ῌ ,**3 RSSJ Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 ) pp. ,+0ῌ ,,/

RReevviieeww

The AMSR-E NT, Sea Ice Concentration Algorithm : its Basis and Implementation

Thorsten MARKUS * and Donald J. C AVALIERI *

Abstract

The U.S. AMSR-E Science Team uses the enhanced NASA Team (NT, ) sea ice concentration algorithm to calculate the standard Arctic and Antarctic sea ice concentration products. The NT, algorithm significantly reduces the problem of a low ice concentration bias associated with surface e# ects apparent in sea ice retrievals from areas of deep snow using the original NASA Team (NT) algorithm. This enhancement is achieved through the use of the AMSR-E23 GHz channels. The NT , accommodates ice temperature variability through the use of radiance ratios as in the original NT algorithm and has the added advantage of providing weather-corrected sea ice concentrations through the utilization of a forward atmospheric radiative transfer model. This paper gives a brief summary of the concept of the algorithm and provides details of its implementation as part of the routine U.S. AMSR-E Science Team data product generation.

Keywords : sea ice, AMSR-E, ice concentration

of radiance ratios provide and the relatively large dynamic +. Introduction range in sea ice concentration that the use of the+3 and -1 GHz channels provide. Ma¨tzler et al.-῍ have shown that the The Advanced Microwave Scanning Radiometer for EOS sensitivity to inhomogeneities of the surface layer on the (AMSR-E) developed and built by the Japan Aerospace Ex- horizontal polarization at2/ GHz is much reduced and there- ploration Agency for NASA was successfully launched on the fore have suggested the use of the2/ GHz channels for ice EOS Aqua spacecraft in May,**, . This new state-of-the-art concentration retrievals if one can handle its higher sensitivity satellite radiometer provides a wider range of frequencies and to atmospheric e#+3-1 ects compared to the GHz and GHz twice the spatial resolution than is currently available with the channels. For the AMSR-E NT, algorithm the higher sensi- Defense Meteorological Satellite Program (DMSP) Special tivity to atmospheric e#23 ects at GHz is handled through Sensor Microwave/Imager (SSM/I) series of radiometers forward calculations with a full atmospheric radiative transfer (see Kawanishi et al.+῍ for a summary of AMSR-E frequen- model.῍ . There exist other studies that investigated the use of cies, beamwidths, and mean spatial resolutions. The resolu- the2/ -GHz channels to derive ice concentrations at a higher tion of the AMSR-E is increased by roughly a factor of two spatial resolution. Svendsen et al./0῍῍ and Lubin et al. used the compared to the SSM/I. The resolution of the+3 to 23 GHz 2/ -GHz data with a simplified radiative transfer model. They channels range from,/ to / km. obtained good results when cloud contamination was small0῍ . Sea ice concentrations for the U.S. Aqua AMSR-E sea ice Others used the2/ -GHz data in successive combination with product suite are operationally calculated using an enhanced the low frequency channels to retrieve high resolution ice version of the NASA Team (NT) algorithm,12῍῍ , henceforth concentration) , or a combination of2/ GHz data with referred to as NT, algorithm. This algorithm was originally model results from numerical weather predictions3῍ . developed for the Special Sensor Microwave Imager (SSM/I) This paper provides an overview of the theoretical basis of on board the U.S. Defense Meteorological Satellite Program the NT, algorithm and then gives some details on the im- (DMSP) satellites. The enhancement to the NT algorithm is plementation of the algorithm by the U.S. AMSR-E Science achieved mainly through the incorporation of the23 GHz Team. Included in this overview is a description of the channels (2/ GHz for the SSM/I), while retaining both the reduction of atmospheric e# ects, the removal of erroneous sea relative insensitivity to ice temperature variations that the use ice concentrations near the coast that result from the large

(Received July++ , ,**2 . Accepted November 3 , ,**2 ) * Cryospheric Sciences Branch NASA Goddard Space Flight Center, Greenbelt, MD,*11+ , USA

ῌῌ216 Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 )

