Comparing Design Ground Snow Load Prediction in Utah and Idaho

Abstract

Snow loads in the western United States are largely undefined due to complex geography and climates, leaving the individual states to publish detailed studies for their region, usually through the local Structural Engineers Association (SEAs). These associations are typically made up of engineers not formally trained to develop or evaluate spatial statistical methods for their regions and there is little guidance from ASCE 7. Furthermore, little has been written to compare the independently developed design ground snow load prediction methods used by various western states. This paper addresses this topic by comparing the accuracy of a variety of spatial methods for predicting 50-year (i.e., design) ground snow loads in Utah and Idaho. These methods include, among others, the current Utah snow load equations, Idaho’s normalized ground snow loads based on inverse distance weighting, two forms of kriging, and the authors’ adaptation of the Parameter-elevation Relationships on Independent Slopes Model (PRISM). The accuracy of each method is evaluated by measuring the mean absolute error using 10-fold cross validation on data sets obtained from Idaho’s 2015 snow load report, Utah’s 1992 snow load report, and a new Utah ground snow load data set. These results show that regression-based kriging and PRISM methods have the lowest cross-validated errors across all three data sets. These results also show that normalized ground snow loads, which are a common way of accounting for elevation in traditional interpolation methods, do not fully account for the effect of elevation on ground snow loads within the considered data sets. The methodologies and cautions outlined in this paper provide a framework for an objective comparison of snow load estimation methods for a given region as state SEAs look to improve their future design ground snow predictions. Such comparisons will aid states looking to amend or improve their current ground snow load requirements.