1. Introduction
Radar monitoring of thick snow cover in polar regions with optical thickness (a product of the depth of the snow and the extinction coefficient) of order 1 or more from elevated above-ground and space platforms is of great importance for registration and for understanding glaciology processes caused by climatic change. Applicable to this issue, the volume component of the backscatter coefficient has a notable contribution to the total backscatter (Noveltis, 2005). Although the idea of the radar-cross section (RCS) term and its derivatives, like the backscatter coefficient, aims to separate as far as possible the sensor (radar) and target parameters, this distinction can rarely be fulfilled in the case of spatially extended geophysical targets (SEGT), such as atmospheric clouds and rain, as well as the thick snow cover that is the focus of this paper. Due to this feature, the analysis of backscatter from SEGT strongly depends on the relationship between the technical and physical-geometrical properties of the radar and target, respectively. The main parameters that govern the radar-target configuration for snow sounding are wavelength, antenna characteristics, pulse duration, sounding direction, extinction coefficient and the geometrical depth of the snow. A correct assessment of the volume component of the backscatter coefficient and an understanding of the realm of applicability of any backscattering model is possible only when the size of the scattering volume within a snow slab is known. This parameter depends on the factors mentioned above and should be distinguished for different situations in the practice of radar sounding of snow-covered terrain by Synthetic Aperture Radar (SAR), a scatterometer or an altimeter. This work attempts, at first, to determine the radar-target configurations inherent to volume scattering estimations of the thick snow cover under the different radar applications mentioned above. Next, we analyze the range of applications of the incoherent approach for backscatter magnitude estimation, currently one of the main techniques for snow characteristics assessment. The simplest incoherent approach, based on the so-called “particle” or “discrete” approximation, leads to the dependence of backscatter on the sixth moment of the particle size distribution function (PSDF) and the mean amount of particles in the scattering volume (Siegert & Goldstein, 1951, Battan, 1959, Ulaby et al., 1982). For a medium with losses (such as a thick snow slab) the modification of this approach is referred to as the semi-empirical model (Attema & Ulaby, 1978, , Ulaby et al.1982, Ulaby et al., 1996). There is some evidence of less backscatter occurring than expected by virtue of the conventional reflectivity factor and the backscatter coefficient for incoherent scatter from homogeneous thick snow slab (e.g., Rott et al., 1993). For dense media (mean distance between particles is less than the wavelength) there are many references in the literature that the backscatter is determined by media inhomogeneities (Naito & Atlas, 1967, Gossard & Srauch, 1983, Fung, 1994). To attempt to evaluate the contributions of these inhomogeneities, the radiative transfer (RT) and the dense media radiative transfer (DMRT) models were developed (e.g., Ulaby et al., 1982, Fung, 1994, Tsang et al., 2007). A review of these models and their modifications provided by Noveltis (2005) stated that in certain aspects, these approaches showed some successes. However, these models have not the close-form solutions that make difficult to use them and analyze results obtained. Kendra et al. (1998), based on experiments with artificial snow of varying depths, concluded that both conventional and dense-medium radiative transfer models fail to adequately explain the observed results.
Finding an appropriate and relatively simple approach to calculate the volume component of backscatter to explain the observable deviations from the classical (incoherent) model is therefore a relevant task. The majority of previous studies, focusing on measurements of a dry snow at temperate latitudes, where the snow depth seldom exceeds 1 m and the corresponding optical thickness is much less than 1, have led to the conclusion that such an approach is not feasible because of the much weaker interaction of the electromagnetic radiation of radar wavebands with ice particles within the snow compared with the backscatter from the soil beneath. The current study focuses on the case of the Greenland ice sheet, however, where the depth of snow significantly exceeds that found in temperate latitudes. This favorable condition, in conjunction with the recent finding that the incoherent approach is only a specific case of a more comprehensive description of electromagnetic wave interactions with spatially-extended individual scatterers (Yurchak, 2009), provides a reason for more detailed investigations of the possibilities of the semi-empirical model for interpreting the observable features of radar backscatter from thick snow cover.
2. Condition of complete burial of the probing pulse into snow medium
To estimate the volume component of backscatter from a snow slab, it is first necessary to understand whether the probing pulse is completely buried within the snow slab or whether the illuminated volume is only part of the probing pulse volume. In weather radar meteorology, applicable to rain and clouds, this problem is known as “partial (or incomplete) beam filling,” with a corresponding factor included in the weather radar equation (e.g., Clift, 1985). For sounding thick snow cover, this issue is practically not discussed and is different compared with the sounding of meteorological targets. Complete burial depends on the sounding configuration and snow slab depth. The main criterion for complete burial is the location of the pulse scattering volume with angular size equal to the angular antenna beam width and with spatial length equal to one half of the actual spatial duration of the transmitted pulse into a snow layer. Thus, for complete burial, the angular and radial sizes of the backscatter volume should be matched with the snow slab depth, which is usually known only roughly for a particular geographical region. This condition poses problems for practical applications. Nevertheless, to better understand possible situations where complete burial is feasible, the requirements for the angular antenna beam size and the probing pulse duration will be analyzed separately for different major sounding configurations.
2.1. Condition of complete burial of the angular (transverse) size of the pulse volume
2.1.1. Oblique sounding, flat surface, plane wavefront
The scheme of this configuration is shown in Figure 1 Here, and everywhere below, the snow slab is assumed to have limited depth and an unbounded horizontal extent.