AMSR-E antenna pattern and a summary of recent validation concentration. In the following discussion, we will refer to results. The algorithm look-up tables needed for implementa- these e## ects as surface e ects. In Figure + a, pixels with tion of the algorithm are given in the appendix. significant surface e# ects tend to cluster as a cloud of points towards higherPR values (labeled C) away from the +**ῌ ,. Heritage ice concentration line A-B resulting in an underestimate of ice concentration by the NT algorithm. The use of horizontally The two ratios of brightness temperatures used in the polarized channels makes it imperative to resolve a third ice original NT algorithm as well as in the NT,$# approach are the type to overcome the di culty of surface e ects on the polarization emissivity of the horizontally polarized component of the brightness temperature. TB῍῎ῐ῍῎nn V TB H PR῍῎ῑn ()+ TB῍῎῏῍῎nn V TB H -. Theoretical Basis and the spectral gradient ratio We make use ofGR (23 V +3 V ) and GR ( 23 H +3 H ) to re- TB῍῎ῐ῍῎nn+, p TB p GR῍῎ῑnn+, p p (), solve the ambiguity between pixels with true low ice concen- TB῍῎῏῍῎nn+, p TB p tration and pixels with significant surface e# ects. A plot of whereTB is the brightness temperature at frequencyn for the these two ratios are found to form narrow clusters except for polarized componentp (vertical (V) or horizontal (H)). areas where surface e#+3 ects result in a decrease inTB ( H ) Figure++323+3+ a shows a typical scatterplot ofPR ( ) versus GR and thus an increase in GR ( H H ) (Figure b). Values of (-1VV +3 ) for September conditions in the Weddell Sea. The high GRVV ( 23 +3 ) and high GRHH ( 23 +3 ) are indicative of NT algorithm identifies two ice types which are associated open water. The range ofGR (23 H +3 H ) values is larger with first-year and multiyear ice in the Arctic and ice types A because of the greater dynamic range between ice and water and B in the Antarctic (as shown in Figure+ a). The A-B line for the horizontally polarized components. With increasing represents+**ῌ ice concentration. The distance from the ice concentration, the two ratios have more similar values. open water point (OW) to line A-B is a measure of the ice The narrow cluster of pixels adjacent to the diagonal shown in concentration. In this algorithm, the primary source of error Figure+ b represents +**ῌ ice concentration with di # erent is attributed to conditions in the surface layer such as surfaceGR values corresponding to di# erent ice types. When surface glaze and layering+*῎ , which can significantly a# ect the hori- e# ects come into play, points deviate from this narrow cluster zontally polarized+3 GHz brightness temperature-῎ leading to towards increasedGR ( 23 H +3 H ) values (cloud of points to increasedPR (+3 ) values and thus an underestimate of ice the right of the diagonal) whileGR (23 V +3 V ) changes little or

Fig.+ a)GR ( -1 V +3 V ) versus PR ( +3 ) for the Weddell Sea on September +/ , ,**2 . The gray symbols represent the tiepoints for the ice types A and B as well as for open water as used by the NT algorithm. Label C indicates pixels with significant surface e# ects.f is the angle between the y-axis and the A-B line. b) GR (2/VV +3 ) versus GRHH ( 2/ +3 ). The ice types A and B are close to the diagonal (dahsed line). The amount of layering corresponds to the horizontal deviation from this line towards label C. Adapted from Markus and Cavalieri,῎ .

ῌῌ217 The AMSR-E NT, Sea Ice Concentration Algorithm : its Basis and Implementation

Fig.,῏῏GR over PR ( +3 )RR (a) and GR over PR ( 23 ) (b) for September +/ , ,**2 . They gray symbols represent the NT, tiepoints.

remains constant. This cloud of points labeled C in Figure + atmospheres for the three surface types (A, C, and OW). They b also corresponds to the cluster of points labeled C in Figure also illustrate that the e# ect of weather is well modeled. For +#a. Therefore. the di erence between these two GRs (῏GR ) example, the cluster of open water is mainly a result of chang- is used as a measure of the magnitude of surface e# ects.ing atmospheric conditions. The modeled atmospheres ad- Based on this analysis we introduce a new ice type C which equately span the lengths of the OW clusters. A comparison of represents ice having significant surface e# ects. For com- Figures,, a and b also shows how much more the 23 GHz data putational reasons we rotate the axes in PR-GR space (Figure are a#+3 ected by the atmosphere compared to the GHz data. +a) by an anglef so the A-B line is vertical. This makes the We then calculate brightness temperatures for all possible rotatedPRSR ( PR (+3 ) and PR R ( 23 )) independent of ice types ice concentration combinations in +ῌ increments and for

