Microwave Emissivity of Sea Ice Estimated from Aircraft Measurements during FIRE-SHEBA

Transcription

1 1 Microwave Emissivity of Sea Ice Estimated from Aircraft Measurements during FIRE-SHEBA Julie A. Haggerty and Judith A. Curry Program in Atmospheric and Oceanic Sciences, University of Colorado Boulder, Colorado Corresponding Address: Julie Haggerty Department of Aerospace Engineering Sciences Campus Box 429 University of Colorado Boulder, CO (303) (303) (fax) Submitted to Journal of Geophysical Research, FIRE-ACE Special Issue (submitted December 1,1999)

2 2 Abstract Passive microwave radiometers with frequencies ranging from 37 GHz to 220 GHz were flown over the Surface Heat Budget of the Arctic (SHEBA) experimental site in May and July, Brightness temperature measurements from these sensors are used to calculate sea ice emissivity at each frequency using ancillary aircraft data to characterize the atmosphere and obtain surface temperature. Surface emissivities on clear sky days during the FIRE Arctic Cloud Experiment (FIRE-ACE) aircraft campaign have been calculated and compared with previous estimates cited in the literature. Average emissivity at nadir for snow-covered multiyear ice in the experimental region during late May is estimated as 0.91 at 37 GHz, 0.78 at 89 GHz, 0.73 at 90 GHz, 0.77 at 150 GHz and 0.89 at 220 GHz. In early July, the average nadir emissivity for melting multiyear ice is 0.89 for 37 GHz and 0.87 for 90 GHz. Estimates at a 50 viewing angle are compared with previous satellite and groundbased measurements at 37 GHz and 90 GHz. Our results in May are higher than previous averages for dry multiyear ice, particularly at 37 GHz. Emissivities calculated for July compare well with previous measurements for summer melt conditions.

3 3 1. Introduction Passive microwave measurements of sea ice have been available since the late 1970s, and have been used extensively for monitoring various surface properties including ice type, concentration, and extent. Limited effort has been given to retrievals of atmospheric properties over sea ice from microwave data for two reasons. First, the high emissivity and heterogeneity of the ice surface at microwave frequencies make separation of the atmospheric contribution more difficult than over a radiatively cold surface such as the ocean. Second, it has been widely assumed that the atmosphere is effectively transparent to the commonly used frequencies (e.g., the 19 and 37 GHz channels of SSM/I) when performing sea ice retrievals. However, if microwave observations are to be used for retrieving atmospheric properties, the surface contribution to the upwelling radiance must be estimated. With a method for estimating the surface contribution, retrieval algorithms such as those used over the ocean for deriving cloud liquid and ice water path, can be developed for application over sea ice. Various researchers have addressed the issue of high background emissivity in the microwave portion of the spectrum with studies over land. Jones and VonderHaar (1997) developed a method for measuring cloud liquid water path over land with SSM/I. Their algorithm includes an estimate of land surface emissivity derived with ancillary visible and infrared satellite observations in clear sky regions. By estimating the surface contribution to the upwelling microwave radiance, they are able to subtract it from the signal and assign the remaining term to cloud liquid contributions over a region of Colorado. Land surface emissivity calculations from SSM/I were also performed by Prigent et al. (1997) in the Meteosat observation area (Europe and Africa). Felde and Pickle (1995) retrieved emissivity at 91 and 150 GHz over various surfaces including snow, bare soil, and vegetation using SSM/T2 data.

4 4 Sea ice and snow emissivity have also been calculated from passive microwave observations. Comiso (1983, 1986) used SMMR brightness temperatures and infrared measurements of surface temperature to calculate emissivity for ice classification applications, though atmospheric contributions were not taken into account. Estimates of microwave emissivity for a variety of snow and ice types in the Baltic Sea were made by Hewison and English (1999) using airborne radiometer data at frequencies ranging from 24 GHz to 157 GHz. In this study, we adapt techniques described by these researchers and apply them to airborne radiometer data from the FIRE Arctic Clouds Experiment (FIRE-ACE) and Surface Heat Budget of the Arctic (SHEBA) aircraft observing period (Curry et al., 1999; Perovich et al., 1999). During May, June, and July 1998, two aircraft carrying microwave radiometers overflew the SHEBA ice camp. Using these passive microwave measurements along with ancillary aircraft data, we are able to examine the variations in emissivity prior to the onset of melting when snow cover is still present and during the summer melt season. The high resolution of the airborne sensors allows us to characterize the variability in emissivity on a much finer scale than in the previous satellitebased studies, so features such as leads can be detected. The two radiometers also provide emissivity estimates over a wide range of frequencies. Overlapping frequencies between the two sensors provide data for intercomparison. The in situ data set from the SHEBA ship is useful for relating emissivity values to observed surface conditions and for understanding differences in emissivity at varying frequencies. 2. Methodology Surface emissivity can be derived from the equation of radiative transfer for microwave frequencies. Assuming a non-scattering (cloud-free) atmosphere, the equation at a given frequency and polarization is

