Date of Award

Spring 1-1-2013

Document Type


Degree Name

Master of Science (MS)

First Advisor

Abbie B. Liel

Second Advisor

George Hearn

Third Advisor

James Harris


Snow loads govern roof design in many parts of the United States. These loads are largely prescribed by the American Society of Civil Engineers ASCE 7 Standard for minimum design loads. Where ASCE 7 does not specify snow loads due to extreme local variability, such as in the West, many state jurisdictions have developed individual roof snow load documents and maps. However, among the western states border discrepancies and a general lack of uniformity in the methodology for developing such loads indicates a need for a unified approach.

This paper proposes a methodology to develop ground snow loads for the western United States, the application of which is illustrated for the state of Colorado. An innovative approach is taken which utilizes a hydrological snowpack model, Snow Data Assimilation System (SNODAS), developed by NOAA. This model provides estimates of ground snow depth and snow water content, easily convertible into loads, at 588 SNODAS weather stations in Colorado. The methodology proposed here then incorporates statistical techniques such as principal component analysis (PCA) and multivariate cluster analyses to regionalize the SNODAS stations by key shared properties. Several types of cluster analyses are evaluated including agglomerative hierarchical clustering (AHC), k-means, and a PCA-based method. Using various statistical and practical measures of quality, a step-wise hybrid method combining both AHC and k-means techniques is found to be the most statistically sound and robust clustering method. A relationship is then developed between ground snow depths and ground snow loads for each cluster of SNODAS weather stations.

This paper proposes the following additional steps. A database of National Weather Service CO-OP stations with snow depth only measurements is gathered for the state of interest. The 50-year ground snow depths are extrapolated by testing the goodness-of-fit of several probability distributions. The ground snow depth-load relationships for each cluster produced by the hybrid method are then coupled with these 50-year ground snow depths to produce 50-year ground snow loads. Finally, these ground snow loads are mapped in GIS software using a Kriging geostatistical interpolation method to create continuous snow load isolines.