Sparse observations induce large biases in estimates of the global ocean CO2 sink: an ocean model subsampling experiment

Abstract

Estimates of ocean [Formula: see text] uptake from global ocean biogeochemistry models and [Formula: see text]-based data products differ substantially, especially in high latitudes and in the trend of the [Formula: see text] uptake since 2000. Here, we assess the effect of data sparsity on two [Formula: see text]-based estimates by subsampling output from a global ocean biogeochemistry model. The estimates of the ocean [Formula: see text] uptake are improved from a sampling scheme that mimics present-day sampling to an ideal sampling scheme with 1000 evenly distributed sites. In particular, insufficient sampling has given rise to strong biases in the trend of the ocean carbon sink in the [Formula: see text] products. The overestimation of the [Formula: see text] flux trend by 20-35% globally and 50-130% in the Southern Ocean with the present-day sampling is reduced to less than [Formula: see text] with the ideal sampling scheme. A substantial overestimation of the decadal variability of the Southern Ocean carbon sink occurs in one product and appears related to a skewed data distribution in [Formula: see text] space. With the ideal sampling, the bias in the mean [Formula: see text] flux is reduced from 9-12% to 2-9% globally and from 14-26% to 5-17% in the Southern Ocean. On top of that, discrepancies of about [Formula: see text] (15%) persist due to uncertainties in the gas-exchange calculation. This article is part of a discussion meeting issue 'Heat and carbon uptake in the Southern Ocean: the state of the art and future priorities'.

Keywords: carbon dioxide; observation system design; ocean carbon sink; pCO2 observations.

Figure 1.

Annual mean partial pressure of ${\mathrm{CO}}_2$ (${\mathrm{p}\mathrm{CO}}_2$) averaged over the period 2009–2018 from the output from the global ocean biogeochemistry model FESOM-REcoM. Upper left figure shows the full model output, and the other three panels show the model output after subsampling according to the three masks, as indicated in the title (SOCAT, $\mathrm{SOCAT}+\mathrm{SOCCOM}$, bgcArgo). In the bgcArgo panel, the white background shading in the Arctic depicts the Arctic region that is excluded from all analyses. See text for more explanation.

Figure 2.

Number of monthly $1^{\circ }\times 1^{\circ }$ grid cells covered per year. Shown is the data coverage in the global ocean (top) and three large-scale regions, top to bottom: North (north of ${30}^{\circ }\mathrm{N}$, excluding the Arctic), Tropics (${30}^{\circ }\mathrm{S}$ to ${30}^{\circ }\mathrm{N}$), South (south of ${30}^{\circ }\mathrm{S}$). The bars illustrate the data coverage from the SOCAT (dark orange: October to March (Southern Hemisphere summer, Northern Hemisphere winter), dark red: April to September) and SOCCOM (yellow: October to March light orange: April to September) data sets. The hypothetical ideal bgcArgo data coverage is constant in time as indicated by the dark blue horizontal line. Note the different axis scales. Average number of filled grid cells is given within the figures for the periods before 2000, 2000–2018 and 2014–2018. Total number of monthly $1^{\circ }\times 1^{\circ }$ ocean grid cells per region is given at the top of the panels.

Figure 3.

Data distribution in ${\mathrm{pCO}}_2$ space for the regions: (top to bottom) global (no Arctic), North (no Arctic), Tropics, and South, and for the periods (from left to right) 1982–2000, 2000–2018 and the last 5 years 2014–2018. The thick grey line depicts the full model output, and the coloured lines the three sampling masks (SOCAT: yellow; $\mathrm{SOCAT}+\mathrm{SOCCOM}$: dashed orange; bgcArgo: red).

Figure 4.

Annual mean air-to-sea ${\mathrm{CO}}_2$ flux (${\mathrm{PgC}\hspace{0.17em}\mathrm{yr}}^{-1}$) from the ${\mathrm{pCO}}_2$ reconstructions (coloured lines) compared with the original FESOM-REcoM ${\mathrm{CO}}_2$ flux (black line). The left column shows the MPI-SOM-FFN and the right column the CarboScope reconstructions for the three sampling schemes as indicated in the figures. MPI-SOM-FFN and CarboScope fluxes are calculated from mapped ${\mathrm{pCO}}_2$ and FESOM-REcoM gas-transfer velocity. From top to bottom: Global without Arctic, North (north of ${30}^{\circ }\mathrm{N}$) with Arctic excluded, Tropics (${30}^{\circ }\mathrm{S}$ to ${30}^{\circ }\mathrm{N}$), South (south of ${30}^{\circ }\mathrm{S}$). Positive fluxes denote a flux into the ocean.

Figure 5.