A and B (first-year and multiyear ice for the Arctic) and each of those solutions calculate the ratiosPRRR (+3 ), PR e#23 ectively eliminates one ice type in the ice concentration ( ), and῏GR . This creates a prism in which each element calculations. In the following ice type A will be the combina- contains a vector with the three ratios. For each AMSR-E tion of ice types A and B (or first-year and multiyear for the pixelPRRR (+3 ), PR ( 23 ), and῏ GR are calculated from the Arctic). The addition of the23 GHz data in the algorithm observed brightness temperatures. Next, we move through requires a correction for atmospheric e# ects. This is accom- this prism comparing the observed three ratios with the plished through an additional AMSR-E variable, the polariza- modeled ones. The indices where the di# erences are smallest tion at23 GHz (PR ( 23 )). determines the final ice concentration combination and weath- The response of the brightness temperatures to di# erent er index. The next section will provide detailed information weather conditions is quantified using an atmospheric radia- about the implementation. tive transfer model.῍ . Input data into the model are the Because of the unique signature of new ice in the micro- emissivities of first-year sea ice under winter conditions taken wave range we solve for new ice instead of ice type C for from Eppler et al.++῍ with modifications to achieve agreement selected pixels. Using aGR (-1 V +3 V ) threshold of῎ * . *, we between modeled and observed ratios. Atmospheric profiles, either resolve ice type C (for pixels whereGR (-1 V +3 V ) is used as another independent variable in the algorithm, having below this threshold) or new ice (for pixels whereGR (-1 V di#+3 erent cloud properties, specifically cloud base, cloud top,V ) is above this threshold). Areas of ice type C and new ice cloud liquid water are taken from Fraser et al.+,῍ and average are mutually exclusive because new ice has little, if any, snow atmospheric temperatures and humidity profiles for summer cover. A limitation, of course, is that we cannot resolve mix- and winter conditions are taken from Antarctic research tures of new ice and thicker ice with layering in its snow cover. stations. These atmospheric profiles are based on climatology and are assumed valid for both hemispheres. .. Implementation

The plots of῏῏GR versus PRR (+3 ) (Figure , a) and GR versusPRR (23 ) (Figure , b) illustrate the algorithm domain. .+. Calculation of ice concentrations The gray symbols indicate the tiepoints with the di# erent In contrast to other operational sea ice concentration algo-

ῌῌ218 Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 ) rithms using daily averaged brightness temperatures as input, whereas for pixels whereGR (-1 V +3 V )῔ῑ * . *, we resolve the AMSR-E NT, concentrations are calculated from indi- for new ice using the standardGR (-1 V +3 V ) as suggested by vidual swath (Level, ) data from which daily maps are made Cavalieri+-῏ , i.e. by averaging these swath ice concentrations. Using swath ῎ῌῌ ῏ΐ brightness temperatures is particularly critical for the NT, LUTDGR C A C C W x algorithm and its atmospheric correction. The atmospheric TBCCWῌῌ῎῏ῑ-1 V TB CCW ῌῌ ῎῏ +3 V AC x AC x ()1 influence on the brightness temperatures is non linear and by ῌῌ῎῏ῐ-1 ῌῌ ῎῏ +3 TBCCWAC x V TB CCW AC x V using average brightness temperatures we would dilute the atmospheric signal. Each of these arrays has the dimensions of +*+ῒῒ +*+ +,

The ice concentration algorithm is implemented as follows : where, of course, the total ice concentration (CCACῐ ) cannot +. Generate look-up tables : For each AMSR channel exceed +** . with frequencyn and polarizationpi calculate brightness tem-- . For each pixel we have the actual measured AMSR-E perature for each ice concentration-weather combination brightness temperatures (TBi (n p ))

