The SSTs retrieved by the TMI and the AMSR-E are used widely in oceanic and climatic research. The TMI microwave sensor, launched on November 26, 1997, flies on a non-sun-synchronous orbit and it has a resolution of 6–50 km. The AMSR-E is a passive microwave sensor with a resolution of 5.4–56.0 km that was launched with the Aqua satellite in May 2002 (but failed in October 2011). The Aqua satellite flies on a sun-synchronous orbit that crosses the equator at 13:30 local time (LT) on northward (ascending) passes and at 01:30 on southward (descending) passes. The SST products from the TMI and the AMSR-E have been validated by in situ measurements from buoys, high-resolution XBTs, hull-mounted thermosalinographs, and Argo NSTs (Gentemann et al., 2004; Dong et al., 2006; Bhaskar et al., 2009).

The SSTs retrieved by the TMI and the AMSR-E were provided by remote sensing systems, Santa Rose, CA, USA (Wentz, 1998; Wentz and Meissner, 1999). The TMI SSTs are released on a 0.25°×0.25° grid over the global tropics (±39.875°), whereas the AMSR-E SSTs are available globally (±89.875°) with the same spatial resolution. The daily SST product (Version 4) from the TMI and the daily passes of the SST product (Version 7) from the AMSR-E were used in this study. These data sets also included wind speed (measured using the 10.65 and 37.0 GHz channels from the TMI and the 10.7 and 18.7 GHz channels from the AMSR-E), atmospheric water vapor, liquid cloud water, and rainfall rate. The TMI instrument was calibrated carefully using SSM/I collocations based on the methodology described by Wentz et al. (2001). Brightness temperatures were corrected for atmospheric effects using a radiative transfer model (Wentz and Meissner, 1999), whereby the algorithm constants were tuned slightly by intercomparison with in situ and climatological observations. The AMSR-E has an onboard calibration system that compensates continuously for variations in the gain and noise temperature of the radiometer. The two vital components of this calibration system are the cold mirror and the hot reference load. Wentz et al. (2003) used coincident measurements from collocated observations from other satellites to develop a system for predicting the temperature of the hot reference. One shortcoming of their procedure is that the in situ data available for the calibration of the satellite data were largely concentrated in the tropics.

The Argo NST data used in this study were non-pumped temperatures collected by the APEX, PROVOR, and ARVOR profiling floats over the global oceans from October 2008 to March 2012. These floats were deployed and are processed by the British Oceanographic Data Centre, Indian National Centre for Ocean Information Services, Japan Meteorological Agency, Scripps Institution of Oceanography, and University of Washington. Their compiled data are made available on the National Oceanography Centre’s (UK) ftp site (ftp://ftp.pol.ac.uk/pub/bodc/ argo/NST/). Overall, 409 floats were used in our study, and the uppermost temperature (above 1 m) of each profile was selected for comparison with the satellite-retrieved SSTs.

Each Argo NST was matched to the TMI/AMSR-E measurements collected on the same day, the prior day, and 1 d later. A spatial window of 0.8°×0.5° was chosen to ensure more than one satellite SST retrieval matched a corresponding Argo profile location for each satellite pass. Those selected TMI/AMSR-E observations measured by ascending and descending passes for each day were then interpolated linearly to the Argo measurement point. To minimize diurnal effects, a temporal window of ±5 h was selected for pairing the Argo and TMI/AMSR-E observations. If more than one of the interpolated TMI/AMSR-E retrievals fell into a specific time window, an averaged TMI/AMSR-E observation was computed; otherwise, the nearest SST was selected. After careful assessment of the matchups, abnormal measurements (probably due to transmission errors) from five NST profiles were discarded. Therefore, 3 191 and 3 379 data pairs for the TMI and the AMSR-E, respectively, were matched for this analysis. Figure 1 shows the geographic distribution of the TMI/Argo and AMSR-E/Argo data pairs employed for our analysis. Most TMI/Argo pairs are located (1) near the Kuroshio, Oyashio, and Kuroshio Extension in the northern Pacific, (2) between 15°–40°S in the southern Pacific, (3) in the southern Indian Ocean, especially near Madagascar, and (4) in the southern Atlantic. The AMSR-E/Argo pairs have similar distribution but with many additional data pairs in the Southern Ocean (south of 40°S) and high latitudes (north of 40°N) of the northern Pacific and northern Atlantic.

To check the diurnal variability in the upper ocean, the near-surface and nominal surface (5 m was selected here) temperatures were selected and compared for each Argo temperature profile. Figure 2 shows the temperature difference between the Argo near-surface and 5 m temperatures, bin-averaged based on the LT. A pronounced diurnal warming can be observed between 10:00 and 19:00 LT, with maximum warming of about 0.1°C occurring at 14:00–15:00 LT. Accompanying the diurnal warming, the variability of the temperature difference increases by 0.33°C with a mean standard deviation of about 0.23°C. A diurnal cooling before 08:00 LT can be seen, with average cooling of about 0.07°C and standard deviation of 0.10°C. On the basis of our analysis, the maximum warming and cooling can be as much as approximately 4.5°C.

