Inferring Distributed Snow Depth by Leveraging Snow Pattern Repeatability: Investigation Using 47 Lidar Observations in the Tuolumne Watershed, Sierra Nevada, California

Abstract

AbstractSnow distribution is controlled by the interaction between local meteorology and static features like topography and vegetation. The resulting spatial pattern of snow in mountainous terrain is often repeatable and can be used to infer snowpack distribution at periods when observations are limited. This study uses a library of airborne lidar surveys (ALS) in California's Tuolumne watershed to analyze snow patterns at extents (1,650 km2), resolutions (25 m), and temporal scales (47 ALS observations over 7 years) unmatched by previous research. Distributed snow depth was inferred from snow depth observations covering a portion of the domain (< 4%) at a period near peak‐snowpack timing and snow distribution patterns from different years. Snow patterns from different years differed as a function of snow extents, variability, and interannual noise (r = 0.30 to 0.90). However, matching criteria like seasonal timing and snow extents were able to identify pairs of snow patterns with increased spatial agreement (median r > 0.84). Distributed snow depth inferred using a strip of observations (<4% coverage), and a well‐correlated snow pattern (r≥ 0.90) had a mean absolute error (MAE) of 0.22 m and snow volume error of −6%. Distributed snow depth inferred using patterns of reduced accuracy (r < 0.80) were often too homogeneous, thereby increasing MAE and decreasing the duration of the simulated snowmelt season. This work has applications in water management, where distributed snow depth observations in watersheds with interannual snow pattern repeatability could decrease the extent of observations necessary in future years.