Geoscience Reference
In-Depth Information
the percentage of snowpits that were noted by fieldworkers as 'wet' in the metadata did not
exceed 3 % in either period.
4.3 Comparison: N-layer Versus Fewer Layers of Stratigraphic Information
Using the same resampled snowpit data to represent realistic profiles as might be output by
an LSM, the DT
B
;
V
values simulated by the 1- to 5-layer simulations were assessed relative
to the N-layer simulations, which were assumed to be truth.
Uncertainty introduced into the DT
B
;
V
by simplification of the model layering was
determined from the difference between outputs for each of the 1- to 5-layer models versus
the N-layer model. Bias and standard deviation of these residuals is reported in Sect.
5.3
for each of the simpler models as a function of the model layer thickness.
5 Results and Analysis
5.1 Snow Properties at CLPX Sites
Table
3
summarises the main snow properties recorded from the snowpits and MODIS data,
including the number of relevant snowpits once those with insufficient data were deleted.
Notably, the snow in IOP4 during March was thicker than during IOP3 in February, although
the snow cover fraction had declined from universal coverage to around 80 %.
It should be noted that the average depth and SWE is not necessarily a good repre-
sentation, as the distribution of snowpit values is not symmetric, with a bias towards thin
pits in IOP3 and a bimodal distribution in IOP4, with a number of pits showing thin snow
(\50 cm) and thicker snow (*200 cm). The overall distribution is shown in Fig.
5
although thicker snow predominated at Rabbit Ears and Fraser and thinner snow at North
Park. Additionally, IOP3 saw generally cooler snow (-3.6 vs. -2.2 C) and marginally
smaller average grain sizes (0.57 vs. 0.60 mm).
5.2 Simulated Scene Brightness Temperatures
Figure
6
shows the simulated scene DT
B
;
V
from Eq. (
15
), using the N-layer and 1-layer
HUT model compared with AMSR-E and SSM/I retrievals. Here, the scene is represented
by the average of all 3 MSAs. The difference between the N-layer and 1-layer simulations
is minimal (0.01 K in IOP3, 0.40 K in IOP4) compared with the difference between
simulations and observations, of approximately 3 K in IOP3 and 2 K in IOP4.
Table
4
shows the brightness temperature difference simulated using different model
layering profiles versus the observations. There is a negligible difference in the mean
simulated by different layering profiles. The overall Chang estimates are close to obser-
vations at IOP3, but are too high during IOP4. The Chang algorithm's poorer performance
at individual MSAs (RMSE = 17 K) versus HUT (RMSE = 8 K) is hidden by the aver-
aging over the 3 MSAs. During IOP3, use of the Chang algorithm results in a large DT
B
;
V
overestimate at Rabbit Ears MSA, which is counteracted by a large underestimate at North
Park MSA. During IOP4, a very large (32 K) overestimate by the Chang algorithm due to
saturation in the deep snowpits is partially offset by a 13-K underestimate at North Park.
The largest contributor to the HUT RMSE was due to a large underestimate during IOP4
at North Park, where simulated DT
B
;
V
were of order 1 K versus observed values of 14 K.