Effect of sampling distribution on statistics of annual mean air-to-sea ${\mathrm{CO}}_2$ flux (${\mathrm{PgC}\hspace{0.17em}\mathrm{yr}}^{-1}$) calculated from the ${\mathrm{pCO}}_2$ reconstructions with FESOM-REcoM gas-transfer velocity (coloured bars) compared with the original FESOM-REcoM ${\mathrm{CO}}_2$ flux (black line). From left to right: Mean ${\mathrm{CO}}_2$ flux 2009–2018 (${\mathrm{PgC}\hspace{0.17em}\mathrm{yr}}^{-1}$), trend 2000–2018 (${\mathrm{PgC}\hspace{0.17em}\mathrm{yr}}^{-1}\hspace{0.17em}{\mathrm{decade}}^{-1}$), amplitude of variability (standard deviation of detrended time-series) 1982–2018, phasing of variability (correlation coefficient of detrended time-series) 1982–2018. The blue bars show the MPI-SOM-FFN and the green bars the CarboScope reconstructions for the three sampling schemes as indicated in the figures (SOCAT: light colour, $\mathrm{SOCAT}+\mathrm{SOCCOM}$: medium colour, bgcArgo: dark colour). These statistics are calculated from the area-integrated time-series for (top to bottom): global excluding the Arctic, North excluding the Arctic, Tropics, South.

Figure 6.

Air–sea ${\mathrm{CO}}_2$ flux in FESOM-REcoM and in the reconstructions by MPI-SOM-FFN (left) and CarboScope (right) products. Positive fluxes (purple) denote a flux into the ocean. In the difference maps, positive numbers (blue) denote a larger flux into the ocean (or a smaller flux out of the ocean) in the reconstruction than in FESOM-REcoM. Top: FESOM-REcoM, then from top to bottom: reconstructed air–sea ${\mathrm{CO}}_2$ flux in SOCAT sampling scheme, in bgcArgo scheme, and difference in air–sea ${\mathrm{CO}}_2$ flux between reconstruction and FESOM-REcoM in SOCAT and bgcArgo schemes, as indicated in the titles. Only the SOCAT and bgcArgo sampling schemes are shown. The differences between SOCAT and $\mathrm{SOCAT}+\mathrm{SOCCOM}$ sampling schemes are small.

Figure 7.

Biases in reconstructed annual mean surface ${\mathrm{pCO}}_2$ ($\mu \mathrm{atm}$) (coloured lines) as calculated from the two ${\mathrm{pCO}}_2$ products minus the original FESOM-REcoM ${\mathrm{pCO}}_2$. The left column shows the MPI-SOM-FFN and the right column the CarboScope reconstructions for the three sampling schemes as indicated in the figures. From top to bottom: Global (excluding Arctic), North (excluding Arctic), Tropics, South.

Figure 8.

Spatial patterns of biases in reconstructed surface ${\mathrm{pCO}}_2$ (average 2009–2018) as calculated from the two ${\mathrm{pCO}}_2$ products minus the original FESOM-REcoM ${\mathrm{pCO}}_2$, for MPI-SOM-FFN (left) and CarboScope (right) in the sampling schemes SOCAT (top) and bgcArgo (bottom).

Figure 9.

Effect of choices in gas-exchange calculations based on the ideal sampling case for (top to bottom:) global without Arctic, North without Arctic, Tropics, South. All panels show the ‘known truth’ of the FESOM-REcoM air–sea ${\mathrm{CO}}_2$ flux (flux integrated on native model mesh, black). Left column: MPI-SOM-FFN ${\mathrm{CO}}_2$ flux calculated from mapped ${\mathrm{pCO}}_2$ based on ideal sampling with either FESOM-REcoM piston velocity (blue) or with MPI-SOM-FFN native gas-exchange formulation (red). Middle column: CarboScope ${\mathrm{CO}}_2$ flux calculated from mapped ${\mathrm{pCO}}_2$ based on ideal sampling with its native prior from OCIM and FESOM-REcoM piston velocity (green, same as in figure 4), with FESOM-REcoM (FR) prior and FESOM-REcoM piston velocity (light grey, dashed), and with FESOM-REcoM prior and CarboScope native gas-exchange (purple). Right column: Test cases to quantify the error introduced by recalculating the air–sea ${\mathrm{CO}}_2$ flux from FESOM-REcoM ${\mathrm{pCO}}_2$ fields (full global coverage) and FESOM-REcoM piston velocity (green), from FESOM-REcoM ${\mathrm{pCO}}_2$ fields (full global coverage) and FESOM-REcoM gas-transfer velocity scaled to a global mean of $16.5\hspace{0.17em}{\mathrm{cm}\hspace{0.17em}\mathrm{h}}^{-1}$ (orange). Also shown is a model experiment with higher gas-transfer coefficient a (0.31 instead of 0.251, light blue).