of
‚entre,
tre, Sir
puts to
s. The
It water
electric
channel
y under
nd SAR
ved the
angles.
ation of
f SWE.
ping of
1) were
ensing.
arch to
w grain
ne SAR
ould be
sites at
cover).
und and
at could
run-off.
A 1981)
RADAR
rom the
of these
n which
rad only
s of the
ut snow
1ulation.
) factors
we used
juisition.
low the
SAR data collection we could use only a limited number of
test sites. These sites were chosen in open areas close to
access roads. Each test site consisted of one or two 60m
to 120m long base lines. Snow parameter data were
obtained at 6m intervals along the base lines to coincide
with the ground resolution element of the CCRS SAR.
The SWE determination through RADAR should be based
on calibrated SAR data. To facilitate absolute calibration,
corner reflectors were manufactured to precise
specifications and were deployed in the field before the
SAR flights. Two sets of two reflectors (one for the X and
one for the C band) were placed in locations corresponding
to 30° and 60° incidence angles in the field.
On the advice of CCRS scientists, we chose a regression
analysis design such that the SWE of ground observations
were correlated with single band, multi-band, and multi-
temporal un-calibrated and calibrated SAR data.
DATA ACQUISITION
Airborne SAR
The snow free data were collected on October 30, 1990
after the freeze-up with less than 10cm snow cover. This
flight provided X and C band data with both HH and VV
polarizations (Figure 1). The maximum snow cover data
were acquired on March 20, 1991. At this time, however,
only the C band SAR was available, but with polarizations
of HH, HV, VH, and VV). The 30° and 60° incidence angle
coverage required the use of the SAR in ‘nadir and ‘narrow’
modes respectively.
Ground Data Collection
Since the SAR mission was carried out on the afternoon
after a snow storm, the ground data collection could not be
started until the next day. In total 13 different sites were
extensively surveyed. In each site one or two parallel lines
were laid out and at 6m intervals The snow depth, density,
and SWE were measured or determined. In addition one
or two snow pits were dug along each test line and the
following data collected in the pit: number, thickness,
temperature, snow density, crystal structure, hardness, and
grain size of each snow layer. All works were extensively
documented with ground and large scale aerial
photographs and with video recording. In October 1991 the
test sites were revisited and detailed field work was
performed to obtain information on the surface covered
previously by snow.
DATA ANALYSIS
Digital data and analogue (pictorial) outputs for each flight
line were provided by CCRS. In addition absolute
calibration functions were worked out by CCRS scientists
using the RADAR returns from the deployed comer
reflectors.
Ground Data
The raw field snow measurements in each station of all test
lines were converted into SWE and plotted over a sketch of
the underlying ground cover types. We had a very good
distribution of SWE within and between test lines (Table 1).
The snow depth should be over 20cm to have an effect on
the SAR return. In our case this varied between 31 and
138cm. The calculated SWE values had a range of 7 to
38cm with averages of 12.9 to 26.9cm of individual test
lines. The snow pit dat were analyzed for snow layer
structure determination. A typical pit cross section is
illustrated in Figure 2. All 22 pits had at least 3 distinct
snow layers. Thirty two percent had four and only one pit
had five layers. The mean thickness of the three main
layers from the bottom to the surface was 22, 25, and 24cm
respectively. In the layer closest to the ground the snow
temperature showed the least variation with an average of
-1.5°C. The temperature of the upper most layer was very
close to the ambient temperature. The SWE variation was
minimal in the layer closest to the substrate and highest in
the next layer with variations further decreasing in the
newer snow on the top of the profile.
SAR Data
Based on flight recordings the incidence angles for each
test site were calculated (Table 2, Caines, 1991). An in-
house computer program was written to read, display, and
dump the digital data provided by CCRS. This program
involved a pictorial display of the image from which the
areas of the test sites could be chosen for dumping of the
digital data. Another program was prepared to carry out the
conversion of digital numbers (DN) to ‘power’ and average
RADAR cross section (0,) for a selected sub sample. A
third program provided the averaging of 9 neighboring pixel
values (3 by 3 kernel).
SAR Data Correlation with Ground Data
The ground survey gave detailed information on snow
distribution (depth, density, SWE) and on the underlying
ground cover (soil, rock, ice, vegetation type). The SAR
data should be correlated with all of the snow parameters,
the underlying surface conditions and with the incidence
angle and polarization of the RADAR.
The C-band SAR return from snow covered areas is made
up of volume scatter from the snow and of return from the
snow-soil interface, with the incidence angle being the most
important factor affecting the strength of the signal. Since
the test sites in this study were selected after the ground
was covered with snow, in most cases we did not have
uniform underlying soil surfaces. In order to obtain all of the
required SAR data (two different modes and polarizations),
the aircraft needed four passes each during the fall and
winter data acquisitions. Table 2 shows that the incidence
angles for a particular test site varied significantly. Another
problem encountered was finding the exact location of the
test lines on the various SAR imagery.
Results of simple regression analysis between raw and
filtered SAR digital data and ground SWE values indicate
that the C-band SAR alone cannot reliably predict the SWE
value of snow pack (low correlation coefficient). Although
the regression line has a positive slope, the individual
observations are widely scattered around it. This scatter is
due not only to the variation of the snow properties and of
the ground surface coverage under the snow, but also to
the inherent 'speckle' of the SAR even over uniform
surfaces. To obtain more precise results, the averaging of
more pixels is required. However, this would require a
more detailed observation of SWE on the ground as it also
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 663
