Radio Science

Retrieval of precipitable water using Special Sensor Microwave/Temperature-2 (SSM/T-2) millimeter-wave radiometric measurements

Authors


Abstract

[1] Four years of Special Sensor Microwave/Temperature-2 (SSM/T-2) measurements over the high-latitude regions of both the Northern and Southern Hemispheres are used to retrieve precipitable water W < 0.8 cm. The retrieved W values compare well with several thousand near-concurrent radiosonde-derived estimates of W. Additionally, very good agreement is found between the retrieved SSM/T-2 W values and continuous ground-based radiometer-derived estimates of W during a couple of extended time periods. A bias (0.03–0.06 cm) exists between the W values retrieved from the SSM/T-2 and the values derived from the radiosonde/radiometer observations. This bias is caused primarily by the assumption of equivalent surface emissivity across the 150–183 GHz frequency range that is made during the development of the retrieval algorithm. The spatial distribution of the SSM/T-2-retrieved winter mean W values is examined together with digital elevation models for both the Greenland and Antarctic ice sheets. This detailed comparison conveys expected climatological characteristics that are quite sensible when the combined effects of the surface elevation and the general atmospheric circulation patterns near and over the ice sheets are considered in unison. The SSM/T-2 W retrieval presented provides accurate W information with a high sampling rate for the potentially climate change-sensitive polar regions of the Northern and Southern Hemispheres.

1. Introduction

[2] Although high-latitude regions receive much less incoming solar radiation than the equatorial regions, the high-latitude contribution to the outgoing longwave radiation budget is substantial [Salby, 1996]. Atmospheric water vapor is an important source of energy that feeds the development of weather systems while also being a strong absorber of infrared radiation with a significant impact on the planetary greenhouse effect. Therefore water vapor in these regions is linked through its local influence on the longwave radiation budget to global energy balance and climate.

[3] When assessing the polar surface radiation budget, uncertainties in the estimation of atmospheric water vapor can create errors in satellite-derived estimates of surface temperature [Key and Haefliger, 1992] as well as surface albedo [Rossow et al., 1989]. Therefore accurate knowledge of seasonal and regional variations of precipitable water for the high-latitude regions should help improve the accuracy of these important retrievals. Additionally, there are numerous passive microwave radiometer retrieval techniques used at high latitudes that could potentially benefit from quantifying the effect of water vapor. For instance, Abdulati and Steffen [1997] note that none of the passive microwave methods used to detect snowmelt over ice sheets has included any quantification of the variability in the input brightness temperatures that can be caused by changing atmospheric conditions. Steffen et al. [1992] make the same statement in reference to several passive microwave sea ice retrieval algorithms. Other passive microwave retrievals such as surface temperature [Shuman et al., 1996] and snow depth [Foster et al., 1997] could also benefit from quantifying the effect of water vapor. The amount of improvement would likely be small for most retrieval cases; however, a recent study demonstrates that the effects should not be ignored, especially when the retrieval method uses frequencies ≥37 GHz [Wang and Manning, 2003a].

[4] Until a few years ago, satellite microwave measurements of precipitable water W were limited to the ocean areas where the cold ocean surface microwave emission can be accurately characterized [e.g., Alishouse et al., 1990; Manning, 1997]; the retrieval technique is based on multichannel measurements near the weak water vapor absorption line of 22.235 GHz and nearby window regions. Recently, algorithms using measurements near the stronger water vapor line of 183.31 GHz have been developed to retrieve low W < 0.8 cm over Antarctica [Conlee, 1994; Moore, 1997; Miao, 1998; Miao et al., 2001]. Miao et al. [2001] validated an independently derived Special Sensor Microwave/Temperature-2 (SSM/T-2) W retrieval method over Antarctica using 1 month of numerical model data while also demonstrating the retrieval method's capability of showing seasonal variations of water vapor over Antarctica. They further asserted that the retrieval method's high sensitivity to water vapor and low sensitivity to surface emissivity characteristics and clouds make the proposed method especially useful in polar regions where atmospheric water vapor measurements are temporally and spatially sparse.

[5] The regression approach of Miao et al. [2001] is straightforward and was utilized by Wang et al. [2001, 2002] in their analysis of the airborne Millimeter-wave Imaging Radiometer (MIR) measurements over the Arctic and midwest regions; more specific details related to the algorithm are provided in section 2.2. Wang and Manning [2003b] further extended the analysis to cover a few days of near-concurrent MIR and SSM/T-2 data over these regions and discussed the error sources caused by variations of surface temperature Ts and emissivity ξ(ν) (at frequency ν). The approach proved to be quite robust after these errors were adequately corrected within the retrieval procedure.