(usingTBow , TB A , TB C as given in the appendix) :. . Calculate same ratios from these brightness tempera-

tures as in step,+3 (PRi ( ), etc.). TBCCWῌῌ ῎nn p ῏ΐ῎+ ῑ C A ῑ C C ῏ῒ TB ow ῎ pW x ῏ AC x /. Compare these observed ratios with each of the ratios ῐῒ῎῏ῐῒ῎῏CAA TBnn pW xCC C TB pW x ()- in the look-up tables looping through all ice concentration- whereCA refers to the combined concentration of ice types A weather combinations, i.e. and B for the Antarctic and first-year andmultiyear ice for the , dΐ῎PRiPRACxi ῎+3 ῏ῑ LUT+3 ῎ C ῌ C ῌ W ῏῏ ῐ῎ PR ῎ 23 ῏ Arctic (because of the axis rotation),CC to ice type C concen- , ῑ῎ῌῌ῏῏LUTPR23 C A C C W x tration, andWx to the weather index. As mentioned above, , ῐ῎῕GRi ῑ LUTD GRACx ῎ C ῌ C ῌ W ῏῏ ()2 for pixels whereGR (-1 V +3 V )῔ῑ * . *, CCC and TB will represent new ice. Ice concentrations are between* and +** 0 . The indices ca, cc, wx whered is minimal determine the in+++,῍ increments, weather indices are between and ice concentration (and weather index), i.e. corresponding to the tables in the Appendix. CCTΐῐ Amindd C Cmin ()3 ,. From these TBs calculate ratios creating the look-up tables, e.g.LUTPR+3 ( ca, cc, w x ) etc. : .,. Land spillover correction Although a land mask is applied to the ice concentration LUT +3῎῏ΐῌῌ PR CCWCC x maps, land spillover still leads to erroneous ice concentrations ῌῌ῎῏ῑ-1 ῌῌ ῎῏+3 TBCCWAC x V TB CCW AC x V along the coast lines. This makes operational usage of these sinf+3 ῌῌ῎῏ῐ-1 ῌῌ ῎῏ +3 TBCCWAC x V TB CCW AC x V maps cumbersome. Therefore, we apply a land spillover cor- rection scheme on the maps. The di$ culty is to delete all ῌῌ῎῏ῑ+3 ῌῌ ῎῏+3 TBCCWAC x V TB CCW AC x H ῐ cosf+3 clearly erroneous ice concentration while at the same time TBῌῌ῎῏ῐ+3 V TB ῌῌ ῎῏ +3 H CCWAC x CCW AC x . () preserving actual ice concentrations, as for example, along the margins of coastal polynyas. We apply a five step procedure : LUT 23῎῏ΐῌῌ PR CAC C W x +. Classify all pixels of the polar-stereographic grid with ῌῌ῎῏ῑ-1 ῌῌ ῎῏ +3 TBCCWAC x V TB CCW AC x V respect to their distance to coast (Figure- ). sinf23 ῌῌ῎῏ῐ-1 ῌῌ ῎῏ +3 TBCCWAC x VTBCCWAC x V ,+,. All pixels with classes or will be assessed for erroneous sea ice concentrations due to land spillover by ῌῌ῎῏ῑ23 ῌῌ ῎῏ 23 TBCCWAC x V TB CCW AC x H ῐ cosf23 analyzing the11 by pixel neighborhood. The area of the TBῌῌ῎῏ῐ23 V TB ῌῌ ῎῏ 23 H CCWAC x CCW AC x / () neighborhood (121/ pixels or . km) needs to be greater than IfGR (-1 V +3 V )ῌῑ῕ * . *, we resolve for ice type C using the AMSR-E antenna pattern. Pixels with values of-* and GR as our third variable, i.e., will not be changed. --1. Check whether all class pixels in pixel neighborhood LUTD ῎῏ΐῌῌ GR CAC C W x are open water (if so, set ice concentration to* ). ῌῌ῎῏ῑ23 ῌῌ ῎῏ +3 TBCCWAC x H TB CCW AC x H .1. Calculate an average sea ice concentration for the by ῌῌ῎῏ῐ23 ῌῌ ῎῏ +3 TBCCWAC x HTBCCWAC x H 1 pixel box assuming all ocean pixels have zero ice concentra- tion and all land pixels have an ice concentration of3*῍ . ῌῌ῎῏ῑ23 ῌῌ ῎῏ +3 TBCCWAC x V TB CCW AC x V ῑ ()0 This approximates a theoretical concentration caused by land ῌῌ῎῏ῐ23 ῌῌ ῎῏ +3 TBCCWAC x V TB CCW AC x V spillover only.

