al., 2016). A satellite TSubskin product can be used to calculate αw and
Sc, and to map fCO2w for the global ocean. TSubskin with a cool skin correction can then be utilized to calculate
global fCO2a, and αi. In situ TSubskin should ideally be used to correct fCO2w from the equilibrator temperature to
the subskin seawater temperature. However, the in situ temperature measured concurrently with the fCO2w in
SOCAT is TBulk, and in situ TSubskin measurements are unavailable to exactly match the SOCAT space and time
stamp. Using in situ TBulk (i.e., SOCAT SST) to correct fCO2w is reasonable in the absence of a warm layer effect,
but it is important to account for the potential warm bias in the SOCAT SST.

Table 1 also summarizes the influence of SST and the corresponding importance for the variables used to make
air-sea CO2 flux estimates (after Woolf et al., 2016). The Sc and fCO2a variations due to the bias in the SST prod-
uct have a small influence on the global air-sea CO2 flux. However, any bias in the SST data used for the calcu-
lation of αw, αi, and especially fCO2w can result in a considerable bias in the flux. The temperature influence on
the fCO2w mapping should be significantly dampened by the interpolation process. The most significant influence
on the CO2 flux due to temperature bias comes from individual fCO2w (∼160% K −1, Table 1). An average bias of
0.1 K could result in a bias in fCO2w of ∼1.6 μatm, which corresponds to ∼16% of the net air-sea CO2 flux for the
last decade (Landschützer et al., 2020).
The skin temperature should be used for the calculation of αi and fCO2a. The TSkin can be obtained from TSubskin
with a cool skin correction. If TSubskin is used rather than TSkin for the calculation of αi, and fCO2a, the ocean CO2
uptake is in theory underestimated by ∼19% for the last decade, with a mean cool skin effect of 0.17 K (Donlon
et al., 2002).

2.2.  Bias Assessment

The in situ bulk SST in SOCAT is generally used to correct individual fCO2w observations from the equilibrator
temperature to the seawater temperature (e.g., studies in Table S1 in Supporting Information S1). However, a
warm bias might exist in the SOCAT SST due to heating in the engine room. Watson et al. (2020) colocated the
DOISST v2.0 (NOAA Daily Optimum Interpolation SST data set; Reynolds et al., 2007; representing the subskin
temperature) with individual in situ SST measurements in SOCAT. They found that the SOCAT SST is on aver-
age 0.13 ± 0.78 K higher than the colocated DOISST v2.0. However, Huang et al. (2021) pointed out that there
might be a cold bias in the DOISST v2.0 and DOISST v2.1 products (the difference between DOISST v2.0 and
v2.1 can be seen in Text S4 in Supporting Information S1).

This study uses accurate SST observed by drifting buoys to assess the potential cold bias in the DOISST v2.1 and
the warm bias in SOCAT SST. A drifting buoy SST (measured at nominally 10–20 cm depth; representing the

DONG ET AL. 4 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

subskin temperature) data set from iQuam (in situ SST Quality Monitor v2.10; Xu & Ignatov, 2014) with high
accuracy (quality level = 5) is used for the assessment. The buoy SST is first gridded (1° × 1°, monthly) and then
compared with the resampled DOISST v2.1 (1/4° × 1/4°, daily data are resampled to 1° × 1°, monthly resolution)
and the gridded SST (1° × 1°, monthly) in SOCAT v2021.

2.3.  Cool Skin Effect Estimate

The cool skin effect is ubiquitous in the ocean (Donlon et al., 2002) and should be considered when estimating
air-sea CO2 fluxes. Watson et al. (2020) used a constant value (−0.17 K) to account for the impact of the cool skin
effect on air-sea CO2 fluxes. However, the cool skin effect is affected by many environmental processes. Donlon
et al. (2002) proposed a wind speed-dependent cool skin effect based on skin and bulk temperature measurements
(Donlon02, hereafter). A physical model for the cool skin effect proposed by Saunders (1967) and developed by
Fairall et al. (1996) considers wind speed, longwave radiation, heat flux, and solar radiation (Fairall96, hereaf-
ter). Fairall96 has been included in the COARE 3.5 model (Edson et al., 2013) and recent studies (Alappattu
et al., 2017; Embury et al., 2012; Zhang et al., 2020) suggest that Fairall96 better accounts for the cool skin effect
than the parameterization dependent upon a single variable (wind speed).

We employ the ERA5 wind speed data (Hersbach et al., 2020) to estimate the Donlon02 cool skin effect. The
COARE 3.5 model is used to estimate the Fairall96 cool skin effect. The following model inputs are used:
CCI SST v2.1 (European Space Agency Climate Change Initiative SST product; Merchant & Embury,  2020;
Merchant et al., 2019), NCEP sea level pressure (Kalnay et al., 1996), ERA5 monthly averaged reanalysis data
sets (Hersbach et al., 2020) for wind speed, 2 m above mean sea level (AMSL) air temperature, relative humidity
(calculated from 2 m AMSL air temperature and dew point temperature using the August-Roche-Magnus approx-
imation), downward shortwave radiation, downward longwave radiation, and boundary layer height.

2.4.  Global Air-Sea CO2 Flux Estimates With the Temperature Correction

We use two different methods to account for the bias in the SOCAT SST for the global air-sea CO2 flux esti-
mates. For the first method, we use the buoy SST as the reference temperature to assess the bias in SOCAT SST
(bias_buoy, hereafter). We correct the 1° × 1°, monthly fCO2w in SOCAT v2021 via Equation 2 (i.e., fCO2w_corrected
= fCO2w e −0.0423 *  ΔSST) by the temperature difference (ΔSST) between SOCAT SST and buoy SST. The ΔSST
varies with latitude (with a 10° latitude running mean, see the orange line in Figure 2b) but we do not consider
the variation of ΔSST over time. The number of matched data points between SOCAT SST and buoy SST is
small in most years, so ΔSST is averaged from 1982 to 2020. In addition, only fCO2w data within 70°S to 70°N
are corrected because of the small number of measurements in the polar oceans. For the second method, the
colocated DOISST v2.1 replaces SOCAT SST in Equation  2 to reanalyze fCO2w (bias_OI, hereafter; Watson
et al., 2020). The reanalyzed fCO2w is used for the flux calculation (see Goddijn-Murphy et al., 2015; Holding
et al., 2019 for the reanalysis process).

