Understanding the Spatial Distribution of Snow Water Equivalent in Paired Basins in Southwest Montana, USA
Title | Understanding the Spatial Distribution of Snow Water Equivalent in Paired Basins in Southwest Montana, USA |
Publication Type | Conference Proceedings |
Year of Conference | 2015 |
Authors | Welz, Jason, Hendrikx Jordy, Challender Stuart, and Stoy Paul |
Conference Name | 83rd Annual Western Snow Conference |
Series Title | Proceedings of the Western Snow Conference |
Date Published | 2015 |
Conference Location | Grass Valley, California |
Keywords | avalanche, binary regression tree, paired basin, Snow water equivalent, spatial distribution |
Abstract | The goal of this research, which was the focus of a master’s thesis, was to build upon previous investigations of the processes controlling the spatial distribution of snow water equivalent (SWE) in alpine environments. This involved taking a comprehensive look at the widely accepted physiographic variables of: elevation, slope, aspect, solar radiation, and wind exposure, but also avalanche activity, which has been given limited explicit inclusion. The paired basin design adopted in this study, between hypothesized avalanche-prone and avalanche-free basins, has been previously used to correlate avalanche activity with snowmelt runoff. However, it has not been used in an attempt to parse out which variables have the dominant influence on SWE distribution between adjacent areas of very similar physiographic character. While most previous studies have focused on the period of peak SWE to study its distribution, this current research considered the evolution of the controlling variables throughout snowpack development, and subsequent melt-out. A robust dataset of snow depth and SWE measurements were collected January 31 - July 10, 2013 on Cedar Mountain near Big Sky, MT. Physiographic variable values were extracted from a 10 m resolution digital elevation model (DEM) at snow sample points and input as predictors of observed SWE in multiple linear regression (MLR) and binary regression tree (BRT) models to spatially distribute SWE across the study area. Optimal models were selected by various measures of goodness of fit and cross-validation criteria. Calculated R2 values for MLR models (0.17-0.57) and BRT models (0.33-0.66) were comparable to previous studies indicating a relative level of success in predictive performance. Subsequent analysis of each optimal model’s variable selection and predicted SWE distributions revealed differences in the spatial and temporal patterns of this metric between the paired basins, confirming some well-understood processes as well as offering new insights.
|
URL | /files/PDFs/2015Welz.pdf |