Nobody Knows How Much Water the Snow Holds

Snow water equivalent (SWE) — the amount of liquid water stored in a snowpack — is one of the most important and least well-measured hydrological variables on Earth. Seasonal snowmelt provides 30-60% of irrigation water in the western U.S. and is the primary water source for over 2 billion people globally. The Earth Science decadal survey designated SWE as a highest-priority observable, yet no satellite mission can measure it with the accuracy, spatial resolution, and temporal coverage needed for water resource management. Current satellite estimates of SWE rely on passive microwave sensors (AMSR-E/AMSR2) that become unreliable when snow is deeper than ~50 cm or when the snowpack is wet, melting, or contains ice layers — precisely the conditions that matter most for water supply forecasting. The fundamental measurement problem is that microwave scattering in snowpacks is governed by snow grain microstructure, which varies enormously with weather history and cannot be sensed remotely.

In the western U.S. alone, snowmelt-fed reservoirs supply water worth $300+ billion annually in agricultural output. Water managers must predict spring runoff months in advance to allocate water among agricultural, municipal, and environmental users. Current predictions rely on ~800 SNOTEL ground stations scattered across the western mountains, providing point measurements that are then extrapolated to basins — but SWE can vary by a factor of 5 over distances of 1 km due to wind redistribution, aspect, and vegetation effects. Climate change is reducing the fraction of precipitation falling as snow, shifting peak snowmelt earlier, and making historical statistical relationships less reliable. In the 2021 western U.S. drought, runoff forecasts overestimated actual flows by 30-50%, contributing to emergency water restrictions affecting millions of people.

Passive microwave satellites (SSM/I, AMSR-E, AMSR2) have provided SWE estimates since the 1980s, but the ~25 km spatial resolution is far too coarse for mountainous terrain where most snow accumulates, and the retrievals fail in deep snowpacks (>~50 cm) and forested areas because the microwave signal saturates. NASA's Airborne Snow Observatory demonstrated that lidar can measure snow depth at 1-3 m resolution from aircraft, but lidar measures depth, not density — SWE = depth × density, and density varies from ~100 to >500 kg/m³. Airborne campaigns are too expensive for operational coverage. SAR (synthetic aperture radar) interferometry can detect snow depth changes at high spatial resolution, but requires repeat-pass observations that are sensitive to wind redistribution between passes and cannot separate dry snow from ice-crusted snow. The proposed NASEM Snow-Satellite concept would use active/passive microwave combinations, but the inversion from microwave signals to SWE is fundamentally underdetermined without independent constraints on snow grain size, layering, and moisture content.

A multi-sensor approach combining active radar (for spatial coverage and penetration), passive microwave (for SWE estimation in shallow/moderate snow), and lidar (for high-resolution depth calibration) with physically-based snowpack models that assimilate all data streams simultaneously. Advances in understanding the relationship between snow microstructure and electromagnetic scattering, validated through ground-based snow pit observations linked to coincident remote sensing data. Machine learning approaches trained on paired ground truth and satellite observations, though these face the same problem as the models: ground truth data are extremely sparse in mountainous terrain. New in-situ sensor networks (low-cost, distributed SWE sensors using cosmic ray neutron sensing or GPS reflectometry) that provide the validation data essential for satellite algorithm development.

A student team could deploy a small network (5-10 units) of low-cost cosmic ray neutron sensors or GPS receivers along an elevation transect and compare their SWE measurements against co-located traditional snow pillow data and AMSR2 satellite retrievals, quantifying the spatial variability that satellites miss. Alternatively, a team could develop machine learning models to estimate SWE from Sentinel-1 SAR imagery using publicly available SNOTEL ground truth data, testing whether ML can extract SWE signal from radar in mountainous terrain. Relevant disciplines: hydrology, remote sensing, electrical engineering, machine learning, environmental science.