We employ the MPI-SOMFFN neural network technique (Landschützer et al., 2013) to interpolate the fCO2w_corrected
and the reanalyzed fCO2w to the global ocean from 1982 to 2020, using a set of input variables. We use the same
data sets as Landschützer et al. (2014) for the neural network inputs, except for the SST product. The CCI SST
(Merchant et al., 2019) represents the subskin temperature and is independent of in situ SST measurements, so we
utilize the 1° × 1°, monthly CCI SST v2.1 for the neural network training process. The CCI SST v2.1 is also used
to calculate Sc and αw, while the CCI SST v2.1 with a cool skin correction is employed to calculate αi and fCO2a.

We use two models (Fairall96 and Donlon02) to estimate the cool skin effect. Both Fairall96 and Donlon02
cool skin effect estimates are applied to the CCI SST v2.1 to calculate αi and fCO2a, respectively. The quadratic
wind speed-dependent formulation (K660 = a U10 2; Ho et al., 2006; Wanninkhof, 2014) is used to calculate gas
transfer velocity. The 1° × 1°, monthly ERA5 wind speed data from 1982 to 2020 is utilized to scale the transfer
coefficient a to match to a global mean K660 of 18.2 cm hr −1 (equal to 16.5 cm hr −1 for K) from the  14C inventory
method (Naegler, 2009). It is worth noting that the cool skin effect and the warm layer effect do not impact the
global mean K660 calculated from the  14C inventory because the air-sea  14C concentration difference (Δ 14C) is
very large (Naegler, 2009; Sweeney et al., 2007), and the upper ocean temperature gradients only result in a minor
change in Δ 14C. In the end, we substitute all the variables above into Equation 1 to calculate the global air-sea

DONG ET AL. 5 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Figure 2.  Latitudinal variation in sea surface temperature (SST) differences, number of matched grid cells, the gas transfer
velocity (K660) and the fraction of the globe's surface area covered by ocean: (a) 1° latitude average temperature difference
between DOISST v2.1 and buoy SST (red line) ± 1 standard error (gray shading). The input data are from 1982 to 2020 and
have a 1° × 1°, monthly resolution. Blue bars show the number of cells (5° latitude bin) containing both DOISST and buoy
SST data (b) 10° latitude running mean of the temperature difference between SOCAT SST (from SOCATv2021) and buoy
SST (orange line, i.e., ΔSST in the main text) ± 1 standard error (gray shading). Green bars correspond to the number of cells
(5° latitude bin) containing both gridded SOCAT and buoy SST; (c) 1° latitude average K660 (purple line) calculated with a
wind speed-dependent parameterization (Ho et al., 2006) using the ERA5 wind speed data (Hersbach et al., 2020) for the
global ocean. The blue-shaded area corresponds to the fraction of ocean area in different latitudes (1° latitude average).

CO2 flux. This study typically adopts 1 standard deviation (i.e., 1 sigma) as a representation of uncertainty unless
specified otherwise.

3. Results
3.1.  Warm Bias in the In Situ SOCAT SST

The temperature assessment using the buoy SST suggests a cold bias in the DOISST v2.1 (0.09 K on average,
standard error 4.7 × 10 −4 K) and a small warm bias (0.02 K on average, standard error 4.1 × 10 −3 K) in the
SOCAT SST, which indicates that while a warm bias exists in the SOCAT SST, using the colocated DOISST
would overestimate this bias in SOCAT SST (Figure 2a).

Figure 2b shows the latitudinal variation of the bias in SOCAT SST. The number of grid cells with both SOCAT
and buoy data (green bars in Figure 2b) is small and the standard error for the temperature difference (gray shad-
ing) is large in the high latitude oceans. Therefore, we only consider data between 70°S and 70°N. The SOCAT
SST minus buoy SST (ΔSST, orange line in Figure  2b) shows apparent variation with latitude. ΔSST is on
average positive, but is slightly negative at 35°N and 30°S. In the northern hemisphere, ΔSST is +0.04 K near
the equator and increases by +0.1 K to a maximum at 25°N and then decreases to −0.05 K at 35°N. ΔSST also
increases from 35°N to a maximum of +0.15 K at 50°N and then decreases further north. The ΔSST pattern in
the southern hemisphere roughly mirrors that in the northern hemisphere with a 5° northward shift.

It is worth noting that under-sampling affects these bias assessments for SOCAT SST. If we consider all paired
cells with both buoy and SOCAT SST measurements, the warm bias is on average +0.02 K. If we only consider
cells with at least 10 buoy SST and 10 SOCAT SST measurements, the warm bias is on average +0.03 K (Figure
S2a in Supporting Information S1). The latitudinal variation of the bias is very similar no matter how many meas-
urements are within a cell (Figure S2b in Supporting Information S1).

DONG ET AL. 6 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Figure 3.  (a) Relationship between the cool skin effect and the 10 m wind speed (U10). Green bars represent the frequency
distribution of the ERA5 monthly averaged reanalysis wind speeds (1° × 1°) over the global ocean for 1982–2020. (b)
Latitudinal variation in U10 (red line) and the cool skin effect (1° latitude bins). Both subplots show the average cool skin
effect estimated by the Fairall96 physical model (Fairall et al., 1996, solid blue line), the Donlon02 wind speed-dependent
empirical model (Donlon et al., 2002, dashed blue line) and a constant value (−0.17 K, gray line; Donlon et al., 2002). The
light blue-shaded area in both subplots indicates one standard deviation of the bin averages in Fairall96 cool skin estimates.
Global ocean 1° × 1° monthly data sets are used to estimate the cool skin effect (see Section 2.3).

It is important to consider latitudinal variation when correcting for bias in SOCAT SST. For instance, SOCAT
SST has a relatively large warm bias (thus a large bias in the fCO2w) in the Southern Ocean (south of 35°S,
Figure 2b), which coupled with a high K660 and a large surface ocean area (Figure 2c) results in a substantial bias
in Southern Ocean CO2 flux estimates. This study uses a latitude-varying temperature bias (i.e., the orange line
in Figure 2b) to correct the air-sea CO2 flux between 70°S and 70°N (see Section 2.4).

