A RADAR-BASED SEVERE WEATHER CLIMATOLOGY
For SOUTHERN QUEBEC
Aldo Bellon
Isztar Zawadzki
J.S. Marshall Radar Observatory
McGill University
July 2001
Work supported by the
Climate Change Action Fund
1
TABLE of CONTENTS
1- INTRODUCTION
2
2- DATA USED
2.1- Historical Background
5
2.2- Data Resolution
6
2.3- Data Acquisition and Selection
6
2.4- Data Gaps
7
2.5- Data Processing
9
3- SEVERE WEATHER MAPS and ALGORITHMS
3.1- VIL (Vertically Integrated Liquid water content)
13
3.2- GUST (Estimate of maximum surface winds)
15
3.3- VPI (Vertical Profile Indicator or Overhang)
17
3.4a- Mesocyclone Detection Algorithm
19
3.4b- Mesocyclone Tracking Algorithm
22
4- RESULTS
4.1- Mesocyclone Statistics
24
4.2- Reflectivity Analysis
34
5- CONCLUSIONS
44
6- ACKNOWLEDGEMENTS
46
7- REFERENCES
47
2
1- INTRODUCTION
It has been suggested that the variability of severe weather phenomena could be the most
sensitive manifestation of climate change. For example, an analysis of 29 years of hail data from
central Alberta indicates that the frequency of hail events has changed dramatically over the past
several decades (Smith et al. 1998). Likewise, the climatology of lightning flashes for Alberta
reveals that there has been a significant increase in the number of cloud-to-ground lightning
strikes during the last decade (Kozak 1998). It may be argued that such trends are the inevitable
result of increasing, (and in some cases decreasing), density of observational capabilities, be it
hardware or human as is the case for the visual reporting of tornado and hail events. In contrast, a
climatology by radar, when properly calibrated and when allowance if made for any missing
periods, would have the advantage of being more consistent over the region of interest and over
the entire period of observation. Note that because of its continuous space coverage, radar is
the only instrument that will not miss any severe weather event within its domain of
observation. Radar-derived statistics are thus not dependent on population density.
A number of climatological studies and data collection programs have been conducted
mostly in the US, but they involved observations of severe weather events by instruments other
than radar. (An exception is a recent preliminary attempt undertaken by Mapes, 2001). They have
identified the nature of the various severe weather phenomena, risk factors for different parts of
the country, and detection system technologies. Unfortunately, Canada still lags behind in this
domain, and a comprehensive climatology of severe weather events has yet to be produced. The
work by Lackmann and Gyakum (1996) represents a first step towards this goal. As a component
of the Global Energy and Water Cycle Experiment (GEWEX), they have successfully identified
large-scale precursors to significant precipitation events in the Mackenzie River Basin. More
recently, Fischer (1997) has underlined the unique large-scale precursors to heavy precipitation
events in the Montreal region. We currently have available to us a readily accessible archive of
gridded reanalysis data for the globe that extends back to 1963. It is in the public interest to
identify and to study the precursors to these dangerous events but we must first have a systematic
procedure for identifying them. We believe that radar archives can now provide this objective
source of information that has not yet been tapped.
3
A climatology of severe weather phenomena from past records is highly desirable since it
provides a base for comparison of future events. An ongoing survey of severe weather events
can be used to monitor climate variability as well as providing information of immediate
practical interest. Observations from a radar network are ideal for gathering the information
necessary to produce such a climatology. The Canadian territory just north of the US border,
where most of the population resides, will be covered by 32 Doppler radars, Lapczak et al.
(1997). Three of these have a relatively long history of operation: the Doppler radar data archives
at King City north of Toronto extend back to 1984, those at Carvel west of Edmonton date back
to 1991, and those at the Marshall Radar Observatory (MRO) of McGill University at Ste-Anne-
de-Bellevue west of Montreal began in 1993. Data from these radars are collected for operational
purposes and a considerable amount of information is already recorded and available for analysis.
Thus, these three Doppler radars are the ideal test-bed for the proposed study.
The overall objectives of our research efforts thus consist of the following:
i) To compile a preliminary severe weather climatology for southern Quebec, southern Ontario,
and Alberta.
ii) To develop a methodology for a future, in-line, automatic climatological compilation to be
implemented in the UNIVERSAL RADAR PROCESSOR (URP) presently developed for
the CANADIAN RADAR NETWORK.
iii) To relate severe weather events to larger scale climatology of weather patterns.
iv) To contribute to the systematic observation of the climate of Canada.
v) To improve our climate models by providing a validation database.
vi) To reduce the uncertainties in our knowledge of the magnitude, rate, and regional distribution
of changes in severe weather climatology.
vii) To determine the planetary-scale precursors that affect the severe weather climatology and
to improve the understanding of key climate processes so that climate models at all scales
might be used to determine future changes in frequency or intensity of severe weather.
Several severe weather detection algorithms are in operational use and their statistics
can be compiled for our purpose. In this report, we will concentrate on presenting the areal,
temporal and geographical distribution of severe weather events as perceived by radar data. We
have completed the derivation of statistics of severe weather events using the archived MRO S-
4
band radar data from 1993 to 2000. Those for the King City and Carvel C-band radars are
currently being processed. The analysis of one month (July 2000) of Carvel radar data is attached
as a separate document to this report. In a subsequent investigation that will follow our
identification and statistical description of severe weather events, we intend to integrate this
severe weather climatology with a study of planetary-scale precursors of these events. Precedent
for this scale interaction approach is seen in the work of Lackmann and Gyakum (1996), in which
a planetary-scale upstream ridge in the Bering Sea, which induces a downstream trough, is
typically observed two days prior to the onset of significant precipitation in the Mackenzie River
Basin. A preliminary analysis using a similar approach based on a set of 22 days of intense
mesocyclonic activity around Montreal has already been carried out and is included in this report
in Appendix A.
In section 2 of this report we discuss the data archive system of the McGill Radar and
estimate all the data gaps during our 8-year study. In Chapter 3, we describe the type of radar
maps selected to identify severe weather, with particular emphasis on the mesocyclone detection
and tracking algorithms. In the first part of the section on "Results", the characteristics and the
geographical distribution of all detected mesocyclones is presented. The second part deals with
the climatology of severe convection as deduced from the reflectivity maps alone, in particular
from the upper level VIL parameter. Section 5 summarizes the report and offers some
suggestions for future work.
5
2- DATA USED
2.1 Historical Background
The J. S. Marshall Radar Observatory (MRO) of McGill University has been in real-time
operation since 1970 and was capable of archiving digital data as early as 1974. 'Reflectivity-
only' records were archived on large magnetic tapes until 1993 when Doppler information was
added. However, these records had numerous data gaps due in part to the limited storage capacity
of that medium (~45 to 70 Mbytes) that also suffered from a rapid deterioration with time, that is,
existing records could not be reread. Consequently, they have not been transferred into another
medium like the smaller 2.5 or 5.0 Gbyte exabyte tape that replaced it before the complete
demise of the cumbersome magnetic tape units. Therefore, our investigation of severe weather
events will use simultaneous reflectivity and Doppler data of the 8 summers from 1993 to 2000
inclusive.
During the course of its history, the scanning strategy of the S-band (10 cm) MRO radar
has undergone some modifications but has kept its fundamental objective of achieving 24
elevation angles during its 5-minute cycle. In 1993, a Doppler capability was implemented
without compromising the existing data: radial velocity measurements were collected at all 24
elevation angles simultaneously with the reflectivity information. This was achieved with a pulse-
pair algorithm applied to the first 16 pulse-pairs of each azimuth, followed by a single pulse for
the reflectivity. Thus, a full three-dimensional reflectivity (up to 480 km) and Doppler volume
scan (up to 125 km with a Nyquist interval of
±
31.2 m/s) was made possible. Shortly afterwards,
in August 1994, the antenna scanning strategy was altered so as to perform two 2.5-minute cycles
of 12 odd and then of 12 even elevation angles in an effort to detect short-lived phenomena like
microbursts. Then, in August 1998, in preparation for the implementation of a dual polarization
capability finalized in July 1999, the scanning and data processing strategy had to be modified in
order to adapt to the requirements of the NCAR VIRAQ board selected to perform the
digitization of the radar signals. Doppler and reflectivity data from the first set of 12 odd
elevation angles were collected at 1200 PRF (maximum range of 125 km) as is the case for the
last 7 angles of the even set where range folding is unlikely. Elevation #2 is sampled at a dual
PRF of 400 and 600 Hz (to achieve long range surveillance up to 375 km) while data from the
remaining 4 even elevation angles (#4 to #10) are obtained at 600 PRF to provide measurements
6
up to 250 km. These changes were accompanied by significant software modifications that
perform the necessary range and velocity unfolding and thus maintain full Doppler (Nyquist
velocity of 31.2 m/s) and reflectivity volume scans up to a 250 km range for all 24 elevation
angles.
2.2 Data Resolution
The resolution of our data sets followed the changes made throughout the years. One
constant is the geometrical progression in the separation of the 24 elevation angles, from 0.5
degrees above the horizon for elevation #1 to 34.4 degrees for the last elevation. Since 1993, the
azimuth resolution has been fixed at 1 degree, slightly larger than the radar beam width of 0.86
°
.
The 8-bit intensity resolution of the Doppler data has been 0.5 m/s, while for reflectivity, it is
1.28 dBZ for data prior August '98 and of 0.40 dBZ afterwards. The range resolution is of 1 km
up to 120 km in range, of 2 km between 120 and 240 km and of 4 km beyond 240 km.
