Surface air temperature (TA), sea surface temperature (TO), and surface specific humidity (qa) satellite retrievals from the Atmospheric Infrared Sounder (AIRS) are compared with shipboard measurements across Drake Passage for the period from September 2002 to June 2007. The objective is to evaluate whether AIRS retrievals, in conjunction with microwave sea surface temperatures from the Advanced Microwave Scanning Radiometer (AMSRE), can provide sufficiently accurate parameters to estimate sensible and latent heat fluxes in the data-limited Southern Ocean. The collocated data show that both AIRS TA and TO are colder than those from shipboard measurements, with a time mean bias of −2.03°C for TA and −0.22°C for TO. Results show that air-sea temperature difference (TA − TO), qa, and relative humidity (RH) are the major factors contributing to the differences between satellite and shipboard temperature measurements. Differences in AIRS and shipboard TA (ΔTA) decrease with increasing TA − TO, and ΔTA increases with increasing RH, whereas differences in AIRS and shipboard TO (ΔTO) increase with both increasing TA − TO and increasing qa. The time mean qa from AIRS is lower than the shipboard qa by 0.69 g/kg. Statistical analyses suggest that TA − TO, cloud, and qa are the major contributors to the qa difference (Δqa). Δqa becomes more negative with increasing TA − TO and increasing cloud fraction. Δqa also becomes more negative as qa increases. Compared with TA, TO, and TA − TO, from the National Centers for Environmental Prediction/National Center for Atmospheric Research Reanalysis (NCEP), AIRS-derived and AMSRE-derived variables show more small-scale spatial structure, as is also typical of the ship observations. Although AIRS qa gives a better representation of the full range of values of shipboard qa, its deviation from shipboard qa is relatively large compared to NCEP qa. Compared with several existing gridded flux products, turbulent fluxes estimated from AIRS and AMSRE data using bulk algorithms are better able to represent the full range of flux values estimated from shipboard parameters.
 Heat exchange between the atmosphere and the ocean is an important process in the climate system and reflects the interaction of the coupled climate system [e.g., Barsguli and Battisti, 1998; Ciasto and Thompson, 2008]. Global air-sea heat fluxes have been used in numerous aspects of climate studies over many time scales [Taylor, 2000; Curry et al., 2004] and are needed to evaluate coupled atmosphere-ocean models and weather forecasting models. Thus, accurate estimates of air-sea heat fluxes are important, particularly in high latitudes, where air-sea exchanges play a key role in deepwater formation, which, in turn, is related to the meridional overturning circulation [e.g., Kuhlbrodt et al., 2007]. However, commonly used surface flux estimates from the National Centers for Environmental Prediction/National Center for Atmospheric Research Reanalysis (NCEP) and the European Centre for Medium-Range Weather Forecasts Reanalysis 40 can differ by 100 W/m2 or more on any given day, compared with typical daily values of 150–250 W/m2 [Dong et al., 2007]. An assessment of the upper ocean heat balance in the Southern Ocean [Dong et al., 2007] suggested that air-sea heat flux is the largest contributor to errors in the heat balance. A better air-sea heat flux product is needed to improve our understanding of upper ocean processes related to the meridional overturning circulation.
 The net air-sea heat flux includes four components: shortwave radiation, longwave radiation, sensible heat flux, and latent heat flux. Standard bulk flux formulas for the sensible (SH) and latent (LH) heat components depend on air-sea temperature difference, specific humidity, and wind speed at the sea surface [e.g., Smith, 1988; Fairall et al., 1996],
where ρa is air density, Ce is the exchange coefficient, Cp is heat capacity, and W is near-surface wind speed relative to a stationary ocean. TO and TA are surface ocean and air temperature, respectively. Le is latent heat of vaporization, and qs and qa are saturated specific humidity at the temperature of the ocean surface and the specific humidity of the air above it, respectively.
 It is impractical to make in situ measurements of these parameters with sufficiently high spatial and temporal resolution to estimate the global heat flux for climate studies. Satellite measurements for sea surface temperature (TO) and wind speed have been readily available since the late 1980s, providing a relatively long time series of these climate variables to compute latent and sensible heat fluxes at global scales [Bentamy et al., 2003; Curry et al., 2004, Singh et al., 2005; Tomita and Kubota, 2006]. Satellite retrievals of near-surface air temperature and humidity have proven difficult, although a set of sensors aboard the NASA Earth Observing System (EOS) Aqua satellite offer new possibilities for measuring these quantities. Jackson et al.  assessed the possibility of combining microwave measurements from the Advanced Microwave Sounding Unit-A with other microwave satellite sensors to retrieve near-surface air temperature and humidity. Here, we examine the potential for improving surface flux estimates by using observations from the Aqua Atmospheric Infrared Sounder (AIRS), which provides three-dimensional maps of air, water, and cloud properties, with the potential to obtain higher spatial resolution than is possible with microwave sensors. We will use AIRS observations in combination with all-weather TO from the Advanced Microwave Scanning Radiometer (AMSRE), also on board Aqua, to estimate sensible and latent heat fluxes in the Drake Passage.
 The goal of AIRS is to provide more accurate information on the vertical profiles of atmospheric temperature and moisture for climate research and weather forecasting. Divakarla et al.  suggested that temperature and water vapor profiles from AIRS are in very good agreement with those from radiosonde observations, which were collocated within 1 h temporal separation and 200 km spatial separation. Although the atmospheric boundary layer is not the focus of the AIRS measurements, near-surface parameters are either provided directly in AIRS products or can be extracted from the vertical profiles. The near-surface parameters that AIRS measures, such as air temperature and specific humidity, are required by the bulk formulas that estimate air-sea heat fluxes (i.e., equations (1) and (2)). However, validation for the AIRS measurements has been done mainly in the nonpolar regions [Fetzer, 2006], and high-latitude conditions are not yet well represented in the AIRS validation/calibration process. Susskind et al.  found a small degradation of AIRS retrieval accuracy with increasing cloud cover, and high-latitude regions such as the Southern Ocean can experience particularly frequent cloud cover. Thus, it is important to first validate AIRS retrievals in the Southern Ocean to determine their potential for calculating sensible and latent heat fluxes. Recently, relative humidity from the AIRS measurements over Antarctica has been validated using radiosonde observations [Gettelman et al., 2006], although the radiosondes in that study were over ice, where the atmospheric boundary layer may behave differently than it does over the open ocean.
