Gayana 68(2): 396-401, 2004

ATMOSPHERIC PARAMETERS OVER THE MARGINAL ICE ZONE RETREIVAL FROM ANSR/AMSR-E DATA

Maia L. Mitnik & Leonid M. Mitnik

V.I. Il'ichev Pacific Oceanological Institute, Far Eastern Branch, Russian Academy of Sciences, Vladivostok, Russia, e-mail: mitnik@poi.dvo.ru

ABSTRACT

A direct radiative transfer model, 478 radiosondes complimented by the cloud liquid water content profiles and the underlying surface temperatures and emissivities were used to simulate the brightness temperatures T_Bs over the Marginal Ice Zone at ADEOS-II AMSR and Aqua AMSR-E frequencies. The precipitable water V varied from 0.63 to 18.5kg/m², the total cloud liquid water content Q£0.25kg/m² and SST£1°C. The mean experimental values of emissivity and its fluctuations for several types of the first-year ice at ice temperature of 263 to 271K were used in a model. The sea surface emissivity was computed taking into account wind action. 10 values of ice concentration C uniformly distributed from 0.0 to 1.0 were selected for each radiosonde. The radiometer noises of 0.5K were added to the modeled T_Bs. Standard regression techniques were applied to retrieve V, Q and C from the simulated T_Bs with vertical and horizontal polarization at 18.7, 23.8 and 36.5GHz. The retrieval errors increase from »0.5 to »1.7kg/m² for V and from »0.02 to »0.06 kg/m² for Q with the increase of C from 0.0-0.3 to 0.7-1.0. The algorithms were used to retrieve V, Q and C in the Okhotsk Sea from the T_Bs measured by AMSR-E.

INTRODUCTION

Satellite passive microwave measurements have been used for atmospheric water vapor content V and total cloud liquid water content Q studies over the ice-free sea surface. The changes of the sea surface emissivity caused by variations of water temperature and salinity and wind action can be estimated reasonably well. Over the marginal ice zone (MIZ) and over compact ice cover, V and Q retrieval presents a difficult problem due to the larger values and higher variability of the underlying surface emissivity compare to the water surface. This variability is mainly explained by the change of sea ice concentration C (from 0.0 to 1.0), types of sea ice, the evolution of snow/sea-ice thermophysical properties including liquid water present in the system, snow pack density and snow grain metamorphism, air temperature and wind [1, 3, 5, 11]. The importance of studying the atmospheric and oceanic parameters and processes in the MIZ is due to scientific and practical requirements. In addition, air-sea interaction is particularly intense here, resulting in abrupt horizontal and vertical gradients of hydrometeorological parameters which promote formation of various mesoscale structures both in the atmosphere (convective rolls and cells) and in the ocean (ice edge waves, ice streets, bands and eddies). Besides, the introduction of new sensors require constant efforts in modeling, algorithm development, and then in validation of passive microwave data of sea ice.

The MIZ is the region from the ice edge (where ice is first encountered) to a point that is sufficiently away from the ocean boundary so as not to be affected by the presence of the open ocean. This definition allows for a marginal ice zone of variable width. A reasonable average width for the marginal ice zone is 100200km. Strong offpack winds can extend the MIZ significantly while on-pack winds can compact the MIZ to a relatively narrow width. The features of the MIZ structure in the Okhotsk Sea are clearly visible on Envisat Advanced Synthetic Aperture Radar (ASAR) image in a wide swath mode (405 km) shown in Fig. 1.

Figure 1: Envisat ASAR image acquired on 8 February 2004 at 11:55 UTC over the Okhotsk Sea showing the backscatter variations for different types of sea ice and the structure of the Marginal Ice Zone (ice bands, etc.).

The Advanced Microwave Scanning Radiometer for EOS (AMSR-E) is a Japanese sensor that was launched on the NASA Aqua satellite in May 2002. The AMSR, a similar instrument was launched on the Japan ADEOS-II satellite in December 2002. AMSR has about twice the spatial resolution of SSM/I with resolution as low as 5 km at frequency of 89.0 GHz [8]. This is a substantial improvement over SSM/I and yields improved benefits from passive microwave imagery both over the ice-free sea and ice areas.