Considering data within the 240 km range, the size of a full volume scan at normal resolution is
about 1 Mbyte. In fact, when recorded, it is substantially less because of the suppression of most
of the 'no-echo' bins. In cases of severe convection, manual intervention allowed archived records
to be at a resolution of 250 meters up to a range of 80 km. This capability is useful in visualizing,
and possibly in improving the detection of mesocyclones.
2.3 Data Acquisition and Selection
Since the advent of digital radar data about 25 years ago, the immense quantity of data
limited most analysis to a few case studies, to a small sample of selected events or to relative
short periods of continuous data. The early international GATE experiment (Hudlow and
Patterson, 1979), is perhaps a notable exception. However, as stated by Kruger et al. (1999),
"recent progress in computer information storage technologies as well as developments of
efficient data formats provided the opportunity to change this situation". The archival procedures
at MRO reflect these improvements. From 1993 until August '98, the volume scans were
immediately written into an exabyte tape at the end of each 5-minute radar cycle. This was
necessary on account of the limited memory capacity of the computer handling the real-time
processing of radar data. Depending on the frequency, extent and location of the radar echoes, the
7
2.5 Gbyte tape would last from a minimum of about 3 days to a maximum of a week or perhaps
more in winter. Over this 5-year period, about 275 exbyte tapes were recorded. In addition to data
interruptions caused by power failures, radar hardware problems, radar maintenance and
upgrades, some data gaps were the result of "reaching the end of tape" and by "writing errors" by
the tape unit. However, the data lost when severe weather affected our region is relatively small
from 1994 onwards. An assessment is provided in the following section 2.4. From August '98 to
the present, the increased storage capacity of the radar data processing system permits the data to
be first safely stored in its memory. It is subsequently downloaded to a 5 Gbyte tape once every
few weeks as required, eliminating gaps due to tape problems. This procedure also permitted a
more efficient storage of radar precipitation data because 'clear' periods with only ground echoes
and/or anomalous propagation can be easily omitted from the archived data set. As a result, only
forty 5 Gbyte tapes were required from August '98 until September 2000, the last month of our
analysis period. Even if only the 5 months from May to September were considered for our
convective study, about 150 2.5Gbyte and 22 5Gbyte tapes would have to be processed. An
initial pass through this huge data set just to generate and visualize basic radar maps for the
selection of the convective periods would have entailed a prohibitive amount of time.
Fortunately, since the end of June '94, we had initiated at MRO a crude separate archival of only
the low level CAPPI maps at a 2-km resolution on a (240 by 240) grid. These maps are stored
every 30 minutes on the computer hard disk if sufficient echoes are detected in addition to the
normal ground echo coverage on the selected CAPPI height. A simple software program enables
the quick animation of all the stored images of any given day. This "log" feature enormously
facilitates the selection of the convective time intervals for an entire summer by visual inspection
by an experienced radar meteorologist. Since the basis of selection is a low level CAPPI map
rather than an upper level CAPPI that would have been more suitable for unequivocally
identifying convection, we have included all doubtful periods for processing and for the
extraction of climatological statistics.
2.4 Data Gaps
For over 30 years, the MRO has been striving to fulfill the dual role of a real-time
operational radar and of a centre for research and development in association with McGill
University. The adherence to the latter goal has often meant a temporary cessation of its routine
8
operation in order to implement the hardware modifications for novel approaches (for example,
the recent polarization upgrade completed in July 1999 and the experimentation with a collocated
X-band transmitter during the '90s). As much as possible, these planned 'down times' have been
scheduled to occur in the spring or fall, before or after the 'severe weather' season. However, the
now aging equipment of our radar knows of no such precaution and there have been the
inevitable interruptions due to hardware failures at the most inopportune times. (Those due to
power failures have been drastically reduced after the installation of a UPS (Uninterrupted Power
Supply) system in the late '80s). Data gaps related to archival problems, common prior 1998,
have now been completely eliminated. Nonetheless, some archived data had to be excluded from
our analysis because of some intermittent problems with the PRF circuitry.
It is difficult to estimate the exact number of hours or days with convection lost when the
radar itself is down. In the assessment that follows, we have relied on some brief notes or
comments written by the MRO staff. It has only been since the start of the 2000 summer season
that a detailed account has been kept of the exact number of hours lost and of the nature of the
precipitation during those outages. The latter has been possible only since the RAPID system (see
below) has been able to also process data from 4 other radars. Thus, for the previous years, we
have attempted to provide at least a general overview of the amount of time lost when some level
of convection would have likely occurred over the radar coverage (within 240 km). The 'lost
time' does not mean that severe weather, as defined in terms of the thresholds in the following
section, did continuously occur over the indicated period. 'Lost time' is given in terms of days and
hours. By the former is meant that the radar was down for the entire day when some form of
convection was confirmed or suspected to occur during the preferred period of 11:00 to 21:00
local time. In the case of data interruption during existing convective activity, the number of
'hours lost' have been estimated on the basis of the storm location from the 240 km boundary and
the speed at the onset of the data gap as well as on the diurnal trend noted above. The year 1993
is not included because, after the dopplerization in March of that year, the data archival system
went through some growing pains whereby data were not routinely archived. The absence of the
"log" facility noted in section 2.3 earlier further rendered difficult the inclusion of all convective
events of the existing 1993 data set. Moreover, due to a software error in the range normalization
procedure, the high reflectivities within about 50 km were reduced by a much larger amount.
Data from 1993 has thus been selected from known severe weather days that were the basis of a
9
study of microbursts made for Transport Canada Aviation, Bellon and Zawadzki (1994). They are
not intended to characterize the level of activity for the entire season and should thus not be used
on a comparative basis. Here is the list of the estimated data loss during convective activity for
the summers of 1994 to 2000 inclusively:
1994: 1 day (6-May) + 2 hours
1995: Radar hardware problems kept the radar out of commission for most of the time between
June 19th and July 17th. Fortunately, this coincided with a very dry period except for
some convection on July 1st, 6th and 7th.
3 days + 2 hours
1996: No convective days were apparently lost
1997: A late season hailstorm on September 29th was lost when the radar was down for repairs.
1 day + 7 hours
1998: 20-May and 8-July could not be processed because the tapes could not be re-read.
2 days + 3 hours
1999: Radar down for polarization upgrade from April 26th until May 31st. Again, this was a
very dry period except for May 8th when a weak tornado may have been responsible for
some wind damage near Ottawa. Then, problems related to the newly installed
polarization equipment persisted for a 2-week period at the end of July, resulting into an
estimated loss of about one 'day'.
2 days + 7 hours
2000: 12 hours
TOTAL: 9 days + 33 hours
The percentage of the entire sample represented by the above 'total losses' cannot be
easily estimated. An attempt is made at the end of the next section.
2.5 Data Processing
Over the years, MRO personnel have developed a number of real-time radar data
processing systems, starting with SHARP (SHort-term Automated Radar Prediction, Bellon and
10
Austin, 1976). It was followed by PPS (Product Processing System, Austin et al. (1986) and
Duncan et al. 1992). Since 1995, we are constantly upgrading our current RAPID (Radar data
Analysis, Processing and Interactive Display) system (Kilambi et al.1997 and Bellon and
Kilambi, 1999). RAPID can also operate in simulation mode using historical data sets as input. It
ingests reflectivity and Doppler data in spherical coordinates and performs some necessary
'cleaning' prior the generation of a user-selectable number of Cartesian radar maps. As already
mentioned in section 2.1, this 'cleaning' involves the range and velocity unfolding of the raw
volume scans. In addition, since the actual time of data collection is different from elevation to
elevation during the execution of the 5-minute radar cycle, the data at the various elevation
angles are shifted by an appropriate amount that also depends on the average velocity of the
precipitation pattern. This spatial adjustment simulates an instantaneous measurement of the
entire volume scan. Other pre-processing involves the elimination of regular ground echoes and
of anomalous propagation. The latter procedure is essential for estimating surface rainfall but is
less crucial in our work that focuses on the statistics of severe convection obtained mainly with
radar maps generated from higher altitudes.
All Cartesian maps in our study are in the form of a (240 by 240) array at 1- and 2-km
resolution. They are compressed and stored into a cyclical file for re-analysis. The present
configuration allows 25600 maps to be stored per cyclical file but this limit can easily be
increased. The time period included in such data set depends on the number of maps generated
for every radar cycle. We have selected the following 5 types of radar maps as being most
suitable for our study of convective weather:
1- Low level CAPPI (at 2.5 km)
- needed to compute the velocity of the precipitation pattern required by the 'shifting' routine, by
the mesocyclone and cell tracking algorithms and by the rainfall accumulation module.
- the header of this map contains the results of the mesocyclone and 'overhang' algorithms
2- Regular VIL, referred to simply as VIL
- vertical integration of reflectivity from a low altitude (2.5 km) up to echo top
- also needed for the computation of the GUST map
11
3- Elevated VIL or upper level VIL, referred to simply as UVIL
- vertical integration of reflectivity from a 'middle' altitude (5 km) to echo top
4- GUST
- combines data from the regular VIL and from an echo top map of the 18 dBZ reflectivity
5- Upper level CAPPI (reflectivity at 7 km)
A (3x3) smoother is applied to the latter in order to reduce the noise caused by the limited
sampling of the reflectivity data. Since the VIL and GUST maps are already the results of an
integration (vertical), no smoothing is performed.