Mathematically, the condition for complete burial in this case is:
where d b is the burial depth of the advance point (p) of pulse edge (leading or trailing) when the lagging point (c) reaches the surface and h is the snow depth. As follows from the geometry of Figure 2.1:
For φ0.5<<1 one can state
If h~10 m and R
0
=800 km, then
2.1.2. Oblique sounding, flat surface, spherical wavefront
A real wavefront within a snow slab has a spherical shape. The impact of this shape on the estimation of d b as the look angle θ (Figure 1.) decreases begins at the moment when the line tangent to the spherical front at point p coincides with the horizontal line pg, Figure 2. This situation takes place when θ becomes equal to the angle between pc and the tangent:
As follows from geometrical sketch, the auxiliary angle β is equal to:
Thus, the estimate (2.3), provided in the previous subsection, is valid for
2.1.3. Nadir sounding, flat surface spherical wavefront
This configuration is shown schematically in Figure 3 and the corresponding condition for complete burial is:
From the geometrical configuration and φ 0.5<<1, it follows that:
This estimate can be accepted for elevated platforms (for example, airborne radars and those mounted above the surface) but needs to be analyzed further for space platforms due to the sphericity of the surface, as the corresponding area illuminated on the Earth’s surface by conventional radar is large.
2.1.4. Nadir sounding, spherical surface, spherical wavefront
In the spherical surface approach, the depth of complete burial in altimeter mode is more than that for a flat surface described above by an increment Δz in the center of the beam. This effect is illustrated in Figure 4, and can be written as:
Based on the geometrical relations of the sketch, the following system of equations may be composed
where
where R
E is the Earth’s radius. Taking into account that
A numerical assessment under the parameters values listed above gives Δz~5 m. The estimation of the parameter
This relationship coincides with the estimate obtained by Barrick (1972) for a spherical wave increment over the spherical mean sea surface. For R
0=800 km, R
E=6400 km and
Based on known assessments of penetration depth of 2-5 m for Ku-band (e.g., Davis, 1996), the wave sphericity in practice does not allow for consideration of the complete burial of the transversal size of the probing pulse for a space-based altimeter and scatterometer. This sphericity causes the area illuminated on the surface to change as the probing pulse is buried into a snow slab. It can be easily shown that the radius of the illuminated area changes with the burial increment ΔR as
2.1.5. Wavefront flattening
The speed of an electromagnetic wave within a snow slab is lower than that in an air by
The peripheral ray (ray Ob, for example) has a different path into a snow slab as compared with the central ray (Og), which that has a portion ag extending into the snow. This feature causes the flattening of the spherical wavefront. Thus, the snow slab works like dielectric lens antenna (e.g., Lo & Lee, 1993). Let us make a rough estimate of the impact of this phenomenon on the increment Δz. For a spherical wavefront (dashed curve in Figure 2.5), the following equation is valid: Ob=Oa+au+ug. Dividing both sides of this equation by speed of the wave propagation, c, yields the time of the wavefront arrival at point g:
Due to delays in snow in part of ag=au+ug the equation for the wavefront arrival at point g is:
Moreover, due to the increase in snow density with depth, the dielectric permittivity also increases, additionally impacting the rays’ path configuration as takes place in a Luneburg lens. Thus, the actual meaning of the burial parameter lies between the two estimates carried out above:
Based on the above estimates, one can say that the surface sphericity practically contributes a relatively small amount to the estimate of the complete burial condition of the transverse dimension of the probing pulse, and consequently, the surface can be roughly considered as flat for any radar configurations regarding volume component assessment.
2.1.6. Modes of volume scattering for spherical wave
By analogy to spherical wave scattering from a surface (e.g., Moore & Williams, 1957) it is also reasonable to distinguish two modes of the volume component in the case of spherical wave scattering, as shown in Figure 6. For simplicity, the bounds of the snow slab are considered flat. Let us call the mode depth-limited (Figure 6a.) if at the moment when the central point of the spherical wavefront reaches the bottom of a snow slab or a penetration depth (D p ), the peripheral point of the wavefront, crossing the surface, is still within the beamwidth footprint on the surface. This condition takes place when:
Let us call the mode beamwidth-limited (Figure 6b.) if the peripheral points of the spherical wavefronts, crossing the surface, are beyond the beamwidth footprint on the surface for a significant part of the scattering volume. This situation takes place when:
Obviously, that the depth-limited mode is inherent to space based altimeter and scatterometr and the beamwidth-limited mode is realized for SAR. If the “significant part” in the definition of this mode is replaced with a more exact term, the “determined part,” the condition (2.15) can be modified and written in the form:
where
2.2. Condition of complete burial of radial size of the pulse volume
Suppose that some area of snow cover is illuminated by a radar located in orbit as shown in Figure 7. For simplicity, this scenario is depicted with a plane wavefront and flat surface.
Let us denote the height of the orbit as R 0 , the distance from the radar to the center of the area as R, the anglular width of the main lobe of the antenna pattern as φ 0.5 and, the look angle as α. In addition, we assume that the cross-section of the main lobe is the circular. The illuminated area is assumed to be flat horizontally, and the bottom boundary surface of the snow slab is also flat and parallel to the top surface. Thus, the incidence angle, θ is equal to the look angle, α (θ=α).
The main scaling parameters determining radial propagation are:
1) The radial length of the pulse scattering volume, equal to half of its spatial extent in the snow medium:
where
2) A one-way path in the snow, where the incidence power is decreased by “e” times, usually called the “penetration depth” in the literature
where k e is the extinction coefficient, which characterizes the attenuation properties of the medium due to scattering and absorption. One important remark is necessary. Since radar sounding of snow is often performed in off-nadir mode, and the main lobe of the antenna pattern has a finite angular size, it should be
Figure 8.
Relationships between the penetration depth (D
p), penetration path (L
p
) and propagation depth (
underlined that the extinction coefficient and the “penetration depth” are measured along the direction of the wave (ray) propagation and not only to the vertical. Therefore, calling the term
Penetration path:
Propagation depth:
Penetration depth:
where
where
All of the above conditions are summarized in Table 1.
Table 1.
Conditions for complete burial of the radial size of the probing pulse
Concluding remarks:
If the conditions for complete burial of the radial size of the probing pulse are completed, it can at least be said that one sample of the return signal is formed by the scattering volume
where
Otherwise, under
2.3. An assessment of the scattering volume under incomplete burial condition
If the parameters of the probing pulse and of the snow slab do not satisfy the conditions presented in the table above, the radial size of the scattering volume is determined by the geometry of the snow slab. Several practically important cases for practical application are discussed below.