5 5 H T b ε s T τ( 0, H) s τ( 0, z) ( 1 ε µ s ) T( z)α( z) τ( z, H) = exp + exp z' T( z)α( z) d + exp dz' µ µ 0 where T b is brightness temperature measured by a radiometer, ε s is surface emissivity, T s is the physical temperature of the surface, τ is atmospheric opacity, H is depth of the atmospheric layer, T(z) is the atmospheric temperature as a function of height, α(z) is the atmospheric absorption coefficient, µ is the cosine of the viewing angle, and dz =dz/µ. The first term on the right hand side represents surface emission, the second term is downwelling atmospheric emission which is reflected by the surface, and the third term is upwelling atmospheric emission. Rearranging the above relationship to solve for emissivity, we obtain H 0 T b T up at down ε s = at ( s T down ) (1) where the following notation has been introduced: a = exp τ( 0, H) µ (2) T up = H 0 τ( z, H) T( z)α( z) exp dz' µ (3) T down = H 0 τ( 0, z) T( z)α( z)exp dz' µ (4)

6 6 Emissivity is determined by applying (1) through (4) at a given frequency and polarization using observations obtained from aircraft and surface based measurements. Each of these data sources is described in the following section, along with assumptions made in applying them for emissivity calculations. 3. Data Aircraft measurements in the vicinity of the SHEBA ice camp were conducted in the spring and summer of Of particular interest for this work are flights by the National Center for Atmospheric Research (NCAR) C-130 aircraft and the NASA ER-2 aircraft, since both carried microwave radiometers. The NCAR C-130 performed 8 research flights in May and 8 in July of that year. The NASA ER-2 flew 11 missions from mid- May through early June. Review of the meteorological observations on each flight day revealed cloud-free conditions over the ice camp on May 20, May 24, and July 8. Both aircraft were operating in the region on May 20; the NCAR C-130 also flew on May 24 and July Sensor Descriptions The Airborne Imaging Microwave Radiometer (AIMR) is a cross-track scanning system which flies on the NCAR C-130. Four channels measure upwelling radiation at two frequencies, 37 and 90 GHz, and two orthogonal polarizations which can be converted to horizontal and vertical components. The AIMR views underlying scenes over an angular swath of 120. Beam widths of 1 at 90 GHz and 2.8 at 37 GHz produce spatial resolutions on the order of 20 m to 300 m at typical flight altitudes and velocities. Corresponding swath widths are approximately 3 to 20 km. Detailed specifications for the AIMR can be found in Collins et al. (1996). Calibration of the AIMR is performed internally as the scanning mirror views two calibration loads after viewing the underlying scene. One load is heated while the other floats at ambient tem-

7 7 perature, so two reference points are obtained relating voltage to radiance. Scene measurements are converted to radiance by interpolating between the reference points. Errors resulting from calibration uncertainties and polarization conversion are discussed by Collins et al. (1996). Brightness temperature errors are estimated as less than 1 K for angles greater than 20 ; at smaller angles the errors may increase to 1.5 K or more. It should be noted that these error estimates assume scene brightness temperatures fall between the hot and cold load temperatures which was not always the case in the FIRE-SHEBA experiment. The uncertainties in this situation have not been quantified. The Millimeter Imaging Radiometer (MIR) was flown on the NASA ER-2 during FIRE- SHEBA. The MIR is a cross-track scanner measuring microwave radiation at several frequencies, including 89, 150, and 220 GHz. It scans over an angular swath of 100 and has a 3.5 beam width. Resulting swath width at the typical 20 km flight altitude of the ER-2 is approximately 68 km and spatial resolution is 1.2 km. The MIR calibration system is similar to that of AIMR with two internal loads viewed during each scan. Racette et al. (1996) give error estimates of better than 1 K for brightness temperatures between 240 and 300 K. Post-flight calibration efforts have demonstrated errors on the order of 2 to 4 K for brightness temperatures below 100 K. A set of Heimann Model KT19.85 pyrometers was flown on the NCAR C-130 for the purpose of remotely sensing surface (skin) temperature. These sensors measure radiation within the spectral range 9.6 to 11.5 µm over a 2 field of view, and relate that measurement to surface temperature. The accuracy of the measurement is affected by absorption and emission of infrared radiation from atmospheric water vapor in the layer between the surface and the aircraft. Errors in surface temperature are also introduced by assuming the surface is a blackbody. In reality, the