(We have also generated 1-hr accumulation maps every 30 minutes, and 2- and 4-hr
accumulation maps every 1 hour, but their analysis is not included in this report.)
Consequently, with 2 resolutions per map, about 130 maps were generated for every hour
of data processed. A typical cyclical product file thus can contain nearly 200 hours of data, which
in many cases, is the entire data set of convective events selected for an entire summer. A total of
10 such files were required to hold all the maps for our 8-year study.
In order to relate the data lost as estimated in section 2.4 to the entire sample processed,
we define a 'convective hour' when one of the following conditions is satisfied on at least one of
the 12 maps of either resolution during that hour:
a) 5 pixels exceeding a VIL threshold of 30 kg/m
2
b) 5 pixels exceeding a UVIL threshold of 15 kg/m
2
c) 5 pixels exceeding a GUST threshold of 15 m/s
d) 10 pixels exceeding a reflectivity of 48 dBZ on the 7 km CAPPI maps
e) the presence of an 'overhang' of any size
12
In table 2.1, we provide the total number of 'convective hours' so defined for each of the
8 years analyzed and within the indicated radar ranges. The first and last day with convection
within 240 km is also given.
Year
12-180 km
12-240 km
First Day
Last Day
1993
*
98
119
27 June
10 September
1994
228
254
31 May
29 August
1995
102
113
21 May
5 September
1996
104
113
20 April
26 August
1997
193
210
1 May
2 September
1998
246
273
7 May
28 September
1999
189
233
3 April
13 October
2000
175
224
7 May
21 September
TOTAL
1335
1539
-
-
Table 2.1: Number of 'convective hours' in our entire data sample for each of the 8 years
analyzed and within the indicated ranges.
*
Data from 1993 is not complete.
In the previous section, we estimated a data loss of 9 'days' and of 33 'hours' over the
summers from 1994 to 2000. From the table above, this 7-year period contains over 1400
'convective hours' if data up to a 240-km range is considered. Depending on the number of
'convective hours' that can be arbitrarily attributed to a 'lost day', the total losses can be evaluated
to be of the order of 5%. This is less than the variability observed among the various years as
seen for example in Figs. 4.1, 4.11 and 4.13. Thus, in section 4, the statistical results will be
presented without attempting any yearly adjustment for these estimated losses.
13
3- SEVERE WEATHER MAPS and ALGORITHMS
In this section we describe in greater detail and exemplify the severe weather maps used
in our study.
3.1 VIL (Vertically Integrated Liquid water content)
This product involves the summation of the radar reflectivities from all the elevation
scans above a sufficiently low altitude and deriving a liquid equivalent as described by Greene
and Clark (1972). Since the liquid water content M is proportional to the (4/7) power of the radar
reflectivity factor Z, the VIL value is heavily dependent on the stronger reflectivities, (> 40
dBZ). In the integration process, the reflectivity of each of the 24 elevation angles is multiplied
by the vertical thickness associated with each elevation angle. This vertical thickness is unrelated
to the actual vertical antenna beam resolution, but depends on the spacing between the 24
elevation angles and, of course, on the range from the radar. The VIL map is an excellent
indicator of severe weather since, as thunderstorms intensify, their updrafts and consequently
their ability to sustain airborne liquid water and other hydrometeors over a deep layer of the
atmosphere drastically increases when compared with stratiform situations or with weak
convective storms. The units of this radar product are expressed in terms of kg/m
2
. Based on
experience with thunderstorms of the Montreal area, values in excess of 30 or 40 kg/m
2
represents a severe storm. Unlike a CAPPI map that can show high rainfall rates at only the
selected altitude, or an ET map with high echo tops that may be associated with relatively light
reflectivities, the VIL product best measures the overall '3-D' strength of a storm by 'weighing'
all the reflectivities above a given point. Fig. 3.1a is an example at a resolution if 1 km. Since
"what goes up must come down", a high magnitude for this parameter represents a great potential
for an imminent collapse of the storm reflectivity core often associated with microbursts. As
shown later in eq. (3.1), it is the primary input into the gust algorithm.
There are two input parameters associated with this product:
b) Lowest Height: This parameter specifies the height above which the radar reflectivity
is integrated. This height should be sufficiently high to avoid contamination by ground echoes. If
14
at distant ranges the height of the lowest elevation angle is already above this 'lowest height', the
observed reflectivity is assumed to extend downward unchanged to the selected height. This
assures that the integration is performed over the same vertical slice of the atmosphere,
regardless of range. For the MRO radar with the Laurentians Hills to the north, the Adirondacks
to the south and the Green Mountains in the southeast, this parameter must be relatively high, at
least 2.5 km. However, spurious values due to the highest ground echoes are present, (around the
120-km range in the southeast in Fig. 3.1a). Therefore, we have also generated an 'elevated', or
'upper level' VIL, or UVIL map, with a lowest height threshold of 5.0 km. This special product
has the advantage of completely eliminating ground echo and anomalous propagation as well
bright band and shadowing effects. It is ideally suited for indicating convection and will thus be
extensively used in our analyses. An example is shown in Fig. 3.1b.
(a)
(b)
Fig. 3.1: Example of a 'regular' VIL in (a), where the vertical integration begins at an altitude of
2.5 km and of a 'upper level' VIL in (b) where the summation begins at a height of 5 km. Both
maps consist of a (240 x 240) array at 1-km resolution. In (a), the tracks of VIL cell greater than
30 kg/m
2
are overlaid. The time in GMT at the beginning and end of each track is indicated. The
width of the track is related to an equivalent cell diameter as indicated on the bottom right
corner of the map. The color of the track gives the 'liquid content' of the entire cell in units of
kilotons according to the scale on the right of the VIL scale.
b) Surface Extrapolation: When this flag is set, the reflectivity at the lowest height is
extrapolated downward to the surface in order to obtain the actual liquid water content above a
given pixel of the map. The GUST product that uses VIL values as input assumes this to be the
case. However, for reasons just discussed above, no surface extrapolation is performed with the
MRO data.
15
3.2 GUST (Estimate of maximum surface winds)
The gust algorithm was proposed by Stewart (1991) and is based on the cloud top
penetrative downdraft mechanism developed by Emanuel (1981). This technique involves the
combination of the vertically integrated liquid water content, VIL, and of the thunderstorm echo
top, ET, in order to estimate surface winds below a thunderstorm solely from reflectivity data.
The magnitude of the maximum surface wind gust W can be expressed as
W = (c
1
VIL - c
2
ET
2
)
1/2
(3.1)
where W is in m/s, VIL in kg/m
2
, ET in km and c
1
and c
2
are two constants with a value of 20.6
m/sec
2
and 3.1 s
-
2
respectively.
The echo top is that corresponding to a reflectivity threshold usually taken as 18 dBZ.
Table 3.1 provides the gust estimates in m/s obtained from (3.1) as a function of typical
combinations of VIL and ET magnitudes. Note that a higher echo top for a given VIL reduces
the gust estimate since the amount of liquid water available for evaporation is spread over a
region of the atmosphere that cannot interact with the mid level entrainment. It is the VIL
concentration in the lower altitudes, about < 8 km, that has the potential to enhance the
evaporative process leading to a wet microburst. A VIL value of 30 kg/m
2
coupled with an echo
top of 8 km yields a gust estimate of 20 m/s.
16
VIL
Height (km) of 18 dBZ echo
kg/m
2
6
7
8
9
10
11
12
13
14
15
10
9.7
7.3
2.5
0.0
-
-
-
-
-
-
15
14.0
12.5
10.5
7.5
0.0
-
-
-
-
-
20
17.3
16.1
14.6
12.6
10.0
5.9
0.0
-
-
-
25
20.1
19.0
17.8
16.2
14.3
11.7
8.1
0.0
-
-
30
22.5
21.6
20.5
19.1
17.5
15.5
13.0
9.5
2.5
0.0
35
24.7
23.9
22.8
21.7
20.2
18.5
16.5
13.9
10.5
4.4
40
26.7
25.9
25.0
23.9
22.6
21.1
19.4
17.2
14.6
11.0
45
28.6
27.8
27.0
26.0
24.8
23.5
21.9
20.0
17.8
15.0
50
30.3
29.6
28.8
27.9
26.8
25.6
24.1
22.4
20.5
18.1
55
32.0
31.3
30.6
29.7
28.7
27.5
26.2
24.6
22.9
20.8
60
33.5
32.9
32.2
31.4
30.4
29.3
28.1
26.6
25.0
23.1
65
35.0
34.5
33.8
33.0
32.1
31.0
29.8
28.5
27.0
25.3
70
36.5
35.9
35.3
34.5
33.6
32.6
31.5
30.3
28.8
27.2
Table 3.1: Gust values in m/s as a function of VIL and echo top heights according to Eq. (3.1).
The MRO radar, with 24 elevation angles every 5 minutes is ideally suited, in terms of the
horizontal and vertical resolution of its reflectivity data, for the generation of accurate VIL and ET
maps required by the gust algorithm. The GUST algorithm is being used extensively even with the
current availability of Doppler data since it easily points out storms that are potential candidates for
strong surface winds and that thus require further analysis with Doppler data by the forecasters.
Moreover, its continued use is assured even with the advent of Doppler data because microburst
detection is difficult at ranges beyond about 80 km due to the broadening of the 1 degree radar beam
and to the radar inability to observe low altitude phenomena. An investigation by Amorim et al. (1997)
showed that the high VIL values, and hence the large GUST estimates, are observable on the average
10 minutes prior to a possible microburst detection with Doppler radar. A GUST map can thus be
considered as a short-term forecast of such an event.