3.2.  The Cool Skin Effect

Figure 3 shows the cool skin effect estimated by Donlon02 and Fairall96. The Fairall96 estimate of the cool skin
effect is stronger than the Donlon02 estimate for low wind speeds (U10 < 9 m s −1) but weaker for high wind speeds
(9 m s −1 < U10 < 16 m s −1) (Figure 3a). The monthly wind speed distribution (green bars in Figure 3a)  shows that
wind speeds less than 9 m s −1 account for 80% of the wind conditions. Therefore, the cool skin effect estimated
by Fairall96 is typically stronger than that estimated by Donlon02. The standard deviation of the Fairall96 cool
skin effect is much higher at low wind speeds than at high wind speeds, which reflects that the drivers (longwave
radiation, heat flux, and solar radiation) can produce substantial variations in the cool skin effect under relatively
calm conditions.

The Donlon02 cool skin effect only has a slight latitudinal variation that is not substantially different from a
constant (−0.17 K) value (Figure 3b), which was used by a previous study for air-sea CO2 flux correction (Watson
et al., 2020). In contrast, the Fairall96 cool skin estimate shows a clear latitudinal variation with two relatively
small cool skin effect regions at around 50°S and 50°N where wind speeds are high. The Fairall96 cool skin effect
is stable in the tropical zone and decreases toward both poles to ∼50° and then increases at even higher latitudes.

In most ocean regions, the Fairall96 cool skin effect follows variations in wind speed. Intriguingly, the Fairall96
cool skin effect is nearly constant within the tropical and subtropical zones, even though the wind speed is much
lower near the equator than in the subtropics. Drivers other than wind speed (i.e., latent and sensible heat fluxes,
and longwave radiation) might counteract the low wind speed effect in this area.

DONG ET AL. 7 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Figure 4.  Sea surface temperature (SST) corrections to the air-sea carbon dioxide (CO2) flux (ΔFlux) (a and b) versus time
and (c and d) versus latitude. SST corrections account for the bias in the (a–c) SOCAT SST and the (b–d) cool skin effect.
Negative ΔFlux values represent increased ocean CO2 uptake. Green and red lines represent ΔFlux due to the bias correction
assessed by drifting buoy SST (bias_buoy) and by colocated DOISST (bias_OI), respectively. Blue and purple lines represent
ΔFlux due to the Fairall96 and the Donlon02 cool skin corrections, respectively. ΔFlux in (a and b) is the global annual mean,
while ΔFlux in (c and d) is the long-term average (1982–2020) in 1° latitude bins. Results are based on the MPI-SOMFFN
fCO2w mapping method (Landschützer et al., 2013) (See Methods). The interannual variation of the global air-sea CO2 flux
with different temperature corrections can be seen in Figure S4 (Supporting Information S1). Our preferred corrections are
bias_buoy for warm bias in SOCAT SST and Fairall96 for the cool skin effect (see Section 4.1).

4. Discussion
4.1.  Variation in the CO2 Flux Correction

In this section, we discuss the impact of the warm bias and cool skin effects on global air-sea CO2 flux estimates.
The corrections are applied over time (between 1982 and 2020, Figures 4a and 4b) and by latitude (Figures 4c
and 4d).

The bias correction using the buoy SST assessment (bias_buoy) leads to an average increase in ocean CO2
uptake of 0.19 Pg C yr −1, while the bias correction utilizing the colocated DOISST (bias_OI) suggests an average
increase of 0.43 Pg C yr −1 (Figure 4a). Adopting the cool skin correction from Fairall96 and Donlon02 increases
the 1982–2020 average ocean CO2 uptake by 0.39 Pg C yr −1 and 0.43 Pg C yr −1, respectively (Figure 4b). A
constant cool skin correction of −0.17 K increases the flux by an amount similar to using the Donlon02 correc-
tion. Zhang et al. (2020) show that the mean difference between the Fairall96 cool skin effect and the observed
cool skin effect (7,239 observations) is 0.04 K. If we take this value as the uncertainty of the Fairall96 cool skin
estimate, the corresponding relative uncertainty in the Fairall96 flux correction is ∼20% (i.e., 0.08 Pg C yr −1). In
total, the flux correction using the bias_buoy and Fairall96 is on average ∼0.3 Pg C yr −1 lower than if the bias_OI
and Donlon02 are used from 1982 to 2020. The interannual variation in the net air-sea CO2 flux with different
temperature corrections is shown in Figure S4 in Supporting Information S1.

Figures 4a and 4c show the change in the air-sea CO2 flux (ΔFlux) generated by correcting for the warm bias in
SOCAT SST. The temporal and latitudinal variation of the two flux corrections (bias_buoy and bias_OI) follow
similar patterns, but the magnitude is different. Using bias_OI creates a ΔFlux that is twofold larger (in absolute
terms) than that using bias_buoy. The data in Figure 2a suggest that using bias_OI may overestimate the bias in
SOCAT SST, which would result in a ∼0.25 Pg C yr −1 overestimation of the air-sea CO2 flux correction. There-
fore, we favor the bias_buoy correction over the bias_OI correction.

While we use the same latitude-varying temperature difference (i.e., bias_buoy) to correct the bias in SOCAT
SST every year, the flux correction shows clear interannual variation (green line in Figure 4a). A possible reason
is that the number of measurements in each year of SOCAT is different (Figure S2 in Supporting Information S1),
and their spatial distribution differs between years. The latitude-dependent bias correction, when applied to the

DONG ET AL. 8 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Table 2
Global Mean Net Air-Sea Carbon Dioxide Fluxes From 1994 to 2007 (Numbers in the Text Are Generally the Mean From 1982 to 2020 Unless Specified Otherwise)
Flux with warm bias correction Flux with warm bias and cool skin correction

Net air-sea CO2 flux estimates (Pg C yr  )

−1
Flux without a temperature correction bias_buoy bias_OI bias_buoy + Fairall96 bias_OI + Donlon02
Ensemble mean of fCO2w-based fluxes  a
−1.7 ± 0.4 −1.8 ± 0.4 −2.0 ± 0.4 −2.2 ± 0.4 −2.4 ± 0.4
Ocean carbon inventory b −2.1 ± 0.4
Note. Here, bias_buoy and bias_OI represent the bias correction (to SOCAT sea surface temperature (SST)) using the assessment from buoy SST and colocated
DOISST, respectively. Fairall96 (Fairall et al., 1996) and Donlon02 (Donlon et al., 2002) correspond to the cool skin effect estimated by the physical and the empirical
models, respectively. We favor the bias_buoy and Fairall96 corrections (see Section 4.1).
 aThe ensemble mean of the fluxes from six fCO2 products and three wind speed products (Fay et al., 2021).  bFrom Gruber √ et al. (2019) (−2.6 ± 0.3 Pg C yr  ) with a
−1

riverine-derived carbon flux adjustment (0.53 ± 0.21 Pg C yr −1). The uncertainty (i.e., ±0.4 Pg C yr −1) is calculated
𝐴𝐴 as 0.302 + 0.212   Pg C yr −1.

different year-to-year spatial distribution in the SOCAT data, results in a time-varying annual mean bias correc-
tion (Figure S2 in Supporting Information S1).