8 8 spectral emissivity of ice and snow is smaller than unity, resulting in reflection of downwelling infrared radiation. Corrections for these effects are discussed in Section Data Analysis Equations (1) through (4) require input from several sources in order to estimate ε s at a given frequency. Brightness temperatures, T b, at 37 and 90 GHz are obtained from the AIMR. The MIR provides T b at 89, 150, and 220 GHz. Straight and level flight segments over the SHEBA ice camp provide two-dimensional images of T b for analysis. Flight patterns for the C-130 tended to focus within 50 km of the ship. The ER-2 covered a wider region, but also overflew the SHEBA site. Figure 1 shows the portion of the C-130 and ER-2 flight tracks used for this study on May 20, On each C-130 flight, a series of 50 km segments oriented east-west and centered on the SHEBA ship are used for emissivity calculations. The ER-2 flight on May 20 included one pass over the SHEBA site, oriented approximately north-south, passing over the C-130 flight pattern. Vertical profiles of temperature and humidity are needed to calculate the atmospheric upwelling and downwelling terms, T up and T down. Soundings were routinely made by the C-130 at the completion of its flight pattern over the ice camp from the surface to about 6 km. Since the C-130 profiles do not extend up to the altitude of the ER-2 (20 km), a radiosonde observation from the SHEBA camp was used for that case. These profiles also serve as input to the model used for calculation of atmospheric absorption coefficient, α, at microwave frequencies (Liu,1998). A measure of the physical temperature of the surface emitting layer is required for the emissivity calculation. Measurements from the KT19 infrared thermometer provide an estimate of skin temperature, however, radiation at microwave frequencies emanates from a layer of finite depth depending on frequency. Table 1 gives penetration depths (calculated following Ulaby, 1981 and Liu and Curry, this issue) for snow and ice at frequencies measured by AIMR and MIR. Given

9 9 that the emitting layers are generally several cm deep, we need an effective temperature for the layer rather than just a skin temperature. Since this measurement is not available over the horizontal scale of the passive microwave measurements, we improvise by trying to relate skin temperature measurements to emitting layer temperature. Perovich et al. (1997) logged an annual cycle of sea ice temperature profiles in a multiyear ice floe in the Beaufort Sea. Their results show a nearly isothermal profile of sea ice temperatures to well below microwave penetration depths for the late May and early July periods of interest for this study. Similar profiles near the SHEBA site were measured in 1998 (Perovich et al., 1999), so it seems reasonable to assume that the skin temperature is a good approximation for the temperature of the emitting layer at this time of year. The assumption is probably less valid when snow covers the sea ice, since the temperature profile through the snow layer cannot be assumed isothermal. A snow layer of about 35 cm was present in late May, 1998 (Perovich et al., 1999), but the differences between the atmospheric temperature and the ice temperature were not large, so the temperature gradient through the snow layer should be relatively small. Data from the SHEBA site (Perovich et al., 1999) give temperature profiles through the snow layer showing a slight increase between the snow-ice interface and the snowatmosphere interface in late May. On the dates of interest here, the temperature difference appears to be about 1 K. Thus, the error involved in using skin temperature to represent emitting layer temperature should be 1 K or less. Errors in skin temperature measurements due to emission and absorption of infrared radiation by the atmosphere can be estimated and removed from the data prior to our emissivity calculations. The upwelling radiance observed by a KT19 sensor can be described by the radiative transfer equation for a non-scattering atmosphere: L m = εl sfc + L atm +( 1 ε)l ref