17
Strictly speaking, the gust algorithm ought to be applied only to pulse-type storms
developing in a weakly sheared environment in which an elevated layer of low equivalent
potential temperature is present above a deep and moist low level air. Nevertheless, forecasters at
the Montreal weather office that have been using it since 1991 consider the gust algorithm to be
reliable even for fast moving thunderstorms forming along a frontal boundary. An example of a
GUST map is shown in Fig. 3.2a. To complete the list of severe weather maps, a 7-km CAPPI is
also provided in Fig. 3.2b for the corresponding time. Note how concisely the GUST map
summarizes the location of the potentially severe weather echoes while the 7-km CAPPI map
emphasizes the extent of the leading edge of anvils associated with developed thunderstorms.
(a)
(b)
Fig. 3.2: Examples of a GUST map in (a) and of a 7-km CAPPI map in (b). Note the extensive
anvils downwind from the two major storms portrayed.
3.3 VPI (Vertical Profile Indicator or Overhang)
The VPI product is not a map by itself but rather some information in the form of an
outline around thunderstorms with a potential of severe weather and/or hail that can be added by
the RAPID software onto horizontal cross-section images like CAPPI, Echo Top and VIL maps.
It is also known as the "overhang" (in English) or "surplomb" (in French) indicator. Unlike the
procedure described by Petrocchi (1982) which identifies as many as 9 hail indicators or criteria
that may be retrieved from 3-D radar data, our identification of an accumulation level is based on
18
the simple comparison of the reflectivities at the mid-level and lower level of the atmosphere.
The user-selectable parameters associated with this module are:
a) Height of the low-level CAPPI that is also used by the forecast and accumulation modules,
selected at 2.5 km in our study
b) Height of mid-level CAPPI (
≥
6 km)
c) Reflectivity threshold at the mid-level CAPPI (
≥
45 dBZ)
d) Difference between mid-level and low level reflectivities (
≥
10 dBZ)
The values of these parameters are undoubtedly climatologically dependent. A mid-level
height of 7 km, an upper level reflectivity threshold of 48 dBZ and a reflectivity difference of 10
dBZ have been selected for our study. They have been used by forecasters of the Montreal area
since 1991. The combination of these thresholds identifies the weak echo region and the
pronounced accumulation of reflectivities aloft that are both associated with the exceptionally
strong updrafts present in severe thunderstorms.
For display purposes, an additional parameter is the radius of the circle (~ 5 km) drawn
around each of the pixels satisfying the above conditions. After eliminating isolated pixels, the
coordinates of the perimeter outlining the combined region of influence of the declared severe
weather pixels are then assigned a special value into any horizontal cross-section maps. Thus,
upon display and/or animation of CAPPIs, Echo Top or VIL maps, only those cells associated
with severe weather are circumscribed by the outline. The irregular outline over the storm in the
west on the CAPPI map of Fig 3.3a indicates the presence of an 'overhang'.
19
3.4a MESOCYCLONE Detection Algorithm
The mesocyclone identification algorithm developed for the McGill Doppler Radar is
based essentially on a procedure outlined by Zrnic et al. (1985) using NSSL data of Oklahoma
storms. Brown and Wood (1991) and Desrochers and Donaldson (1992) are two other relevant
papers on this topic (plus all the references therein). Results of our mesocyclonic detection
algorithm with thresholds that are applicable to storms of the southern region of Quebec are
summarized by Vaillancourt et al. (1997). It would be sufficient to recall here that mesocyclone
identification involves the following main steps:
1- The detection of the so-called pattern vectors (PVs) that are azimuthal runs of increasing
Doppler velocities along a fixed radar range bin. These PVs must exceed either a high
momentum threshold of 60 (m/s)km or a high shear threshold of 6 (m/s)/km. This
combination corresponds to the moderate filtering technique adopted in the Quebec study
reported by Vaillancourt et al. (1997).
2- The sorting of the more significant PVs into two-dimensional shear features according some
radial (4 km) and azimuthal (5 degrees) proximity criteria.
3- The association of shear features pertaining to the same elevation scan into potential
2-D mesocyclone features. These features must exceed a threshold radial diameter of
2 km and have an elongation such that the ratio of the azimuthal and radial diameter is
within the acceptable values of 0.4 and 1.9.
4- The examination of the vertical structure of the latter in order to classify them as "declared
3-D mesocyclones". Essentially, it is required that the 2-D mesocyclonic features be
detected on a sufficient number of elevation angles so as to give a realistic thickness to
the mesocyclone. A threshold of 2 km has been selected in our study. The volume scan
consisting of 24 elevation angles available every 5 minutes with the McGill Doppler
system is particularly useful in this respect.
20
5- The association of a "declared 3-D mesocyclone" with a cell exceeding a selectable minimum
reflectivity. For our purpose, we require a reflectivity greater than 40 dBZ on at least 5%
of the pixels within a 6 km by 6 km square centred over the meso on the two elevation
angles that are just above an altitude of 3 km.
6- Finally, a greater confidence in mesocyclone detection is achieved if a "declared 3-D
mesocyclone" is seen to persist for 2 or more radar cycles (
•
10 min.). The animation
capability of the DISPLAY system can be used to examine this temporal continuity. For
the specific need of our research effort, a mesocyclone-tracking algorithm has been
formulated in order to quantify this temporal continuity.
When a mesocyclone has been detected, its actual location is indicated by a square on
any horizontal cross-section map as in Fig. 3.3a. The size of this square is proportional to the
mean radius of the declared mesocyclone. In Fig. 3.3b and 3.3c, we exemplify how mesocyclonic
features may be better highlighted by smoothing, and especially, by removing the storm
propagation velocity from the measured radial component. This 'Lagrangian' correction does not
affect the appearance of mesocyclones that move in a direction perpendicular to the radar
viewing angle (for example, the mesocyclone in the NNW), but brings into sharper focus the
internal rotation of those moving either directly towards or away from the radar. This occurs
because a significant radial component of the precipitation pattern propagation velocity, which in
this example is in excess of 100 km/h from 250 degrees, is added to the solid body rotation of the
mesocyclones resulting into radial velocities that actually exceeded the Nyquist interval. After a
successful velocity unfolding, the approaching velocities of the vortex in the west reach
magnitudes of the order of 45 to 54 m/s that are not evident from the usual Doppler scale which is
limited by the Nyquist interval at both ends. It is only when the radial component of the overall
motion is subtracted that the mesocyclonic couplet of approaching and receding velocities is
clearly revealed as shown in the zoomed image of Fig. 3.3c.
21
(a)
(b)
(c)
(d)
Fig. 3.3: (a) Low level 2.5-km reflectivity CAPPI map with the mesocyclone and 'overhang'
indicators. (b) 3-km Doppler CAPPI after range and velocity unfolding. (c) Zoomed section of (b)
but after applying a 'Lagrangian correction' and a (3x3) smoother. (d) As in (b) and (c), with the
mesocyclone tracks added. The start and end time of each mesocyclone is indicated. The width of
the track is proportional to the meso diameter while the color provides the meso depth according
to the scale at the right of the Doppler scale
The text in red under the intensity scale of maps warning of the existence of a
mesocyclone provides information on the position of the declared mesocyclones, on the number
of elevation angles over which they have been detected and on their strength in terms of the peak
22
rotational velocity and of the average maximum shear. The latter two parameters are initially
computed at the level of a pattern vector as follows:
V
rot
= (V
2
-V
1
)/2
and
shear = (V
2
-V
1
)/L
where
V
2
is the maximum approaching velocity (positive), V
1
is the maximum receding velocity
(negative) and L is their separation or length of the pattern vector. We have defined the average
maximum shear as the average of the 2-D maxima found at each elevation angle while the peak
V
rot
is taken to be the maximum among all the PVs making up the mesocyclone. Thus, for the
mesocyclone in the NNW, peak(V
rot
) = 20 m/s while the average of the maximum shear found
over the 10 elevation angles for which a mesocyclonic couplet has been identified is computed to
be 11 m/s/km.
3.4b Mesocyclone Tracking Algorithm
Positional information and other characteristics of the declared mesocylones of a given
radar cycle are stored in the header of the low-level CAPPI map of Fig. 3.3a. Using such
information from all CAPPI maps of a RAPID product file, or for any shorter time period, a
simple tracking algorithm attempts to follow each declared mesocyclone and displays the
resulting trajectory on any suitable map as illustrated in Fig. 3.3d. This algorithm first requires
the velocity of the precipitation pattern which is computed by a cross-correlation technique on 9
sub-areas of the low-level CAPPI map, each being (80 km by 80 km). From the known position
of a mesocyclone of the current radar cycle, the computed velocity is used to forecast its position
on the next cycle. A user-defined 'search radius' outlines a circular neighborhood around this
expected position where the presence of a mesocyclone would be considered as the continuation
of the one being tracked. A mesocyclone is considered tracked if it persists for at least 2 cycles,
that is, 10 minutes. A mesocyclone track or lifetime is maintained even if no subsequent
detections are found for up to a selectable time interval but then reappears within the expected
position on the following cycle. In this case, the 'search radius' is enlarged by a suitable factor. In
our analysis, we have allowed up to 10 minutes, or 2 radar cycles, during which an existing
mesocyclone can go undetected or become untrackable. When a mesocyclone lasts for 3 or more
cycles, the user has the choice of using its actual past positions, rather then the velocity of the
23
surrounding precipitation, for the purpose of predicting its future location. As each '5-minute
mesocyclone detection' or (m5) is associated with a particular 'tracked mesocyclone' (tm), it is
'tagged' to prevent it from being considered as a possible member of a subsequent tm. The
tracking algorithm terminates when each one of all the m5's has been either associated with a tm
or ignored because of the 10-minute restriction on a tm lifetime. For each tm, the algorithm
outputs its total duration, path length and the average over its lifetime as well as the '5-minute
values' of the following parameters: diameter, depth (vertical thickness), maximum shear and
peak rotational velocity. Note that in the presentation of the results in section 4.1 we will
concentrate mainly on the m5 characteristics since those averaged over the mesocyclone lifetime
are biased by the great variability of the latter.
