CO2 exchange coefficients from remotely sensed wind speed measurements: SSM/I versus QuikSCAT in 2000



This article is corrected by:

  1. Errata: Correction to “CO2 exchange coefficients from remotely sensed wind speed measurements: SSM/I versus QuikSCAT in 2000” by Mary-Elena Carr, Wenqing Tang, and W. Timothy Liu Volume 30, Issue 2, Article first published online: 17 January 2003


[1] We compare here the air-sea exchange coefficient for CO2 estimated with monthly mean wind speed measured by the Special Sensing Microwave Imager (SSM/I), KS, and by the scatterometer QuikSCAT, KQ, for the year 2000. KS and KQ present the same patterns, although are larger than in ∼65% of the world ocean. Zonal mean KS are consistently larger, except ∼50°S and north of 10°S in the Indian Ocean. Global oceanic uptake, FQ, estimated using KQ and climatological ΔpCO2 ranges from 0.43 (July) to 2.6 Gt C y−1 (December). The global sink estimated from SSM/I is ∼10% larger than FQ for most of the year. This comparison supports the use of SSM/I to quantify interannual variability of the global exchange coefficient of CO2.

1. Introduction

[2] Approximately half of annual anthropogenic CO2 emissions is taken up by the terrestrial biosphere and by the ocean [Feely et al., 2002]. The growth of the atmospheric concentration of CO2 varies considerably from year to year despite relatively invariant emission rates [Keeling et al., 1995]. Atmospheric inversion models ascribe this variability to changes in oceanic uptake of 1–3 Gt C y−1 [Battle et al., 2000]. However, interannual variability in the air-sea flux, estimated from extrapolations of measured oceanic CO2 concentrations based on its relationship with sea surface temperature (SST) [Lee et al., 1998] and from ocean circulation models with simple biogeochemistry [Le Quéré et al., 2000], is at most 0.4 Gt C. In these studies, the air-sea exchange coefficient (K) accounts for less than 20–30% of the interannual variability. We believe that climatological or modeled wind speeds likely underestimate variability in the exchange coefficient. Our ultimate goal is to use the existing 13-year time series of wind speed measurements from the Special Sensing Microwave Imager (SSM/I) to quantify interannual variability in K for this period. The uncertainty in wind speed derived from SSM/I (1–3 ms−1) is greater than that of the scatterometer QuikSCAT (∼1 ms−1), flying since late 1999 [Wentz, 1997; Liu, 2002]. Because of this smaller uncertainty and the ability to measure wind direction, although we do not use it here, we consider QuikSCAT our “best quality” wind speed source. In this study we compare the exchange coefficient estimated from SSM/I, KS, with that estimated from QuickSCAT, KQ, to evaluate the results obtained from SSM/I.

2. Background

[3] The flux of CO2 between ocean and atmosphere is given by F = K* ΔpCO2 where Δ pCO2 is the difference in partial pressure between the ocean and the atmosphere. The exchange coefficient, is given by K = s* kave where s is the solubility of CO2 and kave is the long-term gas transfer velocity [Wanninkhof, 1992], which depends on turbulence in the water boundary layer. There exist a range of parameterizations for the gas transfer velocity as a function of wind speed [Nightingale et al., 2000, and references therein] which vary by a factor of two or more globally [e.g., Feely et al., 2002; Boutin et al., 2002]. Wind-based parameterizations only provide an indirect estimate of K because they neglect, for example, explicit consideration of bubbles and surfactants, but they represent the easiest available approach, especially on the global scale.

[4] Three months of measurements from the Seasat scatterometer in 1978 provided a global estimate of K that was consistent with previous values [Etcheto and Merlivat, 1988]. Boutin and Etcheto [1991] concluded that the error in K due to the spatial integration and temporal sampling inherent to satellite wind measurements was less than the uncertainty associated with the choice of wind speed parameterization. Variability of K, estimated from the Geosat altimeter and SSM/I (1985–1992), was much larger at seasonal time scales than from year to year at high northern latitudes, while interannual variability in the tropics and in the Southern Ocean could match or exceed seasonal changes [Boutin and Etcheto, 1997]. The accuracy of K in the equatorial Pacific derived from SSM/I was order 0.14 · 10−2mol m−2y−1 μatm−1, an 8% relative error, while the scatterometer errors (ERS1/2 and NSCAT) were smaller [Boutin et al., 1999a]. NSCAT-based K estimates were in excellent agreement with those from wind speeds made by the CARIOCA drifter for the month of January 1997 in the equatorial Pacific, while SSM/I overestimated the wind speed [Boutin et al., 1999b].