 In this study we present one of the first attempts to evaluate the performance of the AIRS surface air temperature (TA), sea surface temperature (TO), and specific humidity (qa) in the Southern Ocean. Our focus on the Southern Ocean contrasts with the bulk of satellite calibration and validation studies, which have for the most part made use of buoy and ship observations primarily from the tropics and subtropics [e.g., Mears et al., 2001; Barton et al., 2004; Gentemann et al., 2004]. Our ultimate goal is to determine whether they are accurate enough to use in sensible and latent heat flux estimations. The Antarctic research vessel and supply ship Laurence M. Gould of the U.S. Antarctic Program provides measurements of surface atmospheric parameters to evaluate the accuracy of AIRS surface measurements in the Drake Passage. The data sets used in this study and the process of collocating the various data sets are described in section 2. In section 3 we compare AIRS near-surface temperature retrievals with shipboard measurements and discuss possible causes for their differences. Our results show that both air-sea temperature difference (TA − TO) and specific humidity (qa) affect the near-surface temperature differences between satellite retrievals and shipboard measurements. Comparison of surface qa from AIRS retrieval and shipboard measurements and possible causes for their differences are discussed in section 4. The seasonal and cross-passage variations in near-surface temperature and specific humidity differences between AIRS and shipboard measurements are given in section 5. In section 6, we examine the potential uncertainties in the derived turbulent latent and sensible heat fluxes that use the AIRS near-surface retrievals. Finally, we present a discussion and summary in section 7.
2. Data Description and Processing
2.1. Data Description
2.1.1. Atmospheric Infrared Sounder (AIRS) Data
 AIRS is one of six instruments on board the EOS Aqua satellite, representing the most advanced atmospheric sounding system ever deployed in space. The Aqua satellite, launched in May 2002, follows a Sun-synchronous polar orbit, so observations occur at roughly the same local time each day. Although the observational times of AIRS are not exactly fixed due to precession and changes in the sampling location across the scan, in general the orbit moves northward at 1330 and southward at 0130 at equator local time and crossings at 55°S are around 1500 and 0100 local time. In the Drake Passage (55°S–62°S) the daily observational time varies within a 2 h window. AIRS is a cross-track scanning, high-spectral resolution infrared sounder, which is able to retrieve an entire profile of temperature and water vapor in the presence of up to 70% cloud cover [Aumann et al., 2003]. In this study we evaluate the version 5 daily level 3 (L3) near-surface products [Susskind et al., 2003]. The AIRS L3 products are derived from L2 (satellite swath) data that have been binned and averaged onto a 1° × 1° grid. For this study, we use surface TA at 2 m above sea level and TO directly from the AIRS L3 products, and we determine surface qa from the base of the vertical profile of AIRS qa data, as discussed in section 2.2. The data have been validated for nonpolar latitudes (50°S–50°N), but few high-latitude observations are available to provide validation in polar conditions. We validate AIRS surface parameters (TA, TO, and qa) in the Drake Passage over a nearly 5 year period from September 2002 (the start time of the AIRS data) through June 2007. In nonpolar regions, the estimated uncertainties in AIRS near-surface variables are relatively large: the uncertainty for both AIRS TA and TO is 1 K [Aumann et al., 2006], and for qa it is 15%–25% of qa [Tobin et al., 2006].
 The bulk formula for sensible heat flux (equation (1)) depends on the air-sea temperature difference (TA − TO), not on TA or TO alone. Although AIRS provides TO, the infrared measurements are still strongly influenced by water vapor and cloud [e.g., Merchant and Harris, 1999; Vazquez-Cuervo et al., 2004]. For AIRS products, the error estimates are generated for TA and TO independently although using the same methodology. Error thresholds are applied that are stricter for TO than for TA. This can result in observation times when TO are masked out, mostly due to cloud effects, but TA is provided (E. T. Olson, personal communication, 2010). In our Drake Passage study area, AIRS TO is unavailable ∼60% of the total time period. In contrast to infrared radiation, microwave radiation is capable of penetrating through cloud, which is particularly useful in the perpetually cloudy Southern Ocean. Thus, in addition to evaluating the AIRS retrievals, here we also evaluate TO from AMSRE, a multichannel passive microwave radiometer on board the same Aqua satellite [Wentz and Meissner, 2000, 2007]. A comparison of the availability of the AIRS TO with the availability of the AMSRE TO suggested that 80% of the AIRS gaps were due to the presence of clouds. The daily version 5 AMSRE TO are on a 0.25° longitude by 0.25° latitude grid. In previous work, Dong et al. [2006a] found that, in comparison with infrared measurements, AMSRE provides TO measurements with little bias relative to in situ observations in the Southern Ocean. Finally, we compare Drake Passage shipboard measurements of TA − TO with the satellite TA − TO derived using TA from AIRS and TO, either from AIRS or from AMSRE, and we ask specifically whether these are sufficiently accurate to be useful for sensible heat flux estimates.
2.1.2. Shipboard Measurement
 To date, the deployment of large surface meteorological buoys has been prevented in the Southern Ocean because of the frequency of severe weather conditions. For the same reason, there is relatively little ship traffic in the Southern Ocean. However, in 2002, new highly accurate meteorological and underway seawater intake systems were installed on the R/V Laurence M. Gould (LMG), the principal supply ship of the U.S. Antarctic Program for Palmer Station. The LMG crosses the Drake Passage approximately 20 times per year, providing year-round measurements for determination of the air-sea fluxes in the region. The TA, relative humidity (RH), and pressure are measured at 10 m above sea level, and TO is measured by a hull-mounted thermosalinograph (TSG) at 4 m below sea level. To compare with AIRS qa, we calculated the shipboard qa as the saturated humidity multiplied by the RH, where the saturated humidity is computed as a function of the shipboard TA and pressure. As discussed in section 2.2, these measurements were adjusted slightly to represent the same heights as the AIRS retrievals. The 1 s raw data are averaged to 5 min sampling intervals. The meteorological instrumentation and TSG are generally calibrated on an annual basis. The calibration error specified by the TSG manufacturer is 0.01°C. However, Dong et al. [2006a] suggested that TSG measurements are 0.15°C warmer than measurements from high-resolution expendable bathythermographs in the Drake Passage region. The manufacturer's reported accuracies for TA, RH, and atmospheric pressure from the shipboard automated meteorological system are 0.1°C, 1%, and 0.5 hPa, respectively, which implies a qa accuracy of 0.07 g/kg. A total of 88 LMG crossings were completed over the concurrent period of the AIRS measurements, from September 2002 to June 2007 (Figure 1).
 In addition to evaluating AIRS retrievals, we are also interested in the performance of AIRS surface parameters relative to the NCEP products that have been widely used in estimating air-sea heat fluxes [e.g., Yu and Weller, 2007]. Dong et al.  suggested that, compared to available air-sea heat flux products, the NCEP variables provided the best air-sea heat flux estimates to close the heat budget in the Southern Ocean at the time of the study. Here we compare TO and TA at 2 m above sea level and surface qa from NCEP with shipboard measurements from the LMG. This comparison will show how well the NCEP products in the Southern Ocean capture the variability of shipboard observations and will assess whether the AIRS products provide parameters that may lead to better estimates of latent and sensible heat fluxes. NCEP products have better temporal resolution (6 h) but coarser spatial resolution (1.88° in latitude and longitude) than AIRS L3 products. Thus, we would not expect that the NCEP data capture the strong frontal structure that is characteristic of the Antarctic Circumpolar Current [Dong et al., 2006b].