Figures 4b and 4d show the change in air-sea CO2 flux when accounting for the cool skin effect using Fairall96
and Donlon02 models. Figure 4b indicates an increase over time in both flux corrections (absolute value), which
is driven by the increase in fCO2a (see Equation 1 and Table 1). The impact of the cool skin effect on the air-sea
CO2 flux is through αi * fCO2a. The ever rising atmospheric CO2 concentration and thus fCO2a, result in the grow-
ing cool skin flux correction.

The flux correction using Donlon02 exceeds that by Fairall96 by ∼0.05 Pg C yr −1 (in absolute terms). The largest
difference in flux between the two cool skin corrections occurs in the Southern Ocean (Figure 4d). The Donlon02
cool skin effect has minimal latitudinal variation, so the flux correction is largest at ∼50°S where the gas trans-
fer velocity is maximum and the ocean area is relatively large (Figure 2c). The Fairall96 cool skin effect has an
apparent latitudinal variation and a minimum (absolute) value at ∼50°S (Figure  3). This minimum cool skin
effect offsets the maximum wind speed and large ocean area, resulting in a smaller flux correction (in absolute
terms) at ∼50°S for Fairall96 than for Donlon02. Recent work (Alappattu et al., 2017; Embury et al., 2012; Zhang
et al., 2020) has suggested that the Fairall96 cool skin model is better than Donlon02 at capturing the cool skin
effect at a global scale and this, coupled with our estimates, indicates that using the Donlon02 model may lead to
an overcorrection of the air-sea CO2 flux, especially in the Southern Ocean.

4.2.  Implications for Air-Sea CO2 Flux Estimates

This study deals with the potential bias in the fCO2w-based air-sea CO2 flux estimates due to upper ocean temper-
ature effects. A large amount of uncertainty in this fCO2w-based flux also comes from the gas transfer velocity
(Woolf et al., 2019). The air-sea CO2 flux estimated from the ocean carbon inventory (Gruber et al., 2019) does
not require the gas transfer velocity, is unaffected by upper ocean temperature effects, and provides an independ-
ent estimate of ocean CO2 uptake. To compare the fCO2w-based net air-sea CO2 flux with the anthropogenic
air-sea CO2 flux of the ocean carbon inventory, we need to adjust for river-induced CO2 outgassing. The riverine
carbon flux has been estimated as 0.23 Pg C yr −1 (Lacroix et al., 2020), 0.45 Pg C yr −1 (Jacobson et al., 2007),
0.65 Pg C yr −1 (Regnier et al., 2022) and 0.78 Pg C yr −1 (Resplandy et al., 2018). Here, we adopt the mean of
these values (0.53 ± 0.21 Pg C yr −1).

The net air-sea CO2 flux derived from the ocean carbon inventory from 1994 to 2007 is −2.1 ± 0.4 Pg C yr −1
(i.e., −2.6 Pg C yr −1 anthropogenic flux plus 0.53 Pg C yr −1 river carbon flux; see the footnote of Table 2 for the
propagated uncertainty) (Gruber et al., 2019), which is shown in Table 2 along with the ensemble mean of eight-
een fCO2w-based fluxes (Fay et al., 2021). Fluxes from six fCO2w products and three wind speed products (three
wind products are used for each fCO2w product) are utilized to generate the ensemble mean flux, where missing
fCO2w has been filled with a scaled climatology and gas transfer velocity (K660) has been calibrated to a global
average of 18.2 cm hr −1 over the ice-free ocean based on  14C-bomb flux estimates (Fay et al., 2021). All six fCO2w
products (which include the MPI SOMFFN method) have been developed from the SOCAT v2021 data set. So
the corrections to the ensemble mean flux for the temperature effects should be similar to the corrections in this
study based on the MPI-SOMFFN fCO2w mapping method (Landschützer et al., 2013). Furthermore, an ensemble

DONG ET AL. 9 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

of different data interpolation methods and different wind products provides a more robust flux estimate than
a single interpolation method based on a single wind product. The flux corrections estimated in this study are
applied to the ensemble mean flux.

The ensemble mean air-sea CO2 flux without any bias and cool skin corrections (−1.7  ±  0.4  Pg  C  yr −1) is
0.4 Pg C yr −1 lower than the net flux estimate from the ocean carbon inventory. The ensemble mean CO2 flux
with bias_buoy and Fairall96 cool skin corrections is −2.2 ± 0.4 Pg C yr −1, similar to the ocean carbon inventory
derived net ocean CO2 uptake. The corrections using the bias_OI and the Donlon02 suggested by a previous
study (Watson et al., 2020) push the ensemble mean air-sea CO2 flux (−2.4 ± 0.4 Pg C yr −1) toward the lower
limit of the ocean carbon inventory flux estimate (Table 2). However, these comparisons depend on the choice
of the riverine carbon flux correction. The riverine flux is still an unresolved issue and the flux estimates span
from 0.23 Pg C yr −1 to 0.78 Pg C yr −1 (Jacobson et al., 2007; Lacroix et al., 2020; Regnier et al., 2022; Resplandy
et al., 2018). Without knowing which of the riverine flux estimates is most accurate, an average is simply taken
here. Therefore, an accurate estimate of the river flux is required to increase our confidence for the comparison
above.