Simulation of the AMSR brightness temperatures

Modeling of microwave measurements over the open ocean and the MIZ was carried out with a microwave radiative transfer program. The program and calculations of brightness temperatures T_B of the ocean-atmosphere system at frequency n with the vertical (V) and horizontal (H) polarizations T_B^V,H(n) were described in [7]. To model atmospheric conditions observable near and over the MIZ the radiosonde (r/s) database was built up. Only r/s issued when t_S£1°C were included in the database. Total 478 cases measurements were selected: 69 sets from research vessels and 409 sets from 6 polar coastal and island stations. Every set consists of radiosonde, meteorological data (wind speed W and direction, forms and amount of clouds) and the values of sea surface temperature (SST) t_S. In the database, the V varied from 0.63 to 18.5kg/m², Q did not exceed 0.25kg/m² and W£18.0m/s. R/s atmospheric profiles were complemented by the cloud liquid water content profiles as in [7].

For each r/s, the T_Bs(n) were computed for 10 values of sea ice concentration C = 0.0-1.0. The emissivity of the underlying surface k^V,H was determined by Eq. 1

where k_W and k_I are sea surface and sea ice emissivity, correspondingly, and q is an incidence angle.

Emissivity of the calm sea surface at Aqua AMSR-E frequencies n = 18.7, 23.8 and 36.5GHz was computed from the Fresnel formulas with data from [2]. The values of k_W^V,H(n,q,t_s ) at t_S = -0.6°C and q = 55° are given in Table 1. Increments of emissivity associated with the wind action were found on the basis of data [9, 10, 12].

Simulations of the T_B^V,H(n) were carried out for four types of first-year ice. The computed values of emissivity of the calm sea surface at t_S = 0°C (q = 55°) and the experimental values of emissivity of dark, gray and light nilas [3], thick first-year ice [12] and snow-covered smooth first-year ice approximately 150cm thick under winter conditions (no melting) [5] are given in Table 1. Ice concentrations used in calculation of the emissivity were chosen from a massif of the uniformly distributed values in the range from 0.0 (only water) to 1.0 (only ice). In addition, two values of ice temperature T_I were taken for each ice type: 263 and 268K for first-year thick ice and light nilas, 266 and 270K for gray nilas and 268 to 271K for dark nilas. As a result, the dataset used for the algorithm development comprised 32640 scenes. For each scene emissivity fluctuation was added to the mean k_I values given in Table 1. The magnitude and sign of emissivity fluctuations were taken the same for all frequencies and polarization under consideration since they are determined by the common physical factors [5]. They were chosen from a massif of randomly Gaussian distributed fluctuations with the zero mean and s = 0.01. Maximum fluctuations were limited by ± 2s.

Algorithm development

Satellite microwave radiometric data have been used for ice concentration and type retrieval more than 30 years, however, algorithms of atmospheric parameters retrieval over the sea ice were suggested only recently [4, 6]. These algorithms are based on the usage of the polarization differences or the known values of the underlying surface emissivity.

As follows from the analysis of the simulated T_Bs at n = 18.7, 23.8 and 36.5 GHz the increase in C produces the major contribution to the T_Bs. However the increments of T_Bs caused by the V and Q variations are high compare to radiometer sensitivity especially at C<0.5-0.7. At C£0.8-0.9 the T_Bs do not reach the saturation levels which are observed over the sea surface in a warm period due to high values of V and Q. The brightness temperatures calculated at various gray nilas concentration for three radiosondes with different V and Q and tabulated in Table 2 confirm the above conclusions.

The randomly Gaussian distributed radiometer noises were added to modeled T_Bs. The standard deviation of the radiometer error distribution was set equal to the noise level 0.5K for all channels. Then the T_Bs computed for 32640 scenes were divided into two identical massifs.

Standard regression techniques were applied to retrieve V, Q and C from T_B(18V), T_B(18H), T_B(23V), T_B(23V), T_B(36V) and T_B(36H) for four ice concentration subranges: DC = 0.0-0.3, 0.3-0.6, 0.5-0.8, 0.7-1.0 as well as for the whole range C = 0.0-1.0. (Ice concentration can be also found by application of the known algorithms [1]). The regression coefficients in Eqs. (2) ­ (4) were found by processing T_Bs taken from the first massif. The second massif was withheld for testing the algorithms.

The benefits of including the different channels in V, Q and C retrieval algorithms were determined by a series of regression analyses carried out for each DC at different T_B(n) combinations. The number of the channels was decreased from six to four or three and the retrieval errors were computed for each version. Analysis was carried out both for an ideal radiometer and for a real radiometer with noises.

RESULTS

The developed 6-, 5-, 4- and 3-channel algorithms were applied to the second T_Bs massif that was withheld for an independent performance analysis of the algorithms. The retrieved V, Q and C values were compared with values V_o, Q_o and C_o of the database and the coefficients of the regression equations V = a_o + a₁ V_o, Q = b_o+ b₁ Q_o and C = c_o+ c₁ S_o were found.

