Predicting Severe Snow Loads Using Spatial Extremes

Abstract

Severe snow load, unlike severe snow fall, happens over a longer period of time for which snow accumulates causing it to produce increasing amounts of down force on the structure below it. When a structure is forced to bear a load it was not designed for, structural wear and damage can take place and in the worst case scenario, total failure in which the structure collapses. To avoid such occurrences, extreme snow load should be modelled and the potential risks identified so as to improve laws and regulation pertaining to the maximum load a building must handle. This thesis makes use of spatial statistics and extreme value theory to analyse weather data and create models that can aid in predicting extreme snow depth which is directly linked to extreme snow load. Specifically interpolation by Kriging and non-stationary GEV methods are used to obtain predictions between stations. The results are compared to maps already published by the Swedish Building and Housing authority, Boverket, and the discrepancies between them show that certain regions in Sweden are currently being under-estimated. This under-estimation can lead to buildings being constructed to withstand loads less then what is predicted and therefore are at risk of structural fatigue. However in areas of heavy snowfall where the danger is greater, the map by Boverket actually overestimates predicted extreme values. Hence buildings constructed in areas where the risk of very high snow loads is prevalent, should be well future-proofed, and be able to withstand even more load that what is expected.