Another question is whether the warm bias and cool skin flux corrections conflict with our understanding of
air-sea CO2 fluxes. One might argue that the preindustrial ocean and atmosphere would have been in a natural
equilibrium (i.e., the global total of steady state natural air-sea CO2 fluxes would have been zero; see Hauck
et al., 2020 for details), but the temperature corrections would create a preindustrial ocean carbon sink. However,
the warm bias in SOCAT SST is not a natural phenomenon and should not affect the preindustrial flux estimate.
Furthermore, while cool skin is a natural phenomenon, the flux correction due to the cool skin effect includes
both natural and anthropogenic contributions. Figure 4b shows that the cool skin flux correction decreased almost
linearly by ∼0.1 Pg C yr −1 (from −0.34 to −0.43 Pg C yr −1) due to the increase in atmospheric CO2 (∼70 ppm or
μmol mol −1, from 341 to 414 ppm) from 1982 to 2020 (Dlugokencky & Tans, 2018). Preindustrial atmospheric
CO2 was ∼260–280  ppm (Wigley,  1983), which is ∼70  ppm lower than atmospheric CO2 in 1982. Thus, the
preindustrial natural air-sea CO2 flux correction due to the cool skin effect could be ∼−0.25 Pg C yr −1, with
the remaining correction (∼–0.2 Pg C yr −1 in 2020) due to the increase in atmospheric CO2 by anthropogenic
emissions.

A flux correction for the cool skin effect is only related to the fCO2w observation-based flux estimate, which is
available from the 1980s onwards (Friedlingstein et al., 2022). There were no fCO2w measurements in preindus-
trial times, so the total preindustrial air-sea CO2 flux (the sum of steady state natural flux and river flux) is based
on model studies, theory, and lateral transport constraints (Hauck et  al., 2020). Although the cool skin effect
might result in an ∼−0.25 Pg C yr −1 flux, we can still assume that the ocean and atmosphere were in a natural
equilibrium in preindustrial times. Specifically, the cool skin effect has been implicitly included in the preindus-
trial natural equilibrium assumption. Therefore, this study improves our understanding by suggesting an increas-
ing anthropogenic contribution to the air-sea CO2 flux while there is no contradiction between the temperature
correction and the preindustrial natural equilibrium assumption.

The cool skin effect and its impact on the air-sea CO2 flux have been discussed for decades. While the cool skin
effect itself has been well observed and modeled, its impact on the air-sea CO2 flux is mainly based on theoretical
arguments. We still lack strong observational evidence to confirm the need to include the cool skin effect on esti-
mates of air-sea CO2 flux—an important topic we urge the community to demonstrate experimentally. The eddy
covariance method (e.g., Dong et al., 2021) provides direct flux measurements that could be used as a reference
CO2 flux to assess the accuracy of the bulk CO2 flux. Long-term eddy covariance measurements at a place with
|ΔfCO2| ∼ 0 would be insightful because the relative effect of cool skin on the bulk CO2 flux is in theory more
prominent for regions of low |ΔfCO2|. Appropriate laboratory experiments may yield further insight.

In summary, this work updates the temperature corrections to the fCO2w-based air-sea CO2 flux estimates. It shows
that there is a slight warm bias in SOCAT SST and a latitude-varying cool skin effect, resulting in ∼0.6 Pg C yr −1
additional ocean CO2 uptake from 1982 to 2020. The corrected air-sea CO2 flux for an ensemble of six gap-filled
air-sea CO2 flux products agrees well with the ocean carbon inventory derived net flux. The extreme sensitivity
of the air-sea CO2 flux to the accuracy of SST means that we should carefully choose the reference temperature
to assess any bias in the SOCAT SST. The importance of the Southern Ocean for atmospheric CO2 uptake, and

DONG ET AL. 10 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

the strong winds encountered there mean that large scale assessments need a suitable model for the cool skin
correction to the air-sea CO2 flux.

Data Availability Statement

Data can be accessed as follows. Gridded SOCAT v2021 data: https://www.socat.info/index.php/data-ac-
cess/. Reanalyzed sea surface CO2 fugacity data set using colocated DOISST: https://doi.org/10.18160/
vmt4-4563. In situ SST measurements (including the drifting buoy SST and the ship SST): https://www.
star.nesdis.noaa.gov/socd/sst/iquam/data.html. CCI SST v2.1: https://surftemp.net/regridding/index.html.
DOISST v2.1: https://www.ncei.noaa.gov/data/sea-surface-temperature-optimum-interpolation/v2.1/access/
avhrr/. ECMWF monthly averaged reanalysis data: https://cds.climate.copernicus.eu/cdsapp#!/dataset/
reanalysis-era5-single-levels-monthly-means?tab=form.