2.1.3. Surface Turbulent Heat Fluxes
 A number of other turbulent heat flux products have been developed recently using satellite measurements, including three global products that are available during the first few years of our study period. Objectively analyzed air-sea heat fluxes (OAFlux) [Yu and Weller, 2007] are constructed through a combination of satellite observations, surface moorings, ship reports, and atmospheric model reanalyzed surface meteorology. Daily OAFlux products are available on a 1° grid for the period 1985–2006. Turbulent heat fluxes from the Japanese Ocean Flux data sets with Use of Remote sensing Observations version 2 (J-OFURO2) [Kubota and Tomita, 2007] are estimated mainly using satellite measurements, except TA, which is from NCEP. The J-OFURO2 fluxes are also available on a 1° spatial resolution and one-day temporal resolution from 1988 to 2005. The Hamburg ocean atmosphere parameters and fluxes from satellite data (HOAPS-3) [Andersson et al., 2007] provide satellite-derived turbulent heat fluxes with a 1° spatial grid and twice-daily temporal resolution covering the period July 1987 to 2005. The air temperature used in HOAPS-3 for the derivation of sensible heat flux is estimated from near-surface specific humidity and TO. All three flux products are derived using the Coupled Ocean-Atmospheric Response Experiment bulk flux algorithm (COARE) [Fairall et al., 2003]. COARE version 2.6 is used for HOAPS-3, whereas version 3.0 (COARE3.0) is used for OAFlux and J-OFURO2.
 In section 6 we will make flux estimates using shipboard measurements, AIRS retrievals, and NCEP variables from the COARE3.0 bulk formula. These flux estimates from AIRS and NCEP variables and those from OAFlux, J-OFURO2, and HOAPS-3 will be compared with the flux estimates from shipboard measurements to evaluate the potential of the AIRS surface retrievals in estimating turbulent heat fluxes.
 Scatterometers have been providing accurate global wind fields for a broad range of applications since the 1970s. Wind speeds from ocean scatterometers are a main data source for estimating global turbulent heat fluxes, such as OAFlux, J-OFURO2, and HOAPS-3. In this study, we use wind speed from the Center for Ocean-Atmospheric Prediction Studies (COAPS) in conjunction with AIRS/AMSRE parameters to estimate the turbulent heat fluxes, which will be compared with existing flux products to examine the possibility for flux improvement in the Southern Ocean. The COAPS gridded wind fields are objectively mapped onto a 1° × 1° grid from QuickSCAT scatterometer measurements and are available at 6 h temporal resolution [Pegion et al., 2000].
2.2. Data Collocation
 To compare AIRS/AMSRE retrievals and NCEP data with shipboard measurements, we need to collocate all data to the same space and time grids. The collocation is done in three steps.
 1. First, the 10 m shipboard measurements are adjusted to the same vertical level used in the AIRS and NCEP data. The shipboard TA is adjusted to 2 m and pressure measured to sea level. The dry adiabatic lapse rate γ (γ = −dTA/dz = 9.8°C/km) is used to convert TA from 10 to 2 m. This is equivalent to adding a constant of only 0.076°C to all shipboard TA measurements, and thus the choice of the lapse rate does not influence our results. To convert the 10 m pressure to surface pressure, we added a constant pressure adjustment of 1.2 hPa to the shipboard pressure measurements. The shipboard TSG measures ocean temperature at 4 m depth, while satellites detect the very top of the ocean surface. Infrared and microwave radiometers measure temperature at slightly different depths at the ocean surface. An infrared radiometer measures skin temperature at a depth of order 0.01 mm depending on the wavelength of the measurement, whereas a microwave radiometer measures the subskin temperature at a depth of approximately 1 mm. In addition, the AMSRE microwave sea surface temperature (SST) product from Remote Sensing Systems (RSS) was calibrated to match the Reynolds SST [Wentz and Meissner, 2007], that is, the “bulk” SST defined at a depth of ∼1–5 m. Dong et al. [2006a] found that in the Southern Ocean, TO values from AMSRE show good agreement with in situ TO measurements. Hence here we will make no depth adjustments among the various TO values.
 2. Second, the high-resolution along-track shipboard measurements are averaged in 1° latitude bands to match the seven AIRS latitudinal grid points spanning Drake Passage (Figure 1). The 5 min calibrated underway shipboard data have an equivalent spatial resolution of ∼0.014° latitude, and the ship requires about 6 h to travel 1° latitude. To assess the impact of the ship travel time, we explored differences in the averages obtained for various time spans, ranging from 1 to 6 h. Neither the averages themselves nor the analyses of in situ versus satellite measurements differ significantly as averaging times are varied. Therefore, following Dong et al. [2006a], we required that all the available shipboard data that are averaged to each 1° latitude grid point have a ±2.5 h time separation from each other and a ±0.5° latitude separation. The center point for both the time and latitude separations is at the center of the 1° latitude band. At the seven latitudinal grid points in Drake Passage, a total of 594 data points are generated through this averaging process.
 3. The longitude and time of the data resulting from the latitudinal averaging process in step 2 differ from the AIRS regular 1° longitude grid and observational time. Therefore, the final step is to collocate the satellite data to the same longitude and time as the shipboard data. We extracted all the available AIRS day and night TA, TO, and qa and AMSRE TO data at the same date and latitude as the 594 shipboard observations. Those AIRS data within a ±1° longitude band of the shipboard data are then linearly interpolated to the longitude of the averaged shipboard measurements (Figure 1). Dong et al. [2006a] suggested that the TO diurnal cycle affects the temperature difference between satellite and in situ measurements due to the time separation. However, the TO difference is not significant for time separations of less than 5 h in the Drake Passage, as the diurnal effect is small due to the relatively strong winds in the Southern Ocean [Dong et al., 2006a]. Therefore, for consistency we use only satellite data collected within ±2.5 h of the available shipboard measurements. As in the work of Dong et al. [2006a], we found no significant differences in the comparison among AMSRE, AIRS, and shipboard temperature measurements for time separations of less than 5 h, and for qa measurements there were no significant differences for time separations less than 1 day (not shown).