2.3.1. Flat surface, plane wavefront, long probing pulse and fully scattering snow slab
In this scenario, the illuminated area, A
ill
on the snow cover (Figure 1.) is an ellipse, with the minor semi-axis equal to the radar beam cross-section radius
If one uses the height of the satellite orbit
The pattern of the scattering volume for this case is depicted in Figure 9, where the scattering volume is bounded by the surface and bottom planes.
In this case, the size of the scattering volume is the sum of the volume of a circular cylinder V
0 with cylindrical element d
0 and two similar volumes of the truncated cylinders V
1 that mutually add up to a completed circular cylinder with cylindrical element d
1 (the bases of all volumes are the same and equal to
Because
Thus, the equivalent scattering volume is an elliptic cylinder with a base equal to the illuminated area on a surface and with a height element equal to the slab depth (under the assumption that the slab depth is less than the half of the spatial duration of the probe radar pulse). This result was obtained by Matzler (1987).
The duration of the SAR probing pulse duration is equal to several tens of microseconds (e.g., 42 μs for RadarSAT-1 and 37.1 μs for ERS-1). Due to frequency chirp, the compressed probing pulse duration decreases by many times, resulting in a volume radial size equal to only several meters (5-13 m for RadarSAT-1 and 9.7 m for ERS-1, for example). Numerical data are provided based on Alaska Satellite Facility documents (“RadarSAT-1 Standard Beam SAR Images”, 1999; and “ERS-1 and ERS-2 SAR Images”, 1996).
Figure 10.
Cross-section of the scattering volume for a spherical surface and wavefront. The position of the flattened wavefront at the moment when its central point (a’) reaches the penetration depth is depicted with a convex dashed line
Because SAR images represent the backscatter pulse train from the entire path of a wave into a snow slab for any look angle as a point on a surface of some mean brightness (return power), the radial size of the scattering volume can be considered equal to either the slant size of the snow layer or the penetration path (whichever is smaller). That is, although the SAR pulse has a finite compressed spatial duration; it works like a long pulse due to the absence of radial discrimination in the sense of conventional radar terminology.
2.3.2. Long probing pulse, spherical wave, spherical surface, partially scattering slab
For this case, the bottom of the scattering volume is the geometrical placement of points located at a distance L p from the snow surface along the family of rays within the solid angle of the main antenna lobe, as shown in Figure 10.. The defining feature of this configuration is that the central and peripheral points of the spherical wavefront pass the length L p at different times, resulting in different radii of curvature for the top and bottom bounded surfaces.
3. Volume component of the backscatter coefficient
3.1. Semi-empirical model for flat surface, plane wave and long probing pulse
Based on the considerations provided above, one can formulate an estimation of the volume component of the backscatter within the framework of the so-called semi-empirical model (Attema & Ulaby, 1978), as for the case of incomplete burial of the probing pulse with a flat surface, plane wave and a long probing pulse, which better fits the conditions for sounding of a thick snow slab with SAR.
The total radar backscatter from the illuminated area is composed of four components (Fung, 1994):
where σ as is the radar cross-section (RCS) due to backscatter from the air/snow (top) interface, σ s is the RCS due to backscatter from the snow volume, σ g is the RCS due to backscatter from the snow/ground (bottom) interface and σ gv is the RCS due to rescattering between ground and snow volume irregularities. As was summarized by Koskinen (2001), only the volume component is essential for dry snow. Therefore the essential portion of the total backscatter consists of the volume component and the contribution of the ground beneath:
Obviously, the radar becomes sensitive to the properties of the snow only when
This notation presupposes that the illuminated areas on the slab top and on the ground (bottom of the slab) are the same. This assumption applies for small angle divergences of the radar beam, as is valid for strong directed antennae, low refraction on the air-snow interface and a ground surface that is flat and parallel to the surface of the snow slab. Accordingly, the volume component of the backscatter coefficient is:
In the incoherent approach,
where the attenuation coefficient, inherent to the i-th particle with a distance of ξ i from the coordinate origin O, is equal to:
where k e is the extinction coefficient. The factor α takes into account the two-way distance of forth and back wave propagation. The assessment of the scattering volume described above relates only to the absolute value of the volume. In the case of wave directed propagation and, consequently, directed attenuation, the summation should be performed along the propagation axis, ξ. Because N>>1, the summation in (3.5) can be replaced by integration. As was mentioned above, this integration should be carried out along the direction of wave propagation, i.e., along the ξ axes. Small changes in the propagation direction at the air/snow interface are ignored due to minor differences in the corresponding refraction coefficients. Due to the random spatial distribution of scatterers within the scattering volume, this discussion considers only the mean backscatter characteristics averaged over several illuminated areas, with particles having independent spatial locations and RCSs.
where σ(ξ) is the running RCS (by the unit of a distance along axis ξ, dimension is
If the scatterers’ RCSs are independent of their locations within the scattering volume (condition of incoherent approach), we get:
where <σ> is the mean running RCS of the scattering volume.
Based on (3.4) and taking into account (3.9), the mean backscatter coefficient can be written as:
Taking into account that
Therefore, substituting (3.11) into (3.10), we can state:
Taking into account (3.8) and assuming the homogeneity of snow slab
This expression is the mathematical formulation of the semi-empirical model and has been obtained by Attema & Ulaby (1978) and Ulaby et al. (1982). In this paper, this model will be cited as the “A-U model.” For this case, the backscatter coefficient (3.13) depends only on the specific volume backscatter of the snow medium and not on the pulse volume sizes. This remarkable feature is due to (1) “overcomplete” burial of the pulse length into the snow medium, and (2) the backscattering normalization factor A ill (see 3.4) is the basis of the probing pulse.
