analysis and planning for precipitation augmentation for
TRANSCRIPT
FINAL REPORT
ANALYSIS AND PLANNING FOR PRECIPITATION AUGMENTATION FOR CROPS EXPERIMENT
by
Stanley A. Changnon Robert R. Czys
Steven E. Hollinger Floyd A. Huff
Mary Schoen Petersen Robert W. Scott
Donald W. Staggs and
Nancy E. Westcott
ILLINOIS STATE WATER SURVEY
Cooperative Agreement NA89-RAH09086 for the period September 1, 1989 to August 31, 1990
Atmospheric Modification Program, Environmental Research Laboratory National Oceanic and Atmospheric Administration
Champaign-Urbana, Illinois November 1990
TABLE OF CONTENTS Page
INTRODUCTION 1
ECHO CORE BEHAVIOR STUDIES 4
1. Analytical Techniques and Software Development 4 a. Echo Core Tracking Software 4 b. Echo Core Display and Statistics Software 5 c. Echo Field Display Software .. 6
2. Preliminary Results from the 1989 Radar Data 7 a. Experimental Units 7 b. Aircraft Position 9 c. Comparison of Aircraft Penetrations and Radar Cores 10
3. Results from other studies 12 4. Studies of First and Early Echo Properties 13
a. Introduction 13 b. Data Source and Analysis Procedures 14 c. Conditions on 25 July 86, 15:33-19:01 CDT 15 d. Conditions on 26 Aug 86, 08:00-12:30 CDT 17 e. First echo characteristics - dimensions 18 f. Factors influencing first echo location 22 g. Summary 25
5. Echo Core Properties at Time of Merger 27 a. Introduction 27 b. Data 29 c. Description of merging 30 d. Types of Merger 33 e. Relative age of the merging pair 38 f. Differences between growing and declining echo cores 38 g. Characteristics of echoes with continued growth 44 h. Manner of merging and bridge formation 44 i. Height and depth of bridge formation 45 j. Summary and Conclusions 48
6. References 49
IN-CLOUD STUDIES 53
1. Preliminary Analysis of Rime-Splintering in PACE Clouds 53 a. Synopsis 53 b. Scientific Background 55
c. Data 57 d. Calculation of Ice Production Rates 58 e. Results 60 f. Discussion 64 g. Conclusion 69 h. References 71
2. Development of 2D Image Classification Software for PACE 75 a. Synopsis 75 b. Background 76 c. Data 77 d. Theory 78 e. Application 81 f. Handling Partial Images 84 g. Evaluation 87 h. Summary and Conclusions 93 i. References 94
FORECASTING STUDIES 97
1. Development of Objective Forecasting Procedures 97 a. Introduction 97 b. Diagnostic development 101
1.) Procedure I: Use of Modified K 103 2.) Procedure II. Use of TCCL and PB 106
c. Physical basis for Procedure II 108 d. Evaluation of Procedure I and II 110
1.) Performance of Modified K 110 2.) Performance of TCCL and PB 112
e. Conclusions 116 f. References 117
2. Operational Support for the University of North Dakota 120 3. Transfer of Forecasting Software and Data 121
HOT RADAR SYSTEM 122
WEATHER EFFECTS ON CROP YIELDS 124
1. Description of 1990 Experiment 124 2. Results from Previous PACE Experiments on Crop Yields 124 3. Background 125 4. Materials and Methods 128 5. Results 132
6. Discussion 136 7. Economic Value of Precipitation Augmentation to Agriculture 140 8. References 141
SUMMARY OF HEAVY RAINFALL RESEARCH 146
1. Time Distribution of Heavy Rainfall 146 2. Spatial Distribution of Heavy Storm Rainfall 147 3. References 147
INADVERTENT WEATHER MODIFICATION RESEARCH 148
PROJECT PLANNING AND ASSESSMENT 150
1. Studies of Inadvertent Changes in Clouds and Precipitation 151 2. Studies of Planned Precipitation Modification in the Midwest 152 3. Investigations of the Physical Effects of Altered Clouds and
Precipitation on the Environment and Economy 152
FY90 PUBLICATIONS FROM ILLINOIS PreCCIP 153
1. Formal Publications 153 2. Informal Publications and Presentations 155
APPENDIX A: SELECTED PREPRINTS 156
INTRODUCTION
This 12-month project has involved research, field efforts, and planning. The research
has been focused in four areas including: 1) the analysis of the 1989 field experimentation;
2) the analysis of the 1989 data from the agricultural plots; 3) preliminary studies of heavy
rainfall events; and 4) studies of inadvertent atmospheric modification. The latter two areas
of research involving heavy rainfall and inadvertent modification have evolved under the
broadening dimensions of the PACE effort. An extensive project assessment effort was
undertaken during this 12-month period, resulting in an enlarged focus for the program. As
a result of this analytical and planning effort, the program was retitled "Precipitation-Cloud
Changes and Impacts Project" (PreCCIP). The rational and objectives of PreCCIP are
presented in the planning and assessment section of this report.
The analysis of the 1989 field data, both aircraft and radar data, has proven to be
much more difficult than earlier expectations. The large amount of data and the desire to
very thoroughly analyze all forms of the data have led us to continue extensive analysis of
both data sets through and beyond this 12-month period. The analysis of the agricultural
plots data from the 1989 season also continued in this period. The studies of heavy rainfalls
were launched because the results from the agricultural plots suggested that the greatest
rainfall effects on crop yields were as a result of increased rainfall when heavier rain events
occurred, including those producing 1 or more inches of rainfall.
As a part of the broadening of our research into atmospheric modification, we began
1
to pursue studies of urban modification of clouds and precipitation. Several unanswered
questions from prior research, based on data at St. Louis and Chicago, came under
consideration. The first research that was attempted focused on the potential modification
of precipitation in the fall, winter, and spring seasons. This research is described in a
different section of this report.
As a part of our continuing research into the effects of altered weather, we designed
and conducted another field experiment involving the agricultural farm plots on the
University of Illinois South Farms. During the summer of 1990, four groups of plots were
subjected to a variety of rainfall treatments. Three of these groups were in movable covered
plots where the total amount of rainfall was controlled throughout the summer. There,
differences in levels of rainfall, time of planting, and density of stand were tested. In a
group of uncovered plots, 10 different rainfall treatments above the 1990 actual rainfall were
applied to corn and soybeans in a separate test.
Another activity concerned continuing efforts to improve the HOT radar. This
excellent 10-cm wavelength Doppler radar was briefly used during the 1989 field
experiments. Important modifications affecting the sensitivity of the radar and its mode of
data collection were performed during this 12-month period, attempting to get the radar into
excellent condition for future field operations for PreCCIP.
This report begins with a description of each of the atmospheric research activities
which include echo-core behavior studies, in-cloud studies of cloud structure, and forecasting
2
of summer weather conditions suitable for cloud modification. Next, the engineering and
technical work on the HOT radar is described, followed by a section describing the analysis
and field efforts involved in the studies of crop yields using the agricultural plots. The next
two sections describe research into heavy rainfall conditions and inadvertent weather
modification, respectively. Finally, a section describing the project planning and assessment
is presented, followed by a list of the project's publications and scientific talks.
3
ECHO CORE BEHAVIOR STUDIES
1. Analytical Techniques and Software Development
a. Echo Core Tracking Software
An interactive echo core tracking program has been developed during the past year.
This program works from binary gridded data processed by a bi-linear interpolation program.
The program works on UNIX-based computers, since its functions employ some of the
UNIX editor ("vi") commands. The program can work with any size grid, and in 3-
dimensions.
To track a particular echo core, the user enters the radar time that includes the
aircraft pass, and then examines the 5 or 6 km level for tracking. The echo core of interest
is located by moving the cursor (via key stroke) to the aircraft pass location. A command
is given to indicate the echo core identification number. The tracking can be done by: 1)
selecting each grid point separately, 2) tracing the outline of the core, or 3) an expansion
function. The expansion function requires that a peak reflectivity value be defined within
the core. It examines the field of echoes to determine whether the adjacent grid points
should be part of the core. The function works from peaks in reflectivity to the valleys of
reflectivity that separate the cores. Expansion stops in any direction where the reflectivity
begins to increase in value.
As each grid point is defined as part of a core, a flag is inserted into the binary file.
Grid point reflectivities are displayed on the screen, initially as small letters, with each letter
4
indicating a reflectivity range of 2.5 dBZ. As the grid point is flagged to be included in the
core, the letter displayed becomes capitalized. Grid points also can be flagged as noise,
denoted as an asterisk.
Keyboard commands move the cursor right, left, up and down on the screen. While
the screen only displays a portion of the grid (80 x 24 grid points), the screen pans to the
right, left, up or down when the cursor reaches the edge of the screen. At the bottom of
the screen, the date, radar volume number, and volume begin time are displayed, followed
by the (x,y,z) location of the cursor, the reflectivity of the grid point, the grid point echo core
ID, the echo core ID that is currently being tracked, the velocity of the grid point, and the
ZDR value of the grid point.
Keyboard commands also can cause the screen to display the previous or following
volumes or to display the upper or lower interpolated height field. Thus, it is possible to
move up or down in height and backward or forward in time to track the development and
motion of the echoes. A trajectory function also has been developed, where, knowing the
centroid of the two previous consecutive volumes, a new centroid is projected.
Once the cores have been tracked (i.e. the flags set) the data file can be used as input
to programs that summarize the history of the echo cores, and also as input to our standard
radar field display programs.
b. Echo Core Display and Statistics Software
These programs extract information about the tracked echo cores and displays the
5
information in a variety of summary forms. Growth rates will be computed for the individual
echo cores. Initially, these rates will extend from the time of first detection to the time of
the peak value of the particular radar parameter as in Westcott (1990). The radar
parameters that will be examined first are: 1) peak reflectivity within the echo core, 2) peak
horizontal areal coverage of the echo core, 3) total echo core volume, and 4) echo core
height.
Two displays have been developed. The first is a time-height display of peak
reflectivity which includes areal coverage. The second is a time sequence of modelled echo
cores. The square root is taken of the area of the echo core at each altitude that it is found
within each given radar volume. For each given radar volume when an echo core is found,
a plot of the height versus the square root of the area is drawn. This results in an idealized
echo core at each period of its history. One can then view lifetime sequence of the core as
it: 1) is first found at mid-levels, 2) expands horizontally and vertically, and 3) descends as
it rains out.
c. Echo Field Display Software
These are the standard ISWS radar display programs, based again on the gridded
data from the bi-linear interpolation program. The programs have the following capabilities
to:
1. obey, if desired, the edit flags input from the echo tracking program;
2. display the 28 km circle encompassing the experimental units, and to
6
move the origin of the circle based on either echo motion or the
experimental units' centers that had been defined in real-time during
the field program;
3. display the aircraft flight path tracks;
4. display the location of cloud to ground lightning flashes;
5. subtract the storm motion so that the storm relative air motion can be
displayed.
2. Preliminary Results from the 1989 Radar Data
a. Experimental Units
During the 1989 field program, 25 randomized experimental units were treated either
with silver iodide or with a placebo. Until the data reduction is completed, the treatment
decision for a given unit will remain unknown. A candidate cloud was sampled by the
treatment aircraft to determine whether the unit was suitable for treatment. Once the
decision was made, all clouds within a 28 km radius of the center of the area of treated
clouds were possible treatment subjects. The experimental unit moved in time as with the
motion of the echo cores. The unit was covered by the radar until all convection within the
unit dissipated or until the unit moved off the scope. A second experimental unit could be
treated during the same flight or on the same day, if that unit was upwind or lateral to the
first unit and as long as the center of the unit was 28 km from the edge of an adjacent unit.
A list of the experimental units is presented in Table 1. Three units occurred at the
7
Table 1. Experimental Units: from PACE 1989
Cloud Treated Echo Cores
ExpU Date 1989
Type Clouds Iso rep J-et J-dT J-nT J-p NoE
A-1 5/19 B 1
B-2 5A9 A 8
C-3 5/25 B 1 - out of target
D-4 5/30 B 2
E-5 6/01 A 4 - radar scan problems
F-6 6/03 B 1
G-7 6/03 A/B 2
H-8 6/12 B 2
1-9 6/12 B 1
J-10 6/18 B 4 - out of target
xK-11 6/23 A 9(1) 2 1 2 2(1) 1 1 0
L-12 6/23 A 5 - edge of area (echo beyond 120 km)
xM-13 6/23 A 8(1) 1 0 0 6 1(1) 0 0
N-14 6/27 B/A 3
0-15 6/27 B/A 3
P-16 7/02 B 2
xQ-17 7/08 A 5(1) 0 0 0 5 0(1) 0 0
xR-18 7/08 A 9(1) 2 0 2 3 1(1) 0 1
.S-19 7/11 A 9(1)
.T-20 7/19 A 4(2)
U-21 7/19 B 5
xW-22 7/23 A 14(1) 0 2 7 5(1) 0 0 0
xX-23 7/24 A 6(5) 0 0(5) 4 0 1 1 0
.Y-24 7/25 A 9(2)
.Z-25 7/25 A 3(1)
Iso - isolated echo core rep - repenetration of the same echo core J-eT - joined but easy to track J-dT - joined but difficult to track J-nT - joined, impossible to track, part of another core; generally corresponding to a cloud next to a large parent storm noE - no echo (too close to radar?; area not scanned?; or no ppt sized drops?) candidates are indicated by parentheses x indicates the unit is currently being tracked
8
limit or beyond the range of the radar and will not be examined in the radar analysis. Inten
of the remaining units, small clouds were treated (cloud field <30 kft); these units are
indicated as type B units. The clouds within these units generally were not considered asthe
most optimal seeding targets. These units will be examined at a later date with many of
them as case studies since most were penetrated several times. During one of the units (E-
5), there were problems with the radar scanning program; therefore, this unit will also be
examined at a later date. Thus, analysis was concentrated on a total of 11 experimental
units, where 83 large clouds were found and treated. These clouds were concentrated
mainly in the June 23-July 25 time period, plus one in mid May 1989.
Our work in FY90 has centered on 6 of these experimental units, 11, 13, 17,18, 22,
and 23. (While no tracking has taken place for the remaining 5 units (2, 19, 20, 24, 25),
processing for the other days has begun.) A grid covering the whole history of the
experimental unit including the time before the unit was declared as the echo system was
developing) was selected and interpolated with a 1 x 1 x 1 km resolution. The aircraft
treatment passes were located on these interpolated fields. The echo core corresponding
to the aircraft pass was identified and the tracking of the echo cores commenced.
b. Aircraft Position
The location of the aircraft with respect to the radar echo cores is largely based on
the LORAN-C information from the CIC Baron cloud physics/seeding aircraft. These
positions, recorded approximately every second, appear to be more consistently reliable than
9
the radar automated Research Aircraft Tracking System (RATS) positions which were
obtained about every 12 seconds. That is, growing echo cores typically coincided with the
aircraft position as deduced from the aircraft LORAN-C. From some test flights in May
comparing the aircraft location as painted on the radar scope and the RATS position, it
appears that the RATS locations may vary from 0 to 2 km. The direction of the error is not
consistent and probably depends on the heading and location of the aircraft with respect to
the Hanna City FAA radar.
c. Comparison of Aircraft Penetrations and Radar Cores
Upon examination of the echo fields of the six 1989 cases, it was discovered that in
4 of the first 6 experimental units the treated clouds merged more rapidly and became part
of more complex and more intense storms than those observed in the three 1986
experimental units.
These 6 units encompassed 50 treated clouds. Twenty cores were found to be easily
tracked; some were isolated for much of their history and some later merged. No echo was
found near one of the aircraft penetrations, and three cloud passes were found to be
repenetrations of another echo core.
A new system of categorization has been devised for the echoes (cloud penetrations)
remaining to be tracked. Four of these clouds, were at first a puzzle. No distinct echo core
could be found to be associated with these clouds. After viewing the aircraft tapes, it was
found that these clouds were obviously vigorous feeder clouds, with high liquid water
10
contents. They were adjacent to large mature storms, but in such close proximity that the
radar could not distinguish them. The parent storm could have been tracked, but it was
decided that since the growth of the mature storm would not be representative of the feeder
cloud, these four clouds would not be part of the cloud/core analysis.
Additionally, two clouds were found to have cores that were in clear evidence for only
one to two volumes, and were also close to large parent systems. These clouds were in areas
where the radar resolution was at least 1 km. Thus, the remaining echo cores were divided
into 1) feeders adjacent to a parent storm that are not trackable; and 2) clouds adjacent to
a parent storm that are partially trackable, but probably will not be used in the cloud/core
joint study.
Finally, there were clouds with easily identifiable echoes, but were joined to other
storms and difficult to follow during some portion of their history. The more easily tracked
cores seemed to follow a progressive pattern of 1) formation in the 2 to 6 km height range;
2) expansion in the vertical; 3) expansion in the horizontal, and then finally 4) a drop of the
echo core towards the surface. This pattern was assumed with the more complex cores as
well. The cores were tracked to past the time when they reach their peak height, peak area
and peak reflectivity. Often these more complex echoes were engulfed by an encroaching
neighbor. This encroachment typically began at mid-levels and then expanded vertically.
These cores often ended with high reflectivity values at the surface with the encroaching one
immediately above. This group of difficult but trackable cores make up the majority of the
11
three groups of the difficult cores and are currently being tracked for inclusion in the joint
study.
3. Results from other studies.
Several very damaging hail storms occurred on May 25, 1989. No cloud treatment
occurred within 100 km of these storms. In cooperation with the South Dakota School of
Mines, CHILL 10-cm radar data and T-28 microphysical data were combined to examine
hydrometeor growth on this day. The results are presented in Appendix A (see Musil etal.
1989).
Another study was made of a small scale, but damaging (F0 or F1) tornado.
Meteorological conditions were dissimilar to those of either a classic spring season tornado
day or for any of the small scale vortices detailed in the literature. Instead, the tornado
appeared to be associated with a vorticity maximum (20 x 10-5 s-1) occurring along a trough.
Over the 11/2 hours of observation, the storm never exceeded 20,000 ft. The radial velocity
shear development was similar, however, to those observed in the southern plains, in that
the shear first intensified aloft and then approached the surface prior to the tornado. The
vortex signature lasted for well over an hour. Details of this paper are presented in
Kennedy et al. (1990), found in Appendix A.
12
4. Studies of First and Early Echo Properties
a. Introduction
The importance of the merging of clouds together on the longevity, area, vertical
growth and rainfall output of storms has been established by an number of authors (see
review in Westcott 1984). A number of case studies have investigated factors that result in
the initiation, growth and aggregation of convective cells. Numerical modeling studies in
particular have suggested that aggregation will be predominant among echoes of specific
size, age, and distance criteria (Orville et al. 1980; Tao and Simpson 1989). Orville et al.
(1980) and Turpeinen (1982) found that a favorable pressure gradient must be present for
merging to occur, which would be likely when the merging pair are of different intensity or
age. Tao and Simpson (1989) found that the orientation of the storms with the vertical
shear has an impact on the minimum distance of separation is required for merging to occur.