 Unfortunately, as noted above, because of clouds and gaps between satellite tracks, collocated surface TA and TO are not always available in the AIRS daily product. As a result, a total of 315 TA (137 daytime and 178 nighttime) out of the possible 594 collocated data points are available. The availability of AIRS TO is even lower than that of TA, with a total of 181 TO pairs (89 daytime and 92 nighttime) available. Thus, of the 88 LMG crossings, 26 crossings had five or more collocated AIRS TA data available at the seven latitudinal grid points in Drake Passage (Figure 1), and only 7 crossings had five or more collocated AIRS TO data.
 AIRS qa is given at 12 standard pressure layers, with bottom boundaries ranging from 1000 to 100 hPa. We tested two methods for determining surface qa from its vertical profile. In method 1, we assumed the surface boundary layer to be well mixed and used qa from the lowest standard pressure layer. In method 2, we extrapolated qa to the surface based on the assumption that qa varies exponentially with pressure, using the estimates for layers centered at 962.5 hPa (925–1000 hPa) and 887.5 hPa (850–925 hPa). The two methods yield results with similar statistical relationships to the shipboard qa and similar potential causes for the biases. However, method 1 showed a more negative mean bias of −1.09 g/kg compared to the mean bias of −0.69 g/kg from method 2. Thus, in this study we adopted method 2. Like surface temperatures, qa is not always available in the AIRS daily product because of clouds and gaps between satellite tracks. Out of a total of 594 data points for which we sought qa from AIRS, 315 (137 daytime and 178 nighttime) collocated values are extracted from AIRS within ±2.5 h of the shipboard measurements. Of the 88 LMG crossings, 30% had five or more grid points of collocated AIRS qa data available for comparison to the shipboard data.
 To collocate the 0.25° AMSRE TO with shipboard measurements, we averaged the same-day AMSRE TO within ±0.5° latitude and ±0.5° longitude of the averaged shipboard grid point. As with the AIRS data, we used only AMSRE data collected within ±2.5 h of the shipboard measurements. This resulted in a total of 461 (203 daytime and 258 nighttime) collocated TO from the cloud-penetrating microwave AMSRE measurements for comparison with the 594 available shipboard measurements. Over 70% of the Drake Passage crossings had five or more collocated grid point AMSRE TO data available.
 The collocation of the coarser-resolution 1.88° Gaussian gridded NCEP data and the averaged shipboard measurements was relatively straightforward: we simply interpolated the 6-hourly NCEP data to the 594 averaged shipboard data locations and times.
 The collocated data pairs are not evenly distributed in all seasons. However, for all collocated parameters, their seasonal distributions are similar, with about 25% of the data in austral summer (January–March), 35% in autumn (April–June), 10% in winter (July–September), and 30% in spring (October–December). This seasonal distribution is due to the seasonal bias in shipboard observations. The shipboard measurements in the Drake Passage for conditions when AIRS are available show no significant difference from those when AIRS are not available during our study period (2002–2007), suggesting that AIRS gives a representative sampling of the near-surface parameters in our study region.
Figure 2 shows an example of the collocated shipboard and satellite measurements for an October 2005 LMG Drake Passage crossing when satellite retrievals were mostly available. Spatial structures shown in the 5 min shipboard measurements (black lines) of all parameters and their 1° latitude averages (red lines) are very similar. Also shown in Figure 2 are the collocated AIRS/AMSRE (green/cyan lines) and NCEP (blue lines) data. The NCEP data are consistently too smooth to capture the observed variability. AIRS TA (Figure 2a) and TO (Figure 2b) are consistently colder than the corresponding shipboard measurements. However, compared with the NCEP data, the AIRS retrievals generally better represent the observed meridional variability. The exception is the AIRS qa, which for this particular cruise does not demonstrate good agreement with the shipboard qa (Figure 2d).
3. Evaluation of AIRS Surface Temperature
3.1. Evaluation of Surface TA and TO
 In order to evaluate the agreement between satellite and shipboard temperature measurements, we use statistical analyses of correlation and robust linear regression. Unless specified, we regress satellite parameters to the corresponding shipboard measurements, Xsat = a + bXship, where Xsat and Xship represent satellite and shipboard measurements, respectively, and a and b are regression coefficients. If the regression slope b equals one, then satellite and shipboard measurements simply differ by the constant a. The statistical significance of the regression is tested using the value of the null correlation at 95% significance level.
Table 1 summarizes the mean difference, root-mean-square (RMS) error, and standard error of TA, TO, (TA − TO), and qa between the satellite and shipboard measurements for the daytime, nighttime, and combined data sets. The AIRS TA retrievals have an overall average cold bias of −2.03°C (Table 1). The difference between daytime and nighttime is marginal, with a slightly larger cold bias of −2.23°C during the daytime compared with the cold bias of −1.88°C during the nighttime (Table 1). Despite the large cold bias, a linear regression of AIRS TA to shipboard TA measurements (Figure 3a) shows that the air temperature scatters along a line nearly parallel to the zero-bias line (also known as the 1:1 line), with a regression slope b = 0.94. This suggests that the biases in AIRS TA are largely consistent with a constant offset of a = 1.8°C.
Table 1. Temporal Mean Difference ± Standard Error, Root-Mean-Square Error, and Number of Collocated Data From Satellite and National Centers for Environmental Prediction With R/V Laurence M. Gould Shipboard Measurements of Near-Surface Air Temperature, Sea Surface Temperature, and Specific Humiditya
Number of Data
Number of Data
Number of Data
Statistics are given separately for daytime (ascending satellite passes) and nighttime (descending) and for the total data set. Here “AIRS only” refers to the case in which both TA and TO were derived from AIRS. Abbreviations are as follows: AIRS, Atmospheric Infrared Sounder; AMSRE, Advanced Microwave Scanning Radiometer; NCEP, National Centers for Environmental Prediction.
−2.23 ± 0.16(1.83)
−1.88 ± 0.14 (1.85)
−2.03 ± 0.10 (1.85)
−0.30 ± 0.05 (1.18)
−0.16 ± 0.13 (1.22)
−0.27 ± 0.15 (1.40)
−0.22 ± 0.10 (1.31)
0.01 ± 0.05 (0.66)
−0.17 ± 0.04 (0.63)
−0.09 ± 0.03 (0.64)
−0.05 ± 0.04 (0.84)
TA − TO (°C)
−1.63 ± 0.20 (1.75)
−1.08 ± 0.17 (1.60)
−1.25 ± 0.13 (1.69)
TO from AMSRE
−2.13 ± 0.18 (1.81)
−1.64 ± 0.15 (1.86)
−1.76 ± 0.11 (1.86)
−0.73 ± 0.05 (0.71)
−0.66 ± 0.05 (0.68)
−0.69 ± 0.04 (0.70)
0.12 ± 0.02 (0.46)
Figure 3b shows that the shipboard TO and the AMSRE TO are highly correlated, with a correlation of 0.97, and they scatter closely along the zero-bias line. The nighttime TO from AMSRE is colder than the shipboard value by −0.17°C, whereas during the daytime, AMSRE is warmer than the shipboard value by 0.01°C (Table 1). These results imply that, although the AMSRE SST was set to match the “bulk” SST and the strong wind reduces the diurnal effect in the Southern Ocean, the thermal skin effect of the surface ocean is still a concern. Thus a cool skin correction is applied to the LMG in situ measurements in computing the turbulent heat fluxes (section 6). Like AIRS TA, AIRS TO also has a cold bias compared with shipboard measurements, although there is no significant difference in the cold bias during daytime and nighttime (Table 1). The scatterplot of AIRS TO versus shipboard TO values (Figure 3c) shows relatively large deviations from the zero-bias line, suggesting that, in addition to being limited to cloud-free conditions, the AIRS TO retrieval may also be affected by factors such as the stability of the atmospheric boundary layer.