An extension of this algorithm has also been successfully applied to the tracking of VIL
and GUST cells. The size of these cells is first defined by a user-entered threshold and the peak
values and their locations are tabulated. Secondary peaks are ignored depending on a selectable
proximity threshold and neighboring cells may also be agglomerated into a larger cell. The centre
of gravity of the resulting cells and their displacement, the latter determined by local cross-
correlation, are then used by a tracking algorithm in a manner similar to the one described for
mesocyclones. However, the results for the VIL and GUST maps analysis will not be provided in
terms of cell track statistics because of the inevitable complications arising out of the merging
and splitting of cells that would bias such an approach. Instead, statistics will be presented in
terms of the observed pixel values of these parameters over the specified radar coverage.
24
4- RESULTS
4.1 Mesocyclone Statistics
When applied over our entire 8-year sample, the first 5 steps of the mesocyclone detection
algorithm described in 3.4a yield a grand total of 1682 detections. Some of these are spurious as
when a shear line is not recognized as such by step #3 or when the peculiar configuration of a
small ground echo and of the surrounding precipitation generates a false alarm as occasionally
seen over the Mercier Bridge, 25 km east of the radar. Some others are true mesocyclones but
short-lived, being captured by only one radar cycle. Both the spurious and the short-lived mesos
are eliminated by the tracking algorithm because the former are usually stationary and fail to be
also present at the expected downwind location. Our persistence requirement of at least 2 radar
cycles implemented in the final step (#6) of our algorithm has maintained 1293 (77%) of the 'm5'
detections distributed among 308 trackable mesocyclones (or 'tm's for short). The yearly
frequency of the latter is portrayed in Fig. 4.1a where it is seen that the years 1998, (with 47),
1996, (with 51), and particularly 1994 with as many as 71, rank as the top 3 years. The year 2000
with only 14 'tm's ranks the lowest. Of course, we cannot ascertain what percentage of these
tracked vortices actually did give rise to a confirmed tornado touchdown along their trajectory or
to other kind of documented damage at the ground. In fact, only about 4 to 6 weak (F0 or F1)
tornadoes per year are confirmed in Southern Quebec, Vaillancourt (1999), although some may
escape visual observations due to the low density of the population in the many forested areas
under our radar coverage.
25
(a)
(b)
Fig. 4.1: Yearly distribution of the tracked mesocyclones and of the 'mesocyclone hours'.
In Fig. 4.1b, the number of 'm5' detections comprising the tracked mesocyclones of each
year is simply multiplied by 5 minutes in order to derive the distribution of 'meso hours'. Note
that in this simple exercise, 2 co-existing mesocyclones each lasting for 30 minutes would
account for 1 'meso hour'. The result augments the exceptional nature of the year 1994 since the
average number of 'm5' detections (nearly 5) for the 71 'tm's is also higher than for all the other
years. The year 1998 also had relatively long lived mesos, equivalent to 4.7 'm5's which places it
higher than 1996, the latter being closer to the average of 4 detections per meso. The average of
14.3 'meso hours' per year is obtained by omitting the 1993 data, which, as stated in section 2.4,
is rather incomplete.
26
The distribution of the lifetime of the tracked mesocyclones is provided in Fig. 4.2a while
the distance traveled, or path length, is given in Fig. 4.2b. Here, the 'lifetime' is computed from
the time difference between the initial and final detection and thus includes any intervening gaps.
(a)
(b)
Fig. 4.2: Distribution of the lifetime of the tracked mesocyclones and of their path length.
Five minutes are then added to this difference in order that a mesocyclone observed over say 2
radar cycles yield a lifetime of 10 minutes. The length of the trajectory is simply determined from
the initial and final position, not by summing the 5-minute segments. Unlike the typical
Oklahoma storms lasting for about an hour according to Zrnic et al. (1985), only 22, or 7.1%,
attained this persistence in the Montreal region. However, note that both the duration and the path
length are affected by boundary conditions, that is, by the inability of the algorithm to detect such
relatively small-scale vortices with equal efficiency at the farther ranges. (Refer ahead to Fig. 4.6
for a preview). We have not separated the statistics in terms of mesos whose lifetimes are entirely
within the detectable range and those that appear to emerge or vanish across this undefined
boundary. Since the latter are the stronger storms, it is likely that their path and duration are at the
upper end of the distribution. The typical magnitude for these two parameters for such storms is
27
thus somewhat larger than indicated by the histograms. On the other hand, the weaker
circulations that may have been present at farther ranges were likely missed by the algorithm and
thus are not part of these statistics. Since we have chosen, for better clarity in display, to limit the
upper scale of these histograms to 100 minutes and to 100 km, there are five mesocyclones which
are 'off limits'. These are identified in Table 4.1 where we have also included the one with a
lifetime of exactly 100 minutes in order to underline the exceptional day that was August 4th,
1994, (as was also June 29-30th, 1998 according to Table 4.2).
Date
Time (UTC)
Duration (hh:mm)
Track Length (km)
4-August-1994
18:21-21:01
2:45
153
29-June-1998
18:29-20:44
2:20
93
21-June-1997
18:52-20:47
2:00
110
18-July-1995
16:30-18:20
1:55
95
4-August-1994
19:16-20:51
1:40
81
28-August-1994
19:39-21:04
1:30
124
Table 4.1: Times of mesocyclones with tracks that persisted for at least 100 minutes or of a
trajectory in excess of 100 km.
In order to identify as well as quantify the other prominent days or events in our sample
with substantial meso-cyclonic activity, we have summed all the 'm5' detections of a given day,
or of a sequence when it extends across midnight. The result may be referred to as the 'meso
index' of the day. Those that attain a 'meso index' of 18 or more, equivalent to at least 1.5 hours
of meso-cyclonic activity, are listed in Table 4.2. Under the column 'Mesos', we provide the total
number of tracked mesos and, in brackets, those with at least 3 'm5' detections. The combination
of neighboring days as one event and the exclusion of the one event in the year 2000 constitutes
the sample of 22 cases that have been the object of a separate investigation by John Gyakum of
McGill University. The results of his analysis highlighting particular features of the large-scale
circulation in the days preceding such events appear in Appendix A.
28
Year
Day
Time (UTC)
Mesos
Index (hh:mm)
1993
20-21 July
19:50 - 00:20
6 (3)
20 (1:40)
"
29 July
18:40 - 2040
6 (3)
20 (1:40)
"
31 August
19:20 - 2110
8 (6)
27 (2:15)
1994
31 May
18:50 - 20:30
3 (2)
25 (2:05)
"
12 June
19:20 - 21:30
6 (5)
27 (2:15)
"
29 June
19:00 - 22:00
5 (4)
25 (2:05)
"
21-22 July
19:30 - 01:40
21 (11)
75 (6:15)
"
4 August
18:20 - 24:00
11 (9)
91 (7:35)
"
28 August
15:40 - 21:30
4 (4)
40 (3:20)
1995
18-19 July
16:30 - 01:00
16 (6)
68 (5:40)
1996
19-20 May
22:50 -24:00
13 (8)
40 (3:20)
"
4 June
20:10 - 23:20
5 (4)
28 (2:20)
"
11 June
18:20 - 23:30
15 (10)
69 (5:45)
"
14-15 July
21:10 - 01:00
4 (4)
22 (1:50)
"
16 July
17:40 - 19:40
6 (3)
18 (1:30)
1997
21 June
18:50 - 21:20
4 (2)
31 (2:35)
"
25 June
19:10 - 21:00
4 (4)
32 (2:40)
"
15 July
20:20 - 24:00
9 (4)
29 (2:25)
"
28 July
17:00 - 19:00
7 (6)
31 (2:35)
1998
27 June
22:20 - 23:10
2 (2)
18 (1:30)
"
29-30 June
17:10 - 01:40
19 (12)
100 (8:20)
"
27 September
18:50 - 21:40
5 (3)
24 (2:00)
1999
5 July
06:00 - 07:00
7 (4)
20 (1:40)
"
6 July
15:40 - 23:00
6 (3)
25 (2:05)
2000
7 August
19:20 - 24:00
6 (4)
29 (2:25)
Table 4.2: Identification and quantification of the most meso-cyclonic events of
our 8-year study in terms of the number of trackable mesos and of the 'meso
index' as defined in the text. Those with an index exceeding 50 are highlighted.
