Geoscience Reference
In-Depth Information
Density and temperature were drawn directly from the field measurements, while the mean
of the minimum and maximum extent of the average common grain was taken as the grain size,
following Davenport et al. (
2012
). Grain sizes were reported by the observers by stratigraphic
layer, rather than at 10-cm intervals, and these were mass weighted onto the 10-cm profile.
As LSMs commonly feature a thin top layer to improve numerical handling of the
surface energy exchange, snowpits were resampled from the CLPX observational data with
and without a 2-cm surface layer. This did not affect any of the key results, and so
arbitrarily the case with a 2-cm surface layer is presented.
The snowpits were then resampled to profiles of 1-5 equally sized layers subject to a
minimum layer size of 10 cm, with the mass-weighted grain size, temperature and density
from the relevant observational layers applied to each of the resampled layers. Figure
4
illustrates sample layering profiles, where the 2-cm top layer is excluded from the layer
count, and for a snowpit of depth d, each of the n main layers is of depth
ð
d
2
Þ=
n cm.
The minimum layer size criterion meant that, for example, a snowpit of 12 cm depth would
be identical in all layer cases and consist of a single 10-cm layer and the top 2-cm layer,
even in the 5- or N-layer cases.
This approach attempts to mimic a depth structure that might be output by an LSM. A
key feature is the prescribed layer thicknesses, as each layer depth can be determined
uniquely from the total snow depth, thus removing the need for layer thickness components
in the snow state vector. However, it is not necessarily representative of an individual LSM
snow scheme, as a variety exists, and rather than select some arbitrary combination of layer
sizes for each of the 1- to 5-layer schemes, a more simplistic approach was adopted.
Ground surface temperature was taken to be the temperature measured at 0 cm height.
Missing data were linearly interpolated, or if they were at the top or bottom of the pack,
then the nearest neighbouring value was used. If too many data were missing for this
interpolation, then the pit was removed from the analysis.
4.2 Comparison: Layered HUT Scene Simulation Versus Observations
The HUT performance was first assessed by simulating the scene brightness temperatures
based
on
the
snowpit
information,
and
comparing
these
simulations
with
satellite
Fig. 4 Example of the how snowpit data were restructured. The left-hand bar represents the observed
profile where depth and temperature are recorded for each 10 cm of the snow. The N-layer resampling
maintains 10 cm layer thicknesses but adds a 2-cm interaction layer at the surface, as is common in a
number of land surface models' snow schemes. The other layering schemes apply a 2 cm top layer and then
evenly split the remaining snow depth, with density, snow and grain size mass-weighted according to the
observations. All layer structures from 1 to 5 inclusive were calculated, but only 1 and 5 are shown here for
simplicity
