The Arctic Snow Microstructure Experiment (ASMEx) took place in Sodankylä, Finland in the winters of 2013–2014 and 2014–2015. Radiometric, macro-, and microstructure measurements were made under different experimental conditions of homogenous snow slabs, extracted from the natural seasonal taiga snowpack. Traditional and modern measurement techniques were used for snow macro- and microstructure observations. Radiometric measurements of the microwave emission of snow on reflector and absorber bases were made at frequencies 18.7, 21.0, 36.5, 89.0, and 150.0 GHz, for both horizontal and vertical polarizations. Two measurement configurations were used for radiometric measurements: a reflecting surface and an absorbing base beneath the snow slabs. Simulations of brightness temperatures using two microwave emission models, the Helsinki University of Technology (HUT) snow emission model and Microwave Emission Model of Layered Snowpacks (MEMLS), were compared to observed brightness temperatures. RMSE and bias were calculated; with the RMSE and bias values being smallest upon an absorbing base at vertical polarization. Simulations overestimated the brightness temperatures on absorbing base cases at horizontal polarization. With the other experimental conditions, the biases were small, with the exception of the HUT model 36.5 GHz simulation, which produced an underestimation for the reflector base cases. This experiment provides a solid framework for future research on the extinction of microwave radiation inside snow.
Snow is a vital component of the water cycle, and is critically important for meteorological and climatological studies due to its high albedo, high thermal emissivity, and thermal insulating properties (Cohen and Rind, 1991). In addition, over 1 billion people rely on snowmelt for their fresh water drinking supply (Barnett et al., 2005). To predict and monitor the evolution of potential snowmelt, continuous observations of key parameters such as snow water equivalent (SWE), height of snowpack (HS, as defined by Fierz et al., 2009), and snow extent (SE) are required throughout the year. While traditional snow pit and automatic weather station observations are important, remote sensing observations of snow with passive microwave radiometers are currently the only means in northern countries to provide vital global daily measurements of snow properties. As snow crystals act as scattering centres for upwelling microwave radiation, the size of the snow crystal, the radiation wavelength (and therefore its frequency), and the snow depth all play a role in dictating the amount of scattering present in a snowpack (Chang et al., 1987; Hallikainen et al., 1987).
Over the last 30 years, space-borne passive microwave observations have been
used to estimate snow mass and SWE (Chang et al., 1987; Hollinger et al.,
1990; Kelly et al., 2003; Takala et al., 2011). The basis of snow mass
estimates (Chang et al., 1987) is based on comparison of the observed
brightness temperature at a frequency where scattering by the snow crystals
is dominant (
Both the HUT snow emission model and MEMLS use snow parameters to describe the snowpack and snow microstructure. These parameters include physical temperature, density, and some form of microstructure parameter. This microstructure parameter (describing size, shape, orientation of snow grains) has a large effect on the observed brightness temperature (Foster et al., 1999; Armstrong et al., 1993) because the intensity of scattered microwave radiation is directly linked to snow microstructure (Chang et al., 1987). However, the amount of scattering, described by the scattering coefficient in both the HUT snow emission model and MEMLS, is empirically defined based on observations (Pulliainen et al., 1999; Wiesmann and Mätzler, 1999; Pan et al., 2015). However, MEMLS also includes an option to define the scattering coefficient purely on a physical basis (Mätzler and Wiesmann, 1999).
The Arctic Snow Microstructure Experiment (ASMEx) took place at the Arctic Research Centre of Finnish Meteorological Institute (FMI-ARC) in Sodankylä, Finland in the winter seasons of 2013–2014 and 2014–2015. During the ASMEx, macro-, microstructure, and radiometric measurements of homogeneous snow slabs were made. The snow slabs were extracted from the natural seasonal taiga snowpack. The radiometric measurements were made on two different bases: one assumed perfect absorber and one perfect reflector. Observations of snow macro- and microstructure were made after radiometric measurements. The observed parameters were fed into the HUT snow emission model and MEMLS to produce simulated brightness temperatures. Only homogeneous slabs of dry snow were considered for microwave emission simulation. This was to avoid using wet snow in the radiometric measurements, as the dielectric properties of dry and wet snow are very different. The real and imaginary parts of the dielectric constant of water are much greater than those of ice (Stiles and Ulaby, 1981), increasing the complexity of the behaviour of the dielectric properties of snow.
