Jeff Derry, M.S.

RESEARCH:
Regional Patterns of Snow Water Equivalent in the Colorado River Basin Using Snowpack Telemetry (SNOTEL) Data

EDUCATION:
M.S. (Watershed Science) 2008 Colorado State University, Fort Collins, CO, USA 80523-1472
B.A. (Geography) 1993 University of Colorado, Colorado Springs, Colorado 80918

Derry, J.E., 2008. Regional Patterns of Snow Water Equivalent in the Colorado River Basin Using Snowpack Telemetry (SNOTEL) Data. Unpublished M.S. thesis, Watershed Science, Colorado State University, Fort Collins, Colorado, USA, 79pp + 2 appendices.

Abstract

Identifying regions of homogeneity of precipitation data is often a crucial preliminary step in natural resource investigations. Previous clustering of station based snow water equivalent (SWE) data has typically grouped stations based on spatial proximity, political boundaries, or watershed boundaries, and has been restricted due to the temporal resolution of snow course data. This investigation utilized daily data from 216 snowpack telemetry (SNOTEL) stations located in and around the Colorado River Basin over a 15- year period (1991-2005) to cluster stations, i.e., identify regions of homogeneity, based on the patterns and variability of SWE. To achieve this, data were submitted to a selforganizing map (SOM), a particular application of artificial neural networks. This methodology represents a learning algorithm that is non-linear, non-parametric, unsupervised, and learns through an iterative training process.

The number of clusters can be specified to the SOM based on the level of generalization desired. A SOM consisting of a 4, 6, 9, and 16-cluster were constructed from daily values as well as a 6-cluster derived from snowpack descriptor variables (peak SWE, length of snow season, etc.) and physical variables (elevation, aspect, distance to moisture source, etc.) for each station. Areas of homogeneity derived from daily SWE values, annual peak SWE, and physiography were used for multivariate regression analysis to determine the physical variables that best explain variability in peak SWE.

Results showed an unbiased clustering of stations defined not by station location, but by each station’s specific SWE variability over the period of study. The established snow climatologies derived from daily values show general homogenous coarse-scale clusters along a north/south gradient with spatial coherence improving at finer resolutions, but overall there are no definitive spatial patterns to the climatologies, indicating that complex local-scale variables dominate variability of daily SWE. Climatologies derived from descriptor variables showed improved spatial coherence which reflected larger scale influences. Descriptor variables that best represent daily time-step classifications were peak SWE (50% similarity), April 1st SWE (43% similarity), and physical variables (41% similarity).

Regression results showed a consistent increase in predictability as cluster size went from more general (4-cluster) to less general (16-cluster). Key physical variables are elevation, southwest barrier height, regional northness, and southwest shield height. These key variables were consistently used in the regression model, although the degree of importance of the variable depends on resolution and general location of the climatology.

Last update: SRF, 2008.III.03