῍῍219 The AMSR-E NT, Sea Ice Concentration Algorithm : its Basis and Implementation

/. If the AMSR-E ice concentration is less or equal this the radiative transfer atmospheric correction is that not only value set pixel at center of box to open water. are spurious ice concentrations in the open ocean removed, Figure. shows an ice concentration map with and without but atmospheric corrections are applied to ice covered por- the land spillover correction. tions of the ocean. .-. Reduction of atmospheric e # ects Figure/ shows an example for the Sea of Okhotsk. Figure

As explained above, the NT, algorithm has an atmospheric / a shows the ice concentrations ifPRRR ( +3 ), PR ( 23 ), and ῏ correction scheme as an inherent part of the algorithm. ItGR are used but without any weather correction. Figure/ b provides weather-corrected sea ice concentrations through the shows the ice concentration with the NT, weather correction. utilization of a forward atmospheric radiative transfer model. The di#/ erences between those two concentrations (Figure d) However, to eliminate remaining severe weather e## ects over illustrate the e ect of the weather correction not only over the open ocean, two weather filters based on the spectral gradient open ocean but also over the sea ice. More severe weather ratio are implemented using threshold values similar to those e# ects over the open ocean (for example, in the bottom right used by the NT algorithm+.) +/῍ . However, the advantage of corner) are finally removed by the NT weather filter (Figure /-1c). The threshold of this weather filter (pixels whereGR ( VV+3)῎῎ * . */ or GRVV ( ,, +3 ) * . *./ are set to *ῌ ice con- centration) sets ice concentrations of less than+/ῌῌ to * therefore we also observe a slight change along the sea ice edge (Figure/ e). Even with both the atmospheric correction scheme and the GR filters, we still can have problems with residual weather contamination particularly at lower latitudes. A filter based on monthly climatological sea surface temperatures (SSTs) from the National Oceanic and Atmospheric Administration (NOAA) ocean atlas used earlier+0῍ was employed to elimi- Fig. - Example coastline classification map. Ocean pixels directly along the coast are classified with+ , nate these low-latitude spurious ice concentrations. In the pixels farther away are,- and . Open ocean pixels Northern Hemisphere, in any pixel where the monthly SST is are zero. Land pixels directly along the coast are greater than,12 K, the ice concentration is set to zero classified as. and pixels farther away have throughout the month ; whereas in the Southern Hemisphere, increasing values.

Fig.. Map of ice concentration with and without land spillover correction.

ῌῌ220 Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 )

Fig./ AMSR-E sea ice concentrations for March + , ,**1 . (a) Ice concentrations calculated usingPRRR ( +3 ), PR (23 ),῏GR without applying an atmospheric correction ; (b) ice concentration with atmospheric correction ; (c) final ice concentration with additional clean-up over the open ocean by applying the standard NASA TeamGR weather filter. (d) di### erence between (a) and (b). (e) di erence between (b) and (c). Di erences greater than+*ῌ have been truncated for the erroneous sea ice concentrations in the lower right corner.

wherever the monthly SST is greater than,1/ K, the ice ice algorithms for transects both across the Ross Sea and the concentration is set to zero throughout the month. The Sea of Okhotsk. For both transects, the NT, algorithm had higher SST threshold value in the Northern Hemisphere is the highest correlation and the smallest bias relative to the needed because the,1/ K isotherm used in the South is too AVHRR ice concentrations. close to the ice edge in the North. The closest distance the Markus and Dokken+2῍ investigated the algorithm’s per- threshold isotherms are to the ice edge is more than.** km+0῍ . formance in the Arctic during late summer through the In summary, the order of the processing is as follows : comparison with RADARSAT and ERS SAR ice concentra- +. Calculate sea ice concentrations with atmospheric cor- tions obtained using an automated SAR ice discrimination rection. algorithm. Two study regions were examined. The first was ,. ApplyGR weather filters. located within the Arctic ice pack where ice concentrations -. Apply SST mask ranged from0*ῌῌ to +** . The correlation between the NT .,*21*,. Apply land spill over correction. and SAR concentrations was . with little bias (῎ .ῌ ). The second region was located within the marginal ice zone /. Validation where the agreement was not as good. The correlation was *00. with a bias toward higher SAR concentrations. Nonethe- Several studies exist that evaluate the performance of the less, the NT, algorithm outperformed the NT algorithm NT, algorithm and, in some studies, compare its results with which significantly underestimated ice concentations relative results from other algorithms. The first comparison of NT, to SAR. retrievals with those from other algorithms is presented in the McKenna et al.+3῍ compared several algorithms including NT, algorithm paper,῍ . In this paper the authors compared the NT, the BS, and the NT, algorithms with ice charts from sea ice concentrations derived from AVHRR imagery with the U.S. National Ice Center (NIC). The NIC ice charts are those obtained from the NT, NT, , and Bootstrap (BS)+1῍ sea primarily based on high resolution visible and infrared image-

