Abstract. The combination of numerical weather prediction and snowpack models has
potential to provide valuable information about snow avalanche
conditions in remote areas. However, the output of snowpack models is
sensitive to precipitation inputs, which can be difficult to verify in
mountainous regions. To examine how existing observation networks can
help interpret the accuracy of snowpack models, we compared snow depths
predicted by a weather–snowpack model chain with data from automated
weather stations and manual observations. Data from the 2020–2021 winter
were compiled for 21 avalanche forecast regions across western Canada
covering a range of climates and observation networks. To perform
regional-scale comparisons, SNOWPACK model simulations were run at
select grid points from the High-Resolution Deterministic
Prediction System (HRDPS) numerical weather prediction model to
represent conditions at treeline elevations, and observed snow depths
were upscaled to the same locations. Snow depths in the Coast Mountain
range were systematically overpredicted by the model, while snow depths
in many parts of the interior Rocky Mountain range were underpredicted.
These discrepancies had a greater impact on simulated snowpack
conditions in the interior ranges, where faceting was more sensitive to
snow depth. To put the comparisons in context, the quality of the
upscaled observations was assessed by checking whether snow depth
changes during stormy periods were consistent with the forecast
avalanche hazard. While some regions had high-quality observations,
other regions were poorly represented by available observations,
suggesting in some situations modelled snow depths could be more
reliable than observations. The analysis provides insights into the
potential for validating weather and snowpack models with readily
available observations, as well as for how avalanche forecasters can better
interpret the accuracy of snowpack simulations.