5 Fusion of Satellite SAR with Passive Microwave Data for Sea Ice Remote Sensing
101
moisture than the SAR data. Fluctuations of 1-2 dB in backscatter roughly correspond
to SSM/I-derived concentration fluctuations between 10 and 40%. SSM/I appears to be
more sensitive because of the higher frequencies (19,37 GHz),relative to ERS-l (5.3 GHz).
Since the backscatter fluctuations after freeze-up are small it may be feasible to estimate the multiyear ice fraction with SAR to augment the passive microwave algorithms
for ice classification during the early freeze-up period.
5.4.4
Modification of the NASA Team Algorithm for Multisensor Ice Classification
To fuse the SSM/I and ERS-1 SAR data using the hybrid algorithm outlined above, the
passive microwave algorithm for ice concentration estimates is modified to use SARderived information. The multiyear ice concentration from SAR data constrains the NT
algorithm. If the multiyear ice concentration (CM) is known, the two equations in the
NT solution can be solved for the first-year ice concentration (Cp ). These solutions can
be expressed in the following form,
where aI, a2, gI and g2 are functions of the following: (1) the NASA Team algorithm tie
points, (2) the measured brightness temperatures from the SSM/I, and (3) the multiyear concentration derived from the ERS-1 SAR. Detailed definitions of these functions
are given in Beaven et al. (1996). The multiyear ice concentration is used to obtain the
overconstrained system of equations given by Eq. (2). The first-year ice concentration
is then obtained by solving this using the least-squares solution (Twomey 1977). This
results in an updated first-year ice concentration from the modified NT algorithm.
Therefore, the SAR and SSM/I sensor data are fused by using the SAR-derived multiyear ice concentration to constrain the multispectral inversion process.
5.5
Active/Passive Multisensor Fusion Results
To demonstrate the algorithm we used the SAR images obtained from August 30
through September 22, 1991, in which the DLT algorithm worked well for separating the
multiyear ice from the remainder (presumably young or thin first-year ice and open
water). These SAR images were obtained from a region and in a time frame in which
field measurements of radar backscatter were obtained during the International Arctic Ocean Expedition (IAOE) from aboard the US Coast Guard icebreaker Polar Star
(Beaven and Gogineni 1994). Two examples of the DLT algorithm applied to the SAR
imagery are shown in Figs. 6 and 7. The IAOE cruise track passed almost exactly
through the center of image frame 1719 obtained on September 6 (Julian day 249) and
September 12 (Julian day 255). The Polar Star passed through this region during Julian
day 242 through Julian day 243, and based on observation oflarge features in these SAR
frames there is less than a 25-km movement oflarge features from September 6 through
September 12. Although the use of ship-based observations may not reflect the ice characteristics across the entire SAR frame we will use ice observations recorded aboard
101
moisture than the SAR data. Fluctuations of 1-2 dB in backscatter roughly correspond
to SSM/I-derived concentration fluctuations between 10 and 40%. SSM/I appears to be
more sensitive because of the higher frequencies (19,37 GHz),relative to ERS-l (5.3 GHz).
Since the backscatter fluctuations after freeze-up are small it may be feasible to estimate the multiyear ice fraction with SAR to augment the passive microwave algorithms
for ice classification during the early freeze-up period.
5.4.4
Modification of the NASA Team Algorithm for Multisensor Ice Classification
To fuse the SSM/I and ERS-1 SAR data using the hybrid algorithm outlined above, the
passive microwave algorithm for ice concentration estimates is modified to use SARderived information. The multiyear ice concentration from SAR data constrains the NT
algorithm. If the multiyear ice concentration (CM) is known, the two equations in the
NT solution can be solved for the first-year ice concentration (Cp ). These solutions can
be expressed in the following form,
where aI, a2, gI and g2 are functions of the following: (1) the NASA Team algorithm tie
points, (2) the measured brightness temperatures from the SSM/I, and (3) the multiyear concentration derived from the ERS-1 SAR. Detailed definitions of these functions
are given in Beaven et al. (1996). The multiyear ice concentration is used to obtain the
overconstrained system of equations given by Eq. (2). The first-year ice concentration
is then obtained by solving this using the least-squares solution (Twomey 1977). This
results in an updated first-year ice concentration from the modified NT algorithm.
Therefore, the SAR and SSM/I sensor data are fused by using the SAR-derived multiyear ice concentration to constrain the multispectral inversion process.
5.5
Active/Passive Multisensor Fusion Results
To demonstrate the algorithm we used the SAR images obtained from August 30
through September 22, 1991, in which the DLT algorithm worked well for separating the
multiyear ice from the remainder (presumably young or thin first-year ice and open
water). These SAR images were obtained from a region and in a time frame in which
field measurements of radar backscatter were obtained during the International Arctic Ocean Expedition (IAOE) from aboard the US Coast Guard icebreaker Polar Star
(Beaven and Gogineni 1994). Two examples of the DLT algorithm applied to the SAR
imagery are shown in Figs. 6 and 7. The IAOE cruise track passed almost exactly
through the center of image frame 1719 obtained on September 6 (Julian day 249) and
September 12 (Julian day 255). The Polar Star passed through this region during Julian
day 242 through Julian day 243, and based on observation oflarge features in these SAR
frames there is less than a 25-km movement oflarge features from September 6 through
September 12. Although the use of ship-based observations may not reflect the ice characteristics across the entire SAR frame we will use ice observations recorded aboard