Acknowledgments References
We are grateful to H. Zhang (NOAA), C.
Merchant (University of Reading), and H. Alappattu, D. P., Wang, Q., Yamaguchi, R., Lind, R. J., Reynolds, M., & Christman, A. J. (2017). Warm layer and cool skin corrections for bulk
Beggs (Bureau of Meteorology, Australia) water temperature measurements for air-sea interaction studies. Journal of Geophysical Research: Oceans, 122(8), 6470–6481. https://doi.
for their advice on choosing the appropri- org/10.1002/2017JC012688
ate SST product, as well as to J. Kennedy Bakker, D. C. E., Alin, S., Castaño-Primo, R., Cronin, M., Gkrizalis, T., Kozyr, A., et al. (2021). SOCAT version 2021 for quantification of ocean
(Met Office Hadley Centre), and S. Zhou CO2 uptake. Retrieved from https://www.socat.info/index.php/data-access/
(British Antarctic Survey) for suggestions Bakker, D. C. E., Pfeil, B., Landa, C. S., Metzl, N., O’Brien, K. M., Olsen, A., et al. (2016). A multi-decade record of high-quality fCO2 data
on assessing the bias in the SOCAT in version 3 of the Surface Ocean CO2 Atlas (SOCAT). Earth System Science Data, 8(2), 383–413. https://doi.org/10.5194/essd-8-383-2016
SST. We also greatly appreciate the Dlugokencky, E., & Tans, P. (2018). Trends in atmospheric carbon dioxide. National Oceanic & Atmospheric Administration. Earth System
inspirational and helpful discussions with Research Laboratory (NOAA/ESRL). Retrieved from https://www.socat.info/index.php/data-access/
R. Wanninkhof (NOAA) and J. Shutler Dong, Y., Yang, M., Bakker, D. C. E., Kitidis, V., & Bell, T. G. (2021). Uncertainties in eddy covariance air–sea CO2 flux measurements and
(University of Exeter). The surface ocean implications for gas transfer velocity parameterisations. Atmospheric Chemistry and Physics, 21(10), 8089–8110. https://doi.org/10.5194/
CO2 Atlas (SOCAT) is an international acp-21-8089-2021
effort, endorsed by the International Donlon, C. J., Minnett, P.  J., Gentemann, C., Nightingale, T. J., Barton, I. J., Ward, B., & Murray, M. J. (2002). Toward improved valida-
Ocean Carbon Coordination Project tion of satellite sea surface skin temperature measurements for climate research. Journal of Climate, 15(4), 353–369. https://doi.
(IOCCP), the Surface Ocean Lower org/10.1175/1520-0442(2002)015<0353:tivoss>2.0.co;2
Atmosphere Study (SOLAS), and the Donlon, C. J., Robinson, I., Casey, K. S., Vazquez-Cuervo, J., Armstrong, E., Arino, O., et al. (2007). The global ocean data assimilation exper-
Integrated Marine Biogeochemistry and iment high-resolution sea surface temperature pilot project. Bulletin of the American Meteorological Society, 88(8), 1197–1214. https://doi.
Ecosystem Research program (IMBER), org/10.1175/BAMS-88-8-1197
to deliver a uniformly quality controlled Edson, J. B., Jampana, V., Weller, R. A., Bigorre, S. P., Plueddemann, A. J., Fairall, C. W., et al. (2013). On the exchange of momentum over the
surface ocean CO2 database. The many open ocean. Journal of Physical Oceanography, 43(8), 1589–1610. https://doi.org/10.1175/JPO-D-12-0173.1
researchers and funding agencies respon- Embury, O., Merchant, C. J., & Corlett, G. K. (2012). A reprocessing for climate of sea surface temperature from the along-track scanning
sible for the collection of data and quality radiometers: Initial validation, accounting for skin and diurnal variability effects. Remote Sensing of Environment, 116, 62–78. https://doi.
control are thanked for their contributions org/10.1016/j.rse.2011.02.028
to SOCAT. In this study, Y. Dong has Fairall, C. W., Bradley, E. F., Godfrey, J. S., Wick, G. A., Edson, J. B., & Young, G. S. (1996). Cool-skin and warm-layer effects on sea surface
been supported by the China Scholarship temperature. Journal of Geophysical Research, 101(C1), 1295–1308. https://doi.org/10.1029/95JC03190
Council (CSC/201906330072). The Fay, A. R., Gregor, L., Landschützer, P., McKinley, G. A., Gruber, N., Gehlen, M., et al. (2021). SeaFlux: Harmonization of air-sea CO2 fluxes
Natural Environment Research Council from surface pCO2 data products using a standardized approach. Earth System Science Data, 13(10), 4693–4710. https://doi.org/10.5194/
(NERC) has enabled D. C. E. Bakker's essd-13-4693-2021
work (PICCOLO, NE/P021395/1, and Friedlingstein, P., Jones, M. W., O’Sullivan, M., Andrew, R. M., Bakker, D. C. E., Hauck, J., et al. (2022). Global carbon budget 2021. Earth
CUSTARD, NE/P021263/1 projects). System Science Data, 14(4), 1917–2005. https://doi.org/10.5194/essd-14-1917-2022
The contributions of T. G. Bell and Gentemann, C. L., & Minnett, P. J. (2008). Radiometric measurements of ocean surface thermal variability. Journal of Geophysical Research,
M. Yang have been made possible by 113(C8), C08017. https://doi.org/10.1029/2007JC004540
support from the NERC (ORCHESTRA, Goddijn-Murphy, L. M., Woolf, D. K., Land, P. E., Shutler, J. D., & Donlon, C. (2015). The OceanFlux greenhouse gases methodology for deriv-
NE/N018095/1, and PICCOLO NE/ ing a sea surface climatology of CO2 fugacity in support of air-sea gas flux studies. Ocean Science, 11(4), 519–541. https://doi.org/10.5194/
P021409/1 projects) and the European os-11-519-2015
Space Agency AMT4oceanSatFluxCCN Gruber, N., Clement, D., Carter, B. R., Feely, R. A., van Heuven, S., Hoppema, M., et al. (2019). The oceanic sink for anthropogenic CO2 from
(4000125730/18/NL/FF/gp). In the end, 1994 to 2007. Science, 363(6432), 1193–1199. https://doi.org/10.1126/science.aau5153
we would like to thank both reviewers for Hauck, J., Zeising, M., Le Quéré, C., Gruber, N., Bakker, D. C. E., Bopp, L., et al. (2020). Consistency and challenges in the ocean carbon sink
their valuable comments and suggestions, estimate for the global carbon budget. Frontiers in Marine Science, 7, 1–22. https://doi.org/10.3389/fmars.2020.571720
which helped us to improve the quality of Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz-Sabater, J., et al. (2020). The ERA5 global reanalysis. Quarterly Journal
the manuscript. of the Royal Meteorological Society, 146(730), 1999–2049. https://doi.org/10.1002/qj.3803
Ho, D. T., Law, C. S., Smith, M. J., Schlosser, P., Harvey, M., & Hill, P. (2006). Measurements of air-sea gas exchange at high wind
speeds in the Southern Ocean: Implications for global parameterizations. Geophysical Research Letters, 33(16), L16611. https://doi.
org/10.1029/2006GL026817
Holding, T., Ashton, I. G., Shutler, J. D., Land, P. E., Nightingale, P. D., Rees, A. P., et al. (2019). The fluxengine air-sea gas flux toolbox:
Simplified interface and extensions for in situ analyses and multiple sparingly soluble gases. Ocean Science, 15(6), 1707–1728. https://doi.
org/10.5194/os-15-1707-2019
Huang, B., Liu, C., Banzon, V., Freeman, E., Graham, G., Hankins, B., et al. (2021). Improvements of the daily optimum interpolation sea surface
temperature (DOISST) version 2.1. Journal of Climate, 34(8), 2923–2939. https://doi.org/10.1175/JCLI-D-20-0166.1
Jacobson, A. R., Mikaloff Fletcher, S. E., Gruber, N., Sarmiento, J. L., & Gloor, M. (2007). A joint atmosphere-ocean inversion for surface fluxes
of carbon dioxide: 1. Methods and global-scale fluxes. Global Biogeochemical Cycles, 21, GB1019. https://doi.org/10.1029/2005GB002556