[6] The same regression approach [Wang and Manning, 2003b] is applied to the Defense Meteorological Satellite Platform (DMSP) SSM/T-2 data set over the midlatitude and high-latitude regions of the Northern and Southern Hemispheres, and the output is compared directly with instantaneous radiosonde- and radiometer-derived point measurements of W where applicable. The primary objective of this study is to thoroughly examine the instantaneous performance of the SSM/T-2 W retrieval procedure in an effort to more completely assess the quality of the product generated.

2. Data Sources and the Retrieval Algorithm

2.1. Data Sources

[7] Three data sources covering a 4 year period from 1998 to 2001 are utilized in this study, namely, the microwave measurements from the SSM/T-2, the radiosonde observations archived by National Oceanic and Atmospheric Administration (NOAA)'s Forecast Systems Laboratory (FSL), and the measurements from the Atmospheric Radiation Measurement (ARM) Program's Microwave Radiometer (MWR). The SSM/T-2 aboard the DMSP F12, F14, and F15 satellites is a cross-track scanning, total power microwave radiometer that has five channels at 183.31 ± 1, 183.31 ± 3, 183.31 ± 7, 150, and 91.655 GHz. Measurements from all except the 91.655 GHz channel are used in this analysis. Because there are documented problems with SSM/T-2 data near the scan edge that are caused by interference with the glare obstructor, the three leftmost scan positions are not used to retrieve W values as was done by Miao et al. [2001]. There is good agreement between the MIR and SSM/T-2 measurements as documented by both Falcone et al. [1992] and Wang and Manning [2003b]. Therefore the knowledge gained from the previous MIR-based studies [Wang et al., 2002; Wang and Manning, 2003b] is directly applicable to the global SSM/T-2 data archive.

[8] Both radiosonde observations and MWR measurements serve as a validation tool for the SSM/T-2 W retrievals in this study. The global radiosonde observations poleward of 30°N and 50°S for the 1998–2001 period were obtained from the NOAA FSL website. The radiosonde data undergo extensive gross error and hydrostatic consistency checks by the NOAA FSL before being made available for download; however, a disclaimer is provided that states that the radiosondes have not been subjected to production quality control procedures. In the present study, no further quality control was applied to the radiosonde data. Connolley and King [1993] reported an uncertainty in radiosonde-derived W of ∼20% for the extremely cold and dry Antarctic environment. This is potentially a primary error source realized within the comparison of the radiosonde- and SSM/T-2-derived W values in this study.

[9] Data from the MWR consist of two time series of measurements from separate locations. The first time series of data was collected during April and May 1998 on board the drifting Surface Heat Budget of the Arctic Ocean (SHEBA) icebreaker in the Arctic Ocean. The second time series of data was gathered during the Millimeter-Wave Radiometric (MMWR) Winter Water Vapor experiment that was conducted in March 1999 at the ARM Program's North Slope of Alaska/Adjacent Arctic Ocean Cloud and Radiation Testbed (CART) site near Barrow, Alaska. Details related to the MWR and the W retrieval during the experiment at Barrow, Alaska, are documented by Han et al. [2000]. A more detailed study using the MWR aboard the SHEBA icebreaker is described by Westwater et al. [2001]. When comparing point measurements of both radiosonde-derived and MWR-measured W validation data with SSM/T-2-derived W, the center point of the SSM/T-2 beam and the validation data point location must be within 50 km spatially and within ±2 hours temporally of the SSM/T-2 overpass.

2.2. Algorithm

[10] The algorithm to retrieve W is based on the regression approach developed by Miao [1998] and Miao et al. [2001] using the measurements from the 150, 183.31 ± 3, and 183.31 ± 7 GHz channels (method 1) of the SSM/T-2 as well as the three-channel combination that uses the 183.31 GHz channels (method 2). A large set of radiosonde data is used to calculate brightness temperatures Tb(ν) over a range of surface emissivity ξ(ν) values from 0.55 to 0.95 at the frequency ν = 150 GHz as well as the three 183.31 GHz channels of the SSM/T-2. The regression is made between the calculated Tb(ν) and the corresponding W values. Method 1 requires an assumption of equivalent surface emissivity ξ that is independent of frequency across the range of 150–183 GHz. In addition, both methods are developed at a fixed surface temperature Ts. The end results of the regression for method 1 are given by [Miao, 1998]