ῌῌ221 The AMSR-E NT, Sea Ice Concentration Algorithm : its Basis and Implementation ry from both civilian and defense department satellites, but ice products. The NT, algorithm incorporates both features use passive microwave satellite imagery under cloudy condi- of the NT heritage algorithm and new enhancements includ- tions. A total of+- comparative studies were completed. ing a radiative transfer model to provide an atmospheric They found that the NT, algorithm overwhelmingly out- correction to the retreivals. Several comparative studies have performed the other algorithms and provided consistently shown that the algorithm provides stable and reliable sea ice better estimates of the ice edge and the3*ῌ ice concentration concentrations for both hemispheres. The processing code boundary. also includes weather filters for eliminating spurious weather Meier,*῍ undertook a more comprehensive study comparing e# ects over open ocean, sea surface temperature masks for Arctic ice concentrations from the NT, BS, and NT,# algo- eliminating residual weather e ects particularly at low lati- rithms with AVHRR derived ice concentrations. The regions tudes, and a correction for eliminating spurious sea ice con- of study included Ba$ n Bay, the Greenland Sea, and the Kara centrations along the coast as a result of the broad AMSR-E and Barents seas. His results show that the NT, algorithm antenna pattern. has the smallest mean di# erence for all pixels during both summer and winter seasons separately, although all three Acknowledgement algorithms tend to underestimate ice concentration during summer relative to AVHRR. The authors thank Al Ivano# (Science Systems and Appli- Cavalieri et al.,+῍ undertook a comparison between AMSR-E cations, Inc.) and Je# Miller (ADNET Systems, Inc.) for all NT,1 Arctic sea ice concentrations and Landsat Enhanced their help with the development and implementation of the Thematic Mapper Plus imagery obtained during March,**- . NT, algorithm. They also thank the reviewers for their helpful Good overall agreement was obtained with little bias (+ῌ ) comments on the manuscript. for areas of first year and young sea ice. However, in areas of new ice production there was a negative bias of about/ῌ in References the AMSR-E ice concentration retrievals, with a root mean square error of2ῌ . Some areas of deep snow were also found +῍ T. Kawanishi, T. Sezai, Y. Ito, K. Imaoka, T. Takeshima, Y. to result in an underestimate of the ice concentration (+*ῌ ). Ishido, A. Shibata, M. Miura, H. Inahata and R. W. Spencer : However, for all ice types combined and for the full range of The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), NASDA’s contribu- ice concentrations, the bias ranged from*-ῌῌ to , and the tion to the EOS global energy and water cycle studies, IEEE rms errors ranged from+1ῌῌ to , depending on the region. Trans. Geoscience Rem. Sens.,.+ , pp. +2.ῌ +3. , ,**- . Since these results were reported, the new ice tie-points have ,῍ T. Markus and D. J. Cavalieri : An enhancement of the been adjusted which should reduce the low concentration NASA Team sea ice algorithm, IEEE Trans. Geoscience bias. Remote Sensing,-2 , pp. +-21ῌ +-32 , ,*** . Massom et al.,,῍ compared the NT, algorithm and the -῍ C. Ma¨tzler, R. O. Ramseier, and E. Svendsen : Polarization AMSR-E Bootstrap algorithm with aerial photography in e#3 ects in sea-ice signatures, IEEE J. Oceanic Eng., OE- , pp. ---ῌ --2 +32. East Antarctica. They found very similar ice concentrations ,. .῍ C. Kummerow : On the accuracy of the Eddington approx- for the two AMSR-E ice concentrations with small over- imation for radiative transfer in the microwave frequencies, estimation compared to the aerial photography. J. Geophys. Res.,32 , pp. ,1/1ῌ ,10/ , +33- . ,-῍ Andersen et al. compared seven passive microwave sea /῍ E. Svendsen, C. Matzler and T. C. Grenfell : A model for , ice concentration algorithms (among them the NT ) with retrieving total sea ice concentration from a spaceborne ship-based observations and SAR-based concentrations. Al- dual-polarized passive microwave instrument operating at 3* gorithms which use2/ GHz information consistently had the GHz, Int. J. Remote Sensing,2 , pp. +.13ῌ +.21 , +321 . best agreement with both SAR ice concentrations and ship 0῍ D. Lubin, C. Garrity, R. O. Ramseier, and R. H. Whritner : 2/ / observations. They furthermore found that the greater at- Total sea ice concentration retrieval from the SSM/I . GHz channels during the Arctic summer, Remote Sens. mospheric contribution in the23 GHz data is not as important Environ., vol.0, , pp. 0-ῌ 10 , +331 . as variabilities in surface emissivity. 1῍ K. M. St. Germain : A two-phase algorithm to correct for atmospheric e#2/ ects on the GHz channels of the SSM/I in 0 . Summary the Arctic region, Proc. International Geoscience and Remote Sensing Symposium, Pasadena, CA, pp.+.2ῌ +/+ , +33. . This paper provides an overview of the physical basis and 2῍ L. Kaleschke, C. Luepkes, T. Vihma, J. Haarpainter, A. implementation of the NT, sea ice concentration algorithm Bochert, J. Hartmann and G. Heygster : SSM/I sea ice remote used by the U.S. AMSR-E Science Team for its standard sea sensing for mesoscale ocean-atmosphere interaction analysis,