The obvious imperfection of the model, as applied to sounding of thick snow, is the assumption of a constant extinction coefficient within the snow slab. The problem can be solved by designing an appropriate stratification model for the selected study area and modeling the spatial distribution of the extinction coefficient (e.g., Drinkwater et al., 2001).
The next limitation of the model is due to wave sphericity. Under depth-limited mode conditions (see section 2), the illuminated area changes, and the running RCS in (3.7) can not be assumed to be statistically homogeneous within the scattering volume. The same is true regarding the extinction coefficient in (3.8) as well. Thus, for this case, the A-U model in form (3.13) should be used with care. For the beamwidth-limited mode, the form in (3.13) can be used, taking in mind the “determined part” of the scattering volume (see comments to 2.15a).
The ground component of the backscatter coefficient is localized by the bottom location and can thus be expressed by its backscatter coefficient, taking into account the attenuation:
where
3.2. Effective snow depth
Given the expression
(3.13) can be presented in the simple form:
where H eff is the effective depth of snow sounding (EDS); i.e., the depth such that the backscatter from which occurs as if without attenuation, with backscatter equal to that which would occur from a slab of larger real depth with attenuation due to absorption and scattering.
Parameter
where
is the maximal value of the backscatter coefficient for a given snow depth, h, ignoring both energy losses due to attenuation and scattering and the angular dependence of the scattering volume. The normalized effective depth of snow sounding (nEDS≡
And the main formula of the model (3.2) can be written as:
It is useful to find the dependence of the correction factor on the SOT magnitude and the incidence angle. Let us consider two extreme cases:
a) “shallow” snow:
Given this relation,
meaning that the EDS is approximately equal to the geometrical snow depth. Nevertheless, the small magnitude of the volume component compared with the backscatter from the bottom layer (
b) “thick” snow:
Given this relation, and in accordance with (3.18) and (3.20), one gets:
where
To evaluate the angular dependence of the nEDS, we consider the normalized Snow Depth (nSD), which is the depth of snow normalized to the penetration path:
Comparing nSD with the definition of the SOT yields:
Substituting (3.26) into (3.19), we have
A plot of (3.27) reduced to zero dB at θ=0 for different nSD is shown in Figure 12.. As follows from this plot, the angular dependence of the volume component of the backscatter coefficient in a practically appropriate range of angles, 0-50 is rather weak and equal to ~2dB for the thick snow regime. For the intermediate regime, with nSD~1, its range of variation is about 1 dB. Because this range is of the same order as the errors, it is difficult to expect a notable angular dependence in practical measurements inherent to the intermediate regime.
Figure 12.
The angular dependence of the normalized effective depth of snow sounding for the “thick” (nSD>>1) and “intermediate” regimes
On the other hand, a notable angular dependence of the backscatter coefficient indicates that snow depth is greater than the penetration path. The intermediate regime represents practical interest for snow measurements. Suppose, for instance, the linear dependence of the geometrical snow depth on a horizontal distance S over the flat surface:
where k is the proportionality factor. Given that relation and taking into account (3.18) - (3.20), the normalized backscatter coefficient as a function of a distance is:
To specify the proportionality factor, let us assume that the geometrical snow depth reaches the propagation depth value at a distance, for example, equal to S
0
, i.e.,
Figure 13.
Illustration of the expected relative spatial behavior of the backscatter coefficient under a linear dependence of the snow depth on a distance
where
Thus, in accordance with the semi-empirical model, the expected profile of the backscatter coefficient should have an exponential pattern for a transect taken over terrain with a linear increase in snow depth within the intermediate regime condition.
4. Comparison the A-U model estimates and the experimental radar data on the Greenland ice sheet
For accurate comparison of the model with any experimental data, the study area must satisfy the “thick” snow condition to exclude the impact of the ground. Second, the snow pack must contain primarily Rayleigh ice particles with sizes less than approximately one-tenth of the wavelength. The most appropriate area matching this condition is the dry snow zone of the Greenland ice sheet. This region is not impacted by melt-freezing processes due to permanent sub-zero temperatures year-round. Therefore, the auspicious conditions for creation the ice lenses, tubes and large size aggregates, which inherent, for instance, to the percolation zone, are absent here. Additionally, the snow (firn) is up to several tens of meters deep (e.g., Zwally & Brenner, 2001), corresponding to the intermediate or even the thick snow regime. Due to known difficulties regarding in situ measurements in Greenland, there are no examples of the comparison of backscatter with snow depth in the literature for the intermediate or thick snow regimes.
Figure 14.
Transect across hill (straight line crossing the dark spot on the SAR signature) in the north-eastern region of Greenland (left), and the corresponding backscatter (solid line, convex downwards curve) and elevation (dotted line, convex upwards curve) profiles (right)
Nevertheless, a certain qualitative example of the sensitivity of backscatter to the snow depth relative to the intermediate regime can be provided based on a topographic feature (a hill) in the northeast region of Greenland between 81 and 82 N, as shown in Figure 14..
The counter-phase behavior of the backscatter coefficient and the elevation profiles plotted above can be explained supposing that the snow depth is minimal on the hilltop and increases proportional to the distance from the top of the hill down to its base. This assumption closely matches the field and modeled data provided by Jaedicke et al. (2000).
For quantitative assessment of the semi-empirical model more regular terrain should be chosen. A typical dry snow area is presented on the RadarSAT-1 SAR image of eastern Greenland in Figure 15..
As is clearly seen, the magnitude of the backscatter coefficient is about -20 dB in the dry snow zone. This assessment coincides with data in the known literature (e.g., Drinkwater et al., 2001, Partington, 1998; Baumgartner et al., 1999, Forster et al., 1999) which also gives values of -12…-20 dB. We can estimate the corresponding value using the A-U model. We assume that the thick snow regime is valid, and that the snow slab consists primary of particles of roughly the same sizes; i.e., it is a monodisperse medium.
For these conditions (see 3.24), ignoring backscatter from the bottom surface, we get:
Figure 15.