equation image
equation image

where α, β, X0, and Y0 are coefficients derived from the regression and θ is the radiometer view angle measured away from nadir. For method 2 a different set of coefficients results when Tb(150), Tb(183.31 ± 7), and Tb(183.31 ± 3) in equation (2) are replaced by Tb(183.31 ± 7), Tb(183.31 ± 3), and Tb(183.31 ± 1), respectively. Method 1 is applicable when Tb(183.31 ± 7) < Tb(183.31 ± 3); this happens for Wsec(θ) < 0.8 cm when both channels sense the cold surface background. Similarly, method 2 becomes effective when Tb(183.3 ± 3) < Tb(183.3 ± 1); this usually occurs when Wsec(θ) ≤ 0.2 cm. It is important to note that Wang et al. [1997] discovered that the radiometric signatures seen by the 183.31 GHz water vapor channels over convective storms containing scattering hydrometeors is quite similar to the radiometric signature that is used to determine the application of W retrieval for methods 1 and 2. When sampling intense convective storms, the 183.31 GHz channels become opaque above the freezing level of the storm, so the brightness temperature contribution would primarily come from the water vapor present high in the atmosphere between the scattering hydrometeors and the airplane carrying the MIR instrument. Therefore all three of the water vapor channels are most sensitive to the amount of water vapor present in the atmospheric column between the plane and the opaque layer. The SSM/T-2 data are probably less sensitive to this phenomenon than the MIR instrument simply because the resolution of the SSM/T-2 instrument is roughly 25 times greater than the MIR; however, the size of some storms is large enough that the same radiometric signature is detected within the larger beam of the SSM/T-2 radiometer. There was no attempt to exclude these false retrievals that occur infrequently over convective storms. Because extremely dry air is usually associated with synoptic-scale phenomena and intense convective regions are more spatially isolated events, it is possible that a simple filter based on the size of a supposed valid retrieval region could exclude the potentially erroneous retrievals. It is possible that a few of these erroneous retrievals became part of the database used in this study, but their contribution is not obvious. The issue of properly treating these false retrievals must receive more attention.

[11] Wang et al. [2001] applied this regression approach to the MIR data obtained over the Arctic region and found that the assumption of frequency-independent ξ(ν) across the ν range of 150-183 GHz could lead to errors in W retrievals. Additionally, another analysis that used both MIR and SSM/T-2 data sets [Wang and Manning, 2003b] showed nonnegligible errors in the W retrievals when Ts varies appreciably away from the value used in the derivation of the coefficients in equations (1) and (2). The errors due to Ts variations could be readily corrected if Ts is independently measured [Wang and Manning, 2003b]. This correction procedure is applied as in section 3 whenever Ts is available. The correction for the errors resulting from the assumption of the same ξ(ν) across 150–183 GHz is not trivial because of the limited channels of the SSM/T-2 measurements; MIR has an additional channel of measurements at 220 GHz that is used to correct for the ξ(ν) effect [Wang et al., 2001]. The effect of the equivalent ξ(ν) assumption on the W retrievals when using method 1 is further described in section 3.

3. Retrieval Results and Comparison

3.1. Comparison With Radiosondes

[12] In this section the retrieved W values from the SSM/T-2 are compared with those derived from the radiosonde data obtained from NOAA FSL. First, for extremely dry atmospheric cases where Tb(183.31 ± 1) > Tb(183.31 ± 3) the three-channel combination using only the 183.31 GHz channels should provide a more robust W retrieval because the equivalent ξ assumption for the three frequencies used is more accurate. Table 1 shows the comparison between the 150, 183.31 ± 3, and 183.31 ± 7 GHz channel combination, method 1, and the three-channel 183.31 GHz combination, method 2, for all radiosonde comparison cases where the Tb(183.31 ± 1) > Tb(183.31 ± 3) criterion is met; both are corrected for the Ts effect. Table 1 gives sample number, bias, and standard deviation of the difference between the W retrieval results for each SSM/T-2 sensor available from 1998 to 2001 and the corresponding matching radiosonde W values. The information is separated by year to see if there is any large degradation of any of the SSM/T-2 radiometers with increasing time. Table 1 shows that the F14 W retrievals using both methods provide significantly different results than both the F12 and F15 retrievals. Clearly, the F12 and F15 W retrieval cases described in Table 1 demonstrate that there is a definite advantage to using method 2. The absolute value of the bias is smaller for all sets of data except the F12 1998 collection, which also contains the smallest number of samples. More important, the standard deviation values are significantly smaller when using method 2 compared with method 1. Therefore, when applying this retrieval to current operational radiometers such as SSM/T-2 and the Advanced Microwave Sounding Unit-B (AMSU-B), the three-channel 183.31 GHz combination, method 2, should be used exclusively when Tb(183.3 ± 1) > Tb(183.3 ± 3).

