Statistical Modelling And Estimation Of Basic Ground Snow Pressure For China

Snow load is one of the enviromental loads that should be considered carefully in structural design for buildings in the regions with severe winter climate, especially for the snow-load-sensitive structures such as large-span roofs and light-weight roofing structures. Considering the fact that extreme climate such as snow storms are observed frequently in recent years, some fatal snow-induced structure failures are reported in the media or in literatures, and that the large-span roofs and light-weight roofing structures are widely applied in the building industry, it’s of practical significance and of great interest for one to deepen the study of the snow loads on roofs.On the other hand, basic ground snow pressure is the foundation for the calculation of roof snow loads. This basic ground snow pressure is tabulated in the current Load Code for the Design of Building Structures in China for more than 500 locations across the country. Probability disrtibution models and fitting methods for the extreme value analysis of ground snow pressure are also recommended by the code. However, details on the processing of snow measurement data and extreme value analysis is not available neigher in the code nor in any available references, making it difficult for one to fully understand the uncertainty involved in the tabulated basic ground snow pressure. Besides, some recent developments in the area of extreme value analysis provide some new views and tools for the estimation of basic ground snow pressure. This thesis is assigned to utilize some recent developments in extreme value analysis to estimate the basic ground snow pressure for the country. For this purpose, details of the estimation is discussed and documented. These include:1) Investigation of preferred probability distribution model for the annual maximum snow depths. Some commonly used distribution models and fitting methods were firstly introduced. Then the annual maximum snow depths for 120 meteorological stations, each with at least 40 years of data, were fitted to Gumbel, Log-normal and Generalized Extreme Values distributions; K-S test was carried out to test if the hypothesis that the data follows the three above mentioned distribution models are acceptable. To find the preferable distribution model, the goodness-of-fit was measured by using AIC index and correlation coefficient on probability paper plot. It’s found that Log-normal disrtibution is preferred by most of the stations considered. The influence of using Log-normal distribution instead of Gumbel distribution on the estimated return period value of annual maximum snow depths was investigated by Monte Carlo simulation. The simulation indicated that with cov increases from 0.2 to 0.8, the increase of estimated 50-year return period value of ground snow depths ranges from-4.0% to 17.5%. This non-negligible increase illustrated the importance of using an appropriate distribution model in the extreme value analysis.2) Application of regionalization methods in the estimation of extreme ground snow pressure. Regionalization methods are used to reduce the small samle size effect in the extreme value analysis. Using the annual maximum snow depths from 1981 to 2010 for 83 stations in Heilongjiang province, this study introduced the application of regional frequency analysis(RFA) and region of influence approach(ROI) in the estimation of extreme ground snow depths. Results obtained from Monte Carlo simulations showed that these two regionalization methods can significantly improve the accuracy and stability of the estimation, especially when the sample size is relatively very small.3) Presentation of the procedure and issues to be adressed for estimating basic ground snow pressure by carrying out the estimation of basic ground snow pressure for Heilongjiang province. In the estimtion, Gumbel, Log-normal and Generalized Extreme Value distribution were considered to fit the annual maximum snow depth data in the considered province; it’s verified that the Log-normal distribution is the most preferred model. The influence of the uncertainty of snowpack bulk density was analyzed by assuming the snowpack bulk density is Log-normally distributed. By considering some deterministic methods and some geostatical methods, it’s found that the ordinary co-Kriging with elevation as co-variable is the preferred technique to spatially interpolate the snow pressure.4) Estimation of basic ground snow pressure for the whole country. For this purpose, 5-daily ground snow pressure records from 1999 to 2008 for 733 meteorological stations, as well as daily snow depth records from 1951 to 2010 for 734 stations were used as the basis of estimating the basic ground snow pressure. The estimation was carried out by considering different probability distribution models and different fitting methods, both at-site analysis and ROI approach were applied. To determine the snowpack bulk density, the country was somehow subjectively devided into 6 zones based on the measurements to define the regional average snow density. The estimated basic ground snow pressure showed some discrepancies with those recommended by the Chinese design code, indicating the need to update the basic ground snow pressure recommended in the code using most current snow measurements and fitting techniques.