Additionally, they found that favorable pressure gradients are often associated with low-level
convergence resulting from cold outflows. However, little evidence has been documented
as to the size, age, and location factors that may be important to a field of aggregating
echoes.
One objective of this past year's research effort was to examine more closely for a
population of storms, the initial conditions of echoes at their first detection, to determine
whether clouds which join together have different properties early on, which result in a
higher likelihood of their aggregating together, than those that do not join. Specific
13
conditions which have been examined were their location (orientation and distance) with
respect to the parent cell, and whether they merged with another first echo or with a
complex. The temporal history of 142 first echoes has been traced using three-dimensional
reflectivity data. These echoes occurred on 2 summer days in Dlinois, under very different
environmental conditions.
b. Data Source and Analysis Procedures
The data used were obtained by the ISWS/NSF CHILL 10-cm radar on two summer
days in 1986. Since the CHILL was in the midst of its refurbishment period, only reflectivity
data were available. These data were processed by a program which compressed 3-
dimensional volumetric data into 6 two-dimensional Cartesian grids of: maximum echo top
height, minimum echo height, height of the maximum reflectivity, reflectivity at the echo top,
reflectivity at the base of the echo, and the maximum reflectivity within the echo, for each
(x,y) grid location. A 121 x 121 grid was used, with a horizontal resolution of 1.25 km (150
x 150 km area) and the radar located at the center of the grid. The minimum reflectivity
recorded on the first case day (July 25, 1986) was 20 dBZ, and on the second day (August
26, 1986), 14 dBZ. For comparison purposes, a threshold of 20 dBZ was imposed on all of
the data at the outset.
The echoes were identified and tracked in time, first by computer and then checked
by a radar analyst. The echo had to contain at least two 20 dBZ grid points (3 km2) and
had to exist for at least 2 consecutive time periods to be included in the sample. Care was
14
taken to insure that only echoes that were topped by the radar were included in the sample.
Echoes that moved into the analysis grid were excluded from the first echo study but were
included in the echo bridge characterization study. In the case of the first echoes, the
echoes were tracked to the time when they dissipated or to when they merged with an
adjacent echo. In order for an echo to be considered as joined with its neighbor, it had to
be linked by at least two grid points at the minimum reflectivity level and the bridge had to
be present for at least 2 consecutive time periods. The radar volumes ranged in time from
2 to 5 minutes.
c. Conditions on 25 July 86, 15:33-19:01 CDT
During this afternoon, convection formed along or just ahead of a cold front which
moved through the area at about 1700 CDT. The 1634 CDT sounding launched from the
radar site (CMI) indicated unstable conditions, as well as the presence of the front (Fig. 1).
A strong inversion was observed at about 500 mb; the air was conditionally unstable both
below and above the inversion. Ample moisture was indicated, as the subcloud mixing ratio
was computed as 16 g/kg. Cloud base was estimated as 2.1 km with a temperature of about
17 °C.
Two major lines developed in central Illinois, one about 60 km to the north of CMI
and the second about 50 km to the south (Fig. 2). The echo cores of the northern line
became joined into a solid line. The southern line was broken and included at least 6 storms
separated by 10-15 km. Late in the history of these storms, they also joined. Both lines
15
Figure 1. Skew-T/Log P plot of the 1643 CDT CHILL Radar Site (CMI) sounding on 25 July 86.
Figure 2. Plot of maximum reflectivity found within each echo within 75 km of the CHILL radar during the radar volume from 164438 to 164851 CDT, on 25 July 86. Each tick mark represents 1.25 km. The lower right corner is (-75 km,-75 km) from the radar. Contours levels begin at 20 dBZ and are incremented by 5 dBZ.
were orientated from the WSW to ENE (247.5°), generally parallel (12.5°) to the front
(~235°), to the mean cloud layer winds (18 m s-1, 252°), and to the cloud layer shear (6.4
m s-1 at 244 °). The individual cells were moving to the ENE at about 12 m s-1, while the
system as a whole moved to the east at about 13 m s-1 slower than the mean cloud layer
winds.
The vertical shear of the wind was moderate and concentrated below cloud base.
Computing the vertical shear from the surface to 3 km (700 mb), as suggested by Weisman
and Klemp (1982), the shear was moderate (13.5 m s-1 at 253°, 4.55 x 10"3 sec-1). If the
shear were computed from the surface to cloud base as estimated by the LCL, as suggested
by LeMone (1990), the shear was stronger, 5.7 x 10"3 sec-1. Within the cloud layer (2.3-11.6
km), the winds were fairly constant, averaging 18 m s-1 at 252 degrees.
Along both lines, cloud growth appeared to be by both weak and strong evolution
(Foote and Frank 1983). Here, weak evolution was assumed to occur when an echo expands
horizontally, followed by the appearance of a joined echo core within the area of the
expansion. Strong evolution was assumed when an echo core first appeared as a discrete
echo core, isolated or merged. While these lines passed between most first order and
cooperative NWS observing stations, the NWS storm data indicated that the portion of the
southern storm which passed over Effingham, IL, produced strong winds, one inch of rain
between 1800 and 1830 CDT, and one inch diameter hail.
16
d. Conditions on 26 Aug 86, 08:00-12:30 CDT
On this morning, thunderstorms were forming in a warm air mass about 300 km
ahead of a strong, slow moving cold front. The 0700 CDT sounding at Salem, Illinois (SIX),
160 km to the SSW of CMI), showed a conditionally unstable atmosphere to 500 mb, above
a shallow surface inversion, and substantial moisture to 450 mb (Fig. 3). Visual aircraft
reports indicated a variable cloud base with multiple layers of clouds. The subcloud mixing
ratio was computed as 16 g/kg. The cloud base temperature and height were estimated as
19° C, at about 1 - 1.5 km msl. Surface heating was inhibited by a cirrus overcast and an
analysis of NWS surface station winds indicated only weak convergence in the area.
The area of convection originated in southern Iowa near midnight and moved
southeastward into the study area by morning. This area consisted of a field of many small
convective lines and several large lines (Fig. 4). By 1145 CDT, the clouds were moving out
of the study area and were rapidly diminishing in intensity and areal extent. The lines were
typically oriented SSW to NNE (202.5°), more perpendicular (42.5°) to the mean cloud
layer winds (10 m s-1, 245 °) and to the cloud layer vertical shear vector (2.6 m s-1 245 °).
The storms moved faster than the mean cloud layer winds, to the east at about 19 m s"1, and
the individual cells moved northeast, also at about 19 m s-1.
The winds at the surface were light (~ 2.7 m s-1), and from the south; they veered
to the WSW at 10 m s-1 by 0.7 km agl, and remained steady aloft. The subcloud layer
vertical wind shear was strong, from surface to the LCL at 1.2 km (7.5 x 10-3 sec-1). If
17
Figure 3. Skew-T/Log P plot of the 0700 CDT Salem (SLO) sounding on 26 August 1986.
Figure 4. Plot of maximum reflectivity found within each echo within 75 km of the CHILL radar during the radar volume from 102820 to 103323 CDT, on 26 August 86. Each tick mark represents 1.25 km. The lower right corner is (-75 km,-75 km) from the radar. Contours levels begin at 20 dBZ and are incremented by 5 dBZ.
computing the low-level vertical shear from the surface to 700 mb (3 km agl), the shear was
weak and more in line with what might be expected for strongly evolving storms (7 m s-1,
250°, 1.2 x 10-3 sec-1). Because of the lower cloud bases observed in the central Midwest,
the sub cloud vertical shear of the horizontal wind is somewhat higher than observed in great
plains thunderstorm cases. Within the cloud layer (1.2 - 8.5 km), the winds were from 245 °
at 10 m s-1.
e. First echo characteristics - dimensions
A first look at the storms in this study revealed that merging occurred very rapidly
(Table 2). On July 25, 50% of the first echoes which merged did so within the first 10
minutes of their history, and on August 26, 71% merged within 10 minutes. In many cases,
reflectivity cores when first observed were already joined to the parent storm. On July 25,
the total number of new cores doubled when the 42 reflectivity cores that were joined at the
time of first observation were included in the first echo sample (Table 3). On August 26,
there were 33 echo cores that were joined when first observed, in addition to the 112 first
echoes. On both days, looking at the total sample of cores, 79% of the mergers occur within
10 minutes of first detection (Table 4). In the remainder of the analysis, only those echoes
not joined at first detection are included.
The rapidity with which reflectivity cores joined suggests that the cores which
aggregate may have different initial characteristics that may predispose that echo core to
merge. Thus, it may be possible to predict from first echo conditions whether an echo will
18
Table 2. Age of echo at the time of merger, including only those cores not joined at first detection.
July 25, 1986 August 26, 1986
1-5 min 12 40% 32 41%
6-10 min 3 50% 24 71%
11-15 min 4 63% 8 81%
16-20 min 2 70% 5 87%
21-25 min 4 83% 4 92%
26-30 min 1 87% 1 94%
31-45 min 1 90% 3 97%
46-60 3 100% 2 100%
30 79
Table 3. Number of new echoes.
July 25, 1986 August 26, 1986
First echoes which dissipate 10 12% 33 23%
First echoes which merge 30 37% 79 54%
First echoes already joined 42 51% 33 23%
Total new cores 82 100% 145 100%
merge or dissipate. The area, height and reflectivity of the echo core at first detection were
examined, as well as the distance to the nearest first echo.
19
Table 4. Age of echo at the time of merger including echo cores already joined at first detection.
July 25,1986 August: 26,1986
0 min 42 58% 33 29%
1-5 min 12 75% 32 58%
6-10 min 3 79% 24 79%
11-15 min 4 85% 8 87%
16-20 min 2 88% 5 91%
21-25 min 4 93% 4 95%
26-30 min 1 94% 1 96%
31-45 min 1 96% 3 98%
46-60 min 3 100% 2 100%
Total 72 112
The median dimensions of the echoes which eventually merged were indeed different
from those that did not (Table 5). The difference in the initial height of the core was most
pronounced. Differences were observed in the distribution of the area (Table 6), and
heights (Figures 5-7) of the cores as well. The Mann-Whitney U test indicated that the
merging and dissipating populations from which the height values were drawn were
significantly different.
20
Table 5. Median First Echo Characteristics
July 25, 1986 August 26, 1986
Dissipate Merge Dissipate Merge
Echo Core Sample 10 (25%) 30 (75%) 33 (29%) 79 (71%)
Distance to Nearest Echo (km) 5.0 5.0 5.0 3.8
Area (km2) 4.0 7.0 4.0 6.25
Max Reflectivity (dBZ) 26.3 26.9 27.4 28.7
Top of Echo (km) 5.0 7.5 4.5 5.5
Height of Max dBZ (km) 4.5 6.5 4.5 5.5
Base of Echo (km) 3.5 4.5 3.5 4.5
Depth of Echo (km) 2.0 3.0 1.0 1.0
Time to Merge (min) 9.5 7
Duration (min) 18 22
Table 6. Area (≥20 dBZ) of first echo cores.
July 25, 1986 August 26, 1986
Dissipate Merge Dissipate Merge
3-4 km2 6 60% 10 33% 23 70% 38 50%
5-6 km2 2 80% 3 43% 4 82% 9 62%
7-8 km2 0 80% 5 60% 1 85% 13 79%
9-10 km2 0 80% 5 77% 3 94% 6 87%
11-12 km2 1 90% 3 87% 2 100% 4 92%
13-14 km2 0 90% 1 90% 0 2 95%
15-16 km2 1 100% 1 93% 0 3 99%
17-18 km2 0 0 93% 0 0 99%
19-20 km2 0 0 93% 0 1 100%
>20 km2 0 2 93% 0 0
TOTAL 10 30 100% 33 76
21
A MAX ECHO HEIGHT (KM) 7 / 2 5 / 8 6 MERGING ECHOES
DISSIPATING ECHOES
B MAX ECHO HEIGHT (KM) 8 / 26 / 86 MERGING ECHOES
DISSIPATING ECHOES
Figure 5. Distribution of the height of the echo tops at the time of first detection, for dissipating and merging first echo cores for a) 25 July 1986 and b) 26 August 1986.
A ZMAX HEIGHT (KM) 7 / 2 5 / 8 6 MERGING ECHOES
DISSIPATING ECHOES
B ZMAX HEIGHT (KM) 8 / 2 6 / 8 6 MERGING ECHOES
DISSIPATING ECHOES
Figure 6. Distribution of the height of the peak reflectivity within the echo cores at the time of first detection, for dissipating and merging echoes for a) 25 July 86 and b) 26 August 86.
A MIN ECHO HEIGHT (KM) 7 / 2 5 / 8 6 MERGING ECHOES
DISSIPATING ECHOES
B MIN ECHO HEIGHT (KM) 8 / 2 6 / 8 6 MERGING ECHOES
DISSIPATING ECHOES
Figure 7. Distribution of the height of the echo base at the time of first detection for dissipating and merging echoes for a) 25 July 86 and b) 26 August 86.
f. Factors influencing first echo location
Based on the longer time to merge, and the larger area and greater median echo top
for cores on August 25,1986, the characteristics of the merging cores were different on the
two days. This may be due in part to the closer proximity of the cold front on the afternoon
of July 25, as well as substantial surface heating. Additionally, it was found that on July 25,
the storms were oriented (247.5°) more parallel (within 22.5°) to the storm motion (270°)
and to the low level shear (253 °), and on August 20, the orientation (202.5 °) was more
perpendicular (greater than 45 °) to the storm motion (270 •) and to the low level shear
(250°). The first echoes that eventually merged tended to form at greater distances from
other echoes (83% within 12.5 km on the 25th, as opposed to 83% within 5 km on the 26th
(see Table 7). Likewise, the time from first detection to merger was longer during the
afternoon of July 25, as might be expected from the wider spacing (70% merged within 20
min. on the 25th as opposed to 71% within 10 min. on the 26th (see Table 2).
These results agree with the modeling results of Tao and Simpson (1989). In a
simulation of mergers based on the GATE data, a similar spacing was found for "parallel"
and "perpendicular" mergers, as was found here in actual data. The distance between
perpendicular cells prior to merging in the simulation case was usually about 5 to 6 km. The
distance between parallel cells was often 10 km or more. Additionally, the perpendicular
mergers were found to occur more rapidly in simulation as was the case in the data
presented here.
22
The echo cores were further examined to see if there was a preferred formation site
for echoes which would merge, and if so, to determine significant differences between those
echoes forming in preferred locations and elsewhere. In particular, the echoes were
stratified by the direction from the parent storm and by distance (5 km) from 2 or more
nearby echoes. Additionally, the data were examined according to whether the cores merged
with another first echo or with a complex. In looking at this particular stratification, no
significant differences were found in the dimensions of the echoes.
Table 7. Distance of first echo cores to nearest neighbor.
July 25, 1986 August 26, 1986
Dissipate Merge Dissipate Merge
<2.5 km 0 0% 7 57% 12 36% 34 43%
2.6-5.0 km 7 70% 11 60% 14 79% 32 83%
5.1-7.5 km 0 70% 1 63% 3 88% 7 92%
7.6-10.0 km 2 90% 5 80% 2 94% 2 95%
10.1-12.5 km 0 90% 1 83% 0 94% 0 95%
12.6-15.0 km 0 90% 2 90% 0 94% 3 99%
15.1-17.5 km 0 90% 0 90% 1 94% 1 100%
17.6-20.0 km 1 100% 0 90% 0 97% 0
>20.0 km 0 3 100% 1 97% 0
TOTAL 10 30 33 100% 79
In looking at echoes on 3 summer days in Ohio, Byers and Braham (1949) found that
the most likely location (35-48%) for new echo formation on 2 of the days, was between and
23
within 3 miles of 2 existing echoes. On the 3rd day, no echoes were found in the overlap
area. In our data on August 26, 29% formed within 5 km and between 2 echoes (Figure 8).
On July 25, when the spacing between new echoes was greater, only 12% formed in this
area. About the same proportion (70%) of echoes merged whether forming in the overlap
area or in other quadrants relative to the closest echo. The cores forming between
neighbors on August 26, typically merged more rapidly than the other merging first echoes,
but were similar in size and height to the other echoes.
Other investigators have observed clouds to form on the right flank of an existing
cloud system (Browning and Ludlum 1962; Dennis et al. 1970; and Mauwitz 1972). Typically
these storms developed under conditions of moderate to strong shear and were severe in
nature. Here, the right flank of the parent was the preferred direction of formation, but
echoes were observed to form in all quadrants from their closest neighbor. July 25 was
characterized by weak to moderate low level shear, and August 26 by weak shear. About
25% of the first echoes formed to the southwest on July 25, and about 18% formed to the
south on August 26. Echoes forming to the SW were more likely to merge on the July 25
(90 vs. 70%), and those forming to the S on August 26 (75 vs. 69%).
At the time of first detection, little difference was observed in the time and height of
the echoes which formed in a preferred location. In contrast to the echoes which formed
between two existing nearby echoes, these echoes tended to form further away from the
parent echo with which they would merge and took somewhat longer to join with the
24
A JULY 25,1986
B AUGUST 26, 1986
Figure 8. Distribution of the first echo location with respect to the parent echo for dissipating and merging echoes for a) 25 July 86 and b) 26 August 86.
adjacent echo. On July 25, the SW echoes formed approximately 10 km further from the
adjacent echoes and took more than 15 minutes longer to merge. During the morning case
little difference was found in the initial separation, but those forming in the preferred south
direction took 9 minutes to merge as opposed to 5 minutes for all other cores.
g. Summary
First-echo characteristics of echoes within two different convective periods have been
studied. The echoes were examined to determine whether differences between echoes that
merged and those that dissipated could be observed at the time of first detection. It was
found that those that merged were taller and larger on both days than those that dissipated.
This suggests that echoes which eventually merge are more vigorous and thus, more likely
to grow and merge.
Meteorological conditions on the two days were somewhat different. On July 25,
1986, convection occurred during the afternoon, in association with a cold front passing
through the area. The low level vertical shear of the horizontal wind was weak to moderate
on this day. The storms appeared to grow by both weak evolution (echo expanding
horizontally prior to the appearance of an echo core) and strong evolution (distinct echo
core appearing first either separate or joined to the parent echo). The lines of storms were
generally parallel to the mean cloud layer winds, to the mean cloud layer vertical shear and
to the cold front.
On August 26, 1986, the convection sampled was from a morning period, several
25
hundred km ahead of a cold front. Because it was morning and there was a cirrus overcast
during this period, surface heating was of little importance in driving the convection. The
low level vertical shear of the horizontal wind was weak on this day and the storms appeared
to grow by strong evolution. The lines of storms were more perpendicular to the mean
cloud layer winds and to the cloud layer vertical shear.
The 40 first echoes sampled on July 25 were distinctly taller than those on August 26.