29
We have computed the diameter and vertical extent averaged over the lifetime of each
mesocyclone but such statistics are highly biased by the characteristics of those that are short-
lived. Therefore, we prefer to present in Fig. 4.3 the distribution of these parameters as obtained
from the 'm5' detections that were associated with the tracked mesocyclones. These distributions
are bounded by the imposed threshold of 2 km in both the diameter and vertical extent of the 2-D
features comprising a mesocyclone. It appears from Fig. 4.3b that shallower features should
perhaps be accepted as mesocyclones, particularly if they exhibit a time continuity that will be
eventually verified by the tracking algorithm. This suggestion, together with a re-analysis
involving the modification of the various parameters that affect a mesocyclone detection, should
be the object of future investigations. Surprisingly, a scatter plot of the diameter and depth (not
shown) reveals no correlation between them.
(a)
(b)
Fig. 4.3: Distribution of the diameter and of the vertical depth of the 'm5' detections that are part
of the tracked mesocyclones.
30
The distributions of the peak rotational velocity and of the average of the maximum shear
of the 'm5' detections are likewise presented in Fig. 4.4a and b respectively. (Refer to the end of
section 3.4a for the definition of these parameters). Again, while Oklahoma mesocyclones exhibit
an average rotational velocity in excess of 20 m/s, the peak magnitude reached by our storms is
generally confined within the 10-15 m/s range. The linear correlation coefficient between these
two parameters is of the order of 0.5.
(a)
(b)
Fig. 4.4: Distribution of the peak rotational velocity and of the average maximum shear (as
defined in section 3.4a) of the 'm5' detections that are part of the tracked mesocyclones.
31
In Fig. 4.5, we illustrate the hourly and monthly distributions of mesocyclones. In (a), we
use the times of the 'm5' detections while in (b) only the time at the start of the track is
considered. In (a), the hour shown refers to the beginning of the 1-h interval; thus the peak of 319
(25%) occurs during the 3:00 to 4:00 PM interval. Over half (56%) are included in the 3-h
interval between 2:00 and 5:00 PM while nearly the entire sample (84%) lies within the first 7
hours of the afternoon. The fact that such distribution reflects the diurnal trend of temperature is
not surprising. The secondary peak at (01:00-02:00) is exclusively due to the 'bow' squall line
event of July 5th, 1999 which comprised short-lived mesocyclones.
In Fig. 4.5b, each month has been subdivided into three 10-day periods, allowing 11 days
for the months with 31 days. However, the data shown have not been normalized for this slight
inconsistency. There is a preferred period from about mid-June until early August, but severe
weather associated with mesocyclones may also strike in May and September. The local
minimum in the first 10 days of July may be accentuated by the unfortunate timing of the data
gaps in 1998 and particularly in 1995 as noted in section 2.4.
(a)
(b)
Fig. 4.5: Hourly and monthly distribution of mesocyclones in the Montreal area.
32
Finally, we display the geographical distribution of the mesocylones detected by the
McGill Doppler radar during the 8 years of our study in Fig. 4.6 below. Here, each 'm5' detection
that is part of a tracked mesocyclone generates a circular imprint on our map of size equal to the
measured radial diameter. A subsequent detection is added if it influences the same region. In
Fig. 4.6, the number of times that a particular pixel came under the influence of a mesocyclone is
indicated by the color scale divided by 10. (This unconventional procedure is merely done in
order to use the existing 'fixed' 16-level scale for the display of the RAPID accumulation module.
The rest of the information should be ignored). Thus, the dark green, orange and red correspond
to 1, 2 and 3 mesos respectively. The range limits of the algorithm are from 5 to 120 km but we
Fig. 4.6: Geographical distribution of mesocyclonic vortices around the McGill Doppler radar
for 1993 to 2000 inclusively. Range rings are 20 km apart. Divide the "Mesos" scale by 10 in
order to obtain the number of mesocyclones affecting each pixel of the map.
33
notice that the detections beyond about 80 km are drastically reduced. Since during periods of
cyclonic activity data is often collected in high resolution mode, we need to determine whether
this is the result of the sudden reduction in radial resolution occurring at a range of 80 km, or
whether it is a combination of the gradual loss in radial and azimuthal resolution. This can be
achieved by simulating a lower resolution from the archived higher resolution data. The
detectability loss beyond 80 km should also encourage us to devise an algorithm with range
dependent filtering in order to capture the weaker circulations at farther ranges while avoiding
excessive false alarms at nearer ranges.
There is a distinct greater probability of mesocyclones in the western part of the map,
with many pixels reporting more than 3 occurrences near and just north of the St-Lawrence River
but south of Highway 401. (The latter is drawn as a line between the Ottawa and St-Lawrence
Rivers). Another region of maximum activity is the Mirabel area and the surroundings. The
absence of detections within 5 km corresponds to the start of the search for mesocyclonic features
at that range. However, the almost total absence of vortices up to a range of nearly 20 km appears
to be closely related to the presence of significant water bodies in the vicinity of the radar.
34
4.2 - RESULTS (Reflectivity Analysis)
We now present the results of the severe weather analysis as obtained from the
'reflectivity only' maps selected to identify convection, namely, the VIL, UVIL, GUST,
OVERHANG and the 7-km CAPPIs. All these maps describe the upper level structure of
thunderstorms, but most of the emphasis will be on the UVIL that does not suffer from ground
echoes, anomalous propagation, bright band and shadowing effects. In fact, in deriving the
statistics for the 'regular' VIL and the GUST maps, a significant area influenced by echoes from
the Green Mountains in the southeast starting at about the 120-km range had to be excluded. Data
within a 12-km range is also ignored for all maps because the last elevation at 34.4
°
reaches a
height of 7 km at that range. Recalling that all maps are (240x240) arrays, the maximum range
for 1-km resolution maps has been chosen to be 120 km in order to ignore the data beyond that
range near the corners. All plots obtained from 2-km resolution maps are based on data within a
range of 180 km. Beyond this range, along a small sector in the NW, even the UVIL based on
data above 5 km begins to be affected by some shadowing from the required lower elevation
scans. This range was also selected on the initial assumption that the density of severe weather
beyond it as detected by radar would be diminishing because of the smoothing effect of the
widening beam width. We shall see as in Figs 4.12 and 4.13 that this is surprisingly not the case
but we have nonetheless chosen to retain a maximum range of 180 km for the statistical plots.
However, a re-derivation of all the statistics with the range extended up to 240 km performed
while finalizing this report did obviously increase the 'area' and 'hours' above the thresholds but
does not appear to have significantly altered the relative severity among the various years.
We first present in Figs. 4.7 and 4.8 the frequency distribution of the selected severe
weather parameters as obtained from all the Cartesian maps of the entire 8-year data set. These
log-linear distributions begin at the thresholds chosen in section 2.5 and indicate the frequency of
occurrence in terms of the area in km
2
within each subsequent unit interval. They terminate at a
suitable upper level where the rarity of the occurrences renders the curves erratic. This latter
effect can be eliminated by plotting the same data in terms of cumulative frequency distributions,
but such plots are not presented here. Plots are provided for both the 1- and 2-km resolution
pixels, for ranges up to 120 and 180 km respectively, making the area of analysis for the latter
2.26 larger. (However, we are reminded that because of the 0.86
°
beam width of the McGill
35
radar, or the 1
°
sampling of the archived data, the azimuthal resolution is actually 2 and 3 km
respectively at those ranges. Additional averaging is performed during the polar-to-Cartesian
transformation, prior the application of the thresholds.)
(a)
(b)
Fig. 4.7: Frequency distribution of the logarithm of the area of the indicated severe weather
parameters obtained from 8 years of Cartesian maps. (a): 1-km resolution between ranges of 12
to 120 km. (b): 2-km resolution from 12 to 180 km.
The parameters plotted in Fig. 4.7 exhibit a linear behaviour over a significant portion of
the abscissa, that is, the frequency of occurrence (or area) within any unit interval can be
expressed as
Area = C exp
-
λ
x
We provide in Table 4.3 the values for Ln(C) and
λ
, where Ln is the natural logarithm.
Limits
Ln(C)
λλ
1 km
2 km
1 km
2 km
VIL
40-120 kg/m
2
10.73
10.91
0.072
0.063
UVIL
20-80 kg/m
2
10.31
10.85
0.090
0.082
GUST
15-40 m/s
12.83
13.25
0.188
0.179
Table 4.3: Values of the Ln(C) and
λ
parameters describing the log-linear relationships of the
curves plotted in Fig 4.7.
36
(a)
(b)
Fig. 4.8. (a): As in Fig. 4.7 but for the reflectivity on 7-km CAPPI maps. (b): Number of hours
when the 'overhang' exceeded the indicated area.
In Fig. 4.8a is presented the distribution of the reflectivity on 7-km CAPPI maps. Note
that the 'polar-to-Cartesian' processing alluded earlier causes the area of reflectivities > 58 dBZ
on 2-km resolution maps for ranges 12 to 180 km to be less than that from 1-km resolution maps
that extend only up to 120 km. (A similar effect will be seen in Figs 4.9 and 4.11). The (3x3)
Cartesian smoother applied to all 7-km CAPPI maps further contributed to this behaviour. In
retrospect, it may not have been appropriate to implement it.
The presentation of statistics in terms of area within the unit interval of a parameter has
some practical limitations since forecasters and the public alike do not have an inherent natural
perception of 'area'. Instead, they can more easily relate to the time during which severe weather
may persist. A cumulative frequency parameter is likewise better understood. Therefore, since the
'overhang' parameter is probably more difficult to grasp than the others, we have chosen to
present in Fig. 4.8b the number of hours that an overhang of a certain size has been exceeded on
the 1- and 2-km resolution maps. (Here, the original formulation of the overhang algorithm by the
RAPID software is such that a search is performed over the entire (240x240) array and thus
37
includes the corners of these maps). The smallest area threshold has been selected to be 4 km
2
,
corresponding to 1 pixel of the 2-km resolution maps. Thus, from Fig. 4.8b, we can state that
there were about 20 (~60) hours over the 8-year period when the overhang area on the 1 (2)-km
resolution maps exceeded 40 km
2
.