ῌῌ222 Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 )

Canadian Journal of Remote Sensing,,1 , pp. /,0ῌ /-1 , ,**+ . strap algorithm, NASA Reference Publication, +-2* , .3 pp., 3῍ S. Kern : A new method for medium-resolution sea ice anal- +33/ . ysis using weather-influence corrected Special Sensor Micro-+2῍ T. Markus and S. T. Dokken : Evaluation of Arctic late wave/Imager2/ GHz data, Int. J. Remote Sensing, ,/ , pp. summer passive microwave sea ice retrievals, IEEE Trans. .///ῌ ./2,, ,**. . Geoscience Remote Sensing,.* , pp. -.2ῌ -/0 , ,**, . +*῍῍J. C. Comiso, D. J. Cavalieri, C. L. Parkinson and P. +3 P. McKenna, W. N. Meier, M. L. van Woert, M. Chase and Gloersen : Passive microwave algorithms for sea ice concen- K. Dedrick : SSM/I Sea Ice Algorithm Inter-Comparison : trationῌ A comparison of two techniques, Remote Sens. Operational Case Studies from the National Ice Center, Environ.,0* , pp. -/1ῌ -2. , +331 . Proc. International Geoscience and Remote Sensing Sympo- ++῍ D.T. Eppler and +. others : Passive microwave signatures of sium, Toronto, Canada,,**, . sea ice, in Microwave Remote Sensing of Ice, Geophys.,*῍ W. Meier : SSM/I sea ice concentrations in the marginal ice Monogr. Ser., vol.02 , F. Carsey (ed.), pp. .1ῌ 1+ , AGU, zone : A comparison of four algorithms with AVHRR im- Washington, D. C.,+33, . agery, IEEE Trans. Geosci. and Rem. Sensing,.- , pp. +-,.ῌ +,῍ R. S. Fraser, N. E. Gaut, E. C. Reifenstein and H. Sievering : +--1,. ,**/ Interaction Mechanismsῌ Within the Atmosphere, in Manual ,+῍ D. J. Cavalieri, T. Markus, D. K. Hall, A. Gasiewski, M. of Remote Sensing, R. G. Reeves, A. Anson, D. Landen Klein and A. Ivano# : Assessment of EOS Aqua AMSR-E (eds.), pp.+2+ῌ ,-- , American Society of Photogrammetry, Arctic sea ice concentrations using airborne microwave and Falls Church, VA,+31/ . Landsat 1 imagery, IEEE Trans. Geoscience Rem. Sens., .. , +-῍ D. J. Cavalieri : A microwave technique for mapping thin -*/1ῌ -*03 , ,**0 . sea ice. J. Geophys. Res.,+*. , +, , /0+ῌ +, , /1, , +33. . ,,῍ R. A. Massom, A. Worby, V. Lytle, T. Markus, I. Allison, T. +.῍ P. Gloersen and D. J. Cavalieri : Reduction of weather e# ects Scambos, H. Enomoto, K. Tateyama, T. Haran, J. C. Comiso, in the calculation of sea ice concentration from microwave A. Pfa% ing, T. Tamura, A. Muto, P. Kanagaratnam, B. radiances, J. Geophys. Res.,3+ , pp. -3+-ῌ -3+3 , +320 . Giles, N. Young, G. Hyland and E. Key : ARISE (Antarctic +/῍ D. J. Cavalieri, K. St. Germain and C. T. Swift : Reduction Remote Ice Sensing Experiment) in the East,**- : valida- of Weather E# ects in the Calculation of Sea Ice Concentra- tion of satellite-derived sea-ice data products, Ann. Glaciol. tion with the DMSP SSM/I, J. Glaciology,.+ , pp. .//ῌῌ .0. , .. , pp. ,22 ,30 , ,**0 . +33/. ,-῍ S. Andersen, R. Tonboe, L. Kaleschke, G. Heygster and L. +0῍ D. J. Cavalieri, C. L. Parkinson, P. Gloersen, J. C. Comiso T. Pedersen : Intercomparison of passive microwave sea ice and H. J. Zwally : Deriving Long-Term Time Series of Sea concentration retrievals over the high-concentration Arctic Ice Cover from Satellite Passive-Microwave Multisensor sea ice, J. Geophys. Res.,++, , doi : +* . +*,3 / ,**0 JC **-/.- , Data Sets, J. Geophys. Res.,+*. , pp. +/ , 2*-ῌ +/ , 2+. , +333 . ,**1 . +1῍ J. C. Comiso : SSM/I sea ice concentrations using the boot-

