Skip to main content
eScholarship
Open Access Publications from the University of California

Understanding Snow Process Uncertainties and Their Impacts

  • Author(s): Jeong, Seongeun
  • Advisor(s): Dracup, John A
  • Liang, Xu
  • et al.
Abstract

Prediction of snow in regional and global hydrological models has been a difficult task due to errors in the forcing data, subgrid-scale variability in the snowpack, and uncertain model physics. This dissertation conducts thorough studies of uncertainties that are concerned with snow modeling, in particular for high mountain areas. First, an in-depth analysis of uncertainties associated with meteorological forcing data in the Sierra Nevada was performed. The use of ensemble forcing data with a reasonable degree of uncertainty and model parameter adjustments did not overcome the low-bias in simulating snow states using a simple two-layer (2-L) snow model in the Variable Capacity Infiltration (VIC) land surface model (LSM). To reveal the uncertainty related to model parameterization, a multi-layer (M-L) soil-snow model with more complexity has been developed. This dissertation examines the impact of model complexity on snow simulations in high mountains by comparing the M-L model and the 2-L model. While the current VIC LSM solves state variables for soil and snow separately, the new M-L model solves state variables for the integrated soil-snow system simultaneously. This dissertation has found that the complex M-L model performs better than the 2-L model overall, in particular during the melting season, but the added complexity did not significantly remove the uncertainty, which is similar to some other researchers' findings. This conclusion has led this dissertation research to data assimilation work to investigate the uncertainty problem from a different angle.

The data assimilation approach was taken to discover the hidden facets of uncertain land surface processes that could not be explained by the complex M-L soil-snow model. This research uses a multiscale data assimilation scheme that allows for incorporation of data with different scales. As an extension of the traditional state space model (e.g., Kalman filtering), the multiscale data assimilation incorporates data at different scales by computing their conditional probabilities in a scale-recursive way. The multiscale assimilation scheme has been embedded into the M-L soil-snow model of the VIC LSM. This dissertation applies the new assimilation system to the West Coast region to examine the impact of snow data assimilation at the regional scale as well as at the local scale. The assimilation at the local and regional scales showed promise by reducing biases in simulating snow states in the region. In addition, this research shows the impact of snow data assimilation on energy flux and streamflow simulations.

Main Content
Current View