DONG ET AL. 11 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Jähne, B. (2009). Air-sea gas exchange. Elements of Physical Oceanography: A Derivative of the Encyclopedia of Ocean Sciences, 160–169.
https://doi.org/10.1016/B978-0-12-409548-9.11613-6
Jähne, B., Heinz, G., & Dietrich, W. (1987). Measurement of the diffusion coefficients of sparingly soluble gases in water. Journal of Geophysical
Research, 92(C10), 10767–10776. https://doi.org/10.1029/JC092iC10p10767
Johnson, K. S. (1982). Carbon dioxide hydration and dehydration kinetics in seawater. Limnology & Oceanography, 27(5), 849–855. https://doi.
org/10.4319/lo.1982.27.5.0849
Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, L., et al. (1996). The NCEP/NCAR 40-year reanalysis project. Bulletin
of the American Meteorological Society, 77(3), 437–472. https://doi.org/10.1175/1520-0477(1996)077<0437:tnyrp>2.0.co;2
Kennedy, J. J., Rayner, N. A., Atkinson, C. P., & Killick, R. E. (2019). An ensemble data set of sea surface temperature change from 1850:
The Met Office Hadley Centre HadSST.4.0.0.0 data set. Journal of Geophysical Research: Atmospheres, 124(14), 7719–7763. https://doi.
org/10.1029/2018JD029867
Kennedy, J. J., Rayner, N. A., Smith, R. O., Parker, D. E., & Saunby, M. (2011). Reassessing biases and other uncertainties in sea surface temper-
ature observations measured in situ since 1850: 2. Biases and homogenization. Journal of Geophysical Research, 116(D14), 1–22. https://doi.
org/10.1029/2010jd015220
Kent, E. C., Kennedy, J. J., Smith, T. M., Hirahara, S., Huang, B., Kaplan, A., et al. (2017). A call for new approaches to quantifying biases
in observations of sea surface temperature. Bulletin of the American Meteorological Society, 98(8), 1601–1616. https://doi.org/10.1175/
BAMS-D-15-00251.1
Lacroix, F., Ilyina, T., & Hartmann, J. (2020). Oceanic CO2 outgassing and biological production hotspots induced by pre-industrial river loads of
nutrients and carbon in a global modeling approach. Biogeosciences, 17(1), 55–88. https://doi.org/10.5194/bg-17-55-2020
Landschützer, P., Gruber, N., & Bakker, D. C. E. (2020). An observation-based global monthly gridded sea surface pCO2 and air-sea CO2 flux
product from 1982 onward and its monthly climatology. NCEI Accession. 160558.
Landschützer, P., Gruber, N., Bakker, D. C. E., & Schuster, U. (2014). Recent variability of the global ocean carbon sink. Global Biogeochemical
Cycles, 28(9), 927–949. https://doi.org/10.1002/2014GB004853
Landschützer, P., Gruber, N., Bakker, D. C. E., Schuster, U., Nakaoka, S., Payne, M. R., et  al. (2013). A neural network-based estimate of
the seasonal to inter-annual variability of the Atlantic Ocean carbon sink. Biogeosciences, 10(11), 7793–7815. https://doi.org/10.5194/
bg-10-7793-2013
Liss, P. S., & Slater, P. G. (1974). Flux of gases across the air-sea interface. Nature, 247(5438), 181–184. https://doi.org/10.1038/247181a0
Merchant, C. J., & Embury, O. (2020). Adjusting for desert-dust-related biases in a climate data record of sea surface temperature. Remote Sens-
ing, 12(16), 1–15. https://doi.org/10.3390/RS12162554
Merchant, C. J., Embury, O., Bulgin, C. E., Block, T., Corlett, G. K., Fiedler, E., et al. (2019). Satellite-based time-series of sea-surface tempera-
ture since 1981 for climate applications. Scientific Data, 6(1), 1–18. https://doi.org/10.1038/s41597-019-0236-x
Minnett, P. J., Smith, M., & Ward, B. (2011). Measurements of the oceanic thermal skin effect. Deep-Sea Research Part II Topical Studies in
Oceanography, 58(6), 861–868. https://doi.org/10.1016/j.dsr2.2010.10.024
Naegler, T. (2009). Reconciliation of excess  14C-constrained global CO2 piston velocity estimates. Tellus Series B Chemical and Physical Mete-
orology, 61(2), 372–384. https://doi.org/10.1111/j.1600-0889.2008.00408.x
Nightingale, P. D., Malin, G., Law, C. S., Watson, A. J., Liss, P. S., Liddicoat, M. I., et al. (2000). In situ evaluation of air-sea gas exchange parameter-
izations using novel conservative and volatile tracers. Global Biogeochemical Cycles, 14(1), 373–387. https://doi.org/10.1029/1999GB900091
O’Carroll, A. G., Armstrong, E. M., Beggs, H., Bouali, M., Casey, K. S., Corlett, G. K., et al. (2019). Observational needs of sea surface temper-
ature. Frontiers in Marine Science, 7, 571720. https://doi.org/10.3389/fmars.2019.00420
Pfeil, B., Olsen, A., Bakker, D. C. E., Hankin, S., Koyuk, H., Kozyr, A., et al. (2013). A uniform, quality controlled Surface Ocean CO2 Atlas
(SOCAT). Earth System Science Data, 5(1), 125–143. https://doi.org/10.5194/essd-5-125-2013
Prytherch, J., Farrar, J. T., & Weller, R. A. (2013). Moored surface buoy observations of the diurnal warm layer. Journal of Geophysical Research:
Oceans, 118(9), 4553–4569. https://doi.org/10.1002/jgrc.20360
Regnier, P., Resplandy, L., Najjar, R. G., & Ciais, P. (2022). The land-to-ocean loops of the global carbon cycle. Nature, 603(7901), 401–410.
https://doi.org/10.1038/s41586-021-04339-9
Resplandy, L., Keeling, R. F., Rödenbeck, C., Stephens, B. B., Khatiwala, S., Rodgers, K. B., et al. (2018). Revision of global carbon fluxes based
on a reassessment of oceanic and riverine carbon transport. Nature Geoscience, 11(7), 504–509. https://doi.org/10.1038/s41561-018-0151-3
Reynolds, R. W., & Chelton, D. B. (2010). Comparisons of daily sea surface temperature analyses for 2007-08. Journal of Climate, 23(13),
3545–3562. https://doi.org/10.1175/2010JCLI3294.1
Reynolds, R. W., Smith, T. M., Liu, C., Chelton, D. B., Casey, K. S., & Schlax, M. G. (2007). Daily high-resolution-blended analyses for sea
surface temperature. Journal of Climate, 20(22), 5473–5496. https://doi.org/10.1175/2007JCLI1824.1
Robertson, J. E., & Watson, A. J. (1992). Thermal skin effect of the surface ocean and its implications for CO2 uptake. Nature, 358(6389),
738–740. https://doi.org/10.1038/358738a0
Sabine, C. L., Feely, R. A., Gruber, N., Key, R. M., Lee, K., Bullister, J. L., et al. (2004). The oceanic sink for anthropogenic CO2. Science,
305(5682), 367–371. https://doi.org/10.1126/science.1097403
Sabine, C. L., Hankin, S., Koyuk, H., Bakker, D. C. E., Pfeil, B., Olsen, A., et al. (2013). Surface Ocean CO2 Atlas (SOCAT) gridded data prod-
ucts. Earth System Science Data, 5(1), 145–153. https://doi.org/10.5194/essd-5-145-2013
Saunders, P.  M. (1967). The temperature at the ocean-air interface. Journal of the Atmospheric Sciences, 24(3), 269–273. https://doi.
org/10.1175/1520-0469(1967)024<0269:ttatoa>2.0.co;2
Shutler, J. D., Wanninkhof, R., Nightingale, P. D., Woolf, D. K., Bakker, D. C. E., Watson, A., et al. (2019). Satellites will address critical science
priorities for quantifying ocean carbon. Frontiers in Ecology and the Environment, 18(1), 27–35. https://doi.org/10.1002/fee.2129
Sweeney, C., Gloor, E., Jacobson, A. R., Key, R. M., McKinley, G., Sarmiento, J. L., & Wanninkhof, R. (2007). Constraining global air-sea gas
exchange for CO2 with recent bomb  14C measurements. Global Biogeochemical Cycles, 21(2), GB2015. https://doi.org/10.1029/2006GB002784
Takahashi, T., Olafsson, J., Goddard, J. G., Chipman, D. W., & Sutherland, S. C. (1993). Seasonal variation of CO2 and nutrients in the
high-latitude surface oceans: A comparative study. Global Biogeochemical Cycles, 7(4), 843–878. https://doi.org/10.1029/93GB02263
Takahashi, T., Sutherland, S. C., Wanninkhof, R., Sweeney, C., Feely, R. A., Chipman, D. W., et al. (2009). Climatological mean and decadal
change in surface ocean pCO2, and net sea-air CO2 flux over the global oceans. Deep Sea Research Part II: Topical Studies in Oceanography,
56(8–10), 554–577. https://doi.org/10.1016/J.DSR2.2008.12.009
Wanninkhof, R. (2014). Relationship between wind speed and gas exchange over the ocean revisited. Limnology and Oceanography: Methods,
12(6), 351–362. https://doi.org/10.4319/lom.2014.12.351
Wanninkhof, R., Asher, W. E., Ho, D. T., Sweeney, C., & McGillis, W. R. (2009). Advances in quantifying air-sea gas exchange and environmen-
tal forcing. Annual Review of Marine Science, 1(1), 213–244. https://doi.org/10.1146/annurev.marine.010908.163742