Illustration of the backscatter coefficient values (left) range for an arbitrary transect (straight line on RadarSAT-1 SAR signature, right) from the percolation to the dry snow areas within the east margin of the Greenland Ice Sheet (directions are from left to right for the plot of the backscatter profile and contra versa for the SAR signature; date of data acquisition: November 2, 2007; data granule ID: R1_62609_SWB_271)
Further calculations below aim to discover the specific volume backscatter, σ
v
and the extinction coefficient, k
e
. The estimates listed below will be conducted with the Rayleigh approach. In accordance with Ulaby et al. (1986), this approach bounds the upper electric size of particle,
where
In this formulae
where
For monodisperse media ( r i =r ) equation (3.3) has the form:
where n is the particle concentration.
The extinction coefficient can be calculated as a sum of the absorption (k a ) and scattering (k s ) coefficients (Ulaby et al., 1986):
where Q a and Q s are the absorption and scattering cross-sections respectively. These parameters, under the Rayleigh approach, can be calculated by the following formulas:
andIn accordance with (Ulaby et al., 1982)
The real part of the dielectric permittivity for ice, contained in the formulas above, is equal to 3.15 and in practice does not depend on temperature or wavelength. The imaginary part of the dielectric permittivity for ice is not constant with changes in temperature or the illumination frequency. In accordance with Matzler (1987) it can be expressed through the following empirical formula:
where F is frequency in GHz and A, B and C are the empirical coefficients. For a temperature of -15C, which is more appropriate for the dry snow case, the coefficients are equal to: A=3.5*10, B=3.6*10, C=1.2. Now, the expressions obtained for the backscatter coefficient calculation may be combined. Substituting (4.2) in (4.1), we get:
Taking into account (4.7) and (4.8)-(4.10), one can get:
Simplifying this expression, we have:
An important feature of this expression is its independence with regards to the particle concentration, a reflection of the property inherent to the saturation regime mentioned above in section III. Since the scattering volume is less than the spatial duration of the probing pulse an increasing particle number is equivalent to an increasing particle concentration. Thus, no additional amount of the snow over the snow pack with a thickness greater than the penetration depth can cause a notable increase in the backscatter coefficient. This circumstance is also very useful from a simulation point of view, since there is no microstructure parameter that needs to be assumed. For the numerical calculation listed below we assumed that the snow within the dry snow zone consists of snow grains that are actually grains of ice that look like grains of rice, with a typical size of ~1 mm. Taking into account the fact for ice: ε’=3.15 and ε’>> ε” (Tiuri et al., 1984, Matzler, 1987),
Taking into account (4.11) and (4.12), we get:
The plot of (4.17) is depicted in Figure 16. for C-band radar with F=5.3 GHz (λ≈ 5.6 cm).
Figure 16.
The dependence of the backscatter coefficient on particle size in accordance with the A-U model for C-band radar and snow depth greater than the penetration depth (“thick” snow regime)
It follows from this plot that the observable backscatter coefficient values of -12…-20 dB can be caused by particles with radii of less than 0.35-0.17 mm. The current estimates are slightly lower compared with the same made by Partington (1998), who found that the backscatter coefficient would range from -20 dB for a mono-distribution of grain sizes with a mean radius of 0.25 mm to -2 dB for a mean grain radius of 1 mm. Our results showed a 0.17 mm mean radius for -20 dB and a 1 mm mean radius for -3 dB. The small discrepancy is probably due to the temperature dependence of the imaginary part of the ice’s dielectric permittivity. The grain sizes values assessed do not coincide with the field data. In accordance with Table 2. the mean particle radius within a ~1 m depth surface layer of dry snow is 0.5…2.0 mm. As seen in Figure 16.., these values should produce a backscatter coefficient of approximately from -8 dB to 0 dB. Thus, there is a discrepancy of ~12 dB between the observable data and the values forecast by the semi-empirical model. Therefore, the A-U model overestimates the backscatter coefficient significantly. It should be noted, in addition, that the particle radius can only increases with the snow depth due to the depth-dependent grain radius model (Alley et al., 1982):
where r 0 is the mean radius at the surface, C is the crystal growth rate and H a is the mean annual layer thickness. Given this model, various simulation experiments (e.g., Forster et al., 1999, Drinkwater et al., 2001) and field measurements (e.g., Jezek et al., 1994, Woods, 1994; Lytle & Jezek, 1994), the mean particle radius for the entire scattering volume is even greater than the surface values and the aforementioned discrepancy becomes even more firmly grounded.
Table 2.
Thus, although the A-U model gives a qualitative assessment of the main features of backscatter behavior from a snow slab, the result of quantitative comparisons with field data does not match the theoretical predictions.
5. Enhanced semi-empirical model of the volume component of the backscatter coefficient
The problem highlighted in the previous section can be resolved by considering the statistical properties of the small-scale fluctuations in the particle concentration and its scattering properties within the scattering volume. The appropriate method for doing so that is the so-called “slice” approach, which was primarily suggested in weather radar meteorology (Marshal & Hitchfeld, 1953; Smith, 1964). This approach exploits the known radar feature in accordance with which the particles of a “cloud” located close to the front of the incident radar wave are considered to be approximately at the same distance from the radar and reflect incident electromagnetic wave almost coherently. One can consider that these particles are embedded in a fictitious thin cylindrical volume (“slice”), whose base coincides with the surface of spherical wave front and side-bounded by the main lobe of the antenna pattern. Thus, the scattering volume into a snow slab can be represented as an adjoining series of these slices, as illustrated in Figure 17.. Each slice is much narrower than the radar wavelength in the wave propagation direction (Δs<<λ).
Figure 17.