 The bulk formula for sensible heat flux (equation (1)) depends on the air-sea temperature difference, (TA − TO). Regardless of the mean cold biases in AIRS TA and TO, Figures 3a and 3c show that both TA and TO scatter along lines parallel to the zero-bias line. However, in scatterplots comparing shipboard (TA − TO) with satellite observations, we see substantial deviation from the zero-bias line, regardless of whether we use the AIRS measurements alone (denoted (TA − TO)airs) or AIRS TA and AMSRE TO (denoted (TA − TO)AMSRE). Results are generally better using AMSRE TO, which has a correlation coefficient of 0.57 versus 0.42 for AIRS data alone. On average, the satellite-derived (TA − TO) is more negative than shipboard measurements by −1.76°C for (TA − TO)AMSRE and by −1.25°C for (TA − TO)AIRS, with most of this bias contributed by the cold bias in AIRS TA (Table 1). With a 10 m/s wind speed, the cold bias of −1.76°C for (TA − TO)AMSRE would give an anomalous sensible heat loss to the atmosphere of 26 W/m2.
 Scatterplots of NCEP parameters against shipboard measurements show that both TA (Figure 4a) and TO (Figure 4b) from NCEP compare well with in situ measurements, although NCEP has a cold bias of −0.30°C for TA and a cold bias of −0.05°C for TO (Table 1). The regression slopes (Figure 4) for both TA and TO are slightly lower than those from the AIRS regression analyses (Figure 3), whereas the correlations are slightly higher. Although NCEP TA and TO agree well with shipboard measurements, their difference (TA − TO)NCEP does not agree as well (Figure 4c), much as found for (TA − TO)AIRS. As shown in Figure 4c, the scatter of (TA − TO)NCEP versus (TA − TO)ship deviates from the zero-bias line. This deviation can be attributed to the lower variability in NCEP air-sea temperature difference, and suggests that a sensible heat flux derived from the NCEP temperatures may not capture the flux variability as completely as a combination of AIRS and AMSRE.
3.2. Causes for the Biases in Temperature and Temperature Differences
 For the atmospheric boundary layer, if the ocean surface is colder than the overlying air, the atmosphere is stable. On the other hand, if the ocean surface is warmer than the overlying air, the near-surface air will warm and become less dense than the air above it, inducing convection. Thus, the air-sea temperature difference reflects the stability of the overlying atmosphere, which may also affect the AIRS retrievals [Fetzer et al. 2004]. Here the differences between the shipboard and the satellite data of near-surface air temperature (ΔTA,AIRS = TA,AIRS − TA,ship or ΔTA,NCEP = TA,NCEP − TA,ship), sea surface temperature (ΔTO,AIRS = TO,AIRS − TO,ship, ΔTO,AMSRE = TO,AMSRE − TO,ship, or ΔTO,NCEP = TO,NCEP − TO,ship), and air-sea temperature difference (Δ(TA − TO) = ΔTA − ΔTO) are statistically examined against (TA − TO)ship. We also explore the dependence of these differences on TA, TO, qa, cloud cover, and geographic location. Other factors, such as the vertical structure of atmospheric temperature and moisture, can also potentially influence the accuracy of the surface satellite retrievals. However, no accurate data source can be used to quantify their effects on the biases of near-surface satellite retrievals. Because of the relatively large mean bias in AIRS near-surface temperatures (Table 1), for both AIRS and AMSRE, the mean biases in TA and TO are removed before performing the statistical analysis.
 The statistical regression analyses (Figure 5a) suggest that ΔTA,AIRS has a significant dependence on (TA − TO)ship. Both the negative correlation between ΔTA,AIRS and (TA − TO)ship and our regression analysis suggest that ΔTA,AIRS decreases with increasing (TA − TO)ship. In contrast, ΔTO,AIRS increases with increasing (TA − TO)ship, with twice the sensitivity of ΔTA,AIRS (Figure 5b). This is confirmed by the relatively large positive correlation of 0.31 between ΔTO,AIRS and (TA − TO)ship. In contrast, no significant correlation is found between ΔTO,AMSRE and (TA − TO)ship (not shown). Both Δ(TA − TO)AIRS and Δ(TA − TO)AMSRE show significant correlations with (TA − TO)ship (Figure 5c), which is to be expected considering the dependence of ΔTA,AIRS and ΔTO,AIRS on (TA − TO)ship (Figures 5a and 5b). Our regression analysis suggests that the dependence of Δ(TA − TO)AIRS on (TA − TO)ship is more than twice as strong as the dependence of Δ(TA − TO)AMSRE on (TA − TO)ship, which is also suggested by the strong dependence of ΔTO,AIRS on (TA − TO)ship. Further analysis suggests that the temperature biases depend significantly on TA,ship but not on TO,ship. The dependence of temperature biases on TA,ship is similar to their dependence on (TA − TO)ship, although it is relatively weak.
 Infrared measurements are strongly influenced by water vapor and cloud cover. Susskind et al.  found that AIRS retrieval accuracy degrades slightly with increasing cloud cover. Our statistical analyses suggest that ΔTA, ΔTO, and Δ(TA − TO) are not affected by cloud cover, but all three are affected similarly by AIRS-derived water vapor, column total water vapor from AMSRE, and shipboard RH. Thus, here we describe only the dependence of temperature differences on shipboard RH measurements. Figure 5d shows that ΔTA,AIRS is positively correlated with RH (dashed line). For satellite-derived ocean temperatures the relationship is less pronounced: ΔTO,AIRS does not depend on RH (Figure 5e), and ΔTO,AIRS shows a weak but significant negative correlation with RH (not shown), consistent with the negative correlation between ΔTO,AMSRE and column water vapor reported by Dong et al. [2006a]. Because satellite retrievals of TA and TO depend on RH, both Δ(TA − TO)AIRS and Δ(TA − TO)AMSRE are positively correlated with RH (Figure 5f) at a statistically significant level.
