Microwave remote sensing is promising
because of its ability to penetrate
the snowpack, and its capability to acquire data
in cloudy or nighttime conditions.
Since the microwave signal from the snowpack
is a volume-integrated response to wetness,
density, surface and substrate roughness, and stratigraphy,
one of the challenges is to invert
the signal to recover the various physical properties of
interest.
The many snowpack properties that affect the microwave
signal can be either sources of error if ignored, or
sources of additional information if
exploited [ Carsey, 1992].
Since longer wavelengths penetrate the snowpack
to greater depths, comparison of different
bands can aid in discriminating between
the effects of surface roughness, stratigraphy, and
snow wetness [ Rott and Davis, 1993].
Active microwave sensors such as being deployed on the First
European Remote Sensing Satellite (ERS-1)
and Canadian RADARSAT
offer the possibility to observe
seasonal snowcover characteristics in detail over the entire snow-cover season
[ Way et al., 1993].
In one simulation of RADARSAT data, snow-cover classification accuracy
was 80%, comparable to aircraft
Synthetic Aperture Radar (SAR) [ Donald et al., 1993].
Comparing a classification
of snow-covered area based on SAR with that done using TM suggests
that a SAR-based classification is sufficiently accurate to
substitute for visible-and-near-IR
based estimates when such data are not
available, for example due to cloudiness
[ Shi and Dozier, 1993].
The sensitivity of SAR to wet snow offers the possibility to determine the onset
of snowmelt [ Bernier and Fortin, 1991].
Shi et al. [1991] reported the
capability to discriminate
between snow, glacier, and rock regions in an alpine basin using
C-band SAR imagery when topographic information is not
available.
Both C-band (4.20-5.75 GHz)(1 GHz = 
Hz)
and X-band (5.75-10.90 GHz) SAR are useful to classify seasonal snow versus glaciated
areas in alpine terrain [ Rott and Davis, 1991].
A method of estimating snow depth based on
the ratio of and difference between HH (horizontal transmit; horizontal receive)
and VV (vertical transmit; vertical receive) polarizations
from multifrequency radar was presented
[ Shi et al., 1990];
unfortunately, these multiple polarizations are not available
on current satellites.
Multiple frequencies and polarizations can also
help solve problems such as
mixed pixels (snow and vegetation) encountered in a test in the Rio
Grande basin (3419 
) in southwestern Colorado
[ Chang et al., 1991.
Modeling has shown the importance of using a distribution
of grain sizes when modeling scattering of
microwaves in snowpacks
[ West and Tsang, 1993;
Boyarskij et al., 1993;
Shi et al., 1993].
Kuga et al. [1991] present a radiative-transfer model for three frequencies,
35, 95 and 140 GHz and show that the backscattering coefficient is sensitive
to liquid-water content at all three frequencies, with 35 GHz being
the most sensitive.
Neural networks have been used to invert
passive-microwave data to estimate SWE
[ Chang and Tsang, 1992;
Tsang et al., 1992].
Five microwave brightness temperatures were used to invert three parameters
from passive microwave: mean-grain size of
ice particles in snow, snow density, and snow temperature
[ Tsang et al., 1991].
A decision-tree algorithm was developed for snow-cover classification
with data from the Special Sensor Microwave Imager (SSMI)
using dual-polarized channels at
19 GHz and the vertically polarized channels at 22 GHz and 85 GHz
[ Grody, 1991].
Fiore and Grody [1992] used a decision-tree
algorithm to classify snow cover and precipitation over large areas.
Measurements of
the change in microwave brightness temperature
for ice overlain by an evolving snowpack were made, demonstrating the
change in brightness temperature due to
increased volume scattering as snow grains grow
[ Lothanick, 1993].
Sturm et al. [1993] observed that the reduced microwave emissivity of
snow cover versus soil was entirely due to a coarse-grained depth-hoar
layer; therefore, other stratigraphic features of the snowpack
could not be distinguished by the microwave emmisivity alone,
and Walker and Goodison [1993] report
difficulty classifying wet snow in the presence of
a vegetation canopy using passive microwave measurements.
Passive microwave signals are also sensitive to the liquid-water content snow,
thus offering the potential to develop snow wetness estimates.
The sensitivity of passive microwave signals to
snow wetness aids in determining the onset of spring
melt and the occurrence of multiple melt events during
the winter [ Goodison and Walker, 1993].
Wang et al. [1992] found good agreement between aircraft and microwave
depth estimates for an Alaskan snowpack;
but they also noted that the radiometric correction for the effect of
atmospheric absorption is important at all wavelengths used
for a reliable estimation of snow depth.
Scanning multichannel microwave radiometer
(SMMR) microwave brightness temperatures and ground-based snow-depth
measurements were used with a regression routine to estimate snow depth
over heterogeneous terrain in the Canadian high plains
[ Chang and Chiu, 1991] and western China
[ Chang and Tsang, 1992], and
have the potential to be applied
to mountainous areas such as the Colorado River basin in the U.S.
[ Chang et al., 1991 b;
Josberger et al., 1993].
The difficulties of integrating field, aircraft, and satellite measurements were
illustrated by measurements made on arctic snow covers by
Hall and Sturm [1991].
Good correspondence was observed between satellite-borne and aircraft-borne
microwave brightness temperatures, but the cause of
a regional minimum in brightness temperature
observable in both aircraft and satellite data could not be seen in
ground based measurements.
This may be attributable to the difficulty in capturing
the variability of snow stratigraphy in the field using snow pits.
Wind affects the microwave backscatter
properties of snowpacks by promoting evaporative loss,
decreasing snow wetness [ Koh, 1992].