Table 1. Comparison Statistics Between SSM/T-2-Derived W and Radiosonde-Derived W Using Both Methods 1 and 2 for All Cases Where Tb(183.3 ± 1) > Tb(183.3 ± 3) for Three Separate SSM/T-2 Radiometers for Each Year Available From 1998 to 2001
Defense Meteorological Satellite PlatformYearSamplesMethod 1Method 2
Bias, cmStandard DeviationBias, cmStandard Deviation
F1219982530.0070.067−0.0110.041
F12199911090.0270.0520.0030.035
F12200013340.0290.060−0.0040.045
F141998462−0.0090.080−0.0210.063
F1419991097−0.0070.074−0.0180.058
F14200010400.0000.079−0.0260.060
F14200112420.0080.090−0.0350.062
F15200012980.0280.060−0.0020.040
F15200118750.0310.053−0.0060.040

[13] Since it is proven that method 2 produces better W retrieval results for extremely dry conditions, it should be used when applicable. This is done to all the available SSM/T-2 data, including the subset used for Table 1, and the results are compared in Table 2. The bias and standard deviation values in the fourth and fifth columns are not corrected for the Ts effect, and those in the sixth and seventh columns are corrected for the Ts effect. Clearly, the correction for the Ts effect reduces the bias in the comparison, although the standard deviation values practically remain the same. The bias and standard deviation values are comparable and relatively consistent from year to year for both the F12 and F15 cases. The differences in both bias and standard deviation observed for the F14 data in Table 1 again exist in Table 2. The noticeably larger standard deviation for the F14 cases in both tables suggests a noisier SSM/T-2 radiometer is operating on that satellite platform. The consistently different bias values from those of the F12 and F15 cases imply a possible discrepancy in the sensor calibration. For the reason of maintaining consistency and uniformity of data through the entire 1998–2001 period the retrieved W values from the F14 SSM/T-2 are excluded from further comparison and analysis.

Table 2. Comparison Statistics Between SSM/T-2-Derived W Both Without and With the Ts Correction and Radiosonde-Derived W Using Both Channel Combinations Where Applicable for Three Separate SSM/T-2 Radiometers for Each Year Available From 1998 to 2001
Defense Meteorological Satellite PlatformYearSamplesNot Corrected for Ts EffectCorrected for Ts Effect
Bias, cmStandard DeviationBias, cmStandard Deviation
F12199811380.0280.0880.0180.091
F12199941590.0370.0790.0290.079
F12200053190.0380.0820.0290.080
F14199821290.0120.105−0.0020.109
F14199939020.0080.103−0.0020.104
F14200046530.0080.109−0.0020.109
F14200151310.0080.116−0.0040.116
F15200057180.0410.0790.0300.078
F15200167160.0430.0780.0340.078

[14] The bias and standard deviation values listed for the temperature-corrected F12 and F15 W retrievals in Table 2 are applicable to the SSM/T-2 W retrievals presented throughout the remainder of this study. Figure 1 shows the scatterplot comparing all of the F12 and F15 SSM/T-2 W retrievals with the radiosonde-derived W values for all of the temperature-corrected points given for the two radiometers in Table 2. The retrieval bias found between the SSM/T-2 and radiosonde W values originates mostly from data with W > 0.2 g/cm2 and is quite similar to the bias derived for a significant number of the MIR airborne retrieval cases of Wang and Manning [2003b] between the similar three-channel and the more reliable four-channel retrieval derived by Wang et al. [2001].

Figure 1.

Scatterplot comparing all of the F12 and F15 SSM/T-2 temperature-corrected W retrievals with matching radiosonde-derived W values.

[15] To verify that this bias is mainly caused by the assumption of an independent ξ(ν) across 150–183 GHz for the method 1 retrieval algorithm, the subset of the retrievals with W ≤ 0.2 cm in Table 1 is used to estimate ξ(150) and ξ(183) with the following equation [Wang, 2002]:

equation image

where Ta(ν) and Γ(ν) are the optical depth weighted atmospheric temperature and optical depth evaluated at frequency ν, respectively, and TCB is the cosmic background radiation. Both Ta(ν) and Γ(ν) are approximately linearly related to W and are readily estimated [Wang, 2002]. For the estimation of ξ(183) the most transparent channel, Tb(183.3 ± 7), is used because it allows the most accurate derived value [Wang, 2002]. The results are displayed in terms of histograms in Figure 2c.

Figure 2.

Histograms showing the difference between the temperature-corrected F12 and F15 SSM/T-2- and radiosonde-derived W retrievals (a) for all sample cases where the method 2 channel combination is valid and (b) for all sample cases where the method 1 channel combination is valid. (c) Histogram showing the differences in derived emissivity values at 150 and 183.31 GHz using all samples displayed in Figure 2a.