(a)
(b)
Fig. 4.9:Yearly distributions of the areal coverage of the 'elevated VIL' parameter for the two
indicated ranges, or resolutions.
We momentarily return to the presentation of statistics in terms of area, but expressed as a
cumulative distribution, in order to compare the relative severity of the years analyzed. For this
purpose, we select the parameter least affected by radar artifacts, namely the UVIL. The
equivalent distributions derived with the other parameters exhibit similar features, both in their
linearity in log-linear coordinates and in the general relationship among the various years. We
concentrate on the longer-range data of Fig. 4.9b because the fewer intersections of curves imply
that the relative severity among the various years is not dependent on the magnitude of the
threshold. Ignoring 1993 because of its incompleteness, 1996 and 2000 clearly stand out as the
two most benign years. The lack of extensive convection in 1996 is surprising since, as seen
earlier in Fig. 4.1, it ranked among the top three in terms of mesocyclonic activity. The year 2000
instead is 'quiet' using both barometers. The year 1995 lies in between these 'quiet' years and the
38
remaining 4 active years. Noteworthy is the different level of severity of the year 1999 in (a) and
(b) indicating that the high intensity cells occurred at ranges between 120 and 180 km.
(a)
(b)
Fig. 4.10: Hourly distributions of deep convection derived from the entire 8-year data set given
in terms of the specified intensity and area thresholds of the UVIL parameter. The area
thresholds are 10 and 40 km
2
for the 1- and 2-km resolutions respectively.
We now combine the entire 8-year sample in order to illustrate in Fig. 4.10 the hourly
distribution of convection as was done earlier for mesocyclones in Fig. 4.5a. The intensity of the
convection is quantified in terms of three UVIL thresholds and of an area threshold that in
combination are suitable for the resolution and coverage analyzed. Thus, if on a 1-km resolution
map there are at least 10 pixels exceeding say 20 kg/m
2
, then 5 minutes is added to the 'hour slot'
that includes the time of such map. In Fig. 4.10b, the intensity thresholds have been reduced and
the area threshold increased to 40 km
2
in order to partially compensate for the averaging over the
larger (2 km by 2 km) pixels and the larger area of analysis. The resulting curves are of a similar
order of magnitude for both resolutions, again displaying a peak during the 15:00 to 16:00 EST
interval as was the case with mesocyclones. This peak tends to broaden as the intensity threshold
is increased. No severe convection is observed between 05:00 and 10:00 EST but a few
exceptional events caused the secondary peak around 02:00 EST.
39
(a)
(b)
Fig. 4.11:Yearly distributions of intense convection given in terms of hours for which various
intensity and area thresholds have been exceeded on UVIL maps.
A similar method of analysis is adopted in order to stratify hours of severe weather by
year as illustrated in Fig. 4.11. The outcome resembles the previous results of Fig. 4.9, namely,
the 'quieter' years of 1996 and 2000 and the contrast in statistics between (a) and (b) for the year
1999. In addition, this figure enables us to determine the number of hours that convection of a
particular intensity and size has been observed within radar ranges of 120 and 180 km
respectively. For example, from Fig. 4.11a, the time during which the UVIL values exceeded a
threshold of 30 kg/m
2
over at least 10 km
2
within a range of 120 km was 16.4, 5.9, 1.0, 6.9, 14.4,
6.5 and 2.2 hours for the years 1994 to 2000 respectively.
40
(a)
(b)
Fig. 4.12: Geographical distribution of deep convection for the years 1993 to 2000 inclusively.
In (a), the magnitude of the UVIL measurement is integrated if it exceeds a threshold of 15 kg/m
2
and then the sum is divided by 10 to conform to a typical VIL scale. In (b), we provide the
number of minutes over the 8-year period that the UVIL measurement exceeded the same
threshold. Range rings are 40 km apart.
In Fig. 4.12 above, we display the geographical distribution of deep convection within
240 km from the McGill radar. Deep convection is defined by the 15 kg/m
2
threshold in the UVIL
pixel measurement. In Fig. 4.12a, the actual value is added to each pixel wherever that threshold
is exceeded. The result after integrating the 8 years of data is divided by 10 in order to confine
the sum to be within the range of a typical VIL scale. The other annotations on the right-hand
side of the map should be ignored. In Fig. 4.12b, '5 minutes' is contributed by each pixel affected
by deep convection as just defined. A (3x3) smoother operating on only the non-zero pixels has
been applied to both maps. Similar maps have been derived for 1-km resolution data extending
only up to 120 km but we prefer to comment on the long-range analysis. We observe that these
fields are not spatially uniform but exhibit preferred regions of deep convection or of lack of it.
Even by ignoring a small sector in the NW beyond about 200 km that may still be affected by
some shadowing, there remains a reasonably well-defined north-south corridor across the length
of the map where deep convection is less likely to occur. This region consists mainly of the
forested lands of the Laurentians Hills in the north and of the Adirondacks Mountains in the
south. The nearby minimum at ranges between about 15 and nearly 40 km is not due to
geometrical factors involving the highest elevation angle but appears to denote a genuine lack of
strong convective activity over the Metropolitan area of Montreal and surrounding areas. Regions
41
of enhanced convective activity are observed over the St-Lawrence flatlands, in particular over
the Province of Ontario in the west and southwest, in the east beyond Sherbrooke and in northeast
up to Quebec City. Another area is evident over the New England States in the farther ranges of
the southeast. The convection over Ontario may be considered as an extension of the Canadian
'mini-tornado alley' in southeastern Ontario that extends along a corridor from Windsor up to
Ottawa. The activity that appears to be aligned along a northwest-southeast direction at ranges
beyond 120 km between Montreal and Quebec City is related to the frequent occurrence of a
stationary front that in the summer separates the colder air mass of the lower St-Lawrence from
the warmer and more humid air mass that intrudes just past Montreal. Along this front, severe
thunderstorms are seen to form as was the case for several days in early June 1999, explaining the
behaviour of the curve for 1999 seen earlier in Fig. 4.11b.
The lack of a general loss in detectability as is usually observed with many radar
parameters as the range is increased is somewhat puzzling. On the contrary, the opposite effect
manifests itself on these climatological results in spite of the beam broadening with range. We
must mention that in addition to the standard 1/r
2
range correction performed on our radar data,
we also add an atmospheric gases (mainly oxygen) correction amounting to 1.5 db/100 km, (Fig.
6.1 of Battan, 1973). Is it possible that the extra 3 dBZ and more added to the reflectivity beyond
200 km is mainly responsible for the observed regional distribution ? A careful scrutiny of all
surface reports of severe weather over the past 8 years may be able to answer this question.
Finally in Fig. 4.13 we provide the year-by-year distributions that together contribute to
the overall result shown in Fig. 4.12b. A lower threshold of 10 kg/m
2
has actually been used in
order to increase the number and size of the observed features. We immediately recognize the
relatively inactive years of 1995, 1996 and 2000, the latter displaying a striking inverse-range
effect. Convective activity is seen to be concentrated mainly in the northern half of our radar
coverage in 1994, to the southern half in 1998, mainly in the west in 1997 while the storms that
developed along the stationary front between Montreal and Quebec City in 1999 are well
depicted. Can these features also be related to a predominant large-scale pattern of hemispheric
circulation prevalent over our region in those particular years as it has been confirmed for
mesocyclonic activity in Appendix A? This can be the object of future investigations.
42
43
Fig. 4:13: Geographical distribution of convective precipitation for each of the 8 years analyzed
in terms of the number of minutes that a value of 10 kg/m
2
was exceeded on the UVIL map. Range
rings are 40 km apart.
44
5- CONCLUSIONS
Thanks to a 'summary' log that permits the quick selection of convective periods, to the
efficient extraction of the raw volume scans from exabyte tapes, and to the existing McGill
RAPID software that allows a complete analysis of radar data in simulation mode, we were able
to process 8 summers (1993-2000) of radar data efficiently and effectively. This software
package has also been used to assess the methodology for a similar radar-based severe weather
climatology in Alberta. The result of this effort applied to one month (July 2000) of data from the
radar at Carvel is included in an attached document authored by J. Brimelow and G. W. Reuter of
the University of Alberta.
Approximately 2000 hours of McGill radar data have been analyzed, out of which about
1500 hours had some degree of convection according to the radar parameters selected for its
identification. These were: VIL, upper level VIL or UVIL, GUST, OVERHANG and the
reflectivity on 7-km CAPPI maps. A mesocyclone detection and tracking algorithm also played a
crucial role in determining the characteristics and the relative occurrence of small-scale
circulations. The years 1994 and 1998 were especially active in this respect, in particular 4-
August-1994 and 29-June-1998 when mesocyclones were tracked for periods of the order of 1
and 2 hours and over distances of the order of 100 km, (Table 4.1 and 4.2). A preliminary
investigation on the relationship between 22 mesocyclone events and the characteristics of the
hemispheric circulation on days prior such events, prepared by Prof. J. Gyakum of McGill
University, is presented in Appendix A. Hourly and monthly distributions of mesocyclone
detections in Fig. 4.5 reveal their higher probability of occurrence during the 3-hour period
between 2 and 5 PM local time and from mid-June to early August. Their geographical
distribution portrayed in Fig. 4.6 is the first of its kind produced in Canada. It reveals an
increased relative probability over the western coverage of the McGill radar as well as to the
north near Mirabel, and a reduced probability near the Metropolitan area of Montreal, in
particular over and in the neighborhood of water bodies. This analysis also shows a marked loss
in detectability for ranges beyond 80 km, suggesting an improvement of the algorithm by means
of range dependent thresholds.