10 10 where L m is the upwelling radiance observed by the sensor, ε is infrared emissivity of the surface, L sfc is radiance emitted by the surface, L atm is radiance emitted by the atmosphere, and L ref is atmospheric radiance reflected by a surface with emissivity less than unity. Skin temperature is derived from the KT19 measurement by assuming that observed radiance consists only of surface emission. The data can be corrected by calculating the atmospheric and reflected radiances and including them in the surface temperature derivation. Such corrections have been previously applied to infrared temperature measurements of sea ice by Steffen and Lewis (1988) and Muller et al. (1975). Following their methods, we use a radiative transfer model (Key, 1998) to calculate correction factors for each of the days analyzed here. Temperature and humidity profiles from the aircraft were input into the model, and runs were performed over a range of hypothetical surface temperatures. The differences between top-of-atmosphere infrared brightness temperature calculated by the model and the input surface temperature represent the measurement error. Figure 2 gives results for each of the three days considered here. Temperature errors are plotted as a function of KT19 measured temperatures for typical flight altitudes. Least square fits to the model output are shown; these curves are used to look-up the appropriate correction when using surface temperature measurements in the emissivity calculations. The effect of varying infrared emissivity was also considered with these model runs. Infrared emissivities for snow and ice are discussed by Salisbury et al. (1994). Over the spectral range of the KT19, they report dry snow emissivities ranging from to For ice (distilled water), they give values ranging from 0.97 to for this spectral range. The actual emissivity depends on many factors (e.g., snow grain size, extent of melting) that cannot be characterized from the available data. Infrared surface temperature measurements were also made from towers at the SHEBA site; researchers report using an emissivity of 0.99 for those measurements (Claffey, et al.

11 ). For consistency, we assume the same, and perform model runs to estimate the errors induced by assuming an emissivity of unity versus For the atmospheric conditions on the three days considered here, the effect on surface temperature of changing the emissivity to 0.99 is on the order of 0.2 to 0.3 K. We consider this amount to be within the error of the instrument, and thus have not corrected the surface temperature data for this effect. 4. Results Surface emissivity at AIMR and MIR frequencies has been calculated over an area centered on the SHEBA site as shown in Figure 1. Clear sky days are required for these calculations; flight days satisfying this requirement were May 20, May 24, and July 8 for the C-130 and May 20 for the ER-2. In late May, the multiyear ice floe on which the ship was located was covered by a layer of snow about 35 cm deep. Melting appears to have begun near the end of May, so the snow layer is assumed to consist of dry snow on May 20 and May 24. Numerous leads were present in the vicinity of the ship, but they were largely frozen with almost no open water observed from the aircraft. Conditions had changed drastically by early July when the second series of flights began. No snow was present on the ice during the July 8 flight. There were numerous open leads, and meltponds had formed on the ice. 4.1 Spatial and Temporal Variation Time series of 37 and 90 GHz emissivities along the C-130 flight track on May 20 are plotted in Figure 3. Each of the discrete time segments represents a distance of about 50 km. Spatial resolution of the AIMR at this altitude (4000 m) is about 70 m for the 90 GHz channels and 200 m for 37 GHz, so leads of at least those widths are resolved in these measurements. The inhomogeneity of the surface is apparent in the range of ε s values that occur over small distances. At 37 GHz, ε s ranges from 0.8 to 0.97, while at 90 GHz the values span a larger range from 0.65 to Most of

12 12 the variation at 90 GHz is attributable to leads; the standard deviation of ε s exclusive of leads is on the order of The difference in ε s at the two frequencies may be explained by considering the observed snow layer covering the ice together with microwave penetration depths. It can be seen from Table 1 that the snow layer has little effect on emission at 37 GHz, since penetration depth for dry snow exceeds the measured depth of 35 cm. Thus we surmise that the emitted surface radiation at 37 GHz is coming primarily from the multiyear ice beneath the snow. At 90 GHz, the theoretical penetration depths suggest that the snow layer contributes part of the emitted radiation. Thus the two frequencies are effectively seeing different dielectric materials. In addition, scattering by snow at 90 GHz causes attenuation of the emitted radiation, so ε s is further reduced. Lead signatures are apparent, especially at 90 GHz, where ε s values briefly jump to 0.8 and above as would be expected for first year ice. The leads are more easily detected at 90 GHz than at 37 GHz for two reasons. First, the spatial resolution is higher at 90 GHz, so smaller leads are not resolved in the 37 GHz measurements. Large leads are apparent at 37 GHz, though the difference between ε s for leads versus that for multiyear ice is smaller than at 90 GHz. The reason for the large difference at 90 GHz may be that the snow layer on leads is thinner than on multiyear ice. If the snow layer on a lead were less than 22 cm, then that channel would be receiving radiation emitted from the first year ice. No observations of lead snow cover are available, but it seems plausible that there would be less snow on leads due to heating from below the thinner layer of ice, and perhaps due to wind blowing snow off the smooth ice. By July the ice surface has changed dramatically since melting has occurred and the snow cover is gone. A similar flight pattern consisting of 50 km segments centered on the ship was flown by the C-130 on July 8. Times series plots of ε s are shown in Figure 4. In contrast to the measurements in May, ε s values at the two frequencies are now quite similar. In the absence of

