A Review of Spatially Distributed Modeling of Snow
Robert E. Davis1
Spatially distributed modeling of snow has a wide variety of applications, from providing lower
boundary conditions and elementary hydrology for atmospheric circulation models operating at
coarse spatial scales, to providing details on snow processes, and fluxes of meltwater and chemical
species for biological studies at fine spatial scales. Research and operational efforts to implement
spatially distributed models of snow have some common interrelated issues. These include 1) the
tradeoff between model complexity and computational expense; 2) the estimation of error due to
forcing variables (i.e., surface meteorology) and due to model performance; 3) the approach to seg-
ment landscape and terrain data at suitable scales in relation to surface heterogeneity; and 4) the
challenge of validating and/or updating model predictions over large areas.
Model complexity controls the type of measurements that can be used to validate model predictions.
For example, simple bucket-type approaches, once common among land surface calculations in
general circulation models, may only allow testing the presence or absence of snow and water
equivalents within a cell, polygon, or patch. As models become increasingly complex in process
detail, one has the opportunity to use a variety of measurements for model testing. The separation
between error due to incorrect meteorological variables and due to model performance has usually
been addressed by testing at snow plots. Validation efforts at the study-plot, or local-area scale using
manual measurements, have become the standard approach to building model credibility. However,
this type of validation becomes increasingly cost prohibitive when testing snow model predictions
over larger and larger areas. Many tests of snow model predictions distributed over large areas have
suffered from the difficulty in quantitative evaluation due to variability in snow extent patterns and
snow physical properties. Recent modeling approaches parameterize the effects of land cover and
terrain heterogeneity on surface energy and mass fluxes. The effort is important whether or not the
data are segmented into regular or irregular areas and may help address what can be accomplished
with point measurements of snow properties through geostatistical stratification of sampling proce-
dures. Recent distributed model implementations have also parameterized the relationship between
snow extent and mean snow water equivalents. The albedo of areas that have incomplete snow cover
is thus estimated from predictions of water equivalents, coupling the snow model back to the atmo-
sphere model. This approach assumes that depth information is contained in snow extent, which has
not been tested over a wide variety of land cover, terrain types, or spatial scales. Finally, recent
advances in remote sensing have demonstrated construction of images of grain size, wetness, and
water equivalents. These methods offer the potential to test distributed models at compatible resolu-
tions, comparing spatial data with spatial predictions.
1
U.S. Army Cold Regions Research and Engineering Laboratory, 72 Lyme Road, Hanover, New Hampshire
03755-1290, USA
29