Retrieval of Sea Ice Surface Features - Sea Ice: Physics and Remote Sensing - page 402

stages of melt: onset of melt, ponding (including drain-
age), and ice decay. Ponding can be detected in SAR
images if the surface of the pond is smooth, i.e., generat-
ing a dark signature that contrasts with the bright sig-
nature of the surrounding ice. If the pond surface is
roughened by the wind, it would appear as bright as the
surrounding ice, which makes the pond identification dif-
ficult. What complicates the identification even further
is the difference in backscatter of the pond between the
near and far ranges of SAR image (ponds appear darker
in the far range). Even with these difficulties, SAR is
suitable for melt pond identification because of its fine
spatial resolution.
De Abreu et al. [2001b] investigated the utility of the
visual interpretation of Radarsat‐1 images in operational
monitoring of the spring melt of Arctic FY ice. They pre-
sented a series of images that showed the evolution of the
surface melt during May-July from a site in the central
Arctic. Although the ice signature in the images could be
interpreted in terms of the expected physical changes
of the surface, the data did not offer a great promise for
accurate mapping of melt phases. Theoretically speaking,
the onset of the snowmelt increases the snow liquid con-
tent and consequently the dielectric constant of the over-
laid snow. This results in decreasing the penetration of
the radar signal and increasing the backscatter from the
surface. If the snow surface roughness is significants
the underlying ice will appear bright regardless of its
degree of roughness (i.e., the wet snow masks the contri-
bution from the underlying ice roughness). Based on the
operational analysis of Radarsat images at CIS, De Abreu
et al. [2001b] reported that the drainage stage that follows
the ponding can be identified by an increase in backscat-
ter and the appearance of unique drainage features,
namely cracks and holes (visual observations). In gen-
eral, operational analysis of single‐channel Radarsat
data offers a limited capability to identify ice surface
melt. Dierking and Busche [2006] compared the C‐band
SAR onboard the Canadian Radarsat‐1 and the L‐band
SAR onboard the Japanese platforms JERS‐1 and ALOS
and found that the L‐band is more capable of mapping
surface structure during the melting period. This is
because the L‐band has larger penetration into the snow‐
covered ice even when the snow is wet.
Similar to the passive microwave algorithms, all algo-
rithms that detect snowmelt onset from radar data utilize
the concept of searching for a change in signature that
identifies the melt onset in a temporal record of the
observations. For that reason scatterometer data are pre-
ferred over SAR data because of their wide spatial cover-
age and the more frequent temporal coverage. As a result,
a daily record of backscatter from each pixel can be
obtained from scatterometer data. Several studies found
that snowmelt onset is associated with an increase in the

i 1

(9.18)

GR GR () i GR ( )

ice ice ice

where GR ice is the spectral gradient ratio between radia-
tion from 37 and 19 GHz channels in vertical polariza-
tion, determined using equation (8.11). GR ice takes
negative values for snow‐covered ice in winter (around
−0.06 and −0.01 for MY and FY ice, respectively) and
zero or positive values during the melt‐freeze transition
periods [ Markus et al ., 2009]. In order to account for the
effect of ice concentration, the brightness temperature
from ice in a heterogeneous footprint T b ,ice should be
determined from the following equation, which can then
be used to determine GR ice :

(9.19)

TT CT ( ( 1 ) ) / C

b , ice b , obs b , OW

where T b ,obs is the observed brightness temperature, C
is ice concentration, and T b ,OW is the typical brightness
temperature from OW. The third indicator is

PT T
bV

08 . (9.20)

, 19 bV , 37

For dry snow on MY ice P has values less than 460 K.
This value increases at the onset of melt [ Smith , 1998].
For FY ice, P is greater than 440 K for dry snow and
drops below that value at the onset of melt. The PMA
uses the three parameters presented in equation (9.17),
(9.18) and (9.20) after normalizing each value by the
ranges representing the melt condition. For example, the
range for ΔGR ice when melt is potentially present is
between 0.005 and 0.015. The normalized value of ΔGR ice
between these two limits is called a melt signature weight,
denoted by W ΔGR ice . Similar signature weights are used
for the other two parameters; W Δ T b ,37 V and WP . The sum
of the three weights is calculated for each day:

WWTW WP
bV

GR ice (9.21)

,37

The day with the greatest sum is considered to be the melt
onset day. Markus et al. [2009] divided the Arctic basin
into 10 regions and presented data on onset of melt and
effective melt for each region.

9.3.3. Active Microwave Observations

Physically speaking, melt‐related changes of snow-
covered sea ice properties give rise to increasing snow wet-
ness, decreasing salinity at the snow‐ice interface, altering
the snow grain size and, of course, resulting in wet ice sur-
face. These changes affect the dielectric properties and
therefore the radar backscatter. The change in backscat-
ter seems to have a potential for identification of the three