Simplified slice approach scheme. The arbitrary position of a single slice is shown on the cross section of the scattering volume
A slice’s radial size, Δs can actually be considered as a minimal spatial scale of the backscatter property fluctuation, if this scale is much smaller than the wavelength. Using this approach, it was shown by the author (Yurchak, 2009), that the specific volume component of the backscatter from any spatial extended geophysical target (SEGT), included the snow as well, can be presented as the incoherent sum of the radar cross-sections of individual scatterers only if the number of particles in the slices (n) are distributed in accordance with the Poisson law; i.e., the variance of the number of particles (Var(n) ) is equal to the mean number (
where
andFollowing Shi et al. (1993), if the PSDF for snow can be presented by the lognormal distribution, the parameters of relationship (5.3) are:
Figure 18.
Deviation factor versus particle radius variation coefficient for different deviations of Poisson index χ from the Poisson law
As can be concluded from analysis of the plot, the deviation of ~12 dB emphasized in the previous section can be explained by the homogeneous microstructure of the dry snow zone, with approximately identical particle sizes (ξ
r
<0.1) and negligible variations of the particle concentration (χ<<1) within the scattering volume. As an example of data supporting the current considerations, the results of comparative measurements of C-band backscatter from firn and in situ measurements of statistical parameters of firn dielectric permittivity for a study area in Antarctic (Zahnen et al., 2002) can be taken into account. In this study, dielectric permittivity was considered to be a measure of the heterogeneity of the snow/firn pack. A notable positive correlation was observed between backscattering signal strength and the standard deviation of the dielectric constant collected along 4 and 15 meters depth interval of several drills within a dry firn area. At the same time, a negative correlation takes place for a mean dielectric constant. Because the dielectric constant is approximately linearly proportional to the density of dry snow (e.g., Tiuri et al., 1984), and therefore linearly proportional to the number of ice particles per unit volume as well, these correlations match the behavior of the deviation factor (5.2) on the Poisson index. The index and, hence, scattering increase with increasing variations in particle number and decrease with increasing
The considerations provided above are related to the simplest case of complete burial of the probing pulse into a snow slab in the beamwidth-limited mode. Obviously, that the result is still the same for incomplete burial of the radial size of the scattering volume. The transverse size change of a slice within the scattering volume in the depth-limited mode results in an additional gradient in particle number among slices, and the contribution of this additional gradient to the backscatter should be analyzed in the future.
6. Normalized snow depth
6.1. Derivation of the normalized snow depth from the backscatter coefficient based on the enhanced semi-empirical model
After enhancing the semi-empirical model by the “slice” approach, one resume assessment of the snow depth, the key parameter in all mass-balance studies. To derive this parameter, let us to write down the enhanced semi-empirical model (ESEM) in a form more fitting for the Greenland ice sheet:
where
Here,
The equation above indicates that the estimated normalized snow depth depends on number and kind of scatterers (σ
∞
) and their statistical characteristics (Y) at small (less then wavelength) scales. Thus, to extract the nSD from measured radar data,
Figure 19.
Typical backscatter coefficient profile across a transect within the margin of the Greenland Ice Sheet (in the vicinity of the Swiss camp)
be underlined, additionally, that the nSD
From the point of view of practical measurements, detailed analysis recommends stopping the nSD estimate at a distance when
Figure 20.
The normalized snow depth profile derived from the backscatter coefficient profile in Figure 19.
The profile has an exponential-like increase at the beginning of transect that transitions into a plateau-like, approximately constant level. Most probably, this level represents the saturation level caused by thick snow of depth greater than the propagation depth. The corresponding nSD is shown in Figure 20.
The nSD profile plotted in Figure 20. is obtained under the assumption that the deviation factor and the extinction coefficient are constant across the entire profile distance. Although the approach demonstrated here does not give the absolute value of the snow depth without certain assumptions, its advantage is that the assumptions are clearly delineated and put forth. In turn, if one has a calibration point (spot), the technique allows extension of the remote sensing methodology on vast remote areas with similar electromagnetic properties. In addition, the nSD is a relative indicator of the snow environment stability and redistribution.
The nSD in form (6.3) is derived from the ESEM, which in turn is based on consideration of a flat surface, plane wavefront and long probing pulse. Hence, the result obtained is most appropriate for SAR sounding of a thick snow slab.
6.2. An example of time transformation of nSD profiles within a selected Greenland marginal area
The normalized snow depth is sensitive to the geometrical snow depth and to the extinction coefficient. Assuming the snow microstructure to be homogeneous within a limited area, one can suppose that nSD depends only on the snow depth and thus apply nSD for snow mass-balance assessment. An illustrative example of an nSD profile change after a 10-year period is shown in Figure 21..
Figure 21.
Study area #3 (box with center: 75.07N and 54.30W) within the Greenland ice sheet margin with several transects normal to the coastline (left). Normalized snow depth profiles (right) for transect 3a (lower line in the study area) derived from RadarSAT-1 data. Distance scale: 1 sample=50 m. Solid line is nSD profile on 11/16/1997, dotted line is nSD profile on 11/12/2007
The nSD profiles were derived from the RadarSAT-1 data archive from the Alaska Satellite Facility in Fairbanks (granules R1_10617_SWB_261 from 1997 and R1_62753_SWB_261 from 2007 with coordinates of the center scene 75.23N/54.29W and 75.22N/54.37W respectively, descending modes for both). As follows from the center locations of the data, the SAR shots were performed from practically the same point in the orbit and thus can be analyzed in parallel without additional corrections.
A comparative assessment of the nSD profiles indicates notable differences between the profiles. Generally, these changes can be caused snow mass redistribution within the 10-year period, variation of the extinction coefficient due to possible snow metamorphism, or both factors simultaneously. The detailed analysis of these factors and their “weights” in the deviations of these profiles is beyond the scope of this work but is planning for further consideration.
7. Model spectral dependence of the backscatter
7.1. Spectral dependence of the backscatter in accordance with the A-U model
The spectral (frequency) dependence of the backscatter coefficient within the framework of the semi-empirical model (A-U and ESEM) is the topic of the current section. The aim of this discussion is to find out how spectral features depend on the thick snow properties. Particularly, Davis & Moore (1993), namely assumed that a two-frequency radar system can provide quantitative estimates of snow physical properties.
