The Department of Defense seeks an improved understanding and capacity to respond to potential climate change impacts on built infrastructure in Alaska. Other studies have hypothesized that Arctic amplification, the rapid warming of the Arctic compared to the northern hemisphere, causes more persistent weather patterns at midlatitudes, which increase the probability of extreme weather due to drought, flooding, cold spells, and heat waves. Annual maximum snow loads, resulting from the accumulation of snow throughout the winter season, may be strongly influenced by persistent weather patterns. Their effect on annual maximum snow loads and the resulting design snow loads for buildings was investigated in this project. 

The extent of sea ice is a large-scale symptom of Arctic warming and Arctic amplification (AA). The decrease in the annual minimum Arctic sea ice extent began in 2000, and by 2005 all the minima were less than the pre-2000 minima. Therefore, the characteristics of snow load data through the winter of 2003–2004 was used to be indicative of the conditions prior to AA and the period beginning with the winter of 2004–2005 to be associated with AA.

Design snow loads for buildings were specified for a 50-year return period and based on an extreme value analysis of the annual maximum snow loads. The snow load was described in terms of the snow water equivalent (SWE) of the snowpack. SWE data was acquired from a number of sources that provide automatic or manual observations, reanalysis data, or modeled snow accumulation using downscaled reanalysis data. For each dataset, the maximum SWE for each year at each location was determined and compiled the characteristics of the winter up to the day with the maximum snow accumulation.

The periods of record of the datasets were short, particularly when they are separated into two eras. The records might include, just by chance, years with large snow loads or might have only ordinary loads. This sampling error was reflected in the results of an extreme value analysis. Extremes obtained from long periods of record were more reliable. Therefore, the characteristics of the snow accumulation seasons was used to generate years of synthetic accumulation seasons. For each location, the correlation of the annual maximum SWE with parameters characterizing the accumulation season was calculated. The short records also provided little data to define the distributions of these parameters. Therefore, the chi-square test was used to compare the distributions of the daily SWE increment at each location with the other locations. For locations where the probability that the distributions were the same was high, those daily SWE increments were merged. Two-thirds of the locations had at least one other locations with a SWE increment distribution that was similar. The expanded set of characteristics of the snow accumulation seasons was used to generate synthetic seasons, values were chosen for each day from the distribution of the daily SWE increments for that location. Each winter was characterized by two randomly chosen values of parameters correlated with the annual maximum SWE.

In the extreme value analysis, the regional frequency analysis approach was used with the long period of record of synthetic SWE as input. L-moments were used to characterize the data and to fit extreme value distributions. Three- and four-parameter extreme value distributions were fit to the data to allow the shape of the tail of the distribution to be defined by the data.

It was found that the annual maximum SWE was correlated with the number of days with snowfall and with the 80th percentile (80th %) SWE increment. The location associations based on the chi-square test of the daily SWE increment also provided an expanded set of 80th% SWE increments. To expand the set of the number of days with snowfall, the distribution of this parameter at each location was compared with the other locations by using the Kolmogorov-Smirnov test. At each location, the distributions of the number of days with snowfall and the 80th% SWE increment were separated into two groups: on the pre-AA era and the AA era. These expanded sets of characteristics of the snow accumulation seasons were used to generate synthetic winters, randomly choosing values for each day from the distribution of the daily SWE increments. Each winter was characterized by the number of days with snowfall and 80th% SWE increment, each chosen randomly. For each location in each era, 499 winters of SWE were generated to use in the extreme value analysis. The calculated 50-year snow loads were typically not significantly different between the pre-AA era and the AA era. For the dataset based on observed SWE, snow loads that were different tended to be lower in the AA era. For the two datasets based on reanalysis data, loads that were different tended to be higher in the AA era.

A running extreme value analysis of twenty-first century simulations from two global climate models was completed, analyzing 33-year-long blocks of data moved in 10-year increments. The British model indicated increasing SWE in the north and decreasing in the south with the magnitude of the trend increasing with greater total radiative forcing. The Russian model showed smaller trends with the increasing trend in the north reversing with greater total radiative forcing.

This analysis indicated that changes in design snow loads in Alaska associated with global climate change are not justified by the available data or by simulations of future climate. Redoing the analysis in 10 or 20 years with a longer period of record of SWE data would be useful. Until then, the snow load guidance that is available in national standards and the Uniformed Facilities criterion website continues to be applicable. This study has pointed out the dearth of information on design snow loads in Alaska compared to the lower 48 states. Alaska is a huge state with great variation in terrain and climate over short distances and snow load measurement sites are few and far between. Therefore, standards provide design snow loads for only a few locations in Alaska. The investigation of SWE datasets to use for this project has shed light on some of the issues around measured SWE. At most first-order weather stations on most days, SWE is estimated by dividing the snow depth by ten, rather than measured. Thus, any apparent daily variation in the reported SWE may not be real. Reanalysis products assimilate those measurements and the measurements are also used in validating satellite observations and global climate model simulations of SWE.