Storms formed in all quadrants with respect to the parent echo. The median spacing of the
30 echoes that merged was slightly larger than that on August 26. The difference in spacing
was greater for those echoes that formed in the preferred SW direction, generally parallel
to the vertical shear. The median spacing of these SW echoes was similar (10 km) to the
parallel clouds modeled by Tao and Simpson (1989). The merger of these echoes took
somewhat longer than for those forming in other quadrants.
On August 26,112 first echoes were sampled, and 79 of these subsequently merged.
These echoes merged more rapidly than those on July 25. While echoes again formed in
all quadrants from the parent cell, a larger proportion of these formed between two
neighboring echoes. A preferred direction of formation was also found on this day; the
direction was to the south (S), more perpendicular to the vertical shear. The characteristic
spacing between the S cores and that of the parent cell, and for the merged sample as a
whole on this day was 4 km, again similar to the modeling results of Tao and Simpson
(1989).
26
5. Echo Core Properties at Time of Merger
a. Introduction
It has been established by many past studies that an important element of the growth
of storm systems is that clouds join together to make larger systems, (for review, see
Westcott 1984). A number of statistical studies directed at examining the rainfall output
from populations of clouds were carried out in the 1970s (Houze and Cheng 1977; Lopez
1976,1978; Simpson et al. 1980; Wiggert et al.; 1981). These studies were carried out on
tropical/subtropical systems and were mainly directed at examining the rainfall output of
merged systems. In general, they found that merged systems produce more rainfall than
more isolated systems. Changnon (1976) carried out a similar study in the midwest and
found that merged systems generally grow taller vertically and last longer than systems that
did not merge.
It was not the intent of the research conducted this past year to duplicate these
efforts. Rather, the objective of this research was to describe as thoroughly as possible,
using 3-D reflectivity data for a population of echo cores, the interaction between echo cores
at the time that they join together, and possible factors affecting whether or not these cores
subsequently grow. Thus, the emphasis is on the smaller units which are bridged together,
not the overall history of the storm.
The manner in which clouds aggregate together has been the subject of a number of
case studies involving radar observations (Westcott and Kennedy 1989; Ackerman and
27
Kennedy 1988; Cunning et al, 1986; Cunning and DeMaria 1986; Foote and Frank 1983)
and numerical modeling experiments (Tao and Simpson 1989,1984; Turpeinen 1982; Orville
et al., 1980). These studies have shown that there are a variety of factors resulting in the
initiation and growth of convective cells. Mesoscale convergence and convective scale
outflows generated by downdrafts have been found to be important in the initiation of new
cells in these studies and by Fankhauser (1982), Simpson et al. (1980), and Simpson (1980).
The strength of the dynamic forcing and the local thermodynamic environment in which a
cloud develops has a large impact on its rate of growth (Westcott 1989).
Additionally, low-level convergence, differential cloud motion, the horizontal
expansion of cloud cells, new cell growth between existing clouds, and natural seeding by
adjacent clouds have been observed to be important in the aggregation of cloud cells
(Westcott and Kennedy 1989; Ackerman and Kennedy 1988; Peterson 1984; Cunning et al.,
1986; Cunning and DeMaria 1986; Simpson 1980).
In other research conducted over the past year, 2 populations of merging echo cores
were studied to describe the ways in which merging occurs and the behavior of the storms
following merger, and to improve the understanding of the processes which typically result
in the natural aggregation of cloud units. The 90 merger events were found on 2 very
different summer convective periods in Illinois. The dimensions and age of the merging
echo pair, and the height and reflectivity of the bridge were examined. The merging core
pairs have been grouped by the type of merger event: 1) two cores which have not
28
previously merged; or 2) one core between a complex and a single cored echo; or 3) one
core between two complexes. The merging pairs also were stratified by relative age. The
immediate pre- and post-merger growth of the clouds was examined as well, to determine
under what conditions merger and subsequent growth occurs.
b. Data
The data used were an expansion of those from the first-echo study (see section 4a
this chapter). Here, the echo cores were tracked to beyond the time of merger. An echo
core was defined as a maximum in reflectivity, with boundaries delineated by a minimum in
reflectivity. These echo cores were either isolated, that is a single-celled storm, or part of
a multicelled system. The minimum reflectivity of the radar data as a whole was 20 dBZ,
and the grid spacing 1.25 km.
The area of the echo with reflectivities ≥20 dBZ and the area ≥ 35 dBZ were
recorded, as well as the peak reflectivity and height of the echo cores just before (2-5 min),
at the time of, and just after the merger (2-5 min). Additionally, the base and top height
and reflectivity of the bridge itself were examined. The frequencies were tallied of echo
cores which grew, remained steady, or diminished prior to and after merger. Whether an
echo core was growing prior to or after a merger was based strictly on the area, height and
reflectivity of the two cores that joined together, and not on the character of the echo as a
whole.
29
c. Description of merging
The overall characteristics for the merging echo cores at the time of merger, and for
the bridge are presented in Table 8. Whether the echo core was denoted as echo 1 or echo
2 was determined by the relative age and size of the echo core pair. Echo core 1 was
always the same age or older, and usually the same size or larger than echo core 2.
It was found for the median merger on July 25 that the older echo core 1 was
probably in its mature or dissipating stage, and that echo core 2 was most likely in its growth
stage. Based on results of many other radar studies, it was assumed that these cores are
representative of the cells as defined by the Thunderstorm Project (Byers and Braham 1949)
with their time of existence on the order of 30 minutes.
The slower time for echo 1 to merge on July 25 reflects the wider spacing of the
separate cores and possibly the tendency of the parent systems to grow both by weak as well
as strong evolution processes. On August 26, the median merger was between two echoes
that were closer in age and height. Merging takes place more quickly on August 26, 1986,
and at lower heights. The area >35 dBZ and the peak reflectivity of each core was
somewhat larger on this day. The median bridge forming between the echo pair during this
afternoon had a depth of 5.0 km, and extended from 3 km (somewhat lower than the first
echo base, but 1 km higher than the estimated cloud base) to 8.5 km, 1 km lower than the
height of the younger echo, and 1 km higher than the median first echo top for echoes that
merged. Merging took place at lower heights on August 26 as would be expected from the
30
Table 8. Median values of maximum reflectivity, height, area ≥ 35 dBZ, and age of the echo cores, and the bridge heights and maximum reflectivities at the time of merger.
July 25, 1986 August 26, 1986
Sample Size 24 66
Bridge:
Maximum Height (km) 8.5 5.5
Max Z Height (km) 6.5 4.5
Minimum Height (km) 3.0 1.5
Depth (km) 5.0 4.0
Max Reflectivity 25.0 27.5
Echo 1:
Age at Merger (min) 26 16
Area of dBZ ≥ 35 (km2) 22 29
Maximum Height (km) 12.5 7.5
Maximum Reflectivity (dBZ) 43.8 50.0
Echo 2:
Age at Merger (min) 8 8
Area of dBZ ≥ 35 (km2) 4 8
Maximum Height (km) 9.5 6.5
Maximum Reflectivity (dBZ) 37.5 42.5
first echo tops and may result from the weaker forcing. The bridge during this morning
period was also narrower in depth.
The echo cores in the radar volume 2-5 minutes prior to merger and 2-5 minutes
following merger were examined to determine whether or not merging resulted in echo core
growth following merger. Results are presented in Table 9. On July 25, the frequency of
growth following merger was small for the older echo core 1. The median age for this group
31
of echo cores was 26 min, again suggesting that the cores were in their mature or dissipation
stage.
Table 9. Percent of echo cores which grow following merger.
July 25, 1986 August 26, 1986
Core 1 Core 2 Core 1 Core 2
Max Reflectivity 17 63 53 56
Area ≥ 20 dBZ 38 48 64 67
Area ≥ 35 dBZ 17 57 65 70
Height 29 50 29 34
The frequency of growth in terms of reflectivity and area for the other cores ranged
from 48-70% for area and reflectivity. In terms of height, the younger core 2 on July 25 has
the greatest proportion of echoes that subsequently grew following merger. Westcott (1990)
and Simpson (1980); have suggested that echoes may reach their peak reflectivity first,
followed by a continued horizontal expansion. Perhaps this effect was acting here such that
many of these echoes had reached their peak height, but not peak area and reflectivity. It
appears that the echo cores which were growing prior to merger were the most likely to
grow following merger (Table 10). Only 8-13% of the younger echo cores 2 on July 25 that
were growing prior to merger did not continue to increase in maximum reflectivity, area
≥ 20, or area ≥35 dBZ; on August 26, 1986, 14-22% did not continue to increase.
The following sections look more closely at the merger events with respect to the relative
32
age of the merging pair and the type of merger, to further explore the merging process and
why growth may or may not occur following merger.
Table 10. Of the echo cores which were growing prior to merger, the percent which grew following merger.
July 25, 1986 August 26, 1986
Core 1 Core 2 Core 1 Core 2
Max Reflectivity 50 87 74 78
Area ≥ 20 dBZ 67 91 88 86
Area >35 dBZ 67 92 93 83
Area ≥50 dBZ 75 100 72 96
Height 14 67 33 48
d. Types of Merger
Merging is known to occur on a variety of scales of cloud units. Here we will
examine the characteristics of the merging pair broken down by merger type: Type 1 is a
merger between two echoes that had not merged previously; Type 2 is a merger between
a core that had never merged, and an echo core that had merged (i.e. between a feeder
cloud and its parent storm), and Type 3 is the merger of echo cores that are part of two
different complexes. On July 25,1986, 24 mergers were observed compared to 66 on August
26, 1986. A similar proportion of echoes were in each category type on the two days.
About 25-26% were observed to occur between young first echoes (Type 1); 63-67%
occurred between young echoes and complexes (Type 2); and a 13-18% between complexes
33
(Type 3).
The median age, height, area and bridge characteristics are presented for the merging
pairs and the echo bridge in Table 11a and 11b. The median values of age, area (≥35 dBZ)
and height typically increased from Type 1 to Type 3, as well as, differences in age and area
within the pairs. It appears that the first echoes that merged with a complex tended to be
older and larger than those that merged with a first echo. On July 25, the median core was
very much taller, and on August 26, they were of similar height.
The echo cores and bridges were distinctly taller on July 25, 1989. On July 25, the
top of the bridge was within 1 km of the echo core 2 maximum top height, and on August
26 it was within 1-2 km of the echo core 2 top height. The bridge depths increased from
the Type 1 to the Type 2 mergers, where the second core was again larger.
Because of the small number of cases of the Type 3 mergers, there was a large
variation in their values and the medians should be taken with a large degree of uncertainty.
From the overall results, it was shown that the second, younger echo core grew
somewhat more frequently than echo core 1. Here, we will concentrate on the growth of
echo core 2 only. In terms of height, the younger echo cores (Types 1 and 3) on July 25
grew more often than the older ones of Type 2 (Table 12). On August 26, there was little
difference in height for the three Types.
In comparing differences between Types 1 and 2, similar numbers increased in
reflectivity for each day. On July 25, the younger echo cores grew more in area >20 dBZ
34
and the older echo cores grew more in area >35 dBZ. On August 26, when the median
reflectivity for both groups was ≥35 dBZ, similar numbers of echo cores increased in both
Table 11a. Characteristics at the time of merger for the echo cores on July 25, 1986, by merger type.
Fe-Fe Cmplx-Fe Cmplx-Cmplx
Sample Size 6 15 3
Bridge:
Maximum Height (km) 6.0 9.5 10.5
Max Z Height (km) 4.0 6.5 7.5
Minimum Height (km) 2.0 3.5 2.5
Depth (km) 3.5 5.0 9.0
Max Reflectivity (dBZ) 23.8 25.0 25.0
Echo 1:
Age at Merger (min) 13 35 29
Area of dBZ ≥i35 (km2) 4 22 47
Maximum Height (km) 10.0 11.5 12.5
Maximum Reflectivity (dBZ) 36.8 45.0 50.0
Echo 2:
Age at Merger (min) 5 12 8
Area of dBZ ≥35 (km2) 0 8 2
Maximum Height (km) 6.0 10.5 9.5
Maximum Reflectivity) (dBZ) 32.5 40.0 42.5
35
Table 11b. Characteristics at the time of merger for the echo cores on August 26, 1986, by merger type.
Fe-Fe Cmplx-Fe Cmplx-Cmplx
Sample Size 17 44 5
Bridge:
Maximum Height (km) 4.5 5.5 4.5
Max Z Height (km) 4.5 4.5 3.5
Minimum Height (km) 1.5 1.5 3.5
Depth (km) 3.0 4.0 3.0
Max Reflectivity (dBZ) 27.5 25.0 27.5
Echo 1:
Age at Merger (min) 9 17 52
Area of dBZ ≥35 (km2) 6 40 88
Maximum Height (km) 6.0 8.5 8.5
Maximum Reflectivity (dBZ) 40.0 52.5 60.0
Echo 2:
Age at Merger (min) 4 8 9
Area of dBZ >35 (km2) 5 12 17
Maximum Height (km) 5.5 6.5 6.5
Maximum Reflectivity (dBZ) 37.5 45.0 45.0
36
Table 12. Percent of echo core 2s which grew following merger, by merger type.
Type 1 Type 2 Type 3
a. July 25, 1986 Fe-Fe Cmplx-Fe Cmplx-Cmplx
Sample Size 6 15 3
Max Reflectivity 67 67 33
Area ≥20 dBZ 83 36 33
Area ≥35 dBZ 50 64 33
Height 67 40 67
b. August 26, 1986
Sample Size 17 44 5
Max Reflectivity 53 52 80
Area ≥20 dBZ 65 66 80
Area ≥ 35 dBZ 71 61 60
Height 37 32 40
categories of area. There was only a small sample of complex to complex mergers on each
day. The mean of those on July 25 do not grow, and on August 26 did increase in area and
reflectivity.
37
e. Relative age of the merging pair
From the previous discussion, it appears that the age of the echo tends to be an
important parameter in determining the frequency of core growth. The following
stratification was based on the relative age of the merging pair. While ideally this grouping
should correspond to the Type of merger as discussed in the previous section, differences
are observed.
One interesting difference for July 25 that can be seen in Table 13a is that young
cores merging with older ones tended to be younger, less reflective, and smaller than the
young echo cores merging with another young echo core. The reverse was true on August
26 (Table 13b). This was reflected in the frequency of growth of echo core 2 on these two
days (Table 14). On July 25, the young echoes merging with older ones grew more often
in area and height. On August 26, the young ones merging with other young echo cores
grew more in area, height, and reflectivity.
The bridge heights and reflectivities were similar for each category on a given day,
except for the base height of the bridge between the older cores on July 25. This could
indicate that the bridge between the older cores often was the result of particle transfer from
the expansion of anvils. Little growth was observed for the Type 3 echo cores on July 25.
f. Differences between growing and declining echo cores
This stratification was made based on whether the second echo core was growing in
terms of 3 out of the 4 echo parameters, or if 3 out of the 4 echo parameters were
38
Table 13a. Characteristics at the time of merger for the echo cores on July 25, 1986, by age grouping in minutes.
Both <15 Core 1 ≥15 Core 2 <15
Both >15
Sample Size 7 11 6
Bridge:
Maximum Height (km) 8.0 8.5 9.5
Max Z Height (km) 6.0 4.5 6.5
Minimum Height (km) 2.0 2.5 4.5
Depth (km) 3.0 6.0 5.0
Max Reflectivity (dBZ) 25.0 25.0 25.0
Echo 1:
Age at Merger (min) 6 31 50
Area of dBZ ≥35 (km2) 11 44 47
Maximum Height (km) 10.5 12.5 11.5
Maximum Reflectivity (dBZ) 42.5 42.5 45.0
Echo 2:
Age at Merger (min) 7 4 23
Area of dBZ ≥35 (km2) 5 2 28
Maximum Height (km) 10.5 9.5 10.0
Maximum Reflectivity (dBZ) 42.5 35.0 45.0
39
40
Table 13b. Characteristics at the time of mer on August 26, 1986, by age grouping
ger for the echo in minutes.
cores
Both <15 Core 1 >.15 Core 2 <15
Both ≥15
Sample Size 29 29 8
Bridge:
Maximum Height (km) 5.5 5.5 5.5
Max Z Height (km) 4.5 4.5 3.5
Minimum Height (km) 2.5 1.5 1.5
Depth (km) 2.0 4.0 5.0
Max Reflectivity (dBZ) 25.0 27.5 30.0
Echo 1:
Age at Merger (min) 6 28 31
Area of dBZ ≥35 (km2) 9 50 61
Maximum Height (km) 6.5 8.5 9.5
Maximum Reflectivity (dBZ) 42.5 52.5 55.0
Echo 2:
Age at Merger (min) 4 8 31
Area of dBZ ≥35 (km2) 3 8 20
Maximum Height (km) 5.5 6.5 7.0
Maximum Reflectivity (dBZ) 37.5 45.0 47.5
Table 14. Percent of echo core 2s which grew following merger, by echo pair age.
Both <15 Core 1 ≥15 Core 2 <15
Both ≥15
a. July 25, 1986 6
Sample Size 7 11
Max Reflectivity 86 64 33
Area ≥20 dBZ 57 64 0
Area ≥35 dBZ 57 64 40
Height 57 64 17
b. August 26, 1986
Sample Size 29 29 8
Max Reflectivity 69 45 50
Area ≥20 dBZ 76 69 25
Area ≥35 dBZ 71 61 60
Height 39 26 50
decreasing for the second echo core (Table 15a, 15b). This was done to emphasize the
characteristics of those cores that grew and those that did not grow.
The maximum height and the maximum reflectivity of either of the two echo cores
appears to have little impact on whether the second echo grew. On both days, a large
difference was observed in the median values of the age and area of the first echo core
(Table 15a and 15b). These echo cores tended to be younger and smaller for those where
the second core was growing.
41
Table 15a. Characteristics at the time of merger for the echo cores on July 25, 1989, grouped by whether or not echo 2 was growing.
Growing Declining
Sample Size 14 3
Bridge:
Maximum Height (km) 8.5 10.5
Max Z Height (km) 6.5 6.5
Minimum Height (km) 3.0 2.5
Depth (km) 4.0 6.5
Max Reflectivity (dBZ) 23.8 25.0
Echo 1:
Age at Merger (min) 13 34
Area of dBZ ≥35 (km2) 16 47
Maximum Height (km) 11.0 11.5
Maximum Reflectivity (dBZ) 43.8 45.5
Echo 2:
Age at Merger (min) 5 17
Area of dBZ ≥35 (km2) 3 16
Maximum Height (km) 10.0 8.5
Maximum Reflectivity (dBZ) 37.5 41.0
42
Table 15b. Characteristics at the time of merger for th on August 26, 1989, by whether or not echo 2 was
e echo cores growing.