ῌῌ223 The AMSR-E NT, Sea Ice Concentration Algorithm : its Basis and Implementation

Appendices : Look-up tables for AMSR-E

Table+- Open water; Arctic and Antarctic Table New ice C; Arctic and Antarctic

Table, Ice type A (first-year and multiyear for Arctic); Arctic and Antarctic Table. Ice type C; Arctic

ῌῌ224 Journal of The Remote Sensing Society of Japan Vol.,3 No. + ( ,**3 )

Table/ Ice type C; Antarctic

ῌ Thorsten Markusῌ Donald J. Cavalieri Thorsten Markus received the M.S. and Ph. Donald J. Cavalieri received the B.S. degree D. degrees in physics from the University of in physics from the City College of New Bremen, Bremen, Germany, in+33, and York, New York, in +30. , the M.A. degree +33/, respectively. He was a National Re- in physics from Queens College, New York, search Council Resident Research Associate in+301 , and the Ph.D. degree in meteorolo- with the Goddard Space Flight Center gy and oceanography from New York Uni- (GSFC) from+33/ to +330 before joining versity, New York, in +31. . From +31. to the NASAῌ University of Maryland Baltimore County (UMBC)+310 , he was a National Research Council Postdoctoral Resident Joint Center for Earth Systems Technology, UMBC, Baltimore, Research Associate with the National Oceanic and Atmospheric where he worked until,**, . He is currently the Head of the Administrationyq Environmental Data Service, Boulder, CO, Cryospheric Sciences Branch in the Hydrospheric and Biospheric where he continued his doctoral research on stratospheric- Sciences Laboratory, NASA GSFC, Greenbelt, MD. His re- ionospheric coupling. From+310 to +311 , he was a Visiting search interests include satellite remote sensing, and the utiliza- Assistant Professor with the Department of Physics and Atmos- tion of satellite data to study cryospheric, oceanic, and atmos- pheric Science, Drexel University, Philadelphia, PA, where he pheric processes. worked on stratospheric temperature retrievals from satellite E-mail : Thorsten.Markus῍ nasa.gov infrared radiometers. In the fall of+311 , he was a Sta # Scientist with Systems and Applied Sciences Corporation, Riverdale, MD, where he worked on sea ice retrieval algorithms, in preparation for the launch of Nimbus-1 SMMR. In +313 , he joined the Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt, MD, where he is currently a Senior Research Scientist with the Cryospheric Sciences Branch, Hydrospheric and Biospheric Sciences Laboratory. His current research inter- ests include polar ocean processes and microwave remote sensing of the cryosphere. E-mail : Donald.J.Cavalieri῍ nasa.gov

ῌῌ225