Quantifying seasonal snow using very-high-resolution stereo optical satellite images

Abstract

Seasonal snow is one of the most dynamic components of the cryosphere. This water resource serves multiple roles in ecohydrology, climate modulation, and nutrient cycling. Measuring snow distributions is critical to estimating water storage and to predict melt volumes and timing. Interconnected, multi-scalar processes control snow distributions, and at the mountain ridge scale and smaller, wind redistribution, terrain, and vegetation interactions drive heterogeneity. With a changing climate affecting meteorological conditions and shifting the timing of events on a regional basis, statistical relationships (provided enough data has been collected to generate these relationships) break down, making real-time region-specific monitoring important.In the last two decades, the ability to accurately make fine-scale observations of snow has greatly increased, enabling us to better understand and model snow depth spatial heterogeneity and temporal variability. However, most of these methods involve aerial or terrestrial platforms, are limited in coverage, and require significant technical expertise and investment of resources to acquire instruments and collect data. Very-high-resolution satellite remote sensing is a promising alternative, providing broad coverage, meter-scale resolution, and little cost to end users. However, rigorously evaluated retrieval algorithms are needed for operational adoption of satellite products. This dissertation investigates the potential of spaceborne stereo photogrammetry using commercial optical imagery to make fine-scale measurements of seasonal snow across a variety of snow climates including the maritime snowpack of the Washington North Cascades, the continental snowpack of the Colorado Rockies, and the boreal forest / taiga of interior Alaska. Comparison with an expansive collection of reference datasets (e.g., data from SNOTEL network; ground, air, and satellite datasets from SnowEx campaigns) across a range of scales provide extensive validation of stereo snow depth mapping approaches. Assessments of stereo snow depth illustrate how stereo snow depth measurement accuracy is affected by stereo pair collection conditions (viewing geometry, illumination), surface properties (land cover, topography), and processing choices (stereo reconstruction and co-registration). The large spatial coverage of stereo snow depth maps also highlights the strengths and limitations of other measurement approaches, such as swath alignment in airborne lidar products. Residual artifacts with decimeter magnitudes are the largest remaining source of error in stereo snow depth products. Though this work focused on select snow climates and vegetated regions in the United States, it has global applications for quantifying changes in surface elevation with sub-meter-scale magnitudes. With decimeter-scale accuracy (RMSE 0.3 m), satellite stereo collections may offer accurate basin-scale estimates of snow distributions.