Growing Declining
Sample Size 40 18
Bridge:
Maximum Height (km) 5.5 5.5
Max Z Height (km) 4.5 3.5
Minimum Height (km) 2.5 1.5
Depth (km) 3.0 4.0
Max Reflectivity (dBZ) 27.5 26.3
Echo 1:
Age at Merger (min) 11 24
Area of dBZ >35 (km2) 24 40
Maximum Height (km) 8.5 8.5
Maximum Reflectivity (dBZ) 52.5 50.0
Echo 2:
Age at Merger (min) 7 9
Area of dBZ ≥35 (km2) 8 8
Maximum Height (km) 6.5 6.5
Maximum Reflectivity (dBZ) 45.0 45.0
The median bridge characteristics differed for the declining and growing categories.
On July 25, the bridge tended to be deeper by being both taller and lower between the
declining echo pair. Likewise, on August 26, the bridge was lower and deeper when echo
core 2 was diminishing.
43
g. Characteristics of echoes with continued growth
Two populations of merging echo cores have been examined. Out of the 90 events,
approximately 25% occurred between young first echoes, about 65% occurred between a
young echo and an echo core that was already a part of a larger system (i.e., a parent
storm), and about 10% of the mergers were between echo cores that were a part of two
separate systems. For the sample as a whole, about 50 - 70% of the echo cores increased
in either reflectivity, area ≥20 dBZ, or area ≥35 dBZ. Of those that did grow following
merger, most were growing prior to merger. All were growing in one of the two area
categories. The most prominent factor in determining whether an echo grew following
merger was the age of the echo. Echoes grew more in area than in height following merger.
This probably reflects the tendency of echo cores to reach their peak height prior to the
time when they reach their peak area (Westcott 1990; Simpson 1980).
h. Manner of merging and bridge formation
In many case studies of mergers, the growth of a new cell which then bridges two
adjacent echoes and differential motion were found as the primary mechanisms that lead to
the merger of clouds (Cunning et al. 1982; Tao and Simpson 1989; Westcott and Kennedy
1989). Here, in examining a field of aggregating echo cores, expansion of the echo cores
appeared to be the primary mechanism leading to merger. In only a relatively few instances
were the development of new cells and differential motion involved in the aggregation of
echo cores on these two days. On July 25 differential motion was a primary factor in one
44
case; in two cases new cell growth was a primary cause; and in a fourth case both of these
factors were involved in the merger. On August 26, there were 6 instances where new core
growth bridged two existing cells and in one case differential motion was involved. In all
cases on both days, the horizontal area of one or both of the echo cores was expanding at
the time of merger.
It must be remembered, however, that clouds may not necessarily appear as a
reflectivity core. While the reflectivity data may indicate that the echoes were simply
expanding, this expansion may result not just from particle transfer. In the PACE 1989
results, it was found in a number of instances that cloud passes made through vigorous
updrafts, with large cloud liquid water contents, adjacent to a large parent storm were
aliased by the radar so that the feeder cloud was not visible as a separate radar echo, even
for echoes close enough to the radar where a 1 by 1 by 1 km resolution of the interpolation
grid would be appropriate.
i. Height and depth of bridge formation
In Cunning et al (1982), Tao and Simpson (1989), and Westcott and Kennedy (1989),
low level convergence was reported to be important in initiating the growth of new echo
cores and in enhancing the likelihood of merger. In Westcott and Kennedy (1989), it also
was proposed that intercell flow at mid-levels from an actively growing cloud resulting from
the radial outflow above the maximum updraft level would be a vehicle resulting in the
transport of cloudy material from one cell to another. While here there was no access to
45
velocity data, perhaps some inferences can be made regarding the origin of the bridge, based
on the height of the bridge in comparison with the first echo data and with the height of
merging cores.
If low level convergence from moist downdrafts was the primary factor involved in
the development of the bridge, it would be expected that the bridge heights might be similar
to or lower than that of the first echo. If particle transfer were the primary cause of bridge
formation, one would expect that the bridge be first observed at higher levels where outflow
from mature updrafts would be expected. The median values of first echo and bridge bases
and tops are presented in Table 16. The median values indicate that the bridge and first
echo top heights were similar on August 26. On July 25, the bridges were somewhat taller.
The distribution of the base dimensions (not shown here) indicate that there were cases
where the bridge first occurred well above the surface.
The base height of the bridge and the first echoes, and thus the depth, were
significantly different. The bridges clearly had lower bases than did the first echoes. This
implies that precipitation sized drops were found lower in the bridge. This may occur due
to a number of phenomena: either 1) precipitation drops were falling through the bridge
from above, 2) precipitation sized drops formed lower in the bridge than in the first echoes,
or 3) portions of the bridge resulted from rainout from the merging cores. Because the first
echoes were slightly more reflective, it is not likely that precipitation sized drops were falling
through the bridge. The formation of precipitation lower in the cloud might be possible if
46
Table 16. Bridge and first echoes (which later merged) median characteristics.
July 25, 1986 August 26, 1986
First Echo Bridge First Echo Bridge
Sample Size 30 24 79 66
Max Reflectivity (dBZ) 26.9 25.0 28.7 27.5
Top of Echo (km) 7.5 8.5 5.5 5.5
Height of Max (dBZ (km) 6.5 6.5 5.5 4.5
Base of Echo (km) 4.5 3.0 4.5 1.5
Depth (km) 3.0 5.0 1.0 4.0
the bridge area were moistened by the merging cores of the pre-rain outflow (Byers and
Braham 1949; Goff 1976). In the older cells where subsequent growth was minimal after
merger, the third factor might be operating; that is, the expansion of one or both of the two
echoes, resulted from the rainout of the merging cores (Table 15a and 15b).
The bridge top was often near the height of one of the merging cores. It may be that
the intercell mechanism proposed by Westcott and Kennedy (1989) was also acting. Plots
of bridge height versus the echo top heights for cores 1 and 2 are present in Figures 9 and
10. There was a positive correlation between the bridge height and the echo heights. The
correlation was stronger for echo core 1 on July 25,1986. Several of the outliers in the July
25 echo core 2 plot resulted from the seeding of the bridge area by the anvil from echo core
1. The correlation was slightly stronger for echo core 2 on August 26, 1986.
47
BRIDGE HEIGHT v.s. ECHO CORE 1 HEIGHT
25 JULY 86 - corr. = .552 26 AUGUST 86 - corr. =.481
Figure 9. Scatterplot of the bridge top height v.s. the height of the top of echo core 1.
BRIDGE HEIGHT v.s. ECHO CORE 2 HEIGHT
25 JULY 86 - corr. = .388 26 AUGUST 86 - corr. =.603
Figure 10. Scatterplot of the bridge top height v.s. the height of the top of echo core 2.
j. Summary and Conclusions
A number of past studies have established that an important element of the growth
of large cloud systems is the aggregation of cloud elements, and that these merged systems
are larger, longer-lived, and produce more rain than more isolated systems. The intent of
this study was not to duplicate that effort but rather to 1) look more closely at the behavior
of the individual cloud elements which join together, 2) describe their dimensions, 3) the way
in which they come together, and 4) the behavior of the echo cores immediately following
the merger.
Merging is known to occur on a variety of scales. Here, we found that on each day,
approximately 25-26% occurred between young first echoes, about 63-67% occurred between
a young echo and an echo core that was already a part of a larger system (i.e., a parent
storm), and 8-13% between echo cores that were a part of two separate systems. For the
sample as a whole, about 50-70% of the echo cores increased in either reflectivity, area ≥20
dBZ, or area ≥35 dBZ. Of those that did grow following merger, most were growing prior
to merger. All were growing in one of the two area categories. The most prominent factor
in determining whether an echo grew following merger was the age of the echo. Echoes
tended to grow more in area than in height following merger. This probably reflects the
tendency of echo cores to reach their peak height prior to the time when they reach their
peak area.
Many case studies of mergers, employing both modeling and observational data,
48
indicate that merging is accomplished through differential cloud motion or the growth of a
new echo core between two adjacent echoes. Here, echoes appeared to grow largely by
horizontal expansion. In only 10-17% of the cases did the other two factors play a major
role.
The bridge between the echo cores was typically deeper than the first echoes
observed in Section 4. The bridges were found to form lower than the first echoes. It
appears that in many cases, moisture-laden pre-rain downdrafts resulted in precipitation
forming lower in the bridge. Additionally, a positive relationship was found between the
echo core tops at the time of merger and the top of the bridge. It appears that the intercell
flow mechanism described by Westcott and Kennedy (1989) may be acting in some cases as
well.
6. References
Ackerman, B. and P.C. Kennedy, 1988: On the merger of two small convective radar
echoes. Preprint, 10th International Conf. on Cloud Physics, Bad Hamburg, 440-442.
Browning, K.A. and F.H. Ludlum, 1962: Airflow in convective storms. Quart. J. Roy.
Meteor. Soc, 88, 117-135.
Byers, H.R. and R.R. Braham, Jr., 1949: The Thunderstorm Project. U.S. Government
Printing Office, Washington, D.C., 287 pp.
Changnon, S.A., Jr., 1976: Effects of urban area and echo merging on radar echo behavior.
J. Appl. Meteor., 15, 561-570.
49
Cunning, J.B., and M.DeMaria, 1986: An investigation of the development of
cumulonimbus systems over South Florida. Part 1: Boundary layer interactions. Mon.
Wea. Rev., 114, 5-24.
Cunning, J.B., R.L. Holle, P.T. Gannon and A.I. Watson, 1982: Convective evolutions and
merger in the FACE experimental area: mesoscale convection and boundary layer
interactions. J. Appl. Meteor., 21, 953-977.
Dennis, A.S., C.A. Schrock and A. Koscielski, 1970: Characteristics of hailstorms of western
South Dakota, J. Appl. Meteor., 9, 127-135.
Fankhauser, J.C., 1982: The 22 June 1976 case study: Large-scale influences, radar echo
structure and mesoscale circulations. Hailstorms of the Central High Plains, Vol. 2.
Colorado Assoc. Univ. Press, 1-35.
Foote G.B. and H.W. Frank, 1983: Case Study of a Hailstorm in Colorado. Part III: Airflow
From Triple-Doppler Measurements. J. Atmos. Sci., 40, 686-707.
Goff, R.C., 1976: Vertical structure of thunderstorm outflows. Mon. Wea. Rev., 104,1429-
1440.
Houze, R.A.,Jr., and C-P Cheng, 1977: Radar characteristics of tropical convection observed
during GATE: Mean properties and trends over the summer season. Mon. Wea.
Rev., 105, 964-980.
LeMone, M.A., 1989: The influence of vertical wind shear on the diameter of cumulus
clouds in CCOPE, Mon. Wea. Rev., 117, 1480-1491.
50
Lopez, R.E., 1976: Radar characteristics of the cloud populations of tropical disturbances
in the northwest Atlantic. Mon. Wea. Rev., 104, 268-283.
Lopez, R.E., 1978: Internal structure and the development processes of C-scale aggregates
of cumulus clouds. Mon. Wea. Rev., 106, 1488-1494.
Maurwitz, J.D., 1972: The structure and motion of severe hailstorms. Part II: Multicelled
Storms. J. Appl. Meteor., 11, 180-188.
Orville, H.D., Y-H.Kuo, R.D.Farley and CS.Hwang, 1980: Numerical simulation of cloud
interactions. J. Rech. Atmos., 14, 499-516.
Peterson, R., 1984: A Triple-Doppler radar analysis of discretely propagating multicell
convective storm. J. Atmos. Sci., 41, 2973-2990.
Simpson, J., 1980: Downdrafts as linkages in dynamic cumulus seeding effects. J. Appl.
Meteor., 19, 477-487.
Simpson, J., N.E. Westcott, R. J. Clerman, and R. A. Peilke, 1980: On cumulus mergers.
Arch. Meteor. Geoph. Biokl., Ser. A, 29, 1-40.
Tao W-K. and J. Simpson, 1989: A further study of cumulus interactions and mergers:
Three-dimensional simulations with Trajectory Analysis, J. Atmos. Sci., 46,2974-3004.
Turpeinen, O., 1982: Cloud interactions and merging on day 261 of GATE, Mon. Wea. Rev.,
110, 1238-1254.
Westcott, N.E., 1990: Radar results of the 1986 Exploratory field program relating to the
design and evaluation of PACE, J. Wea. Mod., 22, 1-17.
51
Westcott, N.E., 1984: A historical perspective on cloud mergers, Bull. Amer. Meteor. Soc,
65, 219-226.
Westcott, N. E., and P.C. Kennedy, 1989: Cell development and merger in an Illinois
thunderstorm observed by Doppler radar. J. Atmos. Sci., 46, 117-131.
Weisman, M.L. and J.B. Klemp, 1982: The Dependence of Numerically Simulated
Convective Storms on Vertical Wind Shear and Buoyancy. Mon. Wea. Rev., 110,
504-520.
Wiggert, V.G., J. Lockett, and S.S. Ostlund, 1981: Rain shower growth histories and
variations with wind speed, echo motion, location and merger status. Mon. Wea.
Rev., 109, 1467-1494.
52
IN-CLOUD STUDIES
1. Preliminary Analysis of Rime-Splintering in PACE Clouds
a. Synopsis
In order to adequately understand the effect of silver iodide on the microphysical and
dynamic character of clouds, it is necessary to obtain a good understanding of the natural
sequence of events in glaciation. The evidence in past (Braham 1964) observations as well
as that more recently obtained in PACE atmospheric research (Czys 1988) generally support
the idea that ice initiates by the freezing of drizzle and rain drops. Subsequently, these
drops go on to rime supercooled cloud droplets. Consequently, warm-based convective
clouds are typically composed of a mixture of graupel, frozen drops and supercooled cloud,
drizzle and rain drops while ice forms, large enough to be detected by instrumentation
available to PACE in 1986 and 1989, rarely take the form of vapor grown crystals. Because
graupel concentrations are usually larger than expected for ice nuclei some sort of secondary
or ice multiplication process is suspected to account for the discrepancy between "first" ice
and ice nuclei.
As part of the research activities of the past year, we conducted an analysis which
focused on the possibility that a rime-splintering or Hallett-Mossop process was responsible
for the first ice observed in the summer convective clouds we sampled in 1986. These clouds
are good candidates to be considered because they meet all of the physical criteria (to be
elaborated on later) necessary for the process.
53
In this analysis observed ice production rates were compared to those diagnosed for
a rime-splintering process using data collected in the updrafts of a small sample of warm-
based cumulus congestus observed during PACE86. Diagnosed rates were found to exceed
observed rates in slightly more than half the updrafts. Mean observed and diagnosed rates
for Midwestern cumulus were found to be 38 and 109 s-1 m-3, respectively. The shift of the
distribution of observed rates to values lesser than diagnosed was found to be statistically
significant at the 0.5% level.
The discrepancy was found not to be strictly related to warming of the graupel
surface by the collection of supercooled cloud droplets (D < 50 µm). Some evidence was
uncovered that indicated that diagnosed rates increasingly tend to exceed observed as the
amount of supercooled liquid water in the size range of drizzle and rain drops increased.
Destruction of the surface structure of frail ice by wetting when a graupel collects a
supercooled rain drop was speculated as a mechanism for suppression of splinter production.
However, these findings are subject to major shortcoming in application of current theory
and large measurement uncertainty associated with an inability to adequately resolve small
(D < 100 µm) ice forms.
The theory used in this research would be improved if three temperature dependent
processes were parameterized. These are: (1) the temperature dependence, F(T), of the
rime-splintering process, (2) the change in the microphysical conditions for splintering that
occurs as a parcel rises through the ice multiplication zone, and (3) the change in the parcels
54
ascent rate in the ice multiplication zone. From this research we recognized that a better
comparison to make would be between observed and diagnosed ice concentrations calculated
from ice production rates.
If this issue is to be pursued in future field programs, instrumentation need to be
included on aircraft which will measure small ice. Aircraft performance needs to be
improved to allow tracking vertical cloud growth in a Lagrangian reference frame to monitor
the change of in-cloud conditions (both microphysical and kinematic) from the melting level,
through the ice multiplication zone, to above the -10 °C seeding level.
b. Scientific Background
Attention to the fact that "first" ice particles are found in clouds at temperatures
warmer and concentrations higher than expected from heterogeneous nucleation alone dates
back many years (Palmer 1949; Brewer and Palmer 1949). First ice particle concentrations
in excess of that expected from ice nuclei have been reported for clouds from all over the
world: cumulus in England (Murgatroyd and Garrod 1960), Australia (Mossop et al. 1972),
South Africa (Krauss et al. 1987), and in the United States in cumuli over Missouri (Koenig
1963), Florida (Hallett et al. 1978), West Texas (Jurica and Scro 1985), and Central Indiana
(Czys 1989).
One mechanism often indicated to explain the discrepancy between first ice and ice
nuclei is that of splinter production during riming. This mechanism was demonstrated in
laboratory experiments that showed the ejection of secondary ice crystals when supercooled
55
droplets froze onto an ice substrate (Hallett and Mossop 1974). Subsequent laboratory work
indicated that certain physical criteria must be met for the process to operate. For example,
the process is temperature dependent in that ice splinters are produced at temperatures
between -3 and -8°C, with peak production at -5°C (Hallett and Mossop 1974). Droplets
larger than 25 µm (Mossop and Hallett 1974) and droplets smaller than approximately 13
µm (Mossop 1978) must be present. The rate of ice splinter production has been found to
be proportional to the rate of sweep out of droplets larger than 25 µm (Mossop 1976) and
has been found to be sensitive to the velocity of the riming body (Mossop 1976). Finally,
the rate of splinter production has been found in laboratory experiments to be influenced
by the surface temperature of the riming body (Heymsfield and Mossop 1984; Foster and
Hallett 1982) rather than by the environment temperature. Hence, splinter production may
occur anywhere the surface temperature of the riming body falls in the temperature range
of-3 to-8°C.
Mossop (1978) has provided a relationship for calculating ice production rates. This
expression is, P = aNLNsm, where P is the ice splinter production rate, NL is the accretion
rate for droplets > 25 µm, Ns the accretion rate for droplets < 13 µm and a and m are
constants relating experiment conditions such as temperature and riming rate. However,
because this equation is somewhat inconvenient to use with aircraft data, accepted practice
has been to suspect that a rime-splintering process has operated in a cloud whenever higher
than expected ice concentrations are found and the in-cloud conditions meet the criteria for
56
rime-splintering. An improvement away from this practice was made possible when Harris-
Hobbs and Cooper (1987) introduced an expression for diagnosing ice production rates by
rime-splintering that not only embodies many of the criteria for rime-splintering, but is also
fairly convenient to use with microphysical measurements.
c. Data
Data used in this study were collected as part of the 1986 field program of the
Precipitation Augmentation for Crops Experiment (PACE). In-cloud measurements were
made using a light, twin engine airplane (Beechcraft Baron) equipped to measure standard
meteorological state parameters, such as, temperature, pressure and dew point as well as
equipped with microphysical instrumentation such as an FSSP, and 2DC and 2DP probes.
Vertical air velocity was estimated using a method discussed by Lawson (1979).