 Our analysis also suggests that ΔTO,AIRS depends on qa (Figure 5h). As qa increases, ΔTO,AIRS tends to increase. ΔTO,AMSRE also experiences significant dependence on qa (not shown), but in an opposite sense to ΔTO,AIRS: an increase in qa would cause a decrease in ΔTO,AMSRE. Although TA is closely related to qa, no significant dependence of ΔTA,AIRS on qa was found (Figure 5g). Δ(TA − TO)AIRS decreases with increasing qa because of the dependence of ΔTO,AIRS on qa (Figure 5k). However, Δ(TA − TO)AMSRE does not experience a statistically significant relationship with qa.
 As in the rest of the Southern Ocean, the Antarctic Circumpolar Current that passes through Drake Passage is characterized by strong meridional gradients in temperature that result in sharp fronts. We therefore considered the possibility that ΔTA, ΔTO, and Δ(TA − TO) have a cross-passage bias in response to the frontal positions. No simple linear dependence of the temperature biases on latitude and longitude was found. We will further examine the cross-passage dependence of ΔTA, ΔTO, and Δ(T − TO) in section 5.
4. Evaluation of AIRS Surface Specific Humidity
4.1. Comparison of Near-Surface Specific Humidity
 On average, AIRS surface qa has a negative bias of −0.69 g/kg compared with shipboard measurements (Table 1). The qa difference between AIRS and shipboard measurements shows a marginal difference between daytime (−0.73 g/kg) and nighttime (−0.66 g/kg) (Table 1). The difference between the AIRS qa and shipboard qa for each collocated pair varies between −4.2 and 2.3 g/kg. Regressing AIRS qa to shipboard qa measurements gives a slope of 0.89 (Figure 6a), suggesting that the AIRS qa captures the observed qa variability.
 To examine the relative performance of AIRS qa measurements, we also compared the NCEP qa with shipboard qa measurements. The mean difference between NCEP and the shipboard qa measurements is small, with NCEP qa higher than shipboard measurements by 0.12 g/kg on average (Table 1). The difference between NCEP qa and shipboard qa for each collocated pair varies between −1.1 and +1.5 g/kg (Figure 6b). Like the AIRS measurements (Figure 6a), NCEP qa measurements are also significantly correlated with shipboard qa measurements (ρ = 0.88). The scatterplot of the NCEP qa versus shipboard qa (Figure 6b) shows that the scatter of the data is relatively small compared to that of the AIRS qa versus shipboard qa (Figure 6a). This is consistent with the smaller RMS error of NCEP qa compared with the RMS error of AIRS qa (Table 1). Thus, whether the AIRS qa can improve the latent heat flux estimate still remains in question.
4.2. Causes for the Biases in AIRS Specific Humidity
 We examined the dependence of the difference in near-surface specific humidity (Δqa = qa,AIRS − qa,ship) between AIRS and shipboard measurements on TA, TO, (TA − TO), qa, cloud cover, RH, and geographic location. Similar to the near-surface temperature, the mean bias in AIRS qa is removed before performing the statistical analyses. Results from the statistical analyses suggest that Δqa decreases with increasing (TA − TO)ship, and the dependence of Δqa on (TA − TO)ship is stronger for cases when the air-sea temperature difference is positive (Figure 7a).
 Statistical analyses of Δqa against AIRS cloud fraction measurements suggest that Δqa has a significant dependence on cloud (Figure 7b), with a significant negative correlation of −0.21, exceeding the 95% significance level of 0.09. The dependence of Δqa on cloud is relatively strong for cases with a cloud fraction exceeding 0.6, whereas for cases with cloud fraction lower than 0.6 the dependence is less apparent. Linear regression of Δqa to the cloud fraction suggests that an increase in cloud fraction causes a more negative bias in the AIRS qa.
 We also found a significant dependence of Δqa on qa itself (Figure 7c). As qa increases, Δqa decreases. Because of the overall negative bias in AIRS qa, this suggests that AIRS qa has more negative bias for cases with relatively higher qa. In terms of percentage, we found that the time mean error of qa is about 21% of qa, consistent with the results of Tobin et al. .
5. Seasonal and Cross-Passage Variations in AIRS Biases
5.1. Surface Temperatures
 The biases in the daytime and nighttime temperature retrievals from AIRS and AMSRE only differ marginally from each other, as suggested by their comparison with shipboard measurements (Table 1). However, ΔTA, ΔTO, and Δ(TA − TO) show strong seasonal differences (Figure 8a–8c). The cold bias in AIRS TA is largest in austral spring and smallest in autumn. AIRS TO has a cold bias except in austral summer, when ΔTO,AIRS is positive but very small. In contrast, ΔTO,AMSRE is relatively small and is positive during summer, winter, and spring but negative in autumn. The seasonal biases in temperature cannot be explained simply by their dependence on TA − TO, RH, and qa. Shipboard RH shows little seasonal variations; thus it could not contribute to the seasonal changes in the temperature biases. Specific humidity derived from shipboard measurements is relatively high during summer but not significantly different in the other three seasons, suggesting that the contribution of qa to seasonal biases in temperature is also limited. Although the large, negative TA − TO during autumn can explain the less negative bias in ΔTA,AIRS, the stronger, positive TA − TO during summer implies a more negative bias in ΔTA,AIRS compared to other seasons, the opposite of that shown in Figure 8a.
Figures 9d–9f show the averages of ΔTA, ΔTO, and Δ(TA − TO) at each latitude grid regardless of the longitudinal location. Both TA (Figure 8d) and TO (Figure 8e, solid) from AIRS are cooler than the shipboard measurements at all latitudes from 55.5°S to 61.5°S. The maximum differences occur in the region between 57°S and 59°S, coinciding with the general location of the Polar Front in Drake Passage [Sprintall, 2003]. As shown in section 3, both temperature and humidity affect AIRS temperature retrievals; hence the large difference between AIRS and shipboard temperature measurements in the frontal region may be related to the strong horizontal gradient in both the temperature and humidity fields. Because of the similarity of the latitudinal distribution of ΔTA,AIRS and ΔTO,AIRS, Δ(TA − TO)AIRS does not have significant cross-passage variation (Figure 8f, solid line), although the overall trend is for smaller values to the north. In general, ΔTO,AMSRE (Figure 8e, dashed line) is much smaller than ΔTO,AIRS, with positive values south of 59°S and negative values to the north. The cross-passage distribution of Δ(TA − TO)AMSRE (Figure 8f, dashed line) is similar to that of ΔTA due to the dominance of the cold bias in AIRS TA.