DONG ET AL. 12 of 13
 Global Biogeochemical Cycles 10.1029/2022GB007360

Ward, B., Wanninkhof, R., McGillis, W. R., Jessup, A. T., DeGrandpre, M. D., Hare, J. E., & Edson, J. B. (2004). Biases in the air-sea flux
of CO2 resulting from ocean surface temperature gradients. Journal of Geophysical Research—C: Oceans, 109(8), 1–14. https://doi.
org/10.1029/2003JC001800
Watson, A. J., Schuster, U., Shutler, J. D., Holding, T., Ashton, I. G. C., Landschützer, P., et al. (2020). Revised estimates of ocean-atmosphere
CO2 flux are consistent with ocean carbon inventory. Nature Communications, 11(1), 1–6. https://doi.org/10.1038/s41467-020-18203-3
Weiss, R. F. (1974). Carbon dioxide in water and seawater: The solubility of a non-ideal gas. Marine Chemistry, 2(3), 203–215. https://doi.
org/10.1016/0304-4203(74)90015-2
Wigley, T. M. L. (1983). The pre-industrial carbon dioxide level. Climatic Change, 5(4), 315–320. https://doi.org/10.1007/BF02423528
Woolf, D. K., Land, P. E., Shutler, J. D., Goddijn-Murphy, L. M., & Donlon, C. J. (2016). On the calculation of air-sea fluxes of CO2 in the presence
of temperature and salinity gradients. Journal of Geophysical Research: Oceans, 121(2), 1229–1248. https://doi.org/10.1002/2015JC011427
Woolf, D. K., Shutler, J. D., Goddijn-Murphy, L., Watson, A. J., Chapron, B., Nightingale, P. D., et al. (2019). Key uncertainties in the recent
air-sea flux of CO2. Global Biogeochemical Cycles, 33(12), 1548–1563. https://doi.org/10.1029/2018GB006041
Xu, F., & Ignatov, A. (2014). In situ SST quality monitor (iQuam). Journal of Atmospheric and Oceanic Technology, 31(1), 164–180. https://doi.
org/10.1175/JTECH-D-13-00121.1
Yang, C., Leonelli, F. E., Marullo, S., Artale, V., Beggs, H., Nardelli, B. B., et al. (2021). Sea surface temperature intercomparison in the frame-
work of the copernicus climate change service (C3S). Journal of Climate, 34(13), 5257–5283. https://doi.org/10.1175/JCLI-D-20-0793.1
Zeebe, R. E., & Wolf-Gladrow, D. (2001). CO2 in seawater: Equilibrium, kinetics, isotopes (pp. 85–140). Elsevier Science.
Zhang, H., Beggs, H., Ignatov, A., & Babanin, A. V. (2020). Nighttime cool skin effect observed from infrared SST autonomous radiometer
(ISAR) and depth temperatures. Journal of Atmospheric and Oceanic Technology, 37(1), 33–46. https://doi.org/10.1175/JTECH-D-19-0161.1

DONG ET AL. 13 of 13