The missing drifts: snow density heterogeneity from wind-packing exacerbates systematic underestimation of deep snow storage from the scale of nivation hollows to mountain ranges

Spatial estimates of snow water equivalent (SWE) often depend at least partially on extrapolation from sparse measurements, but wind drifts are frequently missing from the underlying observational datasets. Despite the increasing availability of airborne lidar surveys, which constrain snow depth variability in drifts, the contribution of wind-packing densification to SWE heterogeneity remains unknown. Here, we leverage extensive snow pit density profiles (78 total, including 14 at least 2 m deep) from the early ablation season across the Wind River Range (Wyoming, USA) to strategically constrain snow density in rarely sampled settings, including deep drifts, steep slopes, avalanche runouts, and high elevations. At the most extreme, complete vertical profiling of a 5.9 m drift in a nivation hollow reveals a bulk density of 585 kg m−3, exceeding the predictions of global empirical models by 14 % to 35 %. In contrast, we contemporaneously observe bulk densities as low as 339 kg m−3 in adjacent forested areas. We scale up our snow density observations to full watersheds at the 3 m grid scale using a Bayesian statistical model weighted by a representativeness metric that is derived from airborne lidar snow depth data across multiple watersheds. The most representative density measurements are associated with alpine snow drifts deeper than 2 m. Like most terrestrial monitoring networks, all 88 of the in-situ daily snow monitoring sites in Wyoming (SNOTEL) are located in relatively low-elevation forested areas, structurally excluding alpine wind drifts. Beyond this systematic underestimation of drift-related snow density, wind drifting also drives depth variations, compounding the underestimation of SWE heterogeneity in extrapolated datasets. Across three large-domain near-real-time gridded SWE datasets, the standard deviation of SWE across the mountain range is underestimated by 33 %–75 % relative to our lidar-based SWE map at the same 500 m grid-scale. These extrapolated datasets underestimate SWE by 65 %–73 % in a 1 km2 glacial cirque basin despite only 2 %–17 % underestimation of the landscape mean. Despite the systematic underrepresentation of drifts in observational datasets, drifts are the most representative snowpack components in windy alpine mountains, so it is essential to account for drift impacts on both snow depth and density to capture the spatial distribution of SWE.