13 13 snow and at surface temperatures near 0 C, penetration depths are less than 1 cm for both frequencies. Thus each channel is viewing the same dielectric material in this case. In general, the multiyear ice ε s values are more homogeneous than during May, ranging from 0.85 to Leads now look radiometrically cold, since they are occupied by open water. At both frequencies, the standard deviation exclusive of leads is smaller (~0.025) than in May. Figure 5 shows frequency distributions of ε s for the May 20 and July 8 cases. (Distributions on May 24 are similar to those on May 20). The 37 GHz distributions for the May dates show a double peak, the higher peak presumably consisting of pixels with leads and the other peak containing multiyear ice pixels. The separation is not distinct, though, probably because the 200 m resolution at this frequency causes many pixels to contain mixtures of leads and multiyear ice. Single peaks centered at about 0.71 characterize the 90 GHz distributions on May 20 and May 24. A smaller number of pixels are spread over the 0.8 to 0.95 range. These represent leads detected at this frequency. The July plots show the similarity of the 37 and 90 GHz distributions at this time of year. In this case, the lead pixels are those with ε s values below 0.8. Frequency distributions of ε s for MIR channels on May 20 are shown in Figure 6. The range of values observed is smaller than that seen by AIMR; this is probably a result of the larger footprint of MIR (1.2 km) and its inability to resolve individual leads. Mean values of ε s are 0.78, 0.77, and 0.89 for the 89 GHz, 150 GHz, and 220 GHz channels, respectively. Referring to the penetration depths in Table 1, we infer that a portion of the emission at these frequencies is from the snow layer, though some emission is also coming from the sea ice. Emission by dry snow increases with frequency. Scattering by snow also increases with frequency. Hence the component emitted by the sea ice will be more strongly attenuated due to enhanced scattering by the snow layer at higher frequencies. The net effect is that emission decreases slightly between 89 and 150

14 14 GHz, then increases again at 220 GHz where the higher emission from snow exceeds attenuation of the sea ice signal by scattering. The MIR and AIMR share a nearly common frequency (89 and 90 GHz, respectively), so it is useful to compare the emissivity calculations from both sensors. Because the aircraft did not fly exactly coincident flight paths and were separated in time by a few hours, we have not attempted to make a pixel-by-pixel comparison. We can compare the results in a statistical sense though, since both aircraft flew in the vicinity of the SHEBA site on May 20. The mean ε s in the 50 km 2 box surrounding the ship is 0.73, as calculated from AIMR measurements at 90 GHz. Along the 150 km linear track where the ER-2 passed over the ship, the mean ε s from MIR at 89 GHz is The T b error ranges given in Section 3 translate into ε s error bars of ±0.005 for MIR and ±0.01 for AIMR, so errors in the T b measurements do not completely account for the difference. It is possible that T b errors for one or both sensors are larger than is thought, especially since the brightness temperatures at this frequency tend to fall below the range of the calibration load temperatures for both sensors. (Calibration errors are likely to be larger for the AIMR since brightness temperatures are lower compared to MIR, and because the lower end of the AIMR calibration temperature range is higher due to warmer ambient temperatures at lower flight altitudes). Another source of error that may contribute to the bias is the differing method by which surface temperature is specified for each data set. Whereas each AIMR data sample was matched individually with a surface temperature at its location, an average surface temperature as derived from KT19 measurements was applied to the entire MIR segment. The average was representative of multiyear ice temperatures in the area covered by the C-130, and excluded the higher temperatures of leads. Underestimating the surface temperature would tend to cause an overestimation of ε s according to (1). Finally, the calculation of the atmospheric contribution probably contains

15 15 some uncertainties, though it is unclear how large or of what sign the errors would be. Given that the atmospheric integration is performed over a 20 km layer for the MIR calculations versus a 4 km layer for AIMR, the errors introduced in each result could be considerably different. 4.2 Polarization and Viewing Angle Variations In calculating emissivity for the two-dimensional images produced by the scanning radiometer, surface temperatures were not available for each pixel since KT19 is a nadir-viewing sensor. In the case of the AIMR, it was assumed that the nadir surface temperature at a given point along the flight track represented the surface temperature for each pixel in the cross-track scan, provided the pixel contained multiyear ice. Statistics compiled from the C-130 flight tracks in the SHEBA region showed little variation in surface temperature (standard deviations on the order of K) on a given day at the resolution of the KT19, so this assumption seems reasonable. Lead surface temperature statistics were also compiled. Again, small variations were observed over the region, so a constant lead temperature was assumed for each day analyzed. For the MIR data set, coincident surface temperature measurements were not available. In addition, the resolution of the MIR does not allow for differentiation of most leads. Thus, an average surface temperature derived from the KT19 measurements in the SHEBA region on a given day was applied to the entire MIR image. AIMR images of emissivity on May 20, 1998 are shown in Figure 7. The flight track is along the vertical axis of these images which correspond to the east-west segment shown passing over the ship in Figure 1. As noted when comparing the nadir traces, ε s at 37 GHz and ε s at 90 GHz are substantially different, presumably due to the snow cover and differing penetration depths. Leads are easily detected in the 90 GHz image as high emissivity features. Many of the same leads are also apparent in the 37 GHz image, but substantial variation in ε s outside of leads is also noted.