Data were collected on 8 July 1986 near the tops of seven cumulus congestus clouds
feeding a main storm system located just southwest of Indianapolis, Indiana. Penetrations
were made near the -10° C level directly at the center of cloud with no more than 5000 and
no less than 1000 ft of cloud top above the level of cloud entry. Data used in this study are
only from the updraft portions of each penetration. In all, 22 updrafts were identified,
where an updraft was defined as any period of in-cloud measurement having at least three
consecutive seconds of positive velocity in excess of 1 m s-1. Thus, the data are largely
representative of the result of adiabatic and collection processes in vertical ascent with other
influences minimized.
57
58
d. Calculation of Ice Production Rates
Observed ice production rates were calculated from values of ice concentration and
updraft velocity observed at the -10°C level and using the depth of the -3 to -8°C layer
which was estimated from a CLASS sounding launched near the time of cloud penetrations.
Thus, the observed rate of ice production was simply taken as ice concentration divided by
the amount of time for secondary ice production, where the amount of time for secondary
production was taken as the ratio of mean updraft velocity to the depth of the -3 to -8°C.
On July 8, 1986 the depth of the -3 to -8° C layer was estimated to be about 1 km. Hence,
a typical graupel concentration of 10 per liter with a typical updraft velocity of 4 m s-1 yields
an ice production rate of 40 m-3 s-1.
Of course this calculation is sensitive to graupel size since millimeter size graupel
(which incidentally should be important contributors to splinter production by virtue of their
large collection kernel) have fall speeds comparable to updraft velocity and thus may spend
a longer time in the ice multiplication zone than indicated from the ratio of layer depth and
updraft velocity. However, although graupel of approximately millimeter size was
occasionally identified in the 2DP image records, most often it was in the range of sizes from
approximately 200 to 600 µm diameter. Thus, the graupel can be assumed to be traveling
vertically at about the speed of the updraft. Neglecting to adjust for graupel fall speed has
probably resulted in observed rates which are no more than 15% larger than if the
adjustment was made.
Secondary ice production rates were diagnosed using an expression developed by
Harris-Hobbs and Cooper (1987). The equation has the form:
where Rg, Ng(Rg), and V(Rg) are equivalent radius, size spectrum, and fall speed for
graupel, respectively. The variables r, n(r), and v(r) are respectively, cloud droplet size, size
spectrum, and fall speed. Fall speeds for graupel and droplets were calculated using
relationships provided by Heymsfield (1978) and Beard (1976), respectively. The collection
efficiency E(R ,r) is that for collisions between graupel and cloud droplets, and was taken
as unity in the calculations for Midwestern clouds. The constant C was given as 0.16 in
Harris-Hobbs and Cooper (1987) and was determined by fitting Eq. 1 to the laboratory data
of Mossop (1978).
Examination of Eq. 1 reveals that it takes into account many of the sensitivities and
dependencies of the rime-splintering process. Temperature dependence is accounted for by
the function f(T) which is unity at -5°C and linearly decreases to zero at -3 and -8°C,
respectively. Droplet size dependence is taken into account by the function g(r) which is
defined as:
59
which accounts for the large and small droplet dependence in addition to the sensitivity of
the process to the fraction of the graupel surface covered by droplets less than 13 µm (G<13)
to the total cloud droplet population (Gall). Finally, the dependencies of sweep out rate and
velocity of the riming body are also evident in Eq. 1.
e. Results
Figure 1 is a plot of observed ice production rates versus that diagnosed for rime-
splintering process using Eq. 1. Ice production rates (IPR's) for Midwestern clouds are
denoted by solid circles. Horizontal and vertical lines passing through the solid circles
represent 70% confidence intervals based on the Poisson statistics discussed by Cornford
(1967). Confidence intervals are based solely on the uncertainty of measuring ice
concentration and do not take into account measurement uncertainties associated with, for
example, measuring distributions of supercooled cloud droplets with an FSSP in the presence
of ice (Gardiner and Hallett 1985). To avoid cluttering Fig. 1 approximately half of the
IPR's were chosen randomly to have their confidence intervals plotted. Diagonal lines
indicate IPR ratios of 1:1, 1:10, and 1:100 in favor of diagnosed ratios.
For comparison, diagnosed and observed IPR's for Florida cumuli as reported by
Harris-Hobbs and Cooper (1987) are plotted in Fig. 1 as open circles. Confidence intervals
for these data were not provided in their paper. To spread out the cluster of data over the
60
Figure 1. Observed versus diagnosed ice production rates.
largest possible area, three of the Florida data do not appear in Fig. 1. These data points
are (2.2,0.002), (1.6,0.002), and (0.15,0.002), where the first number of the ordered pair is
an observed rate and the second number is a diagnosed rate.
As can be seen in Fig. 1, observed and diagnosed IPR's for Midwestern and Florida
cumuli scatter over the same general area. This level of agreement is not surprising since
the characteristics of clouds from either region have many similarities and because the
method of calculation used on the Midwestern data is very similar to that used with the
Florida data. An important similarity is that the distribution of convective cloud-base
temperatures for the Midwest is close to that for Florida (Johnson 1982). Furthermore, the
range of main updraft velocities is also similar (Czys 1988; Sax and Keller 1980).
Consequently, clouds from both regions have active coalescence processes that tend to
produce drizzle and rain drops before cloud top reaches the 0°C isotherm (Hallett et al.
1978; Braham 1964). Cloud droplet populations tend to be greater in Florida than in
Midwestern cumulus, but clouds in both regions generally possess many droplets less than
13 µm, in addition to droplets greater than 25 µm,m diameter. Most importantly, clouds from
either region develop ice in excess of that expected from the action of ice nuclei alone. For
example, the observed ice production rates plotted in Fig. 1 for Midwestern cumuli
correspond to initial graupel concentrations which range from 1 to about 15 per liter, roughly
10 to 100 times the activity of ice nuclei at -10°C (Fletcher 1962).
Close inspection of Fig. 1 reveals a tendency for rates diagnosed for Midwestern
61
updrafts to be 5 to 10 times higher than observed, and in a few instances diagnosed rates
are about 100 times larger than observed. Of the 22 Midwestern updrafts, 13 (~60%) have
diagnosed rates at least five times greater than observed. Furthermore, the occurrence of
diagnosed rates in excess of that observed appears to have occurred by more than can be
explained by measurement uncertainty, even in light of the fact that confidence intervals are
generally not as small as desirable. It is also interesting to note that both Midwestern and
Florida IPR's show at least a couple of updrafts with very low observed rates and very high
diagnosed rates, about 1:100. This size of discrepancy suggests that some updrafts can have
very favorable conditions for rime-splintering, but do not necessarily have the highest
observed rates of ice production. Thus, even though the ice in Midwestern clouds tends to
originate in concentrations greater than expected from ice nuclei, the results presented in
Fig. 1 suggests that, in many cases, observed rates of production are much less than expected
than if a rime-splintering process was operating at full potential.
Similar tendency can be found in examination of mean ice production rates. Mean
observed and diagnosed IPR's for updrafts in Midwest cumuli are 38 and 109 m-3 s-1,
respectively. The discrepancy is even larger for the Florida data with 19 and 163 m-3 s-1 for
mean observed and diagnosed rates, respectively (for the Florida data, means include the
three data points not plotted in Fig. 1). Thus, rates diagnosed for the Midwest sample of
updrafts are about 3 times larger than observed while diagnosed rates for the Florida sample
are about 8 times larger than observed.
62
Ice production rates were subjected to the Wilcoxon sign-rank test to determine
whether the differences noted were statistically significant. The Wilcoxon test is a
nonparametric procedure for determining whether or not the distributions of two related
populations are shifted from one another. For the Midwestern sample, the difference
between the distribution of observed and diagnosed rates was found to be statistically
significant at greater than the 0.5% significance level. Similarly, the difference between the
distribution of observed and diagnosed rates for Florida cumulus were found to be shifted
at greater than the 0.5% significance level. Thus, in this small sample, the tendency for
diagnosed rates to be greater than observed appears to have occurred by more than chance.
The Mann-Whitney sign rank test was used to test for similarities between the
Midwestern and Florida observed rates, as well as for similarities between diagnosed rates.
The Mann-Whitney test is a nonparametric procedure for determining whether or not the
distributions of two independent populations are shifted from one another. The Mann-
Whitney test indicated that the distributions of diagnosed rates were different with only 20%
significance. This lower significance level suggests that the distribution of diagnosed rates
are about the same for clouds from either region. Thus, the clouds (updrafts) in this sample
probably had similar potentials for rime-splinter production.
On the other hand, the distributions of observed rates in Midwestern and Florida
clouds were found to be different at the 5% significance level. This level of significance
supports the possibility that the observed rates for Florida clouds were in fact slightly less
63
than observed rates in Midwestern clouds. Hence, whatever acted to cause observed rates
to be below the potential indicated by the diagnosed rates may have been slightly more
effective in the Florida clouds. Thus, if a rime-splintering process was indeed initially
operating in these updrafts, the possibility of a suppression mechanism and factors that
would control the level of suppression should be considered.
f. Discussion
The evidence of this study suggests that even though initial ice concentrations in
Midwestern updrafts are larger than expected from ice nuclei, the rate of ice production in
many cases is significantly less than expected from a rime-splintering process. This is an
unexpected result because in all updrafts the criteria for a rime-splintering process are met:
sufficient numbers of droplets smaller than 13 µm and larger than 25 µm exist, updraft
velocities in comparison to droplet and graupel fall speed are in the range allowable for
rime-splintering, and supercooled drizzle and rain drops are usually present. It is unclear
as to why a rime-splintering process either did not operate or was suppressed below
potential in many of the updrafts in the Midwestern sample. The discrepancy between
observed and diagnosed IPR's can not be completely explained away in terms of
measurement uncertainty even though the uncertainty is larger than desirable. Furthermore,
the diagnostic expression used appears to take into account most of the physical criteria for
the rime-splintering process. The expression was said to produce good agreement with
laboratory data, and indicate observed rates that were at least as large as, if not in excess
64
of that diagnosed for measurements taken in clouds over Montana (Harris-Hobbs and
Cooper 1987).
The first possibility examined was that the lack of splinter production was somehow
related to a change in the surface temperature of the graupel resulting from the collection
of many supercooled cloud droplets, as was indicated in laboratory work of Heymsfield and
Mossop (1984). Table 1 summarizes liquid water content measured by a hot-wire probe and
calculated from the FSSP data along with values of diagnosed, observed and the ratio of
diagnosed to observed rates of ice production for each Midwestern updraft (U-ID) from
each cloud (C-I) in this study. Means and sample deviations are listed at the bottom of
Table 1. Peak and mean liquid water contents from the hot wire probe are listed beneath
the columns labeled PJWC and MJWC, respectively. Peak and mean liquid water content
calculated from the FSSP data are listed beneath PFWC and MFWC, respectively.
In an attempt to apply their laboratory results to natural clouds, Heymsfield and
Mossop (1984) calculated surface temperature elevations expected during graupel growth
based on the particle growth model by Heymsfield (1982). Two sets of results were
presented for a range of liquid water contents up to 5 g m"3: one for graupel densities of
0.3 g cm-3 to simulate small graupel growing from ice crystals and one of 0.9 g cm-3 to
simulate riming growth by newly frozen drops. Their calculations clearly indicate that
surface temperature elevations sufficient to matter to a rime-splintering process do not
generally occur for liquid water contents less than about 1 g m-3.
65
Table 1. Cloud Droplet Liquid Water Contents and Ice Production Rates.
C_ID U_ID PJWC gm-3
MJWC PFWC gm-3
MFWC Diagnosed Observed Ratio
708cl 1 0.4 0 3 0.4 0.4 108 12 9.0
2 0.7 0.5 0.5 0.4 128 27 4.7
3 0.7 0.5 0.5 0.4 154 19 8.1
708c3 4 0.1 0.0 0 3 0.2 227 31 7 3
5 0.0 0.0 0.1 0.1 36 7 5.1
6 0.1 0.1 0 3 0.2 139 5 27.8
7 0.0 0.0 0.1 0.1 33 6 5.5
8 0.0 0.0 0.0 0.0 - - -
708c5 9 2.3 1.2 0.8 0.5 18 10 1.8
10 1.3 1.1 0.6 0.6 87 8 10.9
11 0.8 0.5 0.4 0.4 125 35 3.6
12 0.2 0.1 0.2 0.2 243 27 9.0
13 0.2 0.1 0 3 0.1 52 0.2 260.0
708c7 14 0 3 0.2 0 3 0.2 30 14 2.1
15 0.2 0.1 0.4 0.3 86 227 0.4
16 0.1 0.0 0 3 0.2 129 178 0.7
17 0.1 0.1 0.2 0.1 132 53 2.5
708cll 18 0.2 0.1 0.2 0.1 - - -
19 0.3 0.2 0 3 0.2 - - -
20 2.2 1.1 0.6 0.5 478 37 12.9
21 1.5 1.2 0.5 0.4 - - -
708cl2 22 0.0 0.0 0.2 0.2 - - -
23 0.1 0.0 0.2 0.1 81 0.3 270.0
24 0.1 0.0 0.1 0.1 57 54 1.1
708cl4 25 0.5 4.0 0.5 0.4 11 23 0.5
26 0.3 0.2 0.4 0.3 27 33 0.8
27 0.4 0 3 0.5 0.3 9 20 0.5
Mean 0.5 0.3 0.4 0 3 109 38 29.3
Var. 0.7 0.4 0.2 0.2 105 56 74.8
66
As can be seen in Table 1 the liquid water for supercooled cloud droplets averaged
less than about 0.5 g m-3 and there appears to be no trend between higher cloud droplet
liquid water contents and larger discrepancy between diagnosed and observed ice production
rates. Therefore, no evidence in the data about supercooled cloud droplets could be found
to support the hypothesis that the discrepancy was related to surface temperature elevation
during graupel growth by accretion. However, in exploring the possibility that, surface
temperature elevation in riming may be responsible for the noted discrepancy, a
fundamental difference between the laboratory experiments and Midwestern and Florida
clouds was noted; the laboratory experiments were confined to cloud droplets less than
about 40 µm while warm-based clouds tend to produce supercooled drizzle and rain drops
in their updrafts by the time cloud top reaches 0°C (Battan 1953). Thus, the possibility that
the discrepancy was somehow associated with the presence of supercooled drizzle and rain
drops was explored.
In this exploration, it was hypothesized that if the rime-splintering process was
sensitive to the surface temperature of the riming body, then the occasional collection of
supercooled drizzle or rain drop might be suspected of resulting in the suppression of
splinter production perhaps (1) by increasing the surface temperature from the release of
latent heat as the liquid of the collected drop freezes, or (2) because when a supercooled
drizzle or raindrop is collected, the riming surface may become coated by liquid before
sufficient disposal of latent heat would allow the water to crystalize. Thus, when a graupel
67
collects a supercooled drizzle or raindrop, the structure of the frail ice created in the
accretion of cloud droplets will be altered from that which is favorable for splinter
production. Therefore, the presence of supercooled drizzle and rain drops may not
necessary be advantageous to a rime-splintering process although the presence of these
hydrometeors are closely associated with the onset of ice (Braham 1964).
This hypothesis was tested by comparing ice production rates to the total supercooled
liquid water indicated derived from the 2DC and 2DP image records for each updraft.
Liquid water contents were determined by first constructing a size distribution. A line was
then fit to the size distribution to obtain a slope and intercept. Once these parameters were
known, the amount of supercooled water represented by the distribution was obtained using
the integral form of the equation for liquid. The departure of ice production rates was
represented by the ratio of the diagnosed rate to the observed rate. Thus, a sense of the
extent to which diagnosed rates exceed observed is obtained as ratios become larger than
one.
Figure 2 is a plot of total 2D liquid water versus ice production rate ratio for the 22
Midwestern updrafts. The largest two IPR ratios and the smallest two IPR ratio are denoted
by encircled dots while the bulk of the data is plotted as open circles. Although the data is
poorly correlated (r=.16), Fig. 2 shows a tendency for diagnosed rates to increasingly exceed
observed rates as updraft liquid water content increases. However, a t-test applied to all
data indicated that the positive slope of the fit was different from no trend at the 20%
68
Figure 2. Total precipitation liquid water content versus the ice production rate ratio
(Diagnosed/Observed) for the Midwestern sample of updrafts.
significance level, thus suggesting no trend.
Since the scatter in Fig. 2 is large, the linear regression was reapplied excluding the
out lying data denoted by the encircled dots to obtain a robust sense of trends in the data.
The solid line in Fig. 2 is a least squares fit to the bulk of the data (N = 18 updrafts) and
the dashed lines form a 90% confidence band to the least squares fit. As might be expected,
the correlation coefficient improves to 0.46 and the positive slope, as indicated by the t-test,
is different from no trend at the 2% significance level. Thus, weak evidence exists to suggest
that ice production rates tend to be negatively associated with the presence of supercooled
drizzle and rain drops.
g. Conclusion
In the updrafts of a small sample of warm-based Midwestern clouds, observed rates
of ice production were found to be less than expected from a rime-splintering process even
though initial ice concentrations were greater than expected from ice nuclei. This
discrepancy was found in approximately 60% of the updrafts. A similar discrepancy was
found in ice production rates in Florida clouds. Low liquid contents of supercooled cloud
droplets (D < 50 µm) do not support the possibility that the discrepancy is strictly related
to surface temperature elevation during graupel growth by accretion. Weak evidence was
found that indicated a tendency for larger discrepancies to occur when larger amounts of
supercooled drizzle and rain drops were present. A possible mechanism for splinter
production suppression that can be connected to collection of supercooled drizzle and rain
69
drops is a liquid coating which destroys the structure of the frail ice necessary for splinter-
production, in addition to increasing the surface temperature of the rimer from the slow
disposal of latent heat as the liquid coating crystalizes. Thus, in warm-based clouds that
produce rain drops before cloud top reaches -10° C, rime-splinter production may be limited
to only graupel that have few or no encounters with supercooled rain drops on ascent
through the ice multiplication zone. It is interesting to note that the discrepancy between
diagnosed and observed rates is greater in Florida clouds than in Midwestern since Florida
clouds typically have slightly warmer bases and thus, may typically have a slightly greater
presence of supercooled drizzle and rain drops.
However, caution must be exercised in generally accepting or extending these findings
to other warm-based clouds in the Midwest. The results for Midwestern clouds apply to
measurements taken on the same day for a relatively small sample of clouds and we should
be able to correct this shortcoming with the addition of data collected during the 1989 PACE
field experiment. It was also necessary to assume that the conditions at -10°C were
representative of the conditions for rime-splintering in the updrafts as they passed through
the ice multiplication zone. It is easy to imagine how subtle differences between the
population of supercooled cloud droplets observed at -10° C and the population that existed
in the ice multiplication zone could lead to lesser rime-splintering rates even though
conditions at -10 ° C appear to be favorable. Another factor that will lead to over estimating
rime-splintering rates is that graupel are less numerous and smaller in the ice multiplication
70
zone than at the -10°C level. Yet another important factor is the presence and size of
graupel change dramatically as the air parcel rises from -3 to -10°C. The theory also does
not include updraft velocity as a variable and this may lead to an overestimate of diagnosed
rates for updrafts having favorable conditions for rime-splintering but moving quickly
through the ice multiplication zone and underestimates for updrafts having not very
favorable conditions for rime-splintering but moving leisurely through the ice multiplication
zone. Because the analysis is restricted to only portions of the cloud with continuous
updrafts of 1 m s-1 or larger, a good assumption is that most of the cloud particles in the
sampled parcels followed vertical trajectories originating from below the sample point.
h. References
Battan, L.J., 1953: Observation on the formation and spread precipitation in convective
clouds. J. Meteor., 10, 311-324.
