:'kscatter. In 
lating back- 
cubic poly- 
r coefficient 
a mid-range 
1eterisations 
ser et al, in 
\-SZR_1B”) 
(http://www. 
projection, 
. The study 
:ycle of the 
possible az- 
stable, well 
a thirty day 
/ 240, 2010 
ter, azimuth 
MS error in 
1g was per- 
1, using the 
, Written for 
te. This re- 
|. Fits were 
/ B4 (cubic) 
nts of SMB 
~95 - 130° 
? -65? E). 
"ham et al., 
'averse sec- 
een deploy- 
years (Nov- 
:t al. (2007) 
ats of SMB 
onducted in 
her 3 years 
the Wilkes 
imately ev- 
as the LGB 
ur, whereas 
southward 
d in Figure 
v stake data 
where snow 
km interval 
1e LGB tra- 
ven (1997), 
| filter of 30 
atial noise. 
 scatterom- 
id dry snow 
nSMB = 
nponent of 
ter on inci- 
isation), or 
n incidence 
      
        
   
Glacier Basin 
East Antarctic 
traverse Front 
Ice Sheet 
Wilkes Land 
traverses "West" 
"South" 
  
     
Figure 1: Australian traverse routes for measuring SMB using 
snow stakes. 
angle for the cubic incidence angle parameterisation). These three 
plots are shown in Figure 2. 
All three sub-figures show a relatively robust empirical relation- 
ship between SMB and A, B or B;. Table 1 shows that the order 
of increasing RMS residual is South, LGB, East, West, regardless 
of the parameter being fitted against SMB. Analysis of individ- 
ual, unfiltered snow stake measurements suggests that much of 
the West traverse route crosses a region where accumulation rate 
is low and surface wind glaze conditions are prevalent. Similar 
conditions were also encountered throughout the LGB traverse, 
to a lesser extent (Higham and Craven, 1997), though the lower 
residual can likely be attributed to lower values of SMB encoun- 
tered here. 
Drinkwater et al. (2001) discovered SMB had a stronger rela- 
tionship with the B parameter than with the A parameter, i.e., 
the incidence angle anisotropy is more sensitive to changes in 
SMB than the isotropic component. In the present study, we 
find a lower RMS residual in the empirical fit between SMB 
and A (0.20 m accumulation/annum), than with SMB and B 
(0.24 m/annum) or B; (0.27 m/annum). We note that in the 
Greenland dry snow zone, there are annual layers that exhibit a 
strong dielectric contrast, likely a result of autumnal hoarfrost 
formation (Hawley et al., 2006). Complex interactions between 
precipitation-bearing synoptic systems and local orography are 
important along the East Antarctic coast (e.g., a large snowfall 
gradient exists between the eastern and western sides of Law 
Dome, centred at ~112.5° E, 66.5° S), and may explain the wider 
range of observed A and B parameters in the Wilkes Land West 
traverse than the East. Furthermore, in regions of persistent wind 
glaze and low accumulation (e.g., parts of the Wilkes Land West 
traverse route), annual dielectric layers may not exist in the snow/ 
upper firn. Thus, the relationship between incidence angle aniso- 
tropy and SMB likely varies regionally, particularly along the 
East Antarctic coast. In such regions, it comes as no surprise 
that the A parameter gives a better indication of SMB than the 
B parameters, since A is a strong function of near-surface snow 
grain size (Ulaby et al., 1996), and in the absence of accumula- 
tion, grain size metamorphism/growth continues to occur at the 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
  
  
  
  
  
  
  
  
  
  
  
    
  
     
  
  
  
  
  
  
  
  
  
T T T T 
oid à Wilkes Land "West" traverse TA 
E % Wilkes Land "East" traverse 1 
2 Wilkes Land "South" traverse 7 
s A Lambert Glacier Basin traverse LGB00-35 | | 
E | 
-6.62-0.35A 
S SMB-e + 
wn 4 
9 J 
E J 
Uu 
z A 
o A 
ë 
wn ^: À 
A 4 
zm- 4-1 
A L 1 
-20 -18 -16 -14 -12 -10 
"A" parameter (isotropic component of backscatter, dB) 
TY Ta T T 
ST Standard deviation of * 
E 5 measurements within 4 
= L a satellite pixel (20) À 
£ E 
S L A 
= 
E 4 
1.91 m = 
= À EN 
e T 
4 3l 
x L 4 
© 
= 
| -9.10-32.78 em 7 a 1 
z — =a t 36cm 34cm 7cm cm |-] 
2 95 SMB=e à | West East South LGB 
“i | à. a 4 
[ b) T A à à 1 
| Bp adh Ca 1 
DO x . aet : mie eine 
0.35 —0.30 70.25 —0.20 -0.15 
"B" parameter (slope of backscatter vs incidence angle, dB/°) 
X T T T T 
15r ; 
E - 
3 L J 
c 
c L A 
© 
= TE + 
E 
m db zl 
aea 
€ L 4 
uo L + 
o 
x | 4 
© 
M E 
a -4.19-14.3B1 
2 05[ SMBze^ ^ À 
o 4 
2 L 
mn L J 
Lc) zd 
0.0 A 1 = 5 
-0.35 —0.30 -0.25 —0.20 70.15 
"B1" parameter (linear component of backscatter 
dependence on incidence angle, dB/°) 
Figure 2: a) SMB vs “A” parameter for the linear parameterisa- 
tion (cubic is very similar, so not shown here). b) SMB vs “B” 
parameter for the linear parameterisation. c) SMB vs *B; param- 
eter for the cubic parameterisation. The least-squares fit using a 
model of the form SM B — e^-"? is shown with the dashed line, 
where x is the “A” parameter for panel a), “B” for b), and “B1” 
for c), after Drinkwater et al. (2001) in Greenland. 
surface, giving higher backscatter at sites of lower accumulation. 
The addition of the 4th order Fourier term to the linear azimuth 
angle parameterisation results in a statistically-significant reduc- 
tion in curve-fitting residual from 0.66 to 0.46 dB, averaged over 
the whole AIS, using 30 days’ ASCAT observations. Addition 
of the cubic incidence angle parameterisation results in a further 
drop in residual fit (from 0.46 to 0.43 dB), again statistically- 
significant, as shown by an F-test. Comparison between the lin- 
ear and cubic incidence angle parameterisations (Figures 2b and 
c) reveals that while the residual is slightly lower for the linear 
incidence angle parameterisation case (except for the South tra- 
verse), there is also less sensitivity to differences in SMB in the 
linear case. This is encouraging from a SMB retrieval perspec- 
tive. This parameterisation, and others using a cubic incidence 
angle parameterisation, will be compared in a forthcoming paper 
(Fraser et al., in prep.).