[16] Figures 2a and 2b describe the difference between the SSM/T-2 and radiosonde retrieval for both SSM/T-2 channel combinations as well as for both the Northern and Southern Hemispheres. Figure 2a shows that there is very little difference in the performance of the 183.31 GHz only W retrieval between the Northern and Southern Hemispheres. This is expected because of the more valid assumption of equivalent ξ for the method 2 retrieval. For Figure 2b most of the bias shown for the Northern and Southern Hemispheres is likely related to the 150 and 183.31 GHz emissivity difference displayed in Figure 2c. As expected, the larger W retrieval bias in the Northern Hemisphere corresponds to a large proportion of cases where ξ(150) is significantly smaller than ξ(183). A smaller positive SSM/T-2 W bias for the Southern Hemisphere in Figure 2b is also explained by the fact that ξ(150) is almost equal to but slightly smaller than ξ(183) for most Southern Hemisphere cases. These same relationships in the 150–220 GHz frequency range were described in more detail by Wang and Manning [2003b].

3.2. Comparison With MWR

[17] An ARM MWR was operational during both SHEBA and the MMWR Winter Water Vapor experiment. The MWR operates almost continuously, and it provides another reliable validation source for the SSM/T-2 W retrieval. Figure 3 exhibits the W values derived from radiosonde, MWR, and the F12 SSM/T-2 during April and May 1998 for the SHEBA ice station. It is evident that all three of the W retrieval sources follow the same temporal patterns. Note that there are two periods of time where no SSM/T-2 W retrievals are present; these occur from about 15 to 20 April 1998 and from about 11 to 20 May 1998. During these time periods the MWR and radiosonde W retrievals are primarily >0.6 cm, which is near the limit of the SSM/T-2 retrieval capability. Another interesting characteristic of the samples displayed in Figure 3 is the persistent bias seen between the SSM/T-2 W and the validation data from around 29 April to 10 May 1998. This prolonged bias is again likely due to the suspected difference between ξ(150) and ξ(183).

Figure 3.

Line plot showing temporal patterns of the retrieved W values derived from radiosonde, MWR, and the F12 SSM/T-2 during April and May 1998 over the SHEBA ice station.

[18] Figure 4 directly compares the results of SSM/T-2 and MWR retrievals. Each valid SSM/T-2 overpass is directly compared with the mean MWR retrieved value for all W samples within ±30 min of the SSM/T-2 overpass. The bias and standard deviation of the samples shown in Figure 4 provide information related to the behavior of the SSM/T-2 W retrieval over sea ice. Additionally, we investigated the performance of the W retrieval farther over sea ice by directly comparing all valid SHEBA radiosonde-derived W values to matching SSM/T-2-retrieved values. Because the SHEBA experiment lasted about a year, the approximate annual behavior of the W retrieval over Arctic sea ice for a sample set of 281 is characterized by a bias of 0.04 cm and a standard deviation of 0.06.

Figure 4.

Scatterplot comparing retrieved MWR and the F12 SSM/T-2 W values during April and May 1998 for the SHEBA ice station.

[19] Figure 5 exhibits the W values derived from radiosonde, MWR, and the F12 SSM/T-2 during March 1999 for the MMWR Winter Water Vapor experiment conducted at the ARM Program's CART site near Barrow, Alaska. Similar to the SHEBA comparison in Figure 3 the retrieved W values for each sensor type exhibit good agreement. Unfortunately, Ts data were not readily available for the entire time period, so the Ts correction is not applied.

Figure 5.

Line plot showing temporal patterns of the retrieved W values derived from radiosonde, MWR, and the F12 SSM/T-2 during March 1999 for Barrow, Alaska.

[20] Figure 6 shows the scatterplot between the SSM/T-2 and the MWR W retrievals shown in Figure 5. The SSM/T-2 W retrievals above 0.2 cm again show a wet bias that is likely related to the surface emissivity characteristics where ξ(150) < ξ(183). Although Figure 2c undoubtedly suggests that ξ(150) < ξ(183) over many types of land surfaces, it is not yet acceptable to use the assumption to improve the SSM/T-2 W retrieval. Additional work is necessary to quantify ξ characteristics at these wavelengths over many types of surfaces.

Figure 6.

Scatterplot comparing retrieved MWR and the F12 SSM/T-2 W values during March 1999 for Barrow, Alaska.