Water vapor

An elimination of 36.5V channel and then 18.7H channel tended to increase the retrieval error of the total atmospheric water vapor content s_V slightly. Retrieval errors of 4-channel V-algorithm increase from 0.65kg/m² (C = 0.0-0.3, 4848 scenes) to 1.17kg/m² (C=0.5-0.8, 4984 scenes) and equal to 1.24kg/m² for the whole massif (C = 0.0-1.0, 16360 scenes). The scatter plots of the retrieved V-values versus V_o values of the database for four ice types with C = 0.0-1.0 and 0.3-0.6 are shown in Figs. 2a,b. Regression equation are V = 0.67 + 0.88V_o; and V = 0.27 + 0.95V_o, root mean square error s_V=1.24kg/m², and s_V=0.90kg/m², correlation coefficient R = 0.85 and R = 0.93 for both ranges of concentrations correspondingly.

Figure 2: Scatter plots of the retrieved V values versus values Vo of the database for four types of sea ice (Table 1) at radiometer noises 0.5K at 18.7-, 23.8- and 36.5-GHz channels and sea ice emissivity fluctuations of 0.01. Cases were taken with Vo < 20 kg/m2 and Q = 0.25 kg/m2. Ice concentration C = 0.0-1.0 (a) and 0.3-0.6 (b). The straight lines are the best-fitting linear regression.

An elimination of both 23.8-GHz channels practically did not change retrieval error. An elimination of one more channel (18.7H) resulted in the s_Q increase by »10%. Retrieval errors of 4-channel Q-algorithm increase from 0.019 kg/m² (C = 0.0-0.3, 4898 scenes) to 0.033 kg/m² (C = 0.7-1.0, 4978 scenes) and equal to 0.037 kg/m² for the whole massif (C = 0.0-1.0, 16360 scenes). The scatter plot of the retrieved Q-values versus Q_o values of the database for four ice types with C=0.3-0.6 is shown in Fig. 3a. Regression equations for C = 0.0-1.0 and 0.3-0.6 are Q=0.023+0.56Q_o and Q=0.015 + 0.75Q_o; root mean square error s_Q=0.037 and s_Q=0.032kg/m², correlation coefficient R = 0.60 and R=0.74 for both subranges of concentrations correspondingly.

Figure 3: Scatter plots of the retrieved Q (a) and C (b) values versus values Qo and Co of the database for four types of sea ice at radiometer noises 0.5K at 18.7-, 23.8- and 36.5-GHz channels. Sea ice emissivity fluctuations are of 0.01. Ice concentration is C = 0.0-1.0. The straight lines are the best-fitting linear regression. Cases were taken with Vo < 20 kg/m2 and Q = 0.25 kg/m2.

Ice concentration

An elimination of both 36.5-GHz channels tended to increase the retrieval error s_C slightly. An elimination of both 23.8-GHz channels practically did not change retrieval error. An elimination of two other channels (18.7V and 36.5H) resulted in the s_S increase by »10%. Retrieval errors of four-channel Ñ-algorithm (18.7V, 18.7H, 36.5V and 36.5H) are 0.04 for the whole range and about 0.03 for the subranges of ice concentration. The scatter plot of the retrieved C-values versus C_o values of the database for four ice types with C=0.0-1.0 is shown in Fig. 3b.

Discussion and conclusions

A large data base of simulated brightness temperatures T_B(n) at AMSR-E frequencies was generated by numerical integration of a radiative transfer equation. The r/s profiles and data of hydrometeorological observations obtained by the research vessels, coastal and island stations at SST £1°C as well as the emissivities of different types of sea ice were used as input data. To model real situation, the random fluctuations of emissivity were added to the mean values of the emissivity. Besides, two values of temperature were taken for each type of sea ice.

The statistical algorithms were developed to retrieve V, Q, and C using a half of the T_B(n) massif. Efficiency of the algorithms was estimated by comparison of the retrieval errors computed by their application to the second half of the T_B(n) massif for different combinations of the AMSR-E channels and at the variations of ice concentration and radiometer noises.

A series of regression analyses carried out to investigate the benefits of including the different AMSR channels in V, Q and C retrieval algorithms have shown that the efficiency of 5- or 4-channel algorithms compares well with 6-channel algorithms. The increase of noises from 0.0 to 0.5K leads to the modest (25%) increase of s_V and s_Q retrieved with the 4-channel algorithms.