45
The analysis of deep convection in terms of the 'reflectivity only' parameters, Fig. 4.7 and
4.8, shows an exponential decrease in the probability of the higher thresholds. The UVIL map has
been preferred for comparison of the 8 years analyzed because it is the least affected by radar
artifacts. Figs. 4.9 and 4.11 both classify 1996 and 2000 as relatively inactive while the size and
frequency of occurrence of convection in 1994, 1997, 1998 and 1999 is significantly higher. The
geographical distribution of UVIL measurements has been obtained for both the entire 8-year
period, Fig. 4.12, and for the individual years, Fig. 4.13. In contrast with the mesocyclone
detection, there appears instead to be a higher occurrence of deep convection at the farther
ranges, both west and east of the radar. A relative minimum is quite noticeable along a corridor
running north and south of the radar. The exact explanation of this result in yet uncertain, but it
appears that the heavily forested areas of the Laurentians to the north, of the Adirondacks in the
south and the small lakes near Montreal tend to suppress strong convection while the flatlands
along the St-Lawrence River enhances it.
Many of the initial objectives listed in the introduction have already been achieved,
namely the compilation of a climatology for southern Quebec and the development of a
methodology that is being implemented in southern Ontario and Alberta. We have also identified
the planetary-scale precursors of severe mesocyclonic events. But this work also points out the
need for continuing this effort. For example, we need to investigate whether similar patterns are
observed for other severe weather days that lack small-scale circulations. A convective index for
the days analyzed will need to be developed, as was done for mesocyclones.
Other areas that may warrant additional research are:
- To devise the optimum combination of range-dependent thresholds for the detection of
mesocyclonic vortices so as to increase detection at far ranges while avoiding false alarms at
nearer ranges.
- To eliminate completely the spurious mesocyclone detections caused by shear or other factors.
To separate these spurious detections from the short-lived (< 10 minutes) true mesocyclones.
- To investigate the effects of high spatial resolution on mesocyclonic detection. Likewise, to
reduce the MRO data sets to a temporal and vertical resolution proposed for the Canadian Radar
46
network in order to examine any possible loss in detectability and to devise appropriate
compensating measures.
- To correlate both the radar and surface reports observations of severe weather, at least in terms
of the general features of the geographical distributions noted in this report.
- To determine whether the tracking algorithm can be applied for the derivation of characteristics
of non-rotating storms. Is there a relationship between the number and size of cells observed over
the years?
- To develop the full potential of the tracking algorithm by applying it for short-term forecasting
of individual severe weather storms.
- To determine the exact cause of the inverse range effect seen in the geographical distribution of
deep convection. Can this be related to any prevailing synoptic condition?
- To include the year 2001 into our analysis. Incidentally, as of July 13, 2001, there has been no
data loss for the current year.
-To assess the effect of various degrees of calibration uncertainty on the derived statistics.
- To include short-period rainfall accumulation statistics deduced from radar as a separate
measure of severe weather.
6- ACKNOWLEDGEMENTS
We foremost acknowledge the support of the Climate Change Action Fund for providing
the required financing that made this research possible. We also thank the staff at the McGill
Weather Radar Observatory, in particular Alamelu Kilambi who participated in the development
of the RAPID software, and GyuWon Lee, a Ph.D student who assisted in the preparation of
some of the graphs of this report. The first author also benefited from electronic and personal
discussions with Julian Brimelow of the University of Alberta during his stay at MRO.
47
7- REFERENCES
Amorim, W., O. Massanbani and I. Zawadzki, 1997: An evaluation of the predicted
pulse-type thunderstorm gusts using horizontal divergence field from Doppler
radar data. Preprints 28th Conf. on Radar Meteor., Amer. Meteor. Soc., 378-379.
Austin, G.L., A. Kilambi, A. Bellon, N. Leoutsarakos, A. Hausner, L. Trueman, and M. Ivanich,
1986: Rapid II; An operational, high speed interactive analysis and display system for
intensity radar data processing. 23rd Conf. on Radar Meteor. & Conf. on Cloud Physics,
Snowmass, Co, Amer. Meteor. Soc., JP, p 79-82.
Battan, L. J., 1973: Radar Observations of the Atmosphere, University of Chicago Press,
279pp.
Bellon, A., and A. Kilambi, 1999: Updates to the McGill RAPID (Radar data Analysis,
Processing and Interactive Display) system. 29th Conf. on Radar Meteor., Amer.
Meteor. Soc., 121-124.
Bellon, A. andI. Zawadzki, 1994: Microburst climatology for the Montreal area using the
McGill Radar Observatory. Final report for Transport Canada Aviation, Contract
No. T8080-3-0656/01-33, 25 pages.
Bellon, A., and G.L. Austin, 1976: SHARP (SHort-term Automated Radar Prediction). A
real-time test. Preprints 17th Conf. on Radar Meteor., Amer. Meteor. Soc.,
522-525.
Brown, R.A. and V.T. Wood, 1991: On the interpretation of single-doppler velocity patterns
within severe thunderstorms. Weather and Forecasting, 6, 32-48.
Desrochers, P.R and R.J. Donaldson Jr., 1992: Automatic tornado prediction with an improved
mesocyclone-detection algorithm. Weather and Forecasting, 7, 373-388.
48
Duncan, M.R., A. Bellon, A. Kilambi, G.L. Austin, and H.P. Biron, 1992: PPS and PPS jr: A
distribution network for weather radar products, severe warnings and rainfall forecasts. 8th
International Conference on Interactive Information and Processing Systems for Meteorology,
Oceanography and Hydrology, Atlanta, Georgia, p 67-74.
Emanuel, K. A., 1981: A similarity theory for unsaturated downdrafts within clouds. J.
Atmos. Sci., 36, 2462-2478.
Fischer, A. P., 1997: A synoptic climatology of Montreal precipitation. McGill
University Master's Thesis. 71 pp.
Green, D. R., and R. A. Clark, 1972: Vertically Integrated Liquid Water Content - A New
Analysis Tool. Mon. Wea. Rev., 100, 548-552.
Hudlow, M. D., and V. L. Patterson, 1979: GATE Radar Rainfall Atlas. NOAA Special
Report. 155 pp.
Kilambi, A. et al., 1997: RAPID: A Radar data Analysis, Processing and Interactive
Display system. 28th Conf. on Radar Meteor., Amer. Meteor. Soc., 220-221.
Kozak, S., 1998: Lightning in Alberta thunderstorms: Climatology and case studies.
M.Sc. Thesis, Univ. Alberta, 127 pp.
Kruger, A., W. F. Krajewski, D. J. Seo, and J.P. Breidenbach, 1999: Development of a
large radar database for hydrometeorological studies. Preprints 29th Conf. on
Radar Meteor., Amer. Meteor. Soc., 945-948.
Lackmann, G. M., and J. R. Gyakum, 1996: The synoptic and planetary-scale signatures
of precipitating systems over the Mackenzie River Basin. Atmos-Ocean, 34,
647-674.
Lackmann, G. M., J. R. Gyakum, and R. Benoit, 1998: Moisture transport diagnosis of a
49
wintertime precipitation event in the Mackenzie River Basin. Mon. Wea. Rev.,
126, 668-691.
Lapczak S. and others, 1999: The Canadian National Radar Project. Preprints 29th Conf.
on Radar Meteor., Amer. Meteor. Soc., 327-330.
Mapes, B. E., 2001: Climatological studies with Doppler radar. Preprints 30th Conf.
on Radar Meteor., Amer. Meteor. Soc., Paper No. 12A.11.
Petrocchi, P. J, 1982: Automatic detection of hail by radar. Air Force Geophysics Laboratory
report No. AFGL-Tr-82-0277. 33 pages.
Reuter, G. W., and N. Aktary, 1995: Convective and symmetric instabilities and their
effects on precipitation: seasonal variations in central Alberta during 1990 and
1991. Mon. Wea. Rev., 123, 153-162.
Smith, S.B., G.W. Reuter, and M.K. Yau, 1998: The episodic occurrence of hail in
central Alberta and the Highveld of South Africa. Atmosphere-Ocean, 36,
169-178.
Stewart, S. R., 1991: The prediction of pulse-type thunderstorm gusts using vertically
integrated liquid water content (VIL) and the cloud top penetrative downdraft
mechanism. National Weather Service Office, FAA Academy, Oklahoma City,
NOAA Tech. Memo. NWS SR-136.
Vaillancourt, P., A. Bellon, and I. Zawadzki, 1997: Results from a mesocyclone detection
algorithm over southwestern Québec, Canada. Preprints 28th Conf. on Radar Meteor.,
Amer. Meteor. Soc., 349-350.
Vaillancourt, P., 1999: Severe weather detection in the Quebec region and its limitations.
Preprints 29th Conf. on Radar Meteor., Amer. Meteor. Soc., 125-128.
50
Zrnic, D.S., D.W. Burgess and L.D. Hennington, 1985: Automatic detection of mesocyclonic
shear with Doppler data. J. Atmos. Oceanic Technol., 2, 425-438.
51