Retrieval of Sea Ice Surface Features - Sea Ice: Physics and Remote Sensing - page 403

backscatter coefficient in the case of FY ice and a decrease
in the case of MY ice [e.g., LeDrew and Barber , 1994;
Shokr and Agnew , 2013]. Forster et al. [2001] determined
the progression of snowmelt onset on the Arctic sea ice
using a time series of backscatter from NSCAT satellite
during selected days from the spring of 1997. The prem-
ise was the persistent reduction of the backscatter coef-
ficient from the vertical polarization return. This was
used to define the snowmelt onset regardless of the
underlying ice, FY or MY ice. This criterion is not in
agreement with the increase of backscatter from FY ice
during melt onset as reported in the above two afore-
mentioned references. NSCAT was a scatterometer
operated at the Ku‐band on bord the Japanese Advanced
Earth Observation Satellite (ADEOS) from August 1996
to June 1997. The algorithm calculates the time derivative
of the backscatter at each pixel from the daily record of
NSCAT [ Naderi et al. , 1991]. Its nominal spatial resolu-
tion was 25 km, but the daily multiple observations
allowed the improvement of the resolution to nearly 8 km
using the scatterometer image reconstruction with filter-
ing (SIRF) method [ Long et al. , 1993].

Figure 9.17 shows a time series of daily‐averaged back-
scatter coefficients from NSCAT for selected days in the
winter. The first image in the series is a typical winter
scene showing high backscatter for most of the Arctic
Ocean basin except at the peripherals. The core of the
scene (around the geographic polar circle) is composed of
MY ice while FY ice extends away from the core. The
backscatter from MY ice is supposed to be higher than
FY ice, but this is not clear in the image. The high gradi-
ent associated with a sharp and persistent reduction in
backscatter is used to identify the snowmelt onset. This
means that short‐term (1-2 days) reduction in backscat-
ter should be rejected as being caused by temporary melt
episodes rather than intermittent snowmelt events.
Mathematically speaking, this is equivalent to identifying
an edge in a two‐dimensional noisy signal. This requires
using a threshold on the calculated time derivative to
indicate the day when smoothed backscatter has decreased
quickly enough to be considered the melt‐onset event.
The first sign of snowmelt onset in Figure 9.17
appears on day 147 (27 May) as a remarkable reduction
in backscatter in the western Arctic region (the Beaufort,

25 147 156

161 164 179

Figure 9.17 A time series of daily‐averaged backscatter coefficients from NSCAT (vertical polarization normal-
ized to 40° incidence angle) for selected days in the winter and spring of 1997 in the Arctic. Day of the year is
indicated by the number in each panel [adapted from Forster et al. , 2001].