The model consists of two components that are definitely frequency-dependent: the specific volume component (4.2) and the normalized effective depth of sounding (3.19). Transforming the wavelength (λ) into the frequency
The frequency-dependent core of (7.1) is
where
7.2. Estimation of spectral differences based on field penetration depth data
For further analysis, it is convenient to deal with the dimensionless spectral function, which can be obtained by multiplying the spectral core (7.2) by the cube of the wavelength:
To evaluate (7.3) the spectral dependence of the penetration path (depth) should be estimated. One can state that the high frequency irradiance interacts primarily with the surface layer of snow with the relative small particles. And the low frequency irradiance penetrates deeper and, thus, is affected by particles which sizes are larger in accordance with the grain growth model (e.g., 4.18). For example, the grain radius changes from 0.25 to 0.6 mm within the depth of 10 m is in model of Drinkwater et al., (2001). In model of Forster et al., (1999) the range of the particles’ changes is even more (up to 1 mm) and additionally depends on the accumulation rate. Thus, to get more realistic spectral dependence, we should take into account the depth of penetration of the electromagnetic wave into the snow and the corresponding grain size change within this depth. Therefore, there is an obvious paradox: to calculate frequency dependence of the penetration depth we should know the latter a priory. Moreover, due to chain: “frequency-penetration depth-particle size” the mean particle radius impacted the radar irradiance turns out to be dependent on the irradiance frequency. In additional, the A-U model was derived assuming the constant extinction coefficient, as it was mentioned in section 3. The solution of these questions should be evaluated in further research. Taking in mind the restrictions listed above, the penetration depth is assumed to be dependent only on frequency in the following estimation. Since the significance of the particles’ size change compared with the fixed mean size in the extinction coefficient calculation is currently unclear, it is reasonable to take the field data available in the literature (Table 3.) for rough assessment the spectral dependence of the penetration depth.
Table 3.
Literature data on the experimental penetration path (depth)These field data can be approximated by a simple empirical power function
with the coefficient of determination equals to 0.9. In (7.4) the penetration path is in meters and frequency is in GHz. It should be noted that field data, provided in the first three rows of Table 3., are obtained based on measurements within the surface layer of the snow (up to 3 m). It means that these data also do not take into account adequately the actual stratification of the thick dry snow cover of the Greenland ice sheet and, therefore, the further estimation should be considered only as a first approximation. A plot of (7.3) is shown in Figure 22. This plot demonstrates a notable discrimination in behavior of the spectral function B(F,h θ ) for different snow depths. Particularly, the magnitudes of the curves at a frequency of 5 GHz are notably different. Differences of the backscatter magnitude for frequencies of 5 and 14 GHz as a function of the slant snow depth are depicted in Figure 23.
Figure 22.
Dimensionless spectral characteristics of the backscatter for several values of the slant snow depth based on the empirical spectral dependence of the penetration depth
Figure 23.
Slant snow depth as a function of the magnitude difference between the dimensionless spectral components of the backscatter coefficients at 5 and 14 GHz
As follows from the approximation results,
Figure 24.
Spectral pattern of the experimental data of (Baumgartner et al., 1999) for an incidence angle of 40 for two test sites in Greenland
where the slant snow depth is in meters and the spectral difference
If, for example, F 1 =5 GHz and F 2 =14 GHz, we get:
Differences in the backscatter coefficients of these frequencies for both test sites are approximately the same and equal roughly to -13 dB, as shown in Figure 24.. Assuming the backscatter error of ~±0.5 dB, the corresponding spectral difference (from 7.7) is approximately equal to 0.4±0.5 dB. In accordance with (7.5), the slant snow depth is ~6.7-20.8 m. Since these data are related to an incidence angle of 40, the estimate of actual snow depth is 5.1-15.9 m. This value is in accordance with a note in the paper that the depth of snow was ~7-8 meters for the NASA-U site, at least.
Thus, a rough assessment of the snow depth within several selected dry snow areas, based on the spectral dependence of the backscatter in the frames of the SEM (ESEM), gives results comparable with in-situ measurements.
7.3. Spectral characteristics of the penetration depth for model monodisperse cold snow
It is interesting to compare the empirical (7.4) and theoretical frequency dependence of the penetration path (depth). Calculation of the extinction coefficient has been completed in the same manner as in section 3, with the additional estimation of particle concentration n (see (4.8)):
where
In Figure 7.5, the results obtained from these calculations for T= -15C and the snow density of 0.3 g/m are compared with the known values (Ulaby et al., 1986, Fig. 19.64, p.1608) for T= -1C and density of 0.24 g/m. The plot indicates the increasing of the penetration depth at low temperatures for the low frequency region, a trend most obvious for small particles of 0.5 mm radius. This behavior is explainable because the losses in snow caused by absorption decrease as the temperature decreases (Ulaby et al., 1986, Fig.Е.3, p. 2027), and the absorption mechanism prevails over scattering in the low frequency region. Pointed changes become more apparent for small particles due to more significant role absorption plays for them as compared with large particles.
Figure 26.
