Testing methods for various applications (Working Group 4)
5.21
Case III.4 Circulation types associated with
freezing precipitation over Bulgaria
Authors: Dimitar Nikolov, Latin Latinov
5.21.1
Introduction
Freezing rain events are not very often observed on the territory of Bulgaria but
they could be surprisingly severe. The most affected regions are in the north, and
especially in the northeastern part of the country. This severity is determined
by the simultaneously influence of two factors: very intensive Mediterranean
cyclone, passing south of the country, and a strong cold advection from north
or northeast. This situation is very favourable for freezing rain in the northern
part of Bulgaria. The southern regions are protected from the cold advection
by the mountain Stara planina, which crosses through the whole territory of the
country from east to west. Our previous investigations have revealed the most
common synoptic configuration during freezing rain events. The purpose of this
study now is to see how well these situations are represented in the selected by
COST733 classification of circulation patterns for domain 10.
5.21.2
Data and Methods
We have used data for freezing rains from 2 representative meteorological stations
in northeast Bulgaria for the period 1957–2002. During this period 79 freezing
rain events were registered. We have presented them as Yes (1) and No (0) event
and combined with the ERA40 data for Domain 10. We have used the COST 733
catalogue 2.0 and the software developed by WG2 [Philipp u. a. 2010] – trying to
have at least one classification from the main methods of classifications. The classifications tested so far are: GWT, DKM, KMN and PXE with different number
of classes. Because of the magnitude of the temperature in determination of the
type of precipitation, we have used also multivariate classification including the
sea level pressure, precipitation and air temperature at 850 hPa. Additionally we
have given different weighting factors of these parameters according to their importance. To each classification we have also determined the differences between
the classes and the corresponding event distributions, the standard deviation of
the classes and the event groups, the within-type standard deviation (WSD) and
the correlation coefficients for each event class. Summary of these results are
presented below. The centroid plots from the classes with highest frequency are
finally compared with the reanalysis from NCEP and archive synoptic maps.
5.21.3
Results
The first method which has been tested is GWT prototype – large scale circulation
types [?]. Written Voice: There is no BECK1997, perhaps you meant
Beck2007? This classification uses prototype patterns and pearson correlation
274
5.21 III.4 Circulation types associated with freezing precipitation over Bulgaria
coefficients between each field. This classification has been tested for 9, 18 and
27 (see the table below) number of classes, showing that the higher number of
classes only slightly increases the relative frequency of the events into classes.
Generally the freezing rain events are evenly distributed into the circulations
classes with very low frequency (Tab. 5.15), which should mean that they are not
well recognized by this method.
Table 5.15: Relative frequency of the freezing rain events for GWT27.
CT1
0.00
CT2
0.004
CT3
0.00
CT4
0.00
CT5
0.03
CT6
0.05
CT7
0.03
CT8
0.01
CT9
0.00
CT10
0.00
CT11
0.004
CT12
0.01
CT13
0.003
CT14
0.005
CT15
0.009
CT16
0.005
CT17
0.00
CT18
0.005
CT19
0.00
CT20
0.005
CT21
0.012
CT22
0.024
CT23
0.01
CT24
0.005
CT25
0.021
CT26
0.00
CT27
0.00
The second type of methods that has been tested is based on non-hierarchical
cluster analysis. We have used KMN (kmeans – conventional k-means with random seeds) and DKM (dkmeans). We have tested this method of classification
for various numbers of classes and with additional parameters. The main field
used here is the sea level pressure (slp) and the additional parameters are precipitation (prc), the 850 hPa air temperature (tmp) and information about the
event (ice) – yes or no. Different weights have been given to all these parameters
according to their magnitude. Now we have achieved small number of classes
with icing but instead well-defined freezing rains classes – some of the centroids
show very high frequency and some of them are predominant for the given class
and have frequency very close to 1. Again the results are better when increasing
the number of classes. The best results have been achieved with the following
combination of parameters and weights: slp – 2.0; prc – 1.0; tmp – 1.0 and ice
– 0.03. The main freezing rain centroids for this classification are given below
(Figs. 5.40–5.42).
These results generally correspond to the results of [Latinov und Nikolov
2002], where only 4 types of freezing rain situations were recognized. Below are
presented the corresponding plots in Figure DN.4 for the best results so far – CT
15 for KMN27 slp2.0 prc1.0 tmp1.0 ice0.03.
5.21.4
Summary and Conclusions
Four methods of classifications have been tested in an attempt to find a common
circulation type for freezing rain events over Northeast Bulgaria. Increasing the
number of classes yields a better result and including more parameters and weight
factors seem to be the best way to find the typical freezing rain patterns. The
275
Testing methods for various applications (Working Group 4)
Figure 5.40: DKM11 prc w0.03 classes with the highest relative frequency of the events – left CT
3 with 7.3% and right CT 6 with 7.0%.
Figure 5.41: DKM27 prc w0.05 classes with the highest relative frequency of the events – CT 7
and CT21 both with 100%.
Figure 5.42: KMN27 slp2.0 prc1.0 tmp1.0 ice0.03 classes with the highest relative frequency of
the events – left CT15 with 100% and CT22 with 10.0%.
best results until now have been achieved with the KMN classification with 27
numbers of classes and sea level pressure, precipitation, 850 hPa air temperature
and icing event as additional parameters with corresponding weight factors –
KMN27 slp2.0 prc1.0 temp1.0 ice0.3. This method has revealed up to four types
of freezing rain circulation patterns with one predominant type. This study will
be continued with a more extended classification effort and additional data for
freezing rains e.g. air temperature, precipitation amount etc.
Question Written Voice: ice0.3 or ice0.03?
276
5.21 III.4 Circulation types associated with freezing precipitation over Bulgaria
Figure 5.43:
Upper panel:
Sea level pressure field for the CT 15 for
KMN27 slp2.0 prc1.0 tmp1.0 ice0.03; Middle panel: Precipitation field for the CT 15 for
KMN27 slp2.0 prc1.0 tmp1.0 ice0.03; Lower panel: 850 hPa air temperature field for the CT 15
for KMN27 slp2.0 prc1.0 tmp1.0 ice0.03
5.21.5
Recommendations
5.21.5.1
Family of algorithms
On the bases of the tested methods only: it seems that GWT is not suitable for
classification of freezing rains; CKM and DKM have shown very good results.
277
Testing methods for various applications (Working Group 4)
5.21.5.2
Number of classes
Increasing the number of classes yields to significant improvement of the separation of the freezing events.
5.21.5.3
Input variables
Additional variables have been shown to be helpful for refining the separation.
278