Beard, K.V., 1976: Terminal velocity and shape of cloud and precipitation drops aloft. J.
Atmos. Sci., 33, 851-864.
Braham, R.R., 1964: What is the role of ice in summer rain-showers? J. Atmos. Sci.. 21,
640-645.
Brewer, A.W., and H.P. Palmer, 1949: Condensation processes at low temperatures and the
production of new sublimation nuclei by the splintering of ice. Nature, 220, 687-689.
Changnon, S.A., 1986: Illinois weather modification program: PACE. Preprints Conf. on
Weather Modification, Arlington, VA, 315-319.
71
Cornford, S.G., 1967: Sampling errors in measurements of raindrop and cloud droplet
concentrations. Met. Mag.. London, 96, 271-282.
Czys, R.R., 1989: Ice initiation by collisional forcing in warm-based cumuli. J. Appl.Met..
28, 1098-1104.
Czys, R.R., 1988: Microphysical characteristics of warm-based cumuli: Observations at -
10°C. Proceeding of the 10th International Cloud Physics Conference, Bad
Homburg, Federal Republic of Germany.
Fletcher, N.H., 1962: Physics of Rain Clouds. Cambridge University Press, 390 pp.
Foster, T., and J. Hallett, 1982: A laboratory investigation of the influence of liquid water
content on the temperature dependence of secondary ice crystal production during
soft hail growth. Preprints, Cloud Phys. Conf., Chicago, Ill., 123-126.
Gardiner B A., and J. Hallett, 1985: Degradation of in-cloud forward scattering
spectrometer probe measurements in the presence of ice particles. J. Atmos. and
Oceanic Technol., 2, 171-180.
Hallett, J., R.I. Sax, D. Lamb, and A.S. Ramachandra Murty, 1978: Aircraft measurements
of ice in Florida cumuli. Quart. J. R. Met. Soc, 104, 631-651.
Hallett, J., and S.C. Mossop, 1974: Production of secondary ice crystals during the riming
process. Nature. 249, 26-28.
Harris-Hobbs, R.L., and W.A. Cooper, 1987: Field evidence supporting quantitative
predictions of secondary ice production rates. J. Atmos. Sci.. 44, 1071-1082.
72
Heymsfield, A.J., 1982: A comparative study of the rates of development of potential
graupel and hail embryos in High Plains storms. J. Atmos. Sci.. 39, 2868-2897.
Heymsfield, A.J., 1978: The characteristics of graupel particles in northeastern Colorado
cumulus congestus clouds. J. Atmos. Sci.. 35, 284-295.
Heymsfield, A.J., and S.C. Mossop, 1984: Temperature dependence of secondary ice crystal
production during soft hail growth by riming. Quart. J. R. Met. Soc. 110, 765-770.
Johnson, D.B., 1982: Geographical variations in cloud-base temperature. Preprints AMS
Conf. on Cloud Physics, Chicago, IL, 187-189.
Jurica, G.M., and K.D. Scro, 1985: Microphysical properties of cumulus congestus clouds
observed during Texas HIPLEX. Pro. Fourth WMO Conf. on Weather Modification,
Honolulu, World Meteor. Org., 363-368.
Koenig, L.R., 1963: The glaciating behavior of small cumulonimbus clouds. J. Atmos.Sci.,
20, 29-47.
Krauss, T.W., R.T. Bruintjes, and J. Verlinde, 1987: Microphysical and radar observations
of seeded and nonseeded continental cumulus clouds. J. Climate Appl. Meteor.. 26,
585-606.
Lawson, R.P., 1979: A system for airborne measurement of vertical air velocity. J. Appl.
Met.. 18, 1363-1368.
Mossop, S.C, 1978: The influence of drop size distribution on the production of secondary
ice particles during graupel growth. Quart. J. R. Met. Soc. 104, 323-330.
73
Mossop, S.C., 1976: Production of secondary ice particles during the growth of graupel by
riming. Quart. J. R. Met. Soc. 102, 45-57.
Mossop, S.C., and J. Hallett, 1974: Ice crystal concentrations in cumulus clouds: Influence
of drop spectrum. Science. 186, 632-634.
Mossop, S.C., R.E. Cottis, and B.M. Bartlett, 1972: Ice crystal concentrations in cumulus
and stratocumulus clouds. Quart. J. Roy. Meteor. Soc. 98, 105-123.
Murgatroyd, R.J., and M.P. Garrod, 1960: Observations of precipitation elements in
cumulus clouds. Quart. J. Roy. Meteor. Soc. 86, 167-175.
Palmer, H.P., 1949: Natural ice-particle nuclei. Quart. J. Roy. Meteor. Soc. 75, 15-22.
Sax, R.I., and V.W. Keller, 1980: Water-ice and water-updraft relationships near -10°C
within populations of Florida cumuli. J. Appl. Met.. 19, 505-514.
74
2. Development of 2D Image Classification Software for PACE
a. Synopsis
By the end of the 1989 field program we had obtained over 100 cloud penetration
and were faced with the enormous task of classifying thousands of 2D particle images before
the aircraft data could be used in meaningful ways. These ways are: 1) characterize the
natural microphysical condition of PACE clouds, 2) development conceptual models about
precipitation processes, and 3) determine the seedability and potential initial reaction of the
cloud both microphysically and dynamically to silver iodide seeding. In our attempt to
develop software that might speed the process of image classification, we discovered that a
polynomial approximation technique, applied to the perimeter data, of image data could be
used to give a reliable discrimination between smooth circular images and circular images
having rough outlines. This section reports on our initial attempts in the development and
testing of this software. Although this project is very near to completion, results of our
evaluation of software performance reported herein are preliminary. Research we are
currently conducting has revealed several potential improvements to the technique.
Therefore, we expect our final evaluation of its performance to be better that reported here.
The technique we developed and tested for discriminating between smooth and rough
images in 2D image records involves fitting a fourth order polynomial using the method of
least squares, to the curvature of each half of large symmetric circular images. The variance
of the polynomial approximation is used as the discriminator. So far we have found that if
75
a variance of less than 0.20 is determined for all halves of an image, then to a high
probability the image has a smooth outline. If the image has a variance greater than 0.20,
then it very likely has a rough outline. A method for determining the radius of "center-out"
images was also developed in this research. A preliminary evaluation of technique
performance based on the slope and intercept of image size distributions, as well as on
computed water content for drops and graupel, has been completed. This preliminary
evaluation indicated that the technique generally performs well and may eventually prove
to be useful under certain circumstances.
b. Background
Instrumentation for shadow graphing cloud particle images has created the need for
reliable objective techniques to automatically classify 2-D image patterns from PMS data.
Part of this need stems from the fact that large numbers of images are recorded, even
though relatively small volumes of air are sampled. For 2-D data collected at temperatures
much warmer than 0°C, the problem usually reduces to one in which smooth, circular images
need to be separated from spurious images such as streakers and streamers. For such data
collected in warm portions of clouds, dimensional analysis techniques such as those described
by Cunningham (1978), Heymsfield and Parrish (1978), and Cooper et al. (1978) can
produce adequate results.
When 2-D data are collected in mixed phase portions of clouds, the problem becomes
more difficult because ice particle forms have a wide range of complex shapes which need
76
to be differentiated from shapes produced by liquid particles. Sophisticated techniques and
algorithms for classifying hexagonal crystals have been described by Holroyd (1987), Raham
et al. (1981a, 1981b), and Hunter et al. (1984). However, a technique for distinguishing
between images with smooth and ragged outlines has not received much attention beyond
the circle fitting technique described by Cooper et al. (1980, 1982a, 1982b). In this paper,
we describe the use of polynomial approximations to image perimeter data to objectively
distinguish between circular images having smooth outlines (presumably drops) and those
having ragged ones (presumably graupel).
c. Data
Data used in the development and testing of the technique were collected as part of
the 1989 field program, coined PACE89. The observation platform used in PACE89 was
a light, twin-engine airplane (Beechcraft Baron). The airplane was equipped to measure
standard meteorological state parameters such as temperature, dew point, pressure, and
altitude, as well as measure in-cloud characteristics with a Forward Scattering Spectrometer
Probe (FSSP) and 2DC and 2DP cloud and precipitation imaging probes. Two dimensional
data presented here are for entire cloud penetrations made near cloud top at about the -
10°C level.
Because in-cloud observations were taken early in the cloud's lifetime, glaciation
processes were at their very onset. As is typical for warm-based Midwestern convective
clouds, ice initiates in the presence of supercooled drizzle and rain as frozen drops and
77
graupel, rather than as hexagonal ice forms (Braham, 1964). Consequently, "first" ice shows
up in the 2D records as fairly large (D >200 microns) circular images with rough outlines.
Thus, we are usually not faced with classifying hexagonally shaped images as is the case for
data collected in, for example, winter orographic clouds. Therefore, our problem in
classifying 2D cloud particle images reduces to one in which a distinction must be made
between images which are probably graupel (i.e., circular shapes with rough outlines), and
those which are probably liquid drops (i.e., circular smooth shapes).
In the interactive software procedure that we have developed for classifying 2-D
images, preprocessing (in addition to filtering ice crystals) is performed before the method
of polynomial roughness discrimination is applied. In pre-processing, images such as zero
area, 1 diode, twin diode, etc., as well as spurious images such as splash images and streakers
are filtered (see Heymsfield and Baumgardner 1985, for 2-D probe terminology). Thus, the
subroutine which we use for roughness discrimination has only to consider fairly large (at
least 6 diodes parallel and 6 diodes perpendicular to the time slice), symmetric images
(aspect ratios between 0.35 to 2.0).
d. Theory
One function often chosen to approximate data is a polynomial having the general
form:
78
(1)
where the aj's are coefficients, x the dependent variable, f(x) the approximate value at x, and
m is the order of the polynomial. The process to determine f(x) involves finding some small
order polynomial so that the differences between the values of the function f(x) and the
observed valuesy(x) at x are minimized. The acceptable approximate function may be found
using the least-squares technique in which the variance provides a measure of the goodness
of fit of the polynomial to the data. Therefore, the problem of approximating data using
polynomials centers on finding the smallest order polynomial (m) for which all higher order
polynomials only give marginal improvement in the approximation (i.e., the variance does
not reduce substantially from polynomials of order > m). Thus, a procedure to follow is to
determine the order of the polynomial needed based on the variance, and then to find the
coefficients of the polynomial, if desired.
Following the discussion of the least-squares method in Ralston and Rabinowitz
(1978), a system of equations which will produce approximate values, f(x), for each data
point, x, that can be compared to the observed values, y(x), is:
(2)
where n is the number of data points, and wj is a weighting function, which in our application
is Wj = 1 for i = 1, ...., n. Furthermore, the Pj(xj)'s are defined by the recurrence
79
relationship:
where
and
Hence, the sum of the squares of the differences between the observed (y) and approximate
(f) values at each x data point is given by:
(10)
and the variance is defined as
Therefore, the variance can be found for any polynomial of order m without explicitly
80
(11)
knowing the coefficients (i.e., aj 's, j = 0, , TO). Thus, the order of polynomial to use is
obtained by computing the variance for m = 0,1, 2,...., until the changes in variance (σ2m -
σ2m+1) from one order polynomial to the next higher order is small.
Once the order of polynomial is determined, explicit coefficients, aj's, can be found
from:
(12)
with the bj's given by equation (3), and the qj(x)'s given by:
where the aj's and ßj's are defined in equation (8) and equation (9), respectively, and the
qj(x)'s give algebraic expressions based on the expressions of previous qj(x)'s.
e. Application
In our first approach to using polynomials, we reasoned that the relative smoothness
or roughness of a quasi-circular image may be embodied in the collection of distances
between the image's center and the discrete points which describe the image's perimeter.
Thus, if the image happens to be circle-like, then the distances represent radii (R), and a
81
plot of these as a function of angle (0) around the image should fall on a straight line with
zero slope (the mean ratios) as shown in Figs. 1A and B. On the other hand, if the image
happens to be rough, then the distances are not necessarily equal to the radii, and a plot of
these as a function of angle should show more scatter around the mean than can be
attributed to discretization, as illustrated in Figs. 2A and B. As is implicit in Figs. 1B and
2B, a higher order polynomial should be required to approximate the data in Fig. 2B than
that in Fig. 1B because of irregularity in the outline of the image. Thus, the order of the
polynomial, or more explicitly, the unexplained variance at a fixed polynomial order, could
be used as a roughness discriminator.
For the most part, we found this reasoning to be correct as long as the images were
fairly large; typically having major and minor axes greater than about 15 diodes. The
technique appeared to work for larger images because when they had rough outlines, the
scatter of radii distances was usually large. However, for smaller images (major and minor
axes between 6 and 15 diodes), the technique failed to reliably detect roughness, even
though it was discernible to the naked eye. After visual inspection of many small images,
it appeared that the property of roughness was communicated by irregularity concentrated
along a short portion of the image's perimeter or perhaps by the position of as little as 1
diode. Consequently, we reasoned that this irregularity was possibly overshadowed by the
discretization noise, and therefore sought to modify our approach so that it would include
these smaller images since they can represent a substantial fraction of the total number of
82
Fig. 1. Schematic diagram illustrating the geometry of a smooth, circular image (A)
and corresponding illustrative plot of image geometry in angle (0) and
distance domain.
Fig. 2. Same as Fig. 1 except for a rough, circular image.
images created by legitimate cloud particles.
The technique we eventually found to be more dependable is based on the use of the
image domain border as a reference for exterior distances to the available left, bottom, and
right halves of the image. The geometry for attaining these distances is illustrated in Fig.
3A for an "entire in" image, and one example of its corresponding plot of the data that we
wish to approximate by a polynomial fit is shown in Fig. 3B. While Fig. 3B shows data for
only the bottom half of the schematic image shown in Fig. 3A, in actuality, "entire in" images
produce three data sets for approximation (i.e., one for each available half of the image).
If the image is smooth and circle-like, the collection of distances L1, L2, , Ln, for the left
half of the image approximates a parabola if plotted sequentially, as shown in Fig. 3B for
B1' B2, , Bn. If any of the halves appear to be rough, then the measured distances will not
closely approximate a parabola and a polynomial of order larger than 2 will be required
before the data can be adequately represented. Furthermore, if an irregularity is
concentrated, for example between Li and Ln, then it is not only captured in the left scan
for distances, but is also captured in the bottom scan, which improves the likelihood of
detecting a rough portion on an image's surface.
Therefore, if we fit a polynomial to the distances between the image domain border
and exterior points along any half of the image, as illustrated in Fig. 3B, then a higher order
polynomial should be required for rough images than for smooth ones. We found that
polynomials of order 2 or 4 have very small variance for smooth images, and larger ones for
83
Fig. 3. Schematic to illustrate measured distances along the left, bottom, and right
halves of an image (A), and corresponding data string for the bottom half of
the image to be polynomial approximated (B).
rough images (i.e., the data for any given half is better approximated for a smooth image).
For the smooth and rough images we have encountered, a variance threshold of 0.20 for
polynomials of order 4 has been found to be a fairly reliable discriminator for roughness.
In our scheme, if any curve, either left, bottom, or right, gives a variance greater than the
threshold value for polynomial order 4, then the image is considered to be rough (i.e., it is
suspected to be a graupel). Therefore, by interrogating the image around its exterior, we
obtained a reliable technique for roughness determination, even when images were small,
but still greater than 6 diodes in either direction in order for a polynomial of order 4 to be
used.
f. Handling Partial Images
It is not uncommon to encounter 2D images which partially fall outside the sample
width of the diode array. Although rejection of these images in favor of using only images
which have passed cleanly between the ends of the diode array is acceptable (i.e., "entire
in"), it is very desirable to include images which are either "center-in" or "center-out," to
improve overall sampling statistics.
The problem of handling images which have cut across either end of the diode array
reduces to 1) determining roughness based on fewer available halves than for "entire-in"
images, and 2) determining whether the image is "center-in" or "center-out" in order to
estimate image radius.
In our scheme, we simply used the available halves as we did for an "entire-in" image
84
to detect roughness. As is obvious, edge images typically produce two curves of data (i.e.,
distances between the image perimeter and the image domain border) for polynomial
approximating, one along the bottom half and the other along the side opposite to the side
cut-off. If the image happens to be very large and crosses both ends of the diode array, then
only the bottom half is used. In any event, if the variance for any side is greater than the
threshold, then the image is considered to be rough. Once roughness is determined, an
estimate of image radius can be made after the approximate location of the image's center
is determined. Figures 4A and B show generalized coordinate systems for a left edge and
right edge image, respectively. Each (i,j) coordinate represents a position in the image
domain where a shadowed diode can be indicated. We defined positions (i,j) such that the
i's are parallel to the time slice, and the j's are columns perpendicular to the time slice. In
this system, the upper left corner is coordinate (1,1), and the lower right coordinate is (n,32),
where the n is the number of rows in the image. In such a coordinate system, an image can
be judged to be "center-in" or "center-out" based on the relationship between the major and
minor axes shown as y and x in Fig. 4, respectively. However, the image must be fairly
circular and symmetric, even though it may have a rough outline for the following criteria
to work.
An image can be considered to be "center-in" if the minor axis is greater than one
half the major axis, or:
85
Fig. 4. Illustrations of partial images: left "center-out" (A), and right "center-out" (B).
(15)
For the case of an edge image, the location of its center is coordinate (ic.jc) where:
(16)
regardless of whether the image is left or right edge. Note that Eq. 16 is also true for an
entire image.
For a left edge "center-in" image, jc can be defined as:
(17)
where jmax = x. For the case of a "center-in" right image, jc can be defined as:
(18)
where j m i n = w - x, and w is the width of the diode array.