5.2. Near-Surface Specific Humidity
 Like near-surface temperature differences, Δqa also shows a seasonal variation (Figure 9a). Both daytime and nighttime Δqa are negative for all seasons, with more negative biases during austral winter and spring. This seasonal bias cannot be explained by the dependence of Δqa on the air-sea temperature difference and cloud fraction: the more positive air-sea temperature difference and the relatively high cloud fraction (not shown) during summer imply a more negative bias compared to winter and spring, the opposite of that shown in Figure 9a. The less negative bias during summer might be related to the fact that there are more summer shipboard samples available south of 59°S, where the negative bias is relatively small (Figure 9b). The daytime Δqa is more negative compared to the nighttime Δqa, except during winter. The larger daytime bias is potentially related to the more positive air-sea temperature difference and slightly higher cloud fraction during the daytime (not shown). The average Δqa at each latitude grid point (Figure 9b) suggests that the bias in the AIRS qa becomes more negative from south to the north across the Drake Passage during both daytime and nighttime, which may be related to the increase in the cloud fraction to the north. The daytime Δqa also shows a large gradient between 57°S and 59°S that coincides with the location of the oceanic frontal region where the temperature gradient is large (Figure 2). The reason for this cross-passage variability in the daytime qa bias is unclear and calls for further examination in future studies.
6. Uncertainties in Turbulent Heat Fluxes From AIRS/Advanced Microwave Scanning Radiometer Surface Retrievals
 To examine whether AIRS near-surface TA, TO, and qa retrievals are accurate enough to use in turbulent heat flux estimations, we compute the sensible (equation (1)) and latent (equation (2)) heat fluxes from the COARE3.0 algorithm using the collocated TA, TO, and qa from AIRS, AMSRE, NCEP, and shipboard measurements. Here we focus on the uncertainties in turbulent heat fluxes stemming from using satellite TA, TO, and qa retrievals, and therefore all other variables required in equations (1) and (2) are the same for each estimate. The collocation process for these other NCEP variables (e.g., qs, W, and ρa) is the same as that outlined in section 2. We use either shipboard or NCEP winds in the computation, as discussed below. The COARE algorithm was initially developed for the tropics, without high-latitude data or high wind conditions [Fairall et al., 1996]. Although high-latitude data were used in developing and validating the updated versions 2.6 and 3.0, no data were available from the Southern Ocean [Fairall et al., 2003]. Thus, the algorithm's performance in the Southern Ocean still needs to be assessed. However, since the main purpose of this study is to evaluate the AIRS surface retrievals, for simplicity we use the COARE3.0 algorithm directly, as it is the most widely used algorithm currently available.
 Four different sensible and latent heat fluxes are estimated, including those based on TA, TO, and qa from shipboard measurements (SHship, LHship), from NCEP (SHNCEP, LHNCEP), and from AIRS (SHAIRS, LHAIRS). The fourth is based on AMSRE TO and AIRS TA and qa (SHAMSRE, LHAMSRE). Here SHship and LHship are considered as “truth” and are used to evaluate the uncertainties in the other three estimations. The fluxes are defined as positive out of the ocean. The mean biases in AIRS/AMSRE and NCEP TA, TO, and qa (Table 1) are removed prior to the flux estimation. Figure 10 shows an example of the turbulent heat fluxes for the October 2005 LMG Drake Passage crossing shown in Figure 2. As we found for the individual parameters (Figure 2), both SHNCEP and LHNCEP are consistently too smooth to capture the variability in SHship and LHship. In contrast, both SHAMSRE and LHAMSRE show good representation of the variability in SHship and LHship across the Drake Passage, except at grid points at the southern and northern boundaries, where land/ice effects may bias satellite retrievals. Although mean biases of all the collocated pairs have been removed, there is a residual bias of −1.10°C in AIRS TA and 0.35°C in AMSRE TO for this particular crossing, which explains the positive biases in both SHAMSRE and LHAMSRE. In this case, the available SHAIRS agrees better with SHship than SHNCEP in showing the sharp cross-passage gradients of the fluxes and the negative values south of 58°S. The agreement of LHAIRS with LHship is also better than that between LHNCEP and LHship (Figure 10b). The fluxes increase by a factor of 2–3, and the distinctions are more dramatic when we use more realistic shipboard winds (solid lines) rather than NCEP winds (dashed lines), underscoring the sensitivity of turbulent heat fluxes to wind speed.
 The overall comparison of the turbulent flux estimations from AIRS, AMSRE, and NCEP against those from all the shipboard measurements is given in Figure 11. All comparisons use the shipboard winds. SHAIRS experiences the smallest RMS error compared with SHship (Figure 11a). However, SHAMSRE demonstrates the best performance in terms of capturing the full range of values of the SHship, suggested by the large slope (0.78) of the regression line (Figure 11b). SHAIRS (Figure 11a) performs slightly better than SHNCEP (Figure 11c) in capturing the full range of values of the SHship. The comparison of LH shows results similar to those from the SH comparison: LHAIRS has smaller RMS errors than LHNCEP (Figure 11d,f), and LHAMSRE demonstrates the best performance in terms of capturing the full range of values of the LHship (Figure 11e). However, the RMS error of all three latent heat flux estimates (LHAIRS, LHAMSRE, and LHNCEP) differs only marginally. Considering both the RMS errors and the ability to capture the full range of LHship and SHship values, we conclude that the turbulent heat flux estimates from the AIRS/AMSRE combination perform better than those from AIRS alone and NCEP.
 Finally, to examine the potential of AIRS surface retrievals in improving flux estimates, we also compare existing satellite-derived flux products, OAFlux, J-OFURO2, and HOAPS-3, with the sensible (Figure 10c) and latent (Figure 10d) heat fluxes estimated from shipboard measurements. Because of the strong sensitivity of the turbulent heat fluxes to the choice of wind speed (e.g., Figures 10a and 10b) and the fact that scatterometers are one of the main wind data sources in the existing flux products, we used the QuickSCAT wind speed from COAPS in calculating LH and SH from the shipboard, AIRS, AMSRE, and NCEP variables to give a fair comparison. For the case shown in Figure 10, the sensible heat fluxes from both OAFlux and J-OFURO2 compare well with SHship, although SHAMSRE performs slightly better in terms of the cross-passage variations in SHship. HOAPS-3 fluxes perform similarly to SHNCEP and LHNCEP, which are consistently too smooth to capture the variability in SHship and LHship. The latent heat flux from J-OFURO2 performs similarly to LHAIRS and LHAMSRE and best captures the cross-passage variations in LHship; however, it deviates from LHship near the boundaries and experiences the largest RMS error. LH from OAFlux captures about half of the cross-passage variations in LHship for this cruise.