3. Data and Methods

[5] We derived K using the formulation given above, where kave was estimated from the long-term average wind (monthly mean for 2000) and the Schmidt number. We correct for the dependence on temperature (Schmidt number and solubility) and on salinity (solubility) as in Wanninkhof [1992]. The wind speeds are monthly averages of the wind speed produced within NASA's Pathfinder Program for and those of Tang and Liu for QuikSCAT. The monthly mean temperature fields for 2000 are an optimally interpolated combination of Advanced Very High Resolution Radiometer (AVHRR) and ship observations [Reynolds and Smith, 1994]. We used climatological salinity fields from the World Ocean Atlas 1998 (http://www.nodc. The air-sea flux of was estimated using the monthly mean K for 2000 and climatological maps of Δ pCO2 [; Takahashi et al., 1997] which refer to 1995.

4. Results

[6] K estimated from the two sensors present the same patterns, though KS exceeds KQ in ∼65% of the world ocean. The annual mean (non-area-weighted) KS and KQ are 7.41 and 7.07 · 10−2mol m−2y−1μatm−1 respectively. The global monthly mean KS surpasses KQ by 0.16 (2%) to 0.6 · 10−2mol m−2y−1 μatm−1 (6.5%).

[7] The non-area-weighted zonal mean values (Figure 1a) follow the pattern described by Boutin and Etcheto [1997] with maxima associated with the westerlies (∼50: 8–10 · 10−2 mol m−2 y−1 μatm−1) and the trade winds (∼15: 5–8 · 10−2 mol m−2 y−1 μatm−1) and minima (<4 · 10−2 mol m−2 y−1 μatm−1) within 10° of the equator. The global mean ΔK (KS-KQ) is consistently positive except at 50°S (Figure 1b) and is minimum within 5° of 0°. ΔK in the Indian Ocean exceeds 1 1 · 10−2mol m−2y−1 μatm−1 at 20–25°S and is <−1 · 10−2 mol m−2 y−1 μatm−1 north of 15°N. These KS are consistent with those of Boutin and Etcheto [1997] in spite of different data handling, product version, and study period. The major discrepancy, the higher equatorial values seen here, are likely due to the new Pathfinder algorithm, as Boutin et al. [1999b] found that SSM/I overestimated winds in the equatorial Pacific.

Figure 1.

Zonal mean of the annual average KQ for the globe and for each basin (a) and the difference between KS and KQ for the globe and each basin (b). Figure (a) is not area-weighted; high latitudes represent small oceanic area.

[8] The seasonal cycle of K in the Atlantic and Pacific is weak in the equatorial band (10°N–10°S) and maximum poleward of 40°; peak amplitude is 8–10 · 10−2 mol m−2 y−1 μatm−1 north of 40°N in the Atlantic (Figure 2a). The equatorial (10°S–10°N) and tropical (>10°N) regions of the Indian Ocean reflect the monsoonal forcing. K is maximum in hemispheric winter in the Atlantic and Pacific and southern Indian Oceans (winter storms). The largest ΔK in the Atlantic and Pacific are ∼1 · 10−2 mol m−2 y−1 μatm−1 and occur in hemispheric fall and winter poleward of 10°. In the Indian Ocean, ΔK is consistently negative between 10°S and 25°N. SSM/I underestimates wind speed in the Arabian Sea in the monsoon period [e.g., Meissner et al., 2001], likely due to the geometry of sensor viewing and wind direction, and to the presence of cold upwelled water. Discrepancies at high latitudes in winter may be related to the tendency of SSM/I to overestimate high wind speeds or, to precipitation (which also affects QuikSCAT). Though ΔK is largest at high wind speeds due to the quadratic expression for K, we found no systematic relationship between ΔK and wind speed. ΔK obviously depends on both sensors, neither of which is perfect. However, our assumption that QuikSCAT is the best quality wind product is supported by the similarity of the ΔK distribution with other studies [Meissner et al., 2001].

Figure 2.

Seasonal progression of mean KQ and KS within latitude bands in each basin for 2000 (a). The solid (dashed) lines are the mean KQ (KS). Seasonal progression of ΔK (b). Figure (a) is not area-weighted; high latitudes represent small oceanic area.

