Ice storms, caused by severe freezing rains and freezing drizzles are among the major weather hazards in winter. They influence seriously the air and ground traffic, disturb and even interrupt the distribution of electricity by snapping power lines or standstill the production from wind farms.
The conducted research had two main goals: conducting of an automatic classification of the weather types connected with freezing rains and freezing drizzles and assessment of the possible impact of the expected climate change tendencies on these events for the region of Bulgaria.
Data about 180 freezing precipitation events in the period 1958/59-2014/15 has been used and information for the classification has been taken from the six hourly reanalysis upper-air data for the same variables at five standard air pressure levels (1000, 925, 850, 700 and 500 hPa) from NCEP/NCAR reanalysis, which covers the same period.
Due to the large amount of the classification outputs, it was decided to focus on the classification approaches which already had given good results during a former investigation in COST 733 – these are the methods based on cluster analysis such as KMN, CKM and DKM k-means clustering.
Classifications have been made mainly for five standard air pressure levels and additional meteorological variables have been included as well. Following variables have been combined as the most important for the determination of freezing precipitation: see level pressure, air temperature at 850 hPa, the precipitation amount and the persistence of the event itself. Firstly, we have made classifications with 9, 18 and 27 classes but without including the event. We received 750 classifications, but with very low percentages of freezing precipitations among the classes.
After that we continued only with the DKM and KMN classifications. We have added the freezing precipitation events and have given different weights (from 0 till 1) of the variables and examining the behavior of the classifications. Increasing stepwise the percentage of freezing rains and freezing drizzles in the classes, we have established an upper. limit of 17 % with 206 classifications left.
After that χ2 -test has been applied to each class of these classifications in order to emphasis those classes with freezing cases which are not uniform distributed.
All used classifications have been evaluated by the means of different evaluation metrics such as frequency of the freezing precipitation cases in the classification classes, and correlation matrix of the classes. All of the classifications have been made using the software developed in COST Action 733.
The project was supported by the German Academic Exchange Service (DAAD).