This paper uses both the HUT snow emission model and MEMLS to simulate the microwave emission of homogenous snow slabs extracted from the natural snowpack in FMI-ARC during ASMEx, and compares simulated and observed microwave emission from the snow slabs. The ultimate aim of ASMEx is to improve the understanding of the microwave extinction processes within the snowpack, and their relation to microstructural properties of natural snow cover. This will enable to improve the precision of future and existing snow emission models.
The snow slabs were extracted from the natural snowpack in the intense
observation area (IOA) of the FMI-ARC that is situated in the clearing of a
sparse pine forest, in Sodankylä, Finland. A snow sample of size
80
The preparation and extraction of the snow slabs was a delicate process. Once a homogeneous layer of sufficient thickness was selected, the sample was prepared by pushing a metal plate (surrounded by a microwave transparent plastic sheet to avoid snow freezing to the metal plate, both at ambient temperature) into the snowpack and selecting the snow sample with a plastic frame as shown in Fig. 1. The plastic frame was also allowed to cool to the ambient temperature, in order to reduce the snow melt–freeze problem. Cuts were made to the surface snow, using metal plates and saws, parallel to the sides of the plastic frame. This allowed the frame to sink to the level of the embedded metal frame. All cuts were made outside to the plastic frame, in order to limit the disruption to the potential sample. Once the plastic frame was level with the metal plate, the entire sampling apparatus and snow sample were pulled out of the snowpack. Any snow above the slab sample and plastic frame was removed. The top of the snow sample was carefully smoothed with a metal plate as gently as possible without making artificial features on the slab surface. Immediately after extraction, the slabs were placed in front of the radiometer for brightness temperature measurements. A total of 14 samples were extracted in that manner.
Snow sample was taken from snowpack with a plastic frame, a metal plate and a saw, and a metal bottom plate surrounded by a plastic sheet.
The microwave radiometric measurements were made with two RPG-XCH-DP Dicke
Switch radiometers, installed on top of the radiometric tower in the IOA.
The experimental set up of radiometric measurements is described in Fig. 2.
Five different frequencies (18.7, 21.0, 36.5, 89.0, and 150.0 GHz) at both
horizontal and vertical polarizations were used, although not all
frequencies were working for all slabs. Tables 1 and 2 detail the
radiometric data collected from the ASMEx slabs in 2014 and 2015,
respectively. The radiometric measurements were made at an inclination angle
of 50
Radiometric measurements followed a comparable procedure as in Wiesmann et
al. (1998). The first measurement was made with the snow slab on top of the
reflective metal base. The metal base acts as a perfect reflector by
reflecting the downwelling emission of microwave radiation from the sky.
Once the snow slab had been observed at all frequencies, sky measurements at
an equivalent incidence angle were made. The metal plate was then carefully
removed from the set up, so that the snow slab was upon the assumed perfect
absorber. The radiometric measurements were then repeated. Emissivity tests
of the absorbing material, using the experimental setup in Fig. 2 without
the snow slab and metal plate, proved that the assumption of a near-perfect
blackbody was valid for all slab experiments, with the exception of slabs B05
and B07. For these two slabs, the metal strips in the tape, used to hold
the top-most piece of Styrofoam together, caused a reduction in brightness
temperature at horizontal polarizations at different frequencies. A
correction (none at 18.7 GHz,
Setup for radiometer measurements with a 50
Radiometric data measured from the 2014 ASMEx slabs. Brightness
temperatures from reflective base (
Once the radiometric measurements had been completed, the destructive
sampling of the physical parameters of snow macro- and microstructure took
place. Initially, the stratigraphy of the slab was observed using the
SnowMicroPen (SMP; Schneebeli and Johnson, 1998 and Schneebeli et al., 1999).
The SMP uses a sensitive piezoelectric force sensor on top of a penetrative
rod, which is capable of detecting changes in penetrative resistance at a
high resolution (4
Approximate locations of the macro- and microstructure measurements in the snow slab. Individual SMP and micro-CT measurement locations are also depicted.