16 16 Brightening along the edges of the ε s images is also apparent in the T b signal and is possibly an artifact of the sensor, as noted by Collins et al. (1996). MIR images (not shown) cover a larger area at lower resolution; surface features such as the leads seen in Figure 7 are not resolved. Estimates of ε s given in the literature are generally at 50 viewing angles to correspond with satellite microwave radiometers. AIMR estimates at these angles have been averaged for comparison with previous estimates. Carsey (1992) compiled emissivity values from satellite and groundbased sensors at 37 and 90 GHz for various ice types including dry multiyear ice and summer melting conditions. Comiso (1983) gives SMMR estimates of emissivity for multiyear ice at various Arctic sites over an annual cycle. Results for multiyear ice conditions in late May are given in Table 2. AIMR estimates at 37 GHz are significantly higher than the averages cited by Carsey (1992), but are close to the upper end of the range determined by Comiso (1983) for late May. We note that Comiso did not incorporate atmospheric effects in his estimates; removal of the atmospheric contribution to brightness temperature tended to raise emissivity estimates for the conditions observed in FIRE-SHEBA. Thus, AIMR emissivity estimates using Comiso s technique would be smaller. For example, at typical brightness temperatures observed on May 20, ε s would change from 0.92 with the atmosphere included to 0.87 without the atmosphere. Also, the AIMR manufacturer notes possible errors along the edge of the scan due to energy from the hot calibration load entering the antenna sidelobes, particularly at 37 GHz (Collins et al., 1996). This excess energy is most apparent at scan angles between 55 and 60 where it has been observed to increase T b by about 5-10 K, but there may be some smaller influence at 50 which would artificially raise T b and thereby increase ε s. The effect is not consistently observed, though, and more study is required before this source or error can be quantified. Table 3 shows the comparisons for July 8. In

17 17 this case, AIMR estimates are close to the averages compiled by Carsey at both 37 GHz and 90 GHz, and are again slightly above the range determined by Comiso for July. 5. Conclusions An algorithm for calculating microwave sea ice emissivity from airborne radiometer data has been developed. The method accounts for emission by the surface, atmospheric upwelling, and reflection of downwelling atmospheric radiation. Infrared surface temperatures and vertical profiles of atmospheric temperature and humidity are used as input. Emissivities at a range of frequencies have been calculated around the SHEBA site for clear sky days during the FIRE-SHEBA aircraft campaign in the spring and summer of While the available data set is not large enough to draw conclusions about day-to-day variations in emissivity, the limited cases analyzed point out some of the important considerations for including surface effects in cloud retrievals from passive microwave data: 1) The presence of snow on the ice has a large impact on the microwave emissivity, especially at frequencies above 37 GHz. Specifically, the snow depth in relation to penetration depth at a given frequency determines whether the emissivity will be characteristic of sea ice alone, or of a combination of sea ice and snow. The relationship between emissivity at different frequencies is strongly dependent on the nature of the surface. Spectral differences in emissivity could theoretically be used as an indicator of snow cover and possibly snow depth. 2) Comparisons of AIMR estimates with previous emissivity measurements from satellite and groundbased sensors give good results at both frequencies during the summer melt conditions. Estimates are higher than those cited in the literature for multiyear ice at 37 GHz in May, but are closer for 90 GHz in May. Given the uncertainties in each data set, the relatively good comparisons give us some confidence in the measurements produced by the airborne sensors. Better char-