4. SSM/T-2 W Retrieval Sampling and Mean Daily Values

[21] Figures 7a and 7b show the number of valid W retrievals returned by the F12 and F15 SSM/T-2 radiometers for the year 2000. The largest number of retrievals per 50 × 50 km grid cell occur over Greenland and the Arctic Ocean in the Northern Hemisphere and over Antarctica in the Southern Hemisphere. Within each grid cell in these locations, there are a few thousand valid retrievals per year. When conditions are dry, the SSM/T-2 radiometer is capable of providing polar W retrievals many times per day for a given location, especially poleward of 70° latitude in both the Northern and Southern Hemispheres. Figures 7a and 7b also show the more limited application of the SSM/T-2 W retrieval for the lower-latitude regions. Because the number of SSM/T-2 overpasses that occur at midlatitudes is significantly lower than those near the poles and also because extremely dry air does not reach the midlatitudes as frequently, the SSM/T-2 W retrieval returns a significantly lower number of valid retrievals for these regions. For instance, Figure 7c shows that the SSM/T-2 W retrieval returns valid information ∼50 days per year over the Wisconsin area, which is barely visible near the lower left corner of the image. Figure 7c also shows that the maximum number of days with valid retrievals occurs over Greenland in the Northern Hemisphere; in fact, it appears that the SSM/T-2 W retrieval is valid almost every day for the higher-elevation areas of the Greenland ice sheet. Additionally, a few high-latitude mountain ranges are apparent in Figure 7c. The Brooks Range mountainous region in Alaska has a significantly larger number of days when the retrieval is valid when compared with the immediate surrounding areas in Alaska. Likewise, there are more days with valid W retrievals over several mountain ranges in central and eastern Asia between 60° and 70°N when compared with the immediate surrounding areas at lower elevations.

Figure 7.

Number of F12 and F15 SSM/T-2 W retrievals produced per 50 km grid cell for the year 2000 (a) in the Northern Hemisphere and (b) in the Southern Hemisphere. Number of days for the year 2000 when a valid W retrieval was returned for each grid cell (c) in the Northern Hemisphere and (d) in the Southern Hemisphere. Mean W value computed using all of the applicable daily mean W values for the year 2000 (e) in the Northern Hemisphere and (f) in the Southern Hemisphere.

[22] Figure 7d demonstrates that the SSM/T-2 W retrieval approach is valid throughout the year for a large portion of Antarctica. This is especially true for the interior areas of the continent away from the coast. Additionally, the oceans around Antarctica, which are covered by sea ice often throughout the year, have a significant number of days when the SSM/T-2 produced valid retrievals; the same is also true for the sea ice covered areas in the Arctic Ocean shown in Figure 7c. The high sampling rate in these polar areas and the overall robustness of the SSM/T-2 W retrieval approach for the extremely dry atmospheric conditions often present in these remote locations undeniably suggest that this retrieval approach will be useful during present and future precipitable water studies in the polar regions of the Northern and Southern Hemispheres.

[23] Figures 7e and 7f show the mean W value computed using all of the applicable daily mean W values for the year 2000. Because no temperature correction is applied to the SSM/T-2 values used to compose Figures 7e and 7f or the retrieval data shown in Figure 8, the bias that is applicable to these retrieved values is likely to be ∼0.01 cm higher than the more accurate temperature-corrected retrieval values as demonstrated in Table 2. Serreze and Barry [1995] note that a region of peak poleward water vapor transport at 70°N near the prime meridian is due primarily to high mean specific humidities at low levels and the frequent advection of moisture by transient eddies. By following the prime meridian poleward in Figure 7e the largest mean W value at 80°N is found slightly offset from the prime meridian. This certainly suggests that significant poleward water vapor transport is occurring in this area. Additionally, Serreze and Barry [1995] note that the region of peak equatorward water vapor transport occurs over the Canadian archipelago near 90°W longitude. Figure 7e does show lower mean W values in this same area. In fact, the mean W values shown are <0.3 cm from near Hudson Bay and continue poleward along the 90°W longitude line to the point where SSM/T-2 data are no longer present near the North Pole. Because the mean W values that surround the North Pole increase quickly in every direction except in the general direction of the 90°W longitude line, it is likely that this dry air that originates near the North Pole is the source of the air found near the Canadian archipelago. Therefore the feature shown in Figure 7e is probably associated with the same equatorward transport of water vapor at 70°N that is mentioned by Serreze and Barry [1995].

Figure 8.

(a) Mean SSM/T-2 W values computed over Greenland for December 1999 through February 2000 using F12 and F15. (b) Mean SSM/T-2 W values computed over Antarctica for June through August 2000 using F12 and F15. (c) Greenland digital elevation model [Zwally and Brenner, 2001]. (d) Antarctica digital elevation model [Zwally and Brenner, 2001]. (e) Scatterplot comparing mean SSM/T-2 W and surface elevation for the Greenland ice sheet. (f) Scatterplot comparing mean SSM/T-2 W and surface elevation for the Antarctic ice sheet. (g) Slope parameter over Greenland derived from Figure 8e. (h) Slope parameter over Antarctica derived from Figure 8f.