Radiometric data measured from the 2015 ASMEx slabs (horizontal/vertical
polarization). Brightness temperatures from reflective base (
The SSA is defined as the ratio between the surface area of the ice and its
mass (Legagneux et al., 2002). A new method for measuring the SSA is the
IceCube instrument, which is a commercially available version of DUFISSS
(Gallet et al., 2009). The IceCube instrument uses a 1310 nm infrared laser
to measure the hemispherical reflectance from the sample. The SSA of the
snow slab was measured at two different locations in a vertical profile with
3 cm intervals. The traditional grain size,
Snow samples were taken from the centre of the radiometer footprint to be scanned with microcomputed tomography (micro-CT) apparatus. The cast samples were analysed via three-dimensional x-ray tomography in WSL Institute of Snow and Avalanche Research, SLF, Switzerland, to produce a three dimensional image of the snow (Heggli et al., 2009). From this image, it is possible to measure many important microstructural parameters, especially a vertically highly resolved profile of density and correlation length.
In addition to the different number of microstructural measurements of the
snow slab, vertical profiles of physical temperature and density took place
in other locations within the slab with a vertical resolution of 5 cm. The
density profiles were made using a density cutter with a volume of 500 cm
The Helsinki University of Technology (HUT) snow emission model (Pulliainen et al., 1999; Lemmetyinen et al., 2010) is a semi-empirical model, which uses the radiative transfer approach to model the microwave brightness temperature. It is capable of treating the snow as a single homogeneous layer (Pulliainen et al., 1999) or as a series of homogeneous layers (Lemmetyinen et al., 2010), with the layers being defined by its physical temperature, density, observed grain diameter, and SWE.
The model's basic assumption is that the microwave radiation is scattered
mostly in the forward direction, which allows simplifying the radiative
transfer equation to a single flux. Calculation of the absorption
coefficient in the HUT model is based on empirical models by Mätzler (1987);
while the total extinction coefficient (sum of absorption and
scattering coefficients) was originally calculated by Hallikainen et al. (1987)
from observations of natural snow slabs collected in southern Finland.
The extinction coefficients calculated by Hallikainen et al. (1987) is valid
between 18 and 90 GHz. Calculation of the total extinction coefficient was
originally based on the mean observed grain size (Table 1, Hallikainen
et al., 1987), which can be interpreted to be close to the traditional
measure of grain size,
The HUT model uses up- and downwelling emissions, represented by single-flux approximations, to calculate the total emission at the top of the snowpack. Multiple reflections at layer interfaces are accounted. Separate modules were used to simulate the effect of vegetation and atmosphere to detected emission were published with the original model, but were not applied here.
Averaged results from macro- and microstructure measurements. It should be noted that slab A03 was wet, so was not considered for model simulation.
MEMLS (Wiesmann and Mätzler, 1999; Mätzler and Wiesmann, 1999) is also
based on radiative transfer theory, treating the snowpack as a stack of
horizontal layers, with each layer being characterized by its depth,
physical temperature, density, and exponential correlation length. Although
exponential correlation length
These data are used to calculate absorption and scattering coefficients within the snow, as well as transmissivity and reflectivity between adjacent snow layers. A two-flux (up- and downwelling) model is used to calculate the emitted brightness temperature at the top of the snowpack. However, the absorption and scattering coefficients are adjusted with six flux coefficients (up- and downwelling, and four horizontal directions). The scattering coefficient was empirically defined from radiometric and macro- and microstructure measurements as laid out by Wiesmann et al. (1998) and it is valid between 10 and 100 GHz. An optional feature, originally implemented for coarse-grained snow with a large correlation length, is to use the improved Born approximation (Mätzler and Wiesmann, 1999) for the calculation of the scattering coefficient.
Preliminary analyses of snow macro- and microstructure measurements include
slab thickness, physical temperature, density, SSA, grain size
HUT (light blue) and MEMLS (dark blue) simulated brightness
temperatures plotted against observed brightness temperatures at 18.7 (circle),
21.0 (square), and 36.5 GHz (triangle). The correlation coefficients of the
single-layer HUT model (HUT
The parameters from the eight dry homogeneous slabs in Table 3 were fed into
both the single-layer HUT snow emission model and into MEMLS to produce
simulated brightness temperatures. The ground layer in both of the models
was modified to simulate the absorbing and reflecting bases by altering the
reflecting properties of the ground, to model the reflective properties of
the metal plate (
The values in Fig. 5 show that for the absorbing base, the HUT model simulations tend to have smaller RMSE values than MEMLS, while for the reflective base simulations the RMSE values are comparable at 18.7 and 21.0 GHz. At 36.5 GHz the HUT snow emission model produces larger RMSE values than MEMLS. The RMSE values for the absorbing base of vertical polarization (V-ABS) are the smallest.