18 18 acterization of the errors inherent in AIMR measurements, particularly at the edges of a scan, are recommended. 3) Comparison of 89 and 90 GHz emissivities on the one day that both AIMR and MIR viewed the SHEBA area reveals a difference of about 7% in the mean values. Given the many sources of uncertainty, this sort of bias is not unexpected. Comparison of just one case does not allow us to draw any conclusions regarding the accuracy of either sensor. A more enlightening comparison might be made by georeferencing both data sets, degrading the resolution of the AIMR data, and performing a pixel-by-pixel comparison of the AIMR and MIR estimates. Though there was only one clear sky day on which both aircraft flew in the SHEBA region, the FIRE-SHEBA data set should be examined for other colocated flight segments outside the experimental region. 4) The range of ε s values in summertime is less than in spring, and the horizontal variability (exclusive of leads) is smaller. Thus the summer season may be an easier time for estimating the surface contribution in cloud retrieval schemes, because it may be possible to assume an average value of ε s. Frequent cloud cover in the summer also make it a good time to attempt cloud retrievals. In both May and July, the standard deviation of emissivity values in the multiyear ice is fairly small (at 90 GHz) over the horizontal scales examined here. Thus, in cases where leads can be detected through cloud cover, it might be feasible to assume a constant value of ε s for multiyear ice pixels in cloud retrieval algorithms. This possibility should be explored with a larger data set, and resulting errors in cloud retrievals should be quantified. 5) The spectral variation of ε s is dependent upon surface properties, so using ε s at low frequencies (which are more transparent to the atmosphere) to estimate ε s at high frequencies where cloud retrievals could be performed would require independent information about the surface. Other sensors, such as Synthetic Aperture Radar (SAR), might be useful for this purpose.

19 19 Acknowledgments: This research was supported by the NSF SHEBA program, the NASA FIRE program, and a NASA Earth System Science Fellowship. We would like to thank J. Wang for making the MIR data available to us and C. Walther of the NCAR Remote Sensing Facility for assistance with the AIMR data. Discussions with G. Liu and J. Maslanik were very helpful in conducting this research.

20 20 REFERENCES Carsey, F.D. (Ed), Microwave remote sensing of sea ice, 462 pp.,geophysical Monograph 68, AGU, Washington, D.C., Claffey, K.J., E.L. Andreas, D.K. Perovich, C.W. Fairall, P.S. Guest and P. Ola G. Persson, Surface temperature measurements at SHEBA, Fifth Conference on Polar Meteorology and Oceanography, January, Dallas, TX, Collins, M.J., F.G.R. Warren and J.L. Paul, Airborne imaging microwave radiometer -- Part I: Radiometric analysis, IEEE Trans. Geosci. Rem. Sens., 34(3), , Comiso, J.C., Sea ice effective microwave emissivities from satellite passive microwave and infrared observations, J. Geophys. Res., 88, , Comiso, J.C., Characteristics of Arctic winter sea ice from satellite multispectral microwave observations, J. Geophys. Res., 91, , Curry, J.A., P. Hobbs, M. King, D. Randall, P. Minnis et al.., FIRE Arctic clouds experiment, Bull. Amer. Met. Soc., in press, Felde, G.W. and J.D. Pickle, Retrieval of 91 and 150 GHz Earth surface emissivities, J. Geophys. Res., 100, 20,855-20,866, Grenfell, T.C., Surface-based passive microwave studies of multiyear sea ice, J. Geophys. Res., 97, , Hewison, T.J. and S.J. English, Airborne retrievals of snow and ice surface emissivity at millimeter wavelengths, IEEE Trans. Geosci. Rem. Sens., 37(4), , Janssen, M.A., (Ed.), Atmospheric remote sensing by microwave radiometry, 572 pp., John Wiley and Sons, Inc., New York, 1993.

21 21 Jones, A.S. and T.H. Vonder Haar, Retrieval of microwave surface emittance over land using coincident microwave and infrared satellite measurements, J. Geophys. Res., 102, 13,609-13,626, Key, J. and A.J. Schweiger, Tools for atmospheric radiative transfer: Streamer and FluxNet, Comp. Geosci., 24(5), , Liu, G., A fast and accurate model for microwave radiance calculations, J. Met. Soc. of Japan, 76(2), , Liu, G. and J.A. Curry, Observation and interpretation of microwave "hot spots" over the Arctic Ocean during winter, J. Geophys. Res., submitted Muller, F. H. Blatter and G. Kappenberger, Temperature measurement of ice and water surfaces in the north water area using an airborne radiation thermometer, J. Glaciol., 15(73), , Perovich, D.K., B.C. Elder and J.A. Richter-Menge, Observations of the annual cycle of sea ice temperature and mass balance, Geophys. Res. Lett., 24(5), , Perovich, D.K., T.C. Grenfell, B. Light, J.A. Richter-Menge, M. Sturm, W.B. Tucker III, H. Eichen, G.A. Maykut, B. Elder, Snow and Ice Studies CD-ROM, Perovich, D.K., E.L. Andreas, J.A. Curry et al., 1999: Year on ice gives climate insights. EOS, 80, 481. Prigent, C., W.B. Rossow and E. Matthews, Microwave land surface emissivities estimated from SSM/I observations, J. Geophys. Res., 102, 21,867-21,890, Racette, P., R.F. Adler, J.R. Wang, A.J. Gasiewski, D.M. Jakson and D.S. Zacharias, An airborne millimeter-wave imaging radiometer for cloud, precipitation, and atmospheric water vapor studies, J. Atmos. Ocean. Tech., 13, , 1996.