For the case of a "center-out" image, defined as having x < y/2, the center coordinates
can be found using the relationship:
(19)
where r is an estimate of the image radius based on the visible fraction of the image (see
Beyer, 1988). Having already found ic by Eq. 16, jc for a left edge image is:
86
(20)
and for a right edge image is:
(21)
Once the center has been found, a second way of estimating the image radius is simply to
take the average of the distances between perimeter points and the center. However, the
farther away (ic, jc) is from the actual center, the greater the underestimate of the mean
radius of the image, as can be shown from simple geometry. Nevertheless, when including
these partial images to calculate concentration, the effective sample width can be adjusted
and a procedure for this has been described by Heymsfield and Parrish (1978).
g. Evaluation
Our preliminary evaluation of the technique is based on seventy-eight cloud
penetrations made during PACE89. The first step in the evaluation involved creating the
data set of image types subjectively classified in a rigorous manner. Classifications of image
types (i.e., drops, graupel, possible ice crystals, streamers, splashes, etc.) for each cloud were
determined by inspection; a technician trained to recognize cloud particle and probe artifact
image patterns subsequently recorded this information on paper copy. A meteorologist
reviewed and corrected the hand-labeled image types. The technician then entered the
classifications into the computer using an interactive program that presented the images and
asked for a subjective classification. The software aided the technician by automatically
87
eliminating 31 simple diode configurations such as zero area, 1 diode, twin diode, etc. from
the subjective tagging process. Although this method is extremely time consuming and
laborious, the result is an image classification data set that can serve as a near perfect
standard by which the objective technique can be evaluated.
Figures 5 and 6 show the size distributions for the images subjectively and objectively
determined to be smooth and fairly symmetric (i.e., drops). Data from the 2DC probe is
plotted as solid circles and data from the 2DP probe is plotted as open circles. The vertical
lines through the open and closed circles are error bars based on Poisson statistics as
discussed by Cornford (1967). As can be seen, size intervals with small errors correspond
to intervals having large numbers of images, while large error bars indicate a relatively few
number of images for that size interval. Clearly, as drop size increases and the number in
that drop size decreases, confidence in the calculated concentration greatly deteriorates.
The solid line passing through the data points is the regression line of the size data, and the
dashed lines are regression lines of the upper and lower concentrations, using the upper and
lower values of the error bars. Figures 7 and 8 can be interpreted identically to Figs. 5 and
6, except the latter two figures are subjectively and objectively determined size distributions
for graupel. Figures 5 through 8 were chosen because they typically display the maximum
amount of discrepancy between the two methods of image classification.
We chose the intercept (N0) and slope (λ) of the size distribution, as well as water
content calculated using two different methods as evaluation parameters since these are
88
Fig. 5. A size distribution for smooth symmetric images determined subjectively.
Fig. 6. Same as Fig. 5 except determined using the (objective) polynomial
approximation technique.
Fig. 7. A size distribution for rough symmetric images determined subjectively.
Fig. 8. Same as Fig. 7 except determined using the (objective) polynomial
approximation technique.
likely to be desired variables to be determined from the 2D image records. Liquid water
content based on method I (LWCd) was calculated using the expression:
(22)
where w is the density of water, n is the number of size intervals, and Di, the drop diameter
and N(Dj)∆D the concentration of the ith size interval. Clearly, size intervals that have no
drop concentration make no contribution to the net liquid water content.
Given the slope (λ) and the intercept (N0), if we assume the size distribution
decreases exponentially according to:
(23)
Then, we can substitute Eq. 23 into:
(24)
which can be integrated (by parts twice) and the absolute value taken to yield:
(25)
Equation 25 gives the liquid water content from the area, for example, under the solid line
in Figs. 5 and 6. Clearly, this method will result in a contribution of liquid water from size
intervals that may have zero concentrations. Therefore, we expect method II to typically
give larger values of liquid water content than method I. Solid water contents were similarly
89
calculated by method I and II with the density of water (ρw) replaced by the density of ice
(ρi), where ρx was taken to be 0.9 to represent high density graupel.
Subjectively and objectively determined intercepts for the size distributions are
compared in the scatter diagram of Fig. 9. The solid line is the one-to-one line. The long-
dashed line is the regression line fit to the data, and the short-dashed lines represent the
90% confidence band around the regression line. Similar comparisons are made for the
slope (X) of the size distributions and the liquid water contents determined using method
I and method II in Figs. 10, 11, and 12, respectively. Similar comparisons are provided for
graupel in the scatter diagrams of Figs. 13, 14, 15, and 16. The equation shown in the
upper, left-hand corner of the plots is the general form of the regression line equation.
Tables 1 and 2 list mean values (µ), standard deviations (σ), slope (m), intercept (b), and
correlation coefficient (r) for drops and graupel, respectively, for each of the (least squares)
regression analyses. In Tables 1 and 2, P-values are listed and were determined using the
Wilcoxon Sign-Rank Test. The Wilcoxon Sign-Rank Test is a non-parametric procedure for
determining whether two dependent distributions are shifted from one another. P-values
given in Tables 1 and 2 are for a two-sided test because we had no reason to suspect that
either the subjective or objective technique would lead to greater or lesser than values for
one another.
As can be seen in Table 1, a high degree of agreement exists between all subjectively
and objectively determined parameters. The Wilcoxon test gives no indication that the
90
Table 1. Summary of evaluation parameters for smooth images.
N0 λ LWCd LWCC
µ
SUB
0.33
OBJ
0.30
SUB
30.92
OBJ
28.36
SUB OBJ
0.05 0.06
SUB OBJ
0.06 0.07
σ 0.86 0.77 20.87 16.74 0.12 0.18 0.11 0.15
n 55 55 78 55
m 0.92 0.64 1.45 1.30
b -0.30 8.45 -0.01 -0.01
r 0.91 0.80 0.96 0.94
Zo 0.94 0.88 -0.33 0.10
P-value 0.35 0.38 0.74 0.92
Table 2. Summary of evaluation parameters for rough images.
N0 λ swcd SWCC
SUB OBJ SUB OBJ SUB OBJ SUB OBJ
µ 1.45 0.72 23.57 18.90 0.12 0.20 0.17 0.26
σ 8.37 3.15 11.60 12.07 0.57 0.79 0.54 0.62
n 51 51 78 51
m 0.98 0.78 1.33 1.12
b -0.31 0.60 0.04 0.07
r 0.91 0.75 0.97 0.97
Z0 2.02 3.71 4.77 4.90
P-value 0.04 <0.01 <0.01 <0.01
91
Fig. 9. Scatter diagram comparing size distribution intercepts for smooth images
determined subjectively and objectively.
Fig. 10. Same as Fig. 9 except for size distribution slopes.
Fig. 11. Same as Figure 9 except for water content determined by method I.
Fig. 12. Same as Fig. 11 except water content determined by method II.
Fig. 13. Same as Fig. 9 except for rough images.
Fig. 14. Same as Fig. 10 except for rough images.
Fig. 15. Same as Fig. 11 except for rough images.
Fig. 16. Same as Fig. 12 except for rough images.
objective and subjective distributions are significantly shifted from one another for all of the
parameters. We therefore conclude that, at least for the available data, the objective
process for classifying smooth and relatively symmetric images (i.e., drops) performed
exceptionally well. We believe one reason for this is that when the drops existed, they did
in very large numbers, and therefore occasionally misclassifying one or two drops as graupel
did not significantly affect the size distribution.
On the other hand, as can be seen by examining Table 2 and Figs. 13-16, the
objective technique did not perform as well in identifying graupel. Close examination of
Table 2 shows that the objective technique systematically gave intercepts less than those
which were determined subjectively, and slopes that were flatter than those found by the
subjective method. This arrangement of differences corresponds to a systematic tendency
for the objective technique to under classify small graupel and over classify the occurrence
of larger graupel. This is not unexpected since smooth images tend to give large least-
squares variances by virtue of the potentially larger number of inflections to fit.
Furthermore, because the number of graupel available to construct the size distributions
tend to be smaller than the number available for drops, systematic errors for the very small
and very large graupel has a greater impact on the accuracy of the objective technique in
creating the distributions. Thus, Figs. 7 and 8 show this tendency for the objective method
to falsely find large graupel when none are classified subjectively.
Nonetheless, the fact that a systematic bias exists for graupel identification is not
92
necessarily cause for out-of-hand rejection of this objective technique. We should first keep
in mind that the technique worked extremely well for drops. Additionally, inspection of
Table 2 shows that even though systematic differences exist between the subjective and
objective values for each of the parameters evaluated, the differences are not only
systematic, but are relatively small - no more than a factor of 2 - and certainly within the
measurement uncertainly displayed in the subjective data. If we also consider the
convenience afforded by the objective technique, then differences of these sizes might be
acceptable under certain circumstances, for example, in a field program when immediate
feedback is needed to make operational decisions, or when assessing data quality.
h. Summary and Conclusions
A least-squares polynomial approximation technique is in the process of development
and testing for objectively distinguishing between smooth and rough 2D images. So far we
have found that use of an image's "radii" for detecting roughness or smoothness performed
poorly but warrants further investigation. However, examining each "side" of the image for
roughness worked well. The technique developed and test involved fitting a polynomial
using the method of least squares to the curvature of each half of a large fairly symmetric
image. The variance of the polynomial approximation then was used as the roughness
discriminator. Generally, rough images gave large variances while smooth images gave small
variances. We found that partial or edge images could also be classified as drops or graupel
as easily as "entire-in" images, and that the dimensions of the visible part of the partial image
93
could be used to estimate image radius. A preliminary evaluation of the technique showed
that it performed well for smooth images but displayed a systematic bias to under classify
small, rough images and over classify large ones. This results in a tendency for the objective
technique to over estimate solid water content. Even though this systematic bias exists,
overall differences are small and within the measurement uncertainty indicated by subjective
classification of the image types.
i. References
Beyer, W.H., 1989: Standard Mathematical Tables. CRC Press, 674 pp.
Changnon, S.A., 1986: Illinois Weather Modification Program: PACE. Preprints,
Conference on Weather Modification. Arlington, VA, 315-319.
Cooper, W.A., 1978: Cloud Physics Investigations by the University of Wyoming and
HIPLEX. Report No. AS-119, Department of Atmospheric Sciences, Laramie, WY,
320 pp.
Cooper, W.A., 1980: Estimation of Rainfall using Measurements from Aircraft. Third
W.M.O. Scientific Conference on Weather Modification, Clermont-Ferrand, France,
365-372.
Cooper, W.A., R.P. Lawson, and E.A. Cerni, 1982a: Cloud Physics Investigations by the
University of Wyoming and HIPLEX 1979. Department of Atmospheric Science,
University of Wyoming, Laramie, Wyoming, 301 pp.
Cooper, W.A., R.P. Lawson, T.A Cerni, and A.R. Rodi, 1982b: Cloud Physics
94
Investigations, University of Wyoming. Report No. AS-140. Department of
Atmospheric Science, University of Wyoming, Laramie, WY, 256 pp.
Cornford, S.G., 1967: Sampling errors in measurements of raindrop and cloud droplet
concentrations. Met. Mag., London. 96, 271-282.
Cunningham, R.M., 1978: Analysis of Particle Spectra Data for Optical Array (PMS) 1-D
and 2-D Sensors. Preprints Fourth Symposium on Meteorological Observations,
Amer. Meteor. Soc, Denver, Colorado, 345-350.
Heymsfield, A.J., and J.L. Parrish, 1978: A computational technique for increasing the
effective sampling volume of the PMS 2-dimensional particle size spectrometer. J.
Appl. Meteor., 17, 1566-1572.
Heymsfield, A.J., and D. Baumgardner, 1985: Summary of a workshop on processing 2D
probe data. Bull. Amer. Meteor. Soc. 66, 437-440.
Holroyd, E.W., 1987: Some techniques and uses of 2D-C habit classification software for
snow particles. J. Atmos. and Ocean. Technol.. 4, 498-511.
Hunter, H.E., R.M. Dyer, and M. Glass, 1984: A 2-dimensional hydrometeor machine
classifier derived from observed data. J. Atmos. Ocean. Technol., 1, 28-36.
Raham, M.M., R.G. Jacquot, E.A. Quincy, and R.E. Stewart, 1981a: Two-dimensional
hydrometeor image classification by a statistical pattern recognition algorithm. J.
Appl. Meteor.. 20, 536-546.
Raham, M.M., E.A. Quincy, R.G. Jacquot, and M.J. Magee, 1981b: Feature extraction and
95
selection for pattern recognition of 2-dimensional images. J. Appl. Meteor.. 20, 521-
535.
Ralston, A., and P. Rabinowitz, 1978: A first course in numerical analysis. McGraw-Hill,
556 pp.
96
FORECASTING STUDIES
Work in the past year within the forecasting unit of PRECCIP has been centered in
three areas. The first, and by far the one receiving the greatest emphasis, has been an
analysis of the PACE 1989 forecasting scheme, i.e., ways in which it can be improved and
made more objective. The second area of attention has been the transfer of our
meteorological monitoring system, the many analysis software packages, plus all archived
weather data from a VAX computer system to a SUN system. Finally, the third area of
work was the support we gave to the University of North Dakota with our
forecasting/nowcasting system during their field research, headquartered at CMI during
August 1990. We will report on the activities involved in each of these areas in this section
of the Annual Report.
1. Development of Objective Forecasting Procedures
a. Introduction
In our research, it has been necessary to generate forecasts which are not available
as standard products of the National Weather Service (NWS). The need originated to meet
the objective of providing us with an operational forecast and to enhance our statistical
support by minimizing sources of bias associated with the selection of operational days.
Attempts at objective-decision forecasting of convection are found throughout the
literature with emphasis on the computation of various stability indices from nearby
rawinsonde soundings. The Showalter Index (Showalter, 1953), Lifted Index (Galway, 1956),
97
K-Index (George, 1960), Total Totals Index (Miller, 1967), and Severe Weather Threat
(SWEAT) Index (Bidner, 1970 and Miller, et al., 1971) are some of the more popular static
stability indices used to forecast convection. Peppier (1988) and Peppier and Lamb (1989:
Appendix) provide an excellent review of these and many other indices and related
thermodynamic parameters. Schaefer (1986) reviews the history of severe weather
forecasting itself.
Forecasts tailored to specific needs have been reported in recent papers. Wilde et
al. (1985) used lifted condensation level measurements to diagnose cumulus onset time and
cloud coverage. Colquhoun (1987) devised a "decision tree" method to forecast the
occurrence of severe weather, using a variety of stability and thermodynamic parameters.
Schultz (1989) related 7 stability indices to convective weather events in Colorado while
Peppier and Lamb (1989) equated 40 stability indices and thermodynamic parameters to
growing season rainfall observations across central North America.
In PACE weather modification research, our forecast interests include not only
predictions on the potential of convective precipitation occurrence, but also expectations on
the maximum cloud height. Forecasts on the internal suitability of clouds for seeding would
be highly beneficial. Previous projects have used various meteorological calculations to
generate an objective decision for daily field operations. In general, these have been based
on particular meteorological conditions that are considered necessary for creating the types
of weather events that would be of interest to their research. For example, in Project
98
Whitetop (Braham, 1966), a combination of precipitable water at nearby upper air sites and
the low-level wind direction observed in the target area were used to define operational
days. The Florida Area Cumulus Experiment (Woodley and Sax, 1976) selected potential
operational days from a blend of the predicted enhancement of cloud growth due to seeding
from a 1-D cloud model and precipitation occurrence prior to the operational period.
During the 1986 PACE field study, days selected for operations relied almost entirely
on subjective forecasts, based largely on analyses of the regional conditions: low-level
moisture content, instability, satellite cloud imagery, the presence of airmass boundaries or
short waves useful as triggering mechanisms, etc. A 1-D cloud model was available to the
project to guide researchers on the potential for deep convection and for increased cloud
height by dynamic seeding (ie. its in situ suitability). However, objective use of the model
to forecast these parameters was limited by the need to know the cloud diameter prior to
the development of convection. Thus, in practice, the determination of in-cloud suitability
for seeding was relegated (as in the earlier weather modification experiments) to in situ
measurements via an aircraft well-fitted with microphysical instrumentation.
Development of our objective forecasting procedures began from an evaluation of
data collected during the PACE 1986 experiment. Scott and Huff (1987) reported on the
performance of 5 stability indices in relation to precipitation occurrence used during the
project. Comparisons were made between thermodynamic properties observed in the 0600
LST soundings at two NWS upper air rawinsonde stations in Illinois and the sillgle daily
99
maximum radar echo top observed within the target during the subsequent 1130-1930 time
period and radar summaries from the hourly observations at the local NWS radar sites
surrounding the area.
The positions of rawinsonde sites at Peoria (PIA) and Salem (SLO) were fortuitous
to PACE (Fig. 1); their close location to the target furnished a regional definition of air
masses over central Illinois. This is opposed to observations derived from just one site alone
which may not always be representative of the conditions in areas that are relatively nearby
the rawinsonde site.
Scott and Huff (1987) took advantage of this site closeness and found greater index
performance when index values from both sites exceeded their respective convection
thresholds concurrently. Using the various indices, days forecasted to be convective (radar
echo tops > 22,000 ft or 6.7 km) verified 67%-75% of the time. Those predicted to have
no echoes were accurate at an 80%-88% rate.
Mather et al. (1986), reported on convective research conducted over the eastern
Transvaal of South Africa. Aircraft and rawinsonde data were used to develop an empirical
thermodynamic relationship between the cloud base temperature (CBT) and the potential
buoyancy (PB) of the atmosphere. The relationship appeared to discriminate between
clouds that have an active coalescence process and those that do not. From a physical
standpoint, the CBT gives an implicit estimate of the distance over which coalescence can
operate between cloud base and the seeding level at -10 °C while the PB is an implicit
100
Fig. 1. Map of PACE 1989 target with station locations.
measure of the velocity that an air parcel may possess between these two points. Therefore,
when viewed in combination, the two parameters give an idea of the amount of time that
is available for coalescence to operate within a cloud. For example, a cloud with a relatively
warm CBT but large PB may have the same time for coalescence that has a cloud with a
cold CBT and weak PB.
As a result of these studies and the 1986 field project, we recognized that our
forecasting effort and statistical support of the research would be enhanced by the
development of an objective procedure to anticipate meteorological events important to
weather modification, namely, the occurrence of clouds that would likely develop rain by
coalescence processes and have the potential for vertical growth beyond the -10°C level
where seeding would take place. This section of the Annual Report presents the
development of such a technique which was then tested in the PACE field experiment of
1989. Specific objectives of the procedure were to: (1) forecast the occurrence of convection
(categorizing operational days), and (2) provide an estimate of the depth of the expected
convection.
b. Diagnostic development
Explorations into the development of customized forecasts for convection in Illinois
required the creation of summertime thermodynamic and kinematic climatologies
representative of environmental weather conditions within the PACE target where cloud
physics measurements were to be made during field operations. As an initial analysis, we
101
expanded the data base of the PACE 1986 work to include all summer months of 1986-87.
Specifically, we compared the 0600 LST rawinsonde data from PIA and SLO to radar echo
data observed over central Illinois from the nearby WSR-57 radar sites at Marseilles (IL),
St. Louis (MO) and Evansville (IN).
The radar data were stratified by the single highest radar echo top reported in the
hourly radar summaries between 1130-1930 each day within approximately 190 km of CMI.
