Geoscience Reference
In-Depth Information
modelled by a number of higher resolution regional climate models (RCMs) with snow
telemetry observations in the Upper Colorado River Basin. Though the RCMs remained
too warm and dry with too little SWE, their simulations better matched observations than
the original reanalysis. However, biases remained even with these more computationally
expensive models.
2.2 Products Which Prioritise Snow
2.2.1 Surface Observations Only
Outside of reanalyses that attempt to produce complete time series of land and atmosphere
properties, a number of snow-specific products have been developed. The simplest
approach is to grid weather station snow depth records as performed by Dyer and Mote
(
2006
) in North America and Kitaev et al. (
2002
) in the Former Soviet Union (FSU). In
Kitaev et al.'s work, station number varied from 2 to 25 per 3 9 5 grid square and
snow's spatial variability within such areas means that large uncertainties are associated
with such sparse measurements. Chang et al. (
2005
) estimated that across the Northern
Great Plains, 10 measurements were required per 1 cell to reduce sampling error of snow
depth to ±5 cm, equivalent to a station density between 6 and 75 times higher than
available to Kitaev et al. Furthermore, the sampling distribution of snow stations was found
by Brasnett (
1999
) to be biased to low elevations.
Post hoc assessment of snow mass is possible using river discharge data, although this
approach suffers from large uncertainties due to unknowns related to inter-annual terres-
trial water storage, periods of river ice and non-snow contributions. This approach has
allowed attempts to test seasonal estimates of snow mass in some basins (e.g., Grippa et al.
2005
; Rawlins et al.
2007
; Yang et al.
2007
) and to provide evidence in conjunction with
other snow products for intensification of the Arctic hydrological cycle in response to
global warming since 1950.
2.2.2 Land Surface Models Assimilating In Situ Observations
Simple areal averaging of snow depth observations cannot account for variation in areas
between point measurements, which can be driven by different elevation, meteorological
regime or land surface category. LSMs featuring a snow component are in principle able to
account for these effects and, furthermore, these models may assimilate measured snow
depths when available to improve the analysis.
Brown et al. (
2003
) used the Canadian Meteorological Centre's analysis scheme
developed by Brasnett (
1999
) to generate a gridded time series of North American snow
depth and SWE. A simple snow model was driven by meteorological data from the EC-
MWF 15-year Reanalysis (ERA-15), with assimilation of 8,000 snow measurements per
day from the USA and Canada. This method relies on relatively intensive daily mea-
surements, for which the authors noted that availability drops off rapidly poleward
of 55 N.
A global estimate is published by the ECMWF using a similar approach, and a summary
and assessment is provided by Drusch et al. (
2004
). They note that the observational
stations are biased towards lower latitudes and lower elevations and that, without assim-
ilating remotely sensed information on snow-covered area, there are disagreements
between the estimated snow-covered areas, and from the Interactive Multisensor Snow and
Ice Mapping System (IMS) described in Ramsay (
1998
).