[24] Figure 7f shows the mean SSM/T-2 W values over Antarctica and the surrounding oceans. It is sensible that the largest W values are located at lower latitudes as shown. Figure 7f also shows lower mean W values over the sea ice regions near the Ross Ice Shelf and the Transantarctic Mountains located at ∼75°S, 180°E and near the Amery Ice Shelf near 72°S, 75°E. This characteristic is primarily explained by the fact that both of these regions experience a significant amount of extremely dry katabatic wind drainage from the higher elevations of the Antarctic continent. Additionally, the ocean surface in these areas is frequently covered by sea ice, which essentially caps the upper layer of the ocean and prevents the exchange of heat and moisture between the ocean and the atmosphere [King and Turner, 1997].

5. Relationship Between Mean SSM/T-2 W and Surface Elevation Over Ice Sheets

[25] During the winter months in each hemisphere the SSM/T-2 W retrieval almost always produces valid results in the polar regions, especially over Greenland and Antarctica. Therefore representative mean SSM/T-2 W values are computed and plotted over Greenland for December 1999 through February 2000 in Figure 8a and over Antarctica for June–August 2000 in Figure 8b. Figures 8c and 8d show digital elevation models (DEM) for Greenland and Antarctica, respectively, for the same resolution (50 × 50 km) polar stereographic projection grid used to sort the SSM/T-2 W retrieval data. Detailed information related to the DEM data is given by Zwally and Brenner [2001]. A visual examination comparing Figures 8a and 8c and Figures 8b and 8d reveals the expected inverse correlation between the mean W for each of the time periods and the corresponding surface elevation. The decrease in mean W for higher elevations is partially explained by the presence of less atmospheric column over these areas and partially by the fact that the ambient air temperatures at these higher elevations are extremely cold. Because of these low air temperatures, air parcels present there generally contain less water vapor even when saturated.

[26] Figures 8e and 8f show the relationship between the surface elevation and winter mean W values over both the Greenland and Antarctic ice sheets, respectively. Both of these scatterplots demonstrate considerable scatter when comparing the two variables, so this detected variability is examined more closely by a calculated slope parameter that is specific to the scatterplots shown in Figures 8e and 8f. The slope of the line drawn between a fixed point where [x, y] = [0, 4.5] and each individual data point in Figures 8e and 8f is calculated. Figures 8g and 8h show the calculated slope parameter for all grid cell samples in both Greenland and Antarctica, respectively, and the colors plotted correspond exactly with the colors of the samples displayed in the scatterplots in Figures 8e and 8f.

[27] Figures 8e and 8g show that there are a significant number of grid cells on the eastern half of Greenland that have higher mean W values than those on the western half of Greenland with similar surface elevations. This characteristic is quite sensible for a couple of reasons. During the winter period, Baffin Bay on the western coast of Greenland is mostly covered with ice, which prevents the exchange of heat and moisture between the ocean and the atmosphere. Because many of the air parcels that move into this region travel equatorward in the polar easterlies from the area near the North Pole, in general, very little water vapor is advected toward Greenland from the east and northeast besides that which comes from the north water polynya in Baffin Bay. The water vapor exchange from the polynya is probably a significant source of the slightly higher W values seen for the coastal grid cells along the northwest coast of Greenland, but as shown, this water vapor does not penetrate far upslope. The primary reason for the larger W values on the eastern side of the continent is related to the water vapor and heat transfer advected into the region by transient eddies. The eddies travel poleward as well as eastward in the North Atlantic Ocean and converge with the semipermanent Icelandic low-pressure system [Grotjahn, 1993]. The eddies travel in a direction that is approximately parallel to the southeastern Greenland coast at varying distances from the coast. They are capable of advecting water vapor a significant distance upslope into the atmosphere on the eastern side of the Greenland ice sheet as well as throughout the southernmost third of the ice sheet. This is demonstrated by the larger slope values present for several pixels inland along the eastern coast and throughout the southern third of Greenland, where the elevation is above 3 km. The smallest slope values are located in the northwest corner of Greenland along the downward sloping section of the DEM shown in Figure 8c; these grid cell locations are shielded most efficiently from the transient eddies by the topography of Greenland.