Comparison of the spectral dependence of the penetration depth calculated with Mie formulas for a temperature of -1С (Ulaby et al., 1986) with our calculations under the Rayleigh approach for a temperature of -15С
8. Backscatter depth profile model for snow sounding with broadband pulses of short effective (compressed) length
8.1. Phenomenological approach in the case of very short probing pulse
The above considerations apply when the scattering volume was less then the probing pulse volume. In the case of altimeter sounding of thick snow cover with short broadband chirp pulses this condition is not satisfied, the backscatter coefficient is represented not by the point of a radar signature but in the form of a backscatter profile along the wave propagation path. When
where
where
8.2. Analytical derivation for commensurate length of the scattering volume and snow depth
In general case, when
where P r (t) is the return signal, P T (t) is the transmitted pulse profile and P s (t) is a term that includes the distribution range of the scattering facets, their scattering properties and the effects of the antenna gain. As applied our consideration of volume backscatter, one can write
If, for example, the probing pulse has a Gaussian form, which is usually an adequate approximation for short pulse systems (e.g., Brown, 1977), the weighting function of the scattering volume is
For short pulses, if their envelope can be represented by the delta-function, i.e.,
Taking into account the tabulated integral (Abramowitz & Stegun, 1972):
where
If one denotes
Finally, if one denotes:
Distance normalization in (8.9) is chosen through the parameter S for the purpose of comparing the backscatter profile calculated here with the modeled “clear” backscatter profile derived for a very short pulse (8.2) and governed only by attenuation. The equation obtained is close to the volume component of the SV model of Davis & Moore (1993). The distinctions of the current approach are that it uses a slightly different description of the pulse-snow interaction, the end result is written in terms of the backscatter coefficient depth profile, and the state of homogeneity of the snow mass is taken into account through the deviation factor, Y. It is interesting, that equation (8.9) coincides with the backscatter model from the sea surface, obtained by Barrick (1972), exactly by the form with, of course, other meaning of model parameters.
8.3. Comparison of the results obtained with the phenomenological approach
8.3.1. Very short probing pulse
For the limit case when
This result confirms relationship (8.2) obtained above based on phenomenological considerations.
8.3.2. Long probing pulse
Taking into account the asymptotic value of the error function
For long pulses, the available distance variation is much less than the pulse length, i.e.,
This expression also confirms the result of the A-U model (enhanced) under thick snow condition (3.24). Thus, an important positive feature of the model (8.9) is that its results coincide with those obtained with phenomenological approaches for extreme values of the scattering volume (much less and much more than the penetration path respectively). Nevertheless, it should be noted that ignoring the change in the illumination area during the burial of the probing pulse into snow media is, strictly speaking, valid only for a short probing pulse or for the beamwidth-limited mode. For long and intermediate-length probing pulses with wavefront sphericity in the commonly-existing depth-limited configuration, this effect should be taken into account because there are slices (see sections 2 and 5) of different transverse sizes within
8.4. Numerical calculation of the return signal for intermediate length of the probing pulse
A plot of the normalized backscatter coefficient
Figure 27.
Illustration of the reproducibility of the snow backscatter uniform profile taking into account only attenuation (bold dash line) and finite probing pulse length normalized to the penetration depth (parameter
Legend: solid line, γ=0.01; dot line, γ=0.035; dashed line, γ=0.06; dashed-dot line, γ=0.085; dashed-dot-dot line, γ=0.11; long dashed line, γ=0.135
This figure illustrates the contribution of the pulse length to the return signal envelope pattern. The shorter pulse clearly gives a better approximation of the real backscatter profile. Contemporary altimeters use a probing pulse with a physical duration of 20-50 microseconds and a corresponding effective duration (due to intrapulse frequency chirping) equal to 3.125 nanoseconds (e.g., Rosmorduc et al., 2006). In spatial units, these values are equivalent to
9. Summary
The semi-empirical model of the volumetric component of the backscatter coefficient, proposed by Attema & Ulaby (1978) (SEM or A-U model) is considered in more detail. The basic advantage of the SEM is its mathematical simplicity and clear physical sense but the SEM, derived using the incoherent approach, does not explain some remote sensing experimental data collected over the dry snow zone of the Greenland ice sheet. We discuss the main configurations of radar sounding of snow, including the relationships between the technical and positional parameters of radar on one hand and the geometrical and electromagnetic properties of snow cover on the other. Based on this consideration, the field of applicability of the SEM is figured out for the further assessments. Some qualitative assessments of the configuration of the scattering volume were carried out using plane and spherical surface approaches. More detailed study of the diffraction of space-bounded spherical electromagnetic wave on a large and strongly absorbing dielectric sphere with a rough surface is needed to evaluate this problem further. Detailed SEM estimates and experimental radar data over the dry snow zone of the Greenland ice sheet are provided for the negative temperatures inherent to the winter seasons. Data analysis shows a significant discrepancy between the expected and actually measured values of the backscatter coefficient. Possible reasons for the observed deviations of the SEM results are discussed based on the enhanced semi-empirical model (ESEM) and analysis of the dependence of the backscatter coefficient on the snow optical thickness and its spectral function. The ESEM is obtained on the basis of the so-called "slice" approach, taking into account partially-coherent backscattering due to finite wave front size and considering the contribution of the statistical characteristics of small-scale fluctuations of concentration and of sizes of the ice particles in a snow pack. The known semi-empirical model is only a special case of the ESEM. Using the ESEM, some observable radar signatures of the Greenland ice sheet, including the dry snow zone, can be interpreted in terms of the statistical characteristics of the snow microstructure. The model obtained retains the mathematical simplicity and physical clarity of the known A-U model and significantly expands its application for the interpretation of experimental data. Available field data, which confirm the consistency of the ESEM with observed values of the backscatter coefficient, are also provided. As a consequence of this analysis, the concept of the normalized snow depth (nSD) is formulated, and its practical application for monitoring snow mass displacement is described. The frequency-dependent core of the backscatter has been analyzed as well. Empirical and theoretical spectral characteristics of the penetration depth are also presented. The result allow for improvements in our understanding of opportunities for a two-frequency method of sounding thick snow covers and promote more accurate interpretations of the known experimental data. The extension of the ESEM for sounding thick snow cover with broadband pulses of short effective length is also discussed, resulting in an exact description of the return signal envelope as a function of the electromagnetic and microstructural characteristics of the snow cover. The validity of the model is supported by the coincidence of the model estimates with assessments obtained for several specific cases using phenomenological approach. In conclusion, this work demonstrates that the enhanced version of the semi-empirical model can be applied successfully in the analysis of radar backscatter from thick snow cover.