De-aggregated reliability analysis of freezing rain hazard

This work addresses issues for improving the estimation of the recurrence rate and the distribution in severity of extreme ice events in the Montreal area, which is required in order to determine design criteria for structures such as electric transmission lines. Some of the limitations of current methods for studying extreme freezing rain events are due to the relatively short data records. This results in variability of 'at site' data sets that have only a few large accumulations. The methods developed in this work address these issues. First, de-aggregated analysis is used to obtain better statistical fits by grouping storms according to physical variables that are correlated with the occurrence of ice storms (spatial patterns of sea level pressure (SLP) or 1000 to 500 hPa geopotential height anomalies). And second a procedure to decrease the uncertainty on estimates of the hazard function at high return periods based on solving the CRREL Simple icing model using reliability method is developed. In this procedure, uncertainty is propagated through the model by treating it as a function of random variables.;Environment Canada hourly data was used to identify freezing rain events and obtain measurements of wind speed and precipitation during the events that occurred over Ottawa, Montreal, and Quebec City. General Pareto or Generalized Extreme Value distributions are fitted to the data of total precipitation or total radial ice accumulation for each cluster using a peaks-over-threshold approach. Statistical tests indicate the resulting distributions for precipitation are significantly different from each cluster. This de-aggregated approach improves estimates of the icing hazard by improving statistical fits and by reducing the sensitivity of the results to the choice of threshold.;The second approach used to improve the estimates of the icing hazard function, using reliability methods, considers total precipitation, freezing ratio, and wind speed as the random variables in solving the icing model. The most likely combination of variables associated with high ice accumulations was found to high total precipitation, high freezing ratios, but only slightly higher than average wind speed. The latter is useful for defining load combinations (wind speed and ice accumulation) for structural design purposes.;Finally, the superstation approach of Jones and White (2002a) was investigated by combining Environment Canada data for Ottawa, Montreal and Quebec City. Monte Carlo simulations were performed on the regional set of data using 'at site' indexes. The reliability analysis of empirical icing equation produced results similar to Jones and White at quantiles associated with the 50 year return period. However there were greater differences at higher quantiles. The estimated return periods for radial ice accumulations of 45mm are 160, 210, and 85 years for Montreal, Ottawa, and Quebec City respectively.;Anomaly maps of several meteorological variables were investigated for the objective categorization of ice storms. The NCEP reanalysis data was used to compile spatial patterns for the analysis of the storms identified and categorized by Rauber et al (2001). Several multivariate statistical analysis procedures were used to investigate the effectiveness of sea level pressure, the 1000 to 500 hPa, and the 1000 to 925 hPa geopotential heights for clustering these storms. Results indicated that the k-means algorithm applied to principle component scores of the storm anomaly maps provided the best clustering results. The results indicated that storms with higher precipitations belong to a group associated with the phenomena of cold air damming as a result of the Appalachian Mountains.