[28] Figures 8f and 8h show that most of western Antarctica as well as the coastal grid cells have slope parameter values that are similar to Greenland; however, for most of Antarctica these larger values of slope do not extend more than a few grid cells away from the coast. Miao et al. [2001] demonstrated that the annual mean over West Antarctica is nearly twice as high as that over East Antarctica; the W values shown in Figure 8b agree with this pattern. There is a very subtle difference in the slope values along the upper coastal portion shown in Figure 8h when compared with the slope values for the lower right section. This is explained by considering that transient eddy formation, or cyclogenesis, is linked strongly to sea surface temperature gradients and the location of the strongest surface temperature gradient around Antarctica roughly matches the extent of the sea ice during winter [Grotjahn, 1993]. The sea ice along the lower right portion of Antarctica seen in Figure 8h extends a much smaller distance from the coast when compared with the sea ice extent for the coastal region at the top of the map of Antarctica [Comiso et al., 1992]. Therefore transient eddies that form in locations where the sea ice does not extend as far off the coast contain more energy when they reach the Antarctic coast. Hence these eddies are more capable of advecting more water vapor to higher elevations of the Antarctic ice sheet while traveling primarily poleward and eastward as they decay. Grotjahn [1993] revealed that many mature cyclones persist along the coast of Antarctica and that many of the cyclones decay in the Ross Sea (bottom left of Figure 8h); this feature is apparent because the slope values for the entire Ross Ice Shelf are comparable to those for the eastern side of Greenland. The last feature discussed is the location of the large slope values along the western side of the mountains that are situated on the Antarctic peninsula. The larger mean W values contained within these grid cells result from water vapor convergence from eastward moving cyclones as the storms meet the 2 km mountain barrier. Also, it appears that a few of these cyclones move over Antarctica south of the mountain barrier and advect water vapor over the Ronne Ice Shelf, which is located east of the Antarctic peninsula.

[29] Figures 8a–8h convey expected climatological characteristics that are quite sensible when the combined effects of the surface elevation and the general atmospheric circulation patterns for the locations near the ice sheets are considered in unison. These observations certainly convey patterns that seem accurate, and the grid cells with comparably low and high mean W values are situated in reasonable locations. If the emissivity characteristics of these ice sheets is more precisely understood in the future, retrieval methods similar to that which is presented in this study are capable of producing exceptionally accurate retrievals of W around and over these two important ice sheets.

6. Conclusions

[30] The radiometric measurements from the 150 and 183 GHz channels of the SSM/T-2 on board the DMSP F12, F14, and F15 satellites are used to retrieve precipitable water W in the high-latitude regions of both the Northern and Southern Hemispheres for the period 1998–2001. The retrieval results are validated with those derived from near-concurrent radiosonde and ground-based microwave radiometer observations; good agreement is shown between the SSM/T-2 W retrievals and both validation groups. The retrieved W values, on average, are higher than the radiosonde-derived values by 0.008 cm in the Southern Hemisphere and by 0.05 cm in the Northern Hemisphere. This bias is mostly attributable to the assumption of frequency-independent surface emissivity ξ(ν) across 150–183 GHz in the development of the retrieval algorithm. On the basis of the comparison cases where W ≤ 0.2 cm and the retrievals using the three 183.31 GHz channels are applicable the average differences of ξ(150)–ξ(183) are estimated to be −0.011 and −0.037 for the Southern Hemisphere and Northern Hemisphere radiosonde station locations, respectively. Therefore the observed biases between the retrieved and radiosonde-derived W values are consistent with the magnitude of the emissivity differences shown by Wang and Manning [2003b]. Correction to this emissivity-induced error is a nontrivial problem based on the currently available channels of SSM/T-2 measurements.

[31] The large-scale general circulation features revealed by the SSM/T-2 W retrieval results agree well with the atmospheric water vapor characteristics described by Serreze and Barry [1995] for the area around 70°N. Furthermore, the spatial distribution of the SSM/T-2-retrieved winter mean W values corresponds well with recently derived digital elevation models for both the Greenland and Antarctic ice sheets. Collectively, the comparisons of the SSM/T-2 W retrieval product with other data sources strongly suggest that the retrieval procedure presented provides accurate W information with a high sampling rate for the potentially climate change-sensitive polar regions of the Northern and Southern Hemispheres.

Acknowledgments

[32] We thank Ed Westwater and Yong Han of the NOAA Environmental Technology Laboratory for providing the MWR data products used in this study. We also thank Jay Zwally and the many helpful people in his research group at NASA Goddard Space Flight Center for providing the digital elevation models for Greenland and Antarctica. Finally, we acknowledge all of the people behind the scenes responsible for providing the Website and database management at both http://raob.fsl.noaa.gov/ and http://www.saa.noaa.gov/, where the radiosonde and SSM/T-2 data used for this study were obtained.

Ancillary