The whole range of sea ice concentration was divided on 4 overlapping subranges: 0.0-0.3, 03-0.6, 0.5-0.8 and 0.7-1.0. The regression coefficients and the retrieval errors s_V and s_Q were computed for each subrange. Since s_V and s_Q errors depend heavily on sea ice concentration a two-step technique of V and Q retrieval should be used. At first sea ice concentration is estimated and then V and Q are determined with the coefficients in equations (2) and (3) which correspond to a particular C subrange.

The developed algorithms will be tested by their application to the measured T_Bs over the Okhotsk Sea and tuned by a comparison of the retrieved values of total water vapor content V with radiosonde reports. Retrieved Q values will be compared with satellite visible and infrared cloud images. To check ice concentration algorithm, the retrieved C-fields is supposed to compare with the results of application of the published algorithms [1] as well as with the sea ice images taken by high- and medium-resolution sensors such as Envisat ASAR and/or Aqua and Terra MODIS.

ACKNOWLEDGMENTS

This study has been carried out within the cooperation between the Japan Aerospace Exploration Agency and the POI FEB RAS in the ADEOS-II Research activity (project A2ARF006). This work is partially sponsored by a Russian State contract 10002-251/Ï-14/197-396/200404-061 for a project: "Investigation of ocean-atmosphere system with passive and active microwave sensing from new generation satellites".

REFERENCES

Comiso, J.C. (2003), Sea ice concentration, ice temperature, and snow depth using AMSR-E data, IEEE Transactions Geoscience and Remote Sensing, 41, 243-252.         [ Links ] [1]

Ellison, W., A. Balana, G. Delbos, K. Lamkaouchi, L. Eymard, C. Guillou, & C. Prigent (1998), New permittivity measurements of seawater, Radio Sciences, 33, 639-648.         [ Links ] [2]

Eppler, D.T., M.R. Anderson, D.J. Cavaliery, J.C. Comiso, L.D. Farmer, C. Carrity, P. Gloersen, T. Grenfell, M. Hallikainen, A.W. Lohanik, C. Mätzler, R.A. Melloh, I. Robinson, & C.T. Swift (1992). Passive microwave signatures of sea ice, in: "Microwave Remote Sensing of Sea Ice," ser. AGU Geophysical Monograph, F. Carsey, Ed., vol. 68, ch. 4, pp. 47-71.         [ Links ] [3]

Haggerty, J.A., J.A. Curry, & G. Liu (2002), Potential for estimating cloud liquid water path over sea ice from airborne passive microwave measurements, Journal of Geophysical Research, 107, doi: 10.1029/2001JD000585.         [ Links ] [4]

Harouche, I.P-F., & D.G. Barber (2001), Seasonal characterization of microwave emissions of snow-covered first-year sea ice. Hydrological Processes, 15, 3571-3583.         [ Links ] [5]

Miao, J., K.-P. Johnsen, S. Kern, G. Heygster, & K. Kunzi (2000), Signature of clouds over Antarctic sea ice detected by the Special Sensor Microwave/Imager, IEEE Transactions Geoscience and Remote Sensing, 38, 2333-2344.         [ Links ] [6]

Mitnik, L.M., & M.L. Mitnik (2003), Retrieval of atmospheric and ocean surface parameters from ADEOS-II Advanced Microwave Scanning Radiometer (AMSR) data: Comparison of errors of global and regional algorithms, Radio Sciences, 38, 8065, doi:10.1029/2002RS002659.         [ Links ] [7]

Parkinson, C.L. (2003), Aqua: An Earth-Observing satellite mission to examine water and other climate variables, IEEE Transactions Geoscience and Remote Sensing, 41, 173-183.         [ Links ] [8]

Rosenkranz, P.W. (1992), Rough-sea microwave emissivities measured with the SSM/I, IEEE Transactions Geoscience and Remote Sensing, 30, 1081-1085.         [ Links ] [9]

Sasaki, Y., I. Asanuma, K. Muneyama, G. Naito, & T. Suzuki (1987), The dependence of sea surface microwave emission on wind speed, frequency, incidence angle, and polarization over the frequency range from 1 to 40 GHz, Transactions Geoscience and Remote Sensing, 25, 138-146.         [ Links ] [10]

Svendsen, E., C. Mätzler, & T.C. Grenfell (1987), A model for retrieving total sea ice concentration from a spaceborne dual-polarization passive microwave instrument operating near 90GHz, International Journal on Remote Sensing, 8, 1479-1487.         [ Links ] [11]

Wentz, F. (1992), Measurement of oceanic wind vector using satellite microwave radiometers, IEEE Transactions Geoscience and Remote Sensing, 30, 960-972.         [ Links ] [12]