Geoscience Reference
In-Depth Information
forming ice lenses which increase the effective grain size and cannot be accounted for in
the Chang-based approach which assumes a static snow microstructure.
However, a number of questionable assumptions remain in the Globsnow approach. The
assumption of constant density is not necessarily valid, as snow settles and increases in
density during the season due to metamorphism and overburden (Anderson
1976
), and
variations in density or in the effect of vegetation are included through varying the grain
size parameter which is an unphysical approach. Furthermore, the snow depth forecast is
produced purely from interpolated observations, which as previously noted are biased
towards low latitudes, low altitudes and clearings in forests. These biases could be
accounted for by a LSM, which in addition to producing a snow depth forecast could also
produce a forecast of the density and grain size.
Durand et al. (
2009
) showed that even a relatively simple land surface model coupled to
a microwave emission model improved snow depth estimates once microwave brightness
temperatures were assimilated. The later work of Brucker et al. (
2011
) and Toure et al.
(
2011
) coupled the snow model Crocus (Brun et al.
1992
) to the MEMLS radiative transfer
model and found that point observations of microwave brightness temperatures at both
H-pol and V-pol were generally well simulated. Brucker et al. (
2011
) noted that late season
grain growth was not well modelled in Crocus, and Toure et al. (
2011
) indicated that ice
lenses must be accounted for. On a larger geographical scale, Dechant and Moradkhani
(
2011
) reported that assimilating brightness temperatures with the SNOW-17 snow model
and a soil moisture model showed potential benefit for operational stream flow forecasting.
Naturally, increasing physical complexity leads to increased computational expense and
computational expense is also affected by the number of layers in the snow model. Most
land surface models typically limit the number of snow layers, with the ECMWF's Tiled
ECMWF Scheme for Surface Exchange over Land (TESSEL) limited to one layer, whereas
the Joint UK Land Environment Simulator (JULES) can run up to a user-defined number of
layers, with new layers only introduced beyond certain thickness thresholds.
In reality, snowpacks almost always have distinct physical layers and this stratigraphic
contrast can have important effects on the radiative transfer. Lemmetyinen et al. (
2010
)
compared passive microwave measurements taken in situ with layered snow information
and found that the simulated brightness temperature was affected by whether or not snow
layering was included. These approaches used field-observed layer properties, but did not
consider how the radiative transfer model would perform if provided with profile infor-
mation as it would be output by a model.
This is assessed here through the experiments detailed in Sect.
4
, where measured snow
profiles from CLPX snowpits are resampled to differing layering structures. After a scene
simulation experiment to confirm that the HUT radiative transfer model is able to repro-
duce observations within acceptable uncertainties, the simulated DT
B
;
V
values for each
snowpit when resampled to different layering structures are compared. This comparison
across a large number of snowpits allows estimation of the bias and variance introduced
when the layering structure is simplified.
4 Methods
4.1 The Cold Land Processes Experiment (CLPX) Resampled Snowpits
The CLPX dataset provides snow profiles from a large number of snowpits over four
intensive observation periods (IOPs). Two of these periods, IOP3 and IOP4, coincide with