22 22 Salisbury, J.W., D.M. D Aria and A. Wald, Measurements of thermal infrared spectral reflectance of frost, snow, and ice, J. Geophys. Res., 99, 24,235-24,240, Steffen, K. and J.E. Lewis, Technical note: Surface temperatures and sea ice typing for northern Baffin Bay, Int. J. Rem. Sens., 9(3), , Ulaby, F.T., R.K. Moone and A.K. Fury, (Eds)., Microwave remote sensing active and passive, Vol. 1: Microwave remote sensing fundamentals and radiometry, 456 pp., Artech House, Inc., Norwood, NJ, 1981.

23 23 FIGURE CAPTIONS: Figure 1: Flight tracks of the NASA ER-2 and NCAR C-130 aircraft in the vicinity of the SHEBA ship on May 20, Figure 2: Surface temperature corrections as a function of flight altitude and surface temperature as measured by the KT19 radiometer: (a) corrections for May 20, 1998; (b) May 24, 1998; (c) July 8, Figure 3: Sea ice emissivity at nadir for 37 and 90 GHz as derived from AIMR measurements on May 20, The five discrete segments correspond to the east-west flight segments of the C- 130 shown in Figure 1. Figure 4: As in Figure 3, but for July 8, 1998 Figure 5: Frequency distributions of sea ice emissivity at nadir along a C-130 flight segment: (a) 37 GHz emissivity on May 20,1998; (b) 90 GHz emissivity on May 20,1998; (c) 37 GHz emissivity on July 8, 1998; (d) 90 GHz emissivity on July 8, Figure 6: Frequency distributions of sea ice emissivity at nadir along the ER-2 flight track on May 20, 1998; (a) 89 GHz; (b) 150 GHz; (c) 220 GHz. Figure 7: Images of sea ice emissivity from AIMR at 37 GHz and 90 GHz on May 20, The flight track is along the vertical axis of each image and passes over the SHEBA ship. Approximate dimensions of the images are 17 km by 50 km.

24 24 Table 1: Microwave penetration depths in cm at -10 C 37 GHz GHz 150 GHz 220 GHz Dry Snow Multiyear Ice First Year Ice Table 2: Emissivity at 50 for multiyear ice (MYI) 37 GHz (H) 37 GHz (V) 90 GHz (H) 90 GHz (V) AIMR (late May) Carsey (dry MYI) (0.079) a (0.115) (0.011) (0.105) Comiso (MYI, late May) a. Standard deviations given in parentheses Table 3: Emissivity at 50 for summer melt conditions 37 GHz (H) 37 GHz (V) 90 GHz (H) 90 GHz (V) AIMR (early July) Carsey (summer melt) 0.93 (0.010) a 0.97 (0.015) 0.92 (0.010) 0.95 (0.020) Comiso (MYI, July) a. Standard deviations given in parentheses

25 Latitude (degrees) Longitude (degrees) Figure 1

26 26 2 KT19 corrections for May 20, 1998 (RF06) 1 DT=Ts-Tkt19 (deg K) Tkt19 (deg K) Figure 2(a)

27 27 2 KT19 corrections for May 24, 1998 (RF07) 1 DT=Ts-Tkt19 (deg K) Tkt19 (deg K) Figure 2(b)

28 28 2 KT19 corrections for July 8, 1998 (RF09) 1 DT=Ts-Tkt19 (deg K) Tkt19 (deg K) Figure 2(c)

29 Emissivity at 37 and 90 GHz (May 20, 1998) Emissivity Time (hrs) Figure 3

30 Emissivity at 37 and 90 GHz (July 8, 1998) Emissivity Time (hrs) Figure 4

31 Emissivity Distribution (May 20, 1998) GHz Number of Pixels GHz Emissivity Figure 5 (a,b)

32 Emissivity Distribution (July 8, 1998) Number of Pixels GHz 37 GHz Emissivity Figure 5(c,d)

33 Emissivity Distribution (May 20,1998) 220 GHz Number of Pixels GHz 89 GHz Emissivity Figure 6 (a,b,c)

34 Figure 7 34