Table 2 summarizes the RMS errors and regression slopes of the turbulent flux estimates from AIRS, AMSRE, and NCEP parameters and those from the existing flux products in comparison with the flux estimates from the shipboard parameters. The sensible heat fluxes from the OAFlux and J-OFURO2 have the smallest the RMS errors. However, SHAMSRE performs best at capturing the full range of values of SHship, with a regression slope closest to 1 (Table 2). In the latent heat flux comparison, LHAMSRE and LHJ-OFURO2 perform better in capturing the range of the LHship variations. However, LH J-OFURO2 has the largest RMS errors. LHNCEP has the smallest RMS error, but LHNCEP does the poorest in capturing the full range of variations. Given the different performance in terms of RMS error and different skill in capturing the full range of shipboard estimates (Table 2), it is difficult to determine which flux products best represent the “truth.” The AIRS/AMSRE combination best represents the full range of shipboard values for both LH and SH, whereas the OAFlux and J-OFURO2 products have the smallest RMS error for SH, and OAFlux and NCEP have the smallest RMS error for LH. Our results therefore suggest that for studies focusing on the mean climate state in the Southern Ocean, OAFlux and NCEP may provide better turbulent heat flux values. However, for those studies focused on climate variability, AIRS and AMSRE measurements have the potential to provide the best turbulent flux estimations.
Table 2. Root-Mean-Square Error and Regression Slope of the Turbulent Heat Flux From Various Satellite and Existing Flux Products Against Those Fluxes Calculated From Shipboard Variables of Near-Surface Air Temperature, Ocean Temperature, and Specific Humiditya
Sensible Heat Flux (W/m2)
Latent Heat Flux (W/m2)
QuikSCAT wind speeds from Center for Ocean-Atmospheric Prediction Studies were used in the turbulent heat flux estimates that used the shipboard, AIRS, AMSRE, and NCEP data. Abbreviations are as follows: HOAPS-3, Hamburg ocean atmosphere parameters and fluxes from satellite data; J-OFURO2, Japanese Ocean Flux data sets with Use of Remote sensing Observations version 2; OAFlux, Objectively Analyzed air-sea Fluxes; RMS, root-mean-square.
0.53 ± 0.06
0.79 ± 0.07
0.73 ± 0.06
0.84 ± 0.05
0.47 ± 0.02
0.59 ± 0.02
0.60 ± 0.03
0.78 ± 0.04
0.58 ± 0.03
0.88 ± 0.07
0.22 ± 0.03
0.72 ± 0.07
7. Discussion and Conclusions
 Satellite measurements have played a major role in estimating air-sea heat fluxes because of their spatial and temporal coverage. Remotely sensed sea surface temperature and wind speed have become standard in the estimation of sensible and latent heat flux components. More recently, near-surface air temperature and specific humidity have been made available from AIRS, an infrared sounder on board the Aqua satellite. In this study, we evaluate whether the near-surface air temperature, sea surface temperature, their difference, and specific humidity from AIRS are suitable for estimating latent and sensible heat fluxes from bulk formula. Our evaluation is undertaken in the Southern Ocean, where existing air-sea heat flux products have large uncertainties [Dong et al., 2007]. The evaluation is done by comparing the near-surface AIRS observations with in situ shipboard meteorological observations in the Drake Passage. Since sensible heat flux estimates depend on air-sea temperature differences, we also consider TO from AMSRE with the shipboard observations.
 Overall, our comparison indicates that, in cases with minimal cloud/water vapor contamination when the AIRS near-surface parameters are mostly available for the entire region, AIRS and AMSRE together can provide a better representation of the spatial variation evident in the shipboard measurements in the Drake Passage compared to the much smoother NCEP data. Although the AIRS surface temperature retrievals have relatively large cold biases, our comparison to shipboard measurements indicates that AIRS temperatures follow the variations in shipboard measurements with regression slopes close to 1.0. In particular, the satellite-derived air-sea temperature differences (with AIRS TA minus AMSRE TO) versus those from shipboard measurements follow the zero-bias line more closely than do NCEP air-sea temperature differences. This suggests that satellite measurements have the potential to provide better parameters for the sensible heat flux bulk formula, largely because they capture the full range of values in Drake Passage. However, it is still in question whether the AIRS specific humidity values can improve the latent heat flux estimates in the Southern Ocean, in view of the relatively large deviations of AIRS qa from the shipboard measurements. Nevertheless, turbulent heat flux estimates using the COARE3.0 algorithm indicate that the AIRS/AMSRE combination does a better job than the existing reanalysis or satellite-derived heat fluxes at capturing the full range of turbulent heat flux values from the Drake Passage shipboard observations.
 The present study is focused on the AIRS L3 gridded products. AIRS level 2 (L2, swath data) products have a higher spatial resolution in both vertical and horizontal directions, which may better represent some of the small-scale variability. It is also straightforward to combine the simultaneous measurements of AIRS L2 TA and qa with AMSRE TO and winds for turbulent flux estimates, which may give the best possible agreement with the full range of shipboard observations in Drake Passage (Table 2). Jackson et al.  suggested that combining two or three microwave sensors could improve near-surface air temperature and specific humidity retrievals. Our analysis for the Drake Passage gives the RMS error of 1.83°C in TA and 0.70 g/kg for qa. These RMS errors are lower than the RMS error of 1.96°C for TA and 0.94 g/kg for qa reported by Jackson et al.  when merging two or three microwave sensors. However, their data were all from the tropical and subtropical regions north of 10°S, with a large range of TA from 5°C to 29°C and qa from 3 to 19 g/kg. In contrast, our Southern Ocean study region of colder TA and lower qa covered a smaller range from −3°C to 11°C for TA and from 0.8 to 7 g/kg for qa. Supporting the findings of Jackson et al. , our ongoing analysis also suggests that AIRS performs relatively better in the tropics compared to the Southern Ocean (not shown), suggesting that a multisensor method merging AIRS and microwave SST may yield even better results using near-surface temperature and humidity data to estimate turbulent heat flux.
 We thank the two anonymous reviewers for their insightful comments. The AIRS version 5 level 3 daily products are distributed by the Goddard Earth Sciences Distributed Active Archive Center at http://disc.sci.gsfc.nasa.gov/AIRS/. NCEP/NCAR Reanalysis data are provided by the NOAA/OAR/ESRL PSD from their Web site at http://www.cdc.noaa.gov/. AMSRE data are produced by Remote Sensing Systems and sponsored by the NASA Earth Science REASoN DISCOVER Project and the AMSRE Science Team and are available from http://www.ssmi.com/. The HOAPS data are provided by the World Data Center for Climate, Hamburg, Germany. The OAFlux is from http://oaflux.whoi.edu. J-OFURO2 data are obtained from the J-OFURO2 Web site http://dtsv.scc.u-tokai.ac.jp/J-OFURO2/. This research is supported by NOAA/AOML and by grants from the National Science Foundation (Office of Polar Programs Award 0337998 and OCE Award 0850350).