If the same cell were reported by more than one station, cell heights were taken from the
radar closest to the cell. On hours for which no echo heights within the target were
transmitted by any station but echoes were known to exist, an estimate was made from echo
top data reported on antecedent and subsequent hours, taking into consideration the
progression and continuity of the individual storm systems. The daily maximum height
values were summed and categorized into six groups: no echoes, echoes with tops < 20,000
ft (6.1 km), 20,000 - 30,000 ft (6.1 - 9.1 km), 30,000 - 40,000 ft (9.1 - 12.2 km), 40,000 -
50,000 ft (12.2 - 15.2 km), and tops > 50,000 ft.
A variety of stability and other standard meteorological parameters were calculated
from the rawinsonde data, providing comparisons with the radar data from many different
measures (i.e., low-level moisture content, parcel instability, cloud development potential,
buoyancy, etc.). The greatest agreement occurred with two procedures; one procedure
involved the use of the Modified K-Index and the other procedure was based on the
combined use of estimated cloud base temperatures and buoyancy.
102
1.) Procedure I: Use of Modified K
As stated earlier, the success of stability indices as indicators of convective
precipitation (defined as maximum tops > 20,000 ft) over east central Illinois in 1986 was
enhanced when data from PIA and SLO were combined. Scott and Huff (1987) found that
of several stability indices tested, the Modified K-index (Charba, 1977) performed the best
in a predictive mode during the 1986 PACE field season. The Modified-K (MK) we used
is the sum of the mean temperature and mean dew point from the surface to 850 mb, plus
the 500 mb temperature, minus the 700 mb dew point depression. Peppier and Lamb
(1989) also showed high performance for MK in their study. Results from the 1986-87 data
are presented in Table 1.
On the left-hand side of the table are binary forecasts on the occurrence of convective
rainfall in the target, based on 3 threshold values of MK calculated from the 0600 LST
rawinsonde data at PIA and SLO for each day in the 1986-87 data. The predictor was
calculated first for each site individually, and then again using an ad hoc procedure described
below. In the center of the table, forecasts are sorted into the appropriate categories of the
maximum echo tops. For example, when MK at PIA exceeded a threshold of 20, there were
23 days on which no echoes were observed in the target area and 38 on which at least one
echo top exceeded 50,000 ft.
Forecast accuracies ("acc") were judged in two ways. The first (labeled "fest") is
simply the success of the forecast to predict convection. Using the same threshold of 20 as
103
Table 1. Performance of the modified K-index (C) calculated from morning soundings (0600 LST) at PIA and SLO during June-August 1986-87 in relation to the single maximum cloud top observed during the subsequent afternoon (1130- 1930) from the local NWS radar sites surrounding central Illinois.
Convective Threshold
echo fcst
Observed Maximum Radar Echo Height (x 1000 feet)
ace (%) echo fcst
0 <20 20s 30s 40s ≥50 fcst obsd
PIA MK≥20 yes 23 8 16 17 24 38 75 87
only MK<20 no 29 6 5 3 1 5 71 53
SLO MK.>20 yes 26 11 16 16 23 42 72 91
only MK<20 no 26 3 5 3 1 1 74 52
PIA MK.>25 yes 13 5 13 15 20 37 83 78
only MK<25 no 39 9 8 5 5 6 67 73
SLO MK.>25 yes 21 8 15 13 21 38 77 81
only MK<25 no 31 6 6 6 3 5 65 56
PIA MK >30 yes 7 4 12 13 16 32 87 67
only MK<30 no 45 10 9 7 9 11 60 83
SLO MK >.30 yes 16 6 12 13 20 37 79 77
only MK<30 no 36 8 9 6 4 6 64 67
both * yes 12 8 14 16 22 40 82 87
sites ** no 40 6 7 3 2 2 77 70
* MK ≥30 at one site and MK ≥ 10 at the other ** MK <30 at both sites or MK <10 at one site
104
an example, 75% of the days that were forecasted to be convective using PIA data, verified
as such; likewise, a 71% success rate resulted from forecasts of no echoes (MK <20). With
higher thresholds, forecasts of convection became more successful while predictions of no
echoes failed more often. The second accuracy test documented in Table 1 (labeled "obsd")
asks what percentage of all observed convective (or non-convective) days were forecast as
such. Again using a threshold of 20, 87% of all convective days were forecast to be
convective while 53% of non-convective days were accurately predicted to have no
convection. Raising the threshold here has the opposite effect: reducing "misses" on days
with no echoes, while increasing them on convective days. If an operational experiment
were to use just this method to objectively forecast days of convection, a relatively low
threshold would miss few convective opportunities but an excessive number of days on which
no echoes occurred would be forecast as convective days, unnecessarily holding field
personnel on station. Likewise, a higher MK threshold value would limit errors on days
without convection but at the expense of missing more actual convection opportunities.
For any meteorological field program in which convection is of interest, accurate
forecasts of "no echo" days are of equal importance to forecasts of days with echoes to allow
for adequate maintenance of equipment and rest for field personnel between operations.
Therefore, a benefit to the research would be to devise a procedure that maximized the
percentages of all verifications. From empirical observations, the devised scheme yielded
a forecast in which convection would be expected if the MK from at least on site attained
105
a rather high index value (>30), while the value at the other site could be quite low (≥10).
That is, a high potential for convective precipitation at either site tended to result in
convection in the target unless countered by a very low potential (mk < 10) at the other site.
Results of the combined-site data (indicated as "both sites" in Table 1) suggest that
this procedure maximized the accuracies in forecast verifications on both convective and
non-convective days (82% and 77%, respectively). In addition, on days when convection was
observed, 87% were forecast to be convective; on days with unsuitable radar tops or no
echoes, 70% were forecast to have no echoes. This combined-site procedure considerably
enhances the predictability of MK over that shown at individual sites near the target.
2.) Procedure II. Use of TCCL and PB
The second relationship in the 1986-87 analysis, which found agreement between a
combination of cloud base temperatures and potential buoyancy and the daily maximum
radar echo height, relates back to the work by Mather et al. (1986) in South Africa. Then-
analysis, using quite similar thermodynamic parameters, may have application in Illinois
because of the similarity in the distribution of CBT (Johnson, 1982). In the research we
conducted this past year, we calculated the temperature of the convective condensation level
(TCCL) in place of the CBT. The CCL is defined as the intersection of the sounding curve
and the mean saturation mixing ratio in the lowest 100 mb of the atmosphere. Potential
buoyancy is defined here as the difference between the 500mb ambient temperature and the
temperature obtained by a moist adiabatic ascent from the CCL.
106
_
The TCCL and PB were selected as coordinates on a scatter diagram, with each point
on the graph representing the maximum echo height for that day within specific height
categories (Fig. 2). Points plotted are for all summer days in 1986-87 for which sufficient
rawinsonde data (both mandatory and significant levels) existed at PIA and SLO, and for
which local radar data existed.
The graph reveals a propensity for days with the tallest echoes (categories 4 and 5)
to be located in a small area in the upper portion of the scatter diagram. In general, this
area is bounded by a PB of 0-8°C and a TCCL of 10-20°C. Days with maximum cloud tops
between 20,000 - 40,000 ft (categories 2 and 3) seem to be well-scattered throughout the
central and upper portions of the graph. Days on which very small cells or no echoes
(categories 0 and 1) formed within the target area occurred with all measures of PB and
T C C L .
Figure 2 was divided into 3 subjectively-chosen regions representing operational
decisions for the 1989 experiment, "Go", "Stand By", and "No-Go", (Fig. 3). The area for Go
days was chosen to encompass most occurrences of tall radar echo tops (4 and 5) during the
summers of 1986-87. These were days on which we expected to find echoes that showed
vigorous growth. Maximum cloud heights on Stand By days were mostly small or medium
in size (2 and 3), indicating limited growth potential due to marginal thermodynamics. No-
Go days were most often days on which no echoes were reported.
107
Fig. 2. Scatter diagram of TCCL and PB values calculated daily during June-August 1986-87 from the 0600 LST sounding at PIA and SLO. Points are coded by the height of the daily maximum radar echo top observed in the target during the 1130 - 1930 operational period from the local NWS radar sites. Specifically, these are: 0 = no echoes, 1 = echo tops < 20,000 ft (6.1 km), 2 = tops 20,000 - 30,000 ft (6.1 - 9.1 km), 3 = tops 30,000 - 40,000 ft (9.1 -12.2 km), 4 = tops 40,000 - 50,000 ft (12.2 - 15.2 km), and 5 = tops ≥ 50,000 ft.
Fig. 3. Superimposed on Fig. 1 are the subjectively-chosen divisions used during PACE 1989 for the objective operational decision forecast based on daily TCCL and PB values.
The distribution of maximum daily echo tops in Fig. 2 was analyzed further as shown
in Fig. 4-7. These graphs represent relative frequencies in the target for: no echoes,
maximum echo tops from 0 - 22,000 ft (6.7 km), tops 22,000 - 42,000 ft (6.7 -12.8 km), and
echo tops greater than 42,000 ft, respectively. The graphs were obtained by the following
procedure:
1) the scatter diagram shown in Fig. 2 was overlaid with a Cartesian grid with grid
points located at every intersection of TCCL and PB whole numbers,
2) calculated values of PB and TCCL were rounded to the nearest whole number,
3) the frequency of occurrences of each category was computed at each grid point,
4) the gridded data were smoothed using a 9-point centered scheme, and
5) the fields were contoured.
As can be seen by comparison of Fig. 4 to Fig. 5-7, a very large portion of the TCCL -
PB domain is covered by the occurrence of no echoes. This reflects the fact that in the
Midwest, favorable thermodynamic conditions are often only a necessary condition for
convection, and frequently a kinematic triggering mechanism is also needed. Nevertheless,
when viewed in combination, Fig. 4-7 show a clear tendency for larger clouds (i.e., taller
echo tops) to occur with larger PB and lower cloud bases (i.e., warmer TCCL).
c. Physical basis for Procedure II
When viewed in combination, PB and TCCL reflect physical processes related to the
vigor of a cloud and its ability to produce rain. For example, in summertime over the
108
Fig. 4. Objective analysis of the frequency of occurrence of "no echoes" in the target based on data in Fig. 1. Contours are labeled in percent and were used in PACE 1989 for the probability of occurrence of the particular event, given daily TCCL and PB values.
Fig. 5. Same as Fig. 3, except for small maximum daily radar echo tops (< 22,000 ft).
Fig. 6. Same as Fig. 3, except for medium maximum daily radar echo tops (> 22,000 < 42,000 ft).
Fig. 7. Same as Fig. 3, except for tall maximum daily radar echo tops (≥ 42,000 ft).
Midwest, a relatively low height for the CCL (i.e., warm temperature) is indicative of
relatively large amounts of moisture in the boundary layer. Warmer cloud base
temperatures imply the existence of a greater depth over which coalescence processes can
operate and, in addition, a greater likelihood that echoes will form before cloud top reaches
0°C, assuming that the updraft is strong enough to carry the condensate load aloft.
Potential buoyancy is a measure of the temperature difference between cloud and
environment, unadjusted for the lessening effects of entrainment and condensate loading.
This constitutes a measure of the potential strength of the updraft. In combination, warm
TCCL and large PB should produce big clouds with an active coalescence process, as was
originally indicated by Mather et al. (1986).
These analyses were made with the assumption that the air masses observed by the
morning rawinsonde observations were representative of the afternoon air masses over the
area of investigation. An air mass located on the western border of the target at 0600 LST
(rawinsonde valid time), moving easterly at, say 10 m s-1, would pass completely through the
target in about 6 hours. With rawinsonde sites located just outside the target borders to the
northwest and south southwest of CMI under which most weather situations would be in the
upwind direction, assuming the morning upper air data to be representative of the afternoon
ambient conditions seems quite reasonable.
Experience has shown this to be an acceptable assumption in Illinois due to several
factors. Fairly uniform terrain exists over several hundred square kilometers, i.e., there are
109
no large topographical features or contrasting surface types to generate organized diurnal
forcing (e.g., a dry line, low-level jet, etc.) as is found in the lee of the Rockies and into the
Great Plains. Near-surface moisture fields are frequently homogeneous, which minimizes
the effect of moisture advection to change conditions represented in the morning soundings.
Typically, only the passage of a front or other large scale synoptic feature would remove the
representativeness of the 0600 sounding to the environmental airmass surrounding the
afternoon weather events.
d. Evaluation of Procedure I and II
1.) Performance of Modified K
As stated earlier, the proximity of the NWS rawinsonde sites, PIA and SLO, to the
operational target area was a large benefit to forecasting for the 1986 PACE field program.
Unfortunately for the 1989 experiment, the sounding site at Salem was moved about 160 km
south, to Paducah, Kentucky (PAH) late in the fall of 1988. This resulted in a less
representative measure of thermodynamic conditions over the target, a point illustrated in
Table 2.
Considering Peoria alone, accuracy rates of forecasts in 1989 were comparable to the
1986-87 data in Table 1 at all thresholds of MK tested ("fcst" under "ace"). However, fewer
convective days were accurately predicted to be convective ("obsd" under "ace"), down about
8%, while an average 10% increase was seen in the frequency of days without echoes being
correctly anticipated.
110
Table 2. Same as Table 1, except with data at PIA and PAH for PACE 1989 (8 May -7 August 1989).
Convective Threshold
Echo fcst
Observed Maximum Radar Echo Height (x 1000 ft)
ace (%) Echo fcst
0 <20 20s 30s 40s ≥50 fcst obsd
PIA MK ≥20 yes 5 6 8 13 9 8 78 79
only MK<20 no 16 5 4 3 1 2 68 66
PAH MK>20 yes 12 8 8 12 8 7 63 70
only MK<20 no 8 3 5 4 3 3 42 35
PIA MK ≥25 yes 3 4 6 12 7 8 83 69
only MK<25 no 18 7 6 4 3 2 63 78
PAH MK≥25 yes 10 6 7 10 7 7 66 62
only MK<25 no 10 5 6 6 4 3 44 48
PIA MK ≥30 yes 0 2 5 11 6 7 94 60
only MK<30 no 21 9 7 5 4 3 61 94
PAH MK ≥30 yes 9 5 5 10 6 5 65 52
only MK<30 no 11 6 8 6 5 5 41 55
both * yes 5 4 6 10 8 8 78 67
sites ** no 16 6 6 6 2 2 58 71
* MK ≥30 at one site and MK >10 at the other ** MK < 30 at both sites or MK < 10 at one site
111
_
The 1989 rawinsonde data from PAH were far less associated with convection within
the target than SLO has shown in Table 1, and were even less related with forecasts of no
echoes. Synoptically, a high frequency of fronts were observed along the Ohio River Valley
south of central Illinois during the 1989 summer (data not shown). This generated
conditions at PAH that would be favorable for forecasts of convection across western
Kentucky and extreme southern portions of Illinois and Indiana, but not necessarily
representative of weather in the PACE target. As a result, accuracy rates of forecasts for
both convective and non-convective days that were in excess of 70% at SLO with a mk
threshold of 20, decreased by 10-30%, respectively.
The move likewise adversely affected the combined site indices test in Table 2, where
accuracy rates of the two sites together (PIA and PAH) were no better than PIA alone.
This result underscores the importance of having rawinsonde data that is representatively
located to a research area. Furthermore, even though the combined site findings in Table
1 could not be confirmed, Table 1 data continues to suggest with quantitative evidence that
a substantial enhancement of forecast accuracy can be achieved with the use of multiple
rawinsonde sites during research on convection, even with relatively small target areas.
Further evaluation of the objective forecasting procedures in this report however, will involve
discussion of PIA only.
2.) Performance of TCCL and PB
Go, No-Go and Stand By Forecasts. Daily subjective operational decisions were
112
made based on all current weather information. The criterion for a "Go" day in PACE 1989
was the expectation of radar echo tops in excess of 30,000 ft (9.1 km) to be located within
the target during the operational period (1130 -1930 LST). A "No-Go" forecast was issued
on days that it was most likely that echoes of 30,000 ft would not occur. Days forecast as
"Stand By" were those on which early morning conditions gave marginal support for large
echo formation, thus the monitoring of events throughout the operational period was
warranted.
Daily PB and TCCL measurements from PIA in 1989 were used to objectively-select
an operational decision from the decision categories seen in Fig. 3; results of the experiment
are shown in Table 3. Days within each objective decision (Go, Stand By, and No-Go) were
categorized by the tallest radar echo height observed during the operational period for each
day. Verification rates are shown as the percentage of all days within the two height
categories given compared to all days with a similar objective decision.
Using this procedure, there were 26 days on which a "Go" decision was made; 21 of
these were days with echoes tops in the target during the operational period that exceeded
30,000 ft (9.1 km), an 81% verification. There were only 3 days on which a "Go" objective
decision was selected and subsequently no echoes occurred. Similarly, on days with a "No-
Go" decision, 61% reported either echoes with maximum tops less than 20,000 ft (6.1 km)
or no echoes at all. Only one day in this category had an echo greater than 30,000 ft. Days
with "Stand By" decisions were fairly evenly split in verification; 39% of these days had
113
Table 3. Performance of the objective decision experiment for the target during PACE 1989 based on divisions presented in Fig. 3, using morning rawinsonde data from PIA.
Objective Decision
Observed Maximum Radar Echo Height (x 1000 ft)
Verification (%) Objective Decision
0 <20 20s 30s 40s ≥50 ≥30 kft <20 kft
Go 3 0 2 6 5 10 81 12
Standby 12 6 4 9 5 0 39 50
No go 6 5 6 1 0 0 6 61
suitable convection, half did not.
We conclude that the objective technique did a good job in selecting the operational
decision for the day. Furthermore, the fact that the tall storms of 1989 occurred on days
with PB and TCCL values similar to those of 1986-87 suggest that our sample size supports
usefully stable statistics.
Maximum Height Forecast. The objective probable maximum height experiment was
based on interpreting the observed relative frequency information of Fig. 4-7 as forecast
probabilities of occurrence during PACE 1989. For each day of the experiment,
probabilities of occurrence were extracted from each chart, with the highest of the 4
probabilities designated as the forecast for the day. The forecast was then compared to the
observed maximum radar echo heights for no echoes, short echoes (tops < 22,000 ft, 6.7
km), medium echoes (tops 22,000 - 42,000 ft, 6.7 - 12.8 km) and tall echoes (tops > 42,000
ft, 12.8 km) with the daily verifications summed for the experiment.
114
Results are presented in Table 4. Out of 23 days on which tall echoes received the
highest probability of occurrence of the 4 forecasts, tall echoes were observed on 12, a 52%
verification. Similarly, medium sized echoes were observed on 9 of 17 days (53%) on which
the medium category possessed the highest probability of occurrence. Although a very small
sample, 3 of 5 days on which small echoes had the highest probability of occurrence were
also days on which only small echoes occurred.
Table 4. Performance of the probability forecast experiment. Numbers represent the observations of maximum radar echo heights in the target within each forecasted category when that category reported the highest daily probability of occurrence from the morning TCCL/PB values at PIA.
Forecast Heights
Observed Heights Forecast Heights None Small Medium Tall
None 12 6 11 7
Small 2 3 0 0