It can be seen from Fig. 6 that the reflective base cases have the smallest
bias, with 18.7 and 21.0 GHz only having very small magnitude (
Simulated brightness temperature RMSE at horizontal (H) and vertical (V) polarizations for the absorber material base (ABS) and the reflective metal plate base (REF). Eight slabs were simulated at 18.7 and 21.0 GHz, while seven slabs were simulated at 36.5 GHz.
Simulated brightness temperature bias at horizontal (H) and vertical (V) polarizations for the absorber material base (ABS) and the reflective metal plate base (REF). Eight slabs were simulated at 18.7 and 21.0 GHz, while seven slabs were simulated at 36.5 GHz.
The larger magnitude bias in the REF situations in the single-layer HUT
simulations can be attributed to the way which the simulated emission is
calculated. The bias in HUT for both REF situations is largely reduced
(bias
Micro-CT-derived bulk average and standard deviation values of SSA and density. The values for A and B correspond to positions A and B in Fig. 3. Micro-CT A02A was not analysed, thus values of SSA and density are not given.
Locally calibrated SMP-derived bulk average and standard deviation values of SSA and density. The values for B2 and C2 correspond to positions B2 and C2 in Fig. 3.
The doubling of the snow thickness also reduces the errors introduced by slab B05 (initial thickness 5.4 cm), as the small thickness is not enough for scattering to be correctly simulated. Additional errors will be introduced, due to the slight changes in the snow density and microstructural parameter that were not recorded by the traditional observations, due to the coarse resolution of the method (vertical profile of 3–5 cm). There were variations in density and SSA that were recorded by the micro-CT and SMP observations, but were not recorded with the traditional observation techniques.
The Arctic Snow Microstructure Experiment (ASMEx) consisted of radiometric, macro-, and microstructure measurements of snow slabs upon absorbing and reflecting bases. Brightness temperatures of the homogeneous snow slabs were simulated with the HUT snow emission model and with MEMLS. Results of the comparison of simulations and observations are described in Sect. 3.2. The HUT model produced smaller RMSE across all three frequencies for the simulations upon an absorbing base. The reflective base simulations produced RMSE values that were comparable with the HUT model and MEMLS at 18.7 and 21.0 GHz. Both models overestimated the brightness temperature at H-ABS, and at V-ABS the single-layer HUT model slightly overestimated the brightness temperature while MEMLS underestimated it. Both models produced very small biases for the reflective base cases, with the exception of the HUT model at 36.5 GHz.
The RMSE and bias is influenced by internal extinction processes within the snow slabs, which are imperfectly simulated by the model physics. The relatively high errors, especially at H pol, considering the highly controlled measurement setup, highlight the requirement for further development of the models, as well as the need to better quantify the snow microstructural properties themselves. These preliminary brightness temperature simulations will be repeated in the future using the physical snow properties collected by the modern techniques including SMP and micro-CT measurements. Ultimately, a revised extinction model will be created for the HUT snow emission model, and implemented with the aim to improve the model inversions of SWE from radiometric measurements of microwave emission. This revised extinction coefficient, based on the data collected during the ASMEx campaign, will be a function of microstructural parameter and frequency.
Juha Lemmetyinen, Mel Sandells, and Martin Schneebeli planned the experiment with help of William Maslanka. William Maslanka had the main responsibility in organizing and carrying out the measurements. Leena Leppänen had the responsibility of organizing measurements as the local operator, including implementation of measurement setup and participation to measurement procedure. Anna Kontu and Henna-Reetta Hannula participated in the experimental measurements. William Maslanka conducted data processing and simulations. Margret Matzl provided the micro-CT data set. Martin Proksch provided the SMP analysis. William Maslanka and Leena Leppänen prepared the manuscript with contributions from all co-authors.
We thank the staff of FMI Arctic Research Centre in Sodankylä for performing the ground-based radiometer measurements and macro- and microstructure measurements. We also thank the staff of WSL Institute of Snow and Avalanche Research SLF for the SMP instrument and for the SMP and micro-CT analyses of the snow samples. The manuscript preparation was supported by the EU 7th Framework Program project “European–Russian Centre for cooperation in the Arctic and Sub-Arctic environmental and climate research” (EuRuCAS, Grant no. 295068). Edited by: C. Ménard