[9] The global mean flux is always into the ocean; it is lowest in July and peaks in December. Flux from QuikSCAT, FQ, is −0.43 to −2.56 Gt C y−1, and that from SSM/I, FS, between −0.61 and −2.76 Gt C y−1. The consistently larger KS values translate into ΔF from −0.02 to −0.25 Gt C y−1 (Figure 3a). FS is at most 11% larger than FQ except in July and August (FS is 40–50% larger) when flux is minimum. The Atlantic and Indian Oceans are basin averaged sinks throughout 2000 (Figure 3b), while the Pacific is a weak source (<0.4 Gt C y−1) in boreal summer (June through September). The Atlantic basin is the strongest sink: FS and FQ are <−0.8 Gt C y−1 from October to April. Mean flux into the Indian Ocean is <−0.5 Gt C y−1 between April and December. FS on the basin average is consistently a larger sink than FQ in the Atlantic and Indian Oceans, while it is weaker than or close to FQ most of the year in the Pacific (Figure 3b).

Figure 3.

Monthly global average of CO2 flux (a), and for each basin (b) for 2000. The solid (dashed) lines are the mean FQ (FS) in (b).

[10] The seasonal cycle of flux is most pronounced between 10°N and 40°N (all basins) and between 10°S and 40°S in the Indian Ocean (Figure 4a). Major sinks (∼−0.5 Gt C y−1) are found year-round south of 40°S (all basins) and poleward of 40°N in the Atlantic. The equatorial band (10°S–10°N) is a source year-round: ∼0.1 Gt C y−1 in the Atlantic and Indian and >0.5 Gt C y−1 in the Pacific. The Indian Ocean is a strong sink between 10° and 40° (<−0.5 Gt C y−1 between June and November) and a year-round source north of 10°N, peaking June through September (∼0.35 Gt C y−1). Our basin estimates (including the southern ocean) of FQ (FS) of −0.69 (−0.74), −0.20 (−0.19), −0.58 (−0.68) Gt C y−1 for the Atlantic, Pacific, and Indian respectively are in reasonable agreement with those of Feely et al. [2002]: (north of 50°S) of −0.69, −0.21, −0.33 Gt C y−1, plus −0.58 Gt C y−1 for the Southern Ocean.

Figure 4.

Monthly global average of CO2 flux for each basin for 2000 (a). The solid (dashed) lines are the mean FQ (FS) in the upper. Seasonal progression of ΔF (b).

[11] ΔF exceeds 0.05 Gt C y−1 in the equatorial Pacific, between 10°N and 40°N in the Pacific and Indian, and between 10°S and 40°S in the Indian Ocean (Figure 4b). In the Atlantic, ΔF is always within ±0.05 Gt C y−1. The ΔK distribution in the Atlantic and Pacific basins is comparable in both source and sink regions, so flux compensates in the basin mean. By contrast in the Indian Ocean, ΔK is negative (north of 10°S) in regions of positive Δ pCO2 and positive in the region of negative Δ pCO2 (10–40°S). The net result is that uptake estimated by SSM/I is consistently larger than that estimated from QuikSCAT north of 40°.

5. Conclusions

[12] The product versions of the two sensors used in this study provide comparable estimates of K and flux. Future changes in the products will likely further decrease the discrepancies (Wentz, pers. comm.). Although KS is larger than KQ in the comparison year 2000, the flux estimate in the Atlantic and Pacific is compensated. In the Indian Ocean, ΔK changes sign with Δ pCO2, leading to a greater mismatch of flux estimates. Globally, the discrepancies in K of ∼0.3 · 10−2 mol m−2 y−1 μatm−1 (5%) translate to uncertainties in flux on the order of 0.12 Gt C (9%). These results validate the use of SSM/I to study interannual variability in the global exchange coefficient of CO2. Ongoing comparison of KQ and KS will improve our understanding of the regional differences and enable quantitative use KS of for periods in which we do not have scatterometer data.


[13] We thank Frank Wentz for the SSM/I data and several valuable suggestions and Taro Takahashi for the climatological Δ pCO2 fields. Richard Feely and Rik Wanninkhof provided helpful discussions and comments on an earlier version. We are grateful to Xiaosu Xie for providing monthly mean QuikSCAT wind speed data from PODAAC to compare with the Tang and Liu product and to two anonymous reviewers for thorough comments that improved the original manuscript. The research described in this paper was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration.