Improving Snow Deposition Magnitude and Heterogeneity Using Historic Snow Patterns in the California, USA, Sierra Nevada

Abstract

Mountainous snow-covered landscapes in the Western United States behave like natural reservoirs, storing water during cold winter periods and sustaining snowmelt-driven streamflow vital for agriculture, municipalities, hydropower generation, and local ecosystems. In these regions, the timing and duration of spring snowmelt, and the resulting streamflow, are driven by both total snow volume and its spatial distribution across the landscape. Yet, our ability to model mountainous snow magnitude at hillslope spatial scales (< 100 m resolution) is hindered by uncertainties in snowfall and misrepresentations of snow processes like wind redistribution, preferential deposition, and avalanching. Fortunately, snow deposition in mountainous landscapes is driven by the interaction between prevailing snowstorm characteristics and static features like terrain and vegetation, often resulting in interannually repeatable snow distribution patterns. This dissertation investigated the value of repeatable snow patterns in the California Sierra Nevada using an unprecedented collection of ground-based, airborne, and satellite-based snow observations. We investigated how historic information about snow distribution and real-time snowpack observations could be combined to infer snow magnitude at hillslope spatial scales using both statistical and numerical modeling approaches. In Chapter 2, we began by calculating snow depth pattern repeatability at 25 m spatial resolution using a set of 47 airborne lidar snow depth observations in the Upper Tuolumne river watershed spanning water-years 2013 through 2019. Our results showed that normalized snow depth patterns between observation dates with similar relative amounts of snow accumulation and depletion, similar seasonal timing, and similar snow extents, were well-correlated in space (median r > 0.84). This pattern repeatability could be used to infer watershed-scale snow depth distribution using the relationship between a snow depth pattern from a different year and a small subset of real-time observations covering < 4% of the watershed. In Chapter 3, airborne lidar snow depth patterns were used to downscale 25 m snow deposition, as a substitution for more complex snow accumulation processes, from coarser-scale (6 km) snowfall estimates. Snow models were used to evaluate the accuracy of pattern-based snow deposition downscaling in the Upper Tuolumne river watershed. Snow input downscaled using snow depth patterns resulted in simulations with improved snow depth spatial heterogeneity (0.52 < r < 0.76), as compared to more-common statistical methods that downscaled snow input using terrain elevation (r = 0.27). However, snow input downscaled using historic snow patterns were still subject to snowfall biases from atmospheric models, which can be as large as ±60% in mountainous terrain. Winter losses (snowmelt and sublimation), snow density spatial variability, and interannual changes to snow depth patterns also influenced the spatial variability of the snow depth patterns and the resulting accuracy of pattern-based snow downscaling. In the last phase of this dissertation (Chapter 4), we focused on correcting the pattern-based snow downscaling strategy using snow accumulation patterns from a 31-year snow water equivalent (SWE) reanalysis. Using the relationship between historic snow accumulation patterns, and ground observations of SWE accumulation, watershed-scale mean snowfall was inferred to within ±13% accuracy (interquartile range) between 1985 – 2016 in the Upper Tuolumne, Upper Kings, and Sagehen Creek Sierra Nevada watersheds. Additionally, as compared to snow depth simulations that downscale snowfall from a popular meteorological product (NLDAS2) using terrain elevation (r = 0.54, on average), simulations inferring snow input using the relationship between historic snow accumulation patterns and real-time ground observations had significantly improved snow depth spatial distribution (r = 0.76). These results show that, where and when interannually repeatable snow patterns emerge, historic snow information can be used to overcome model and forcing data constraints and improve representations of snow magnitude and hillslope spatial heterogeneity in a way that is practical for both operational and scientific purposes.