A seasonal dependence of the temperature difference exists in both the northern and southern Hemispheres (Fig. 3). The warm bias is most pronounced in the summer months, which is a result of increasing stratification in the upper ocean. This warm bias is more significant in the Southern Hemisphere (about 0.05°C on average) than in the northern Hemisphere. Furthermore, the standard deviation of diurnal warming in the southern Hemisphere (mean value of 0.26°C) is much lower than in the Northern Hemisphere. A cold bias can be observed in the winter months, which is attributable to seasonal changes in upper-ocean stratification. However, the standard deviation of the temperature difference (~0.08°C) is smaller than in the summer months because stronger winter winds reduce the background stratification.

Scatter plots of the TMI and the AMSR-E SSTs and the Argo NSTs are shown in Fig. 4. Overall, the TMI/AMSR-E SSTs and the Argo NSTs show good agreement over the global oceans, with correlation coefficients of 0.986 and 0.995, respectively. The root mean square errors (RMSEs) of the TMI and the AMSR-E SSTs with respect to the Argo observations are 0.74 and 0.73°C, respectively, which are higher than found by previous studies (e.g., Wentz et al., 2000; Senan et al., 2001; Bhaskar et al., 2009). These earlier studies used smaller ranges for validation; however, the moored buoy time series they used had more data points than the Argo NSTs used in our analysis.

The TMI/Argo and AMSR-E/Argo temperature differences were bin averaged for a range of physical parameters including not only satellite retrievals of wind speed and atmospheric water vapor (Dong et al., 2006), but also liquid cloud water. The largest temperature differences occur at low wind speed (Fig. 5). For example, on average, for wind speeds less than 5 m/s, both TMI and AMSR-E SSTs are warmer than the Argo NSTs, with maximum warming of about 0.5 and 1.5°C and mean values of 0.13 and 0.63°C, respectively (Table 1). Previous studies (Dong et al., 2006; Donlon et al., 2002; Gentemann et al., 2004) have found that the difference between skin and bulk temperatures have large variations at low wind speeds. Stronger winds tend to mix the water, causing small subskin-bulk SST differences. At low wind speeds, especially during daytime with solar insolation, subskin-bulk SST differences tend to be higher because of the effects of diurnal warming. For wind speeds greater than 5 m/s, the subskin-bulk SST differences become smaller, because wind-induced mixing homogenizes the upper few meters of the ocean. For wind speeds greater than 11 m/s, the standard deviation of the TMI/Argo temperature difference tends to increase (Fig. 5a) more rapidly than that of the AMSR-E/Argo. This implies that the AMSR-E SSTs might have smaller errors than the TMI SSTs under strong wind conditions. This strong wind effect is likely linked to a surface roughness, particularly to the formation of foam and spray, which in turn, leads to high sea surface emissivity. On average, for wind speeds greater than 15 m/s (here, wind speed measured using the 10.65 GHz channel), the TMI SSTs tend to be smaller than the Argo NSTs by about 0.36°C (Fig. 5, Table 1). This cool bias is more distinct with an average bias of –0.56°C if wind speeds are measured using the 37.0 GHz channel. However, the cool bias (about –0.08°C) is not distinct for the AMSR-E retrievals at high wind speeds. This result is consistent with the cool bias found by Gentemann and Wentz et al. (2001), and it implies that the TMI SSTs might be poorly calibrated under strong wind conditions.

Columnar water vapor is the second most important factor that affects the temperature differences, as has been found by some previous studies (Dong et al., 2006; Gille, 2012). Figure 6 shows that both TMI and AMSR-E SSTs are warmer on average than Argo NSTs when columnar water vapor is less than 8 mm, with mean warming of 0.80 and 0.64°C, respectively (Table 1). Radiation is affected by the absorption and emission of water vapor and cloud liquid water within the atmosphere and therefore, brightness temperatures measured by satellites differ from ground-based brightness temperature measurements (Brisson et al., 2002; Merchant and Le Borgne, 2004). Columnar water vapor is normally at its lowest in cold conditions; therefore, biases for low water vapor conditions are expected to have impact at high latitudes. This suggests there are calibration problems for the TMI and AMSR-E SSTs at high latitudes. Compared with low latitudes, there are fewer calibration data available for high latitudes, which might lead to poorer calibration under low water vapor conditions (Dong et al., 2006). The temperature differences become smaller when water vapor values are greater than 8 mm. By excluding data pairs for low water vapor conditions (< 8 mm), the average temperature differences between the TMI/AMSR-E SSTs and Argo NSTs are reduced from 0.05 and 0.20°C (Table 1) to –0.01 and 0.15°C, with mean standard deviations of 0.72 and 0.69°C, respectively. Therefore, the extensively distributed Argo NSTs could provide a data set with potential for eliminating the low water vapor effect in the TMI and AMSR-E SSTs.

Given the impact of columnar water vapor, liquid cloud water could also be suspected as a source of bias. However, the TMI/Argo and AMSR-E/Argo temperature differences hardly vary with liquid cloud water (Fig. 7). The mean standard deviations of the temperature differences are 0.75 and 0.69°C for the TMI/Argo and the AMSR-E/Argo, respectively, with cloud water varying in a range of 0–0.2 mm. However, on average, the TMI SSTs tend to be warmer than the Argo NSTs by 0.2°C when liquid cloud water is low (< 0.06 mm). AMSR-E retrievals are colder than the Argo NSTs when liquid cloud water is high (> 0.15 mm) (Table 1), with a maximum value of about 0.4°C. This is consistent with the results from a comparison of the AMSR-E and global Argo 5 m temperatures (Gille, 2012).

Bin-averaged temperature differences based on latitude were calculated for both TMI/Argo and AMSR-E/Argo SSTs (Fig. 8). The standard deviations between the TMI/AMSR-E SSTs and the Argo NSTs are less than 1°C for most latitudes (with mean standard deviations of 0.56 and 0.54°C for the TMI/Argo and AMSR-E/Argo temperature differences, respectively), except for the mid-latitudes (35°–40°N) in the Northern Hemisphere for the TMI, and for 55°–60°S and 36°–52°N for the AMSR-E. The maximum standard deviation of the temperature differences can exceed 1.4 and 1.2°C between 35°N and 40°N for the TMI/Argo and the AMSR-E/Argo, respectively. Most Argo profiles between 35°N and 52°N are located near the Kuroshio, Oyashio, and Kuroshio Extension regions (Fig. 9), where the SST has distinct spatiotemporal variabilities (Miller et al., 1994; Deser and Blackmon, 1995; Qiu, 2000; Nonaka and Xie, 2003). Between 55°S and 60°S in the Southern Ocean, the presence of the Antarctic Circumpolar Current can lead to considerable variability of the SST (Hall and Visbeck, 2002; Verdy et al., 2006). Although there are few TMI/Argo and AMSR-E/Argo data pairs in the tropics (between 10°N and 15°N), their standard deviations of the temperature differences are smaller than in the mid- and high latitudes, with mean values of about 0.46 and 0.30°C, respectively.

To compare the TMI and the AMSR-E SSTs with the Argo NSTs in the Kuroshio and Kuroshio Extension regions, we selected the Argo NSTs in two regions: 20°–30°N, 125°–135°E (Box B in Fig. 1) and 30°–40°N, 135°–168°E (Box A in Fig. 1). Both TMI and AMSR-E SSTs compare well with the in situ observations in these regions. The correlation and the RMSE are slightly better for the AMSR-E SSTs than for the TMI SSTs (Table 2). The RMSEs for the TMI and AMSR-E SSTs with respect to the Argo NSTs in the region of the Kuroshio Extension are higher than in the Kuroshio region, with values of 1.27 and 1.21, respectively, i.e., the variability of the SST in the Kuroshio Extension region is greater than in the Kuroshio region. Some seasonal dependence of the temperature difference can still be seen in the Kuroshio Extension region (Fig. 10). The satellite-retrieved SSTs tend to be warmer than the in situ observations from February to August, with mean warming of 0.41 and 0.52°C for the TMI and the AMSR-E, respectively. However, on average, the satellite-retrieved SSTs are lower than the Argo NSTs in October and November. The maximum standard deviation of the temperature difference (up to 2.21°C) was observed in spring.

On average, the TMI and AMSR-E SSTs in the Kuroshio Extension region are warmer than the Argo NSTs when the wind speed is less than 8 m/s, with mean warming of 0.78 and 1.02°C, respectively (Fig. 11). A similar warming tendency by as much as 2.5°C is found in the Kuroshio region when the wind speed is less than 5 m/s (not shown). This implies that neither the TMI nor the AMSR-E SSTs in the regions of the Kuroshio and Kuroshio Extension are well calibrated at low wind speeds, and that the TMI might perform better than the AMSR-E. Similar to the results over the global oceans, the TMI and AMSR-E SSTs in the Kuroshio Extension region are slightly warmer than the Argo temperatures when columnar water vapor is less than 8 mm, with mean warming of 0.82 and 0.81°C, respectively (not shown); note, columnar water vapor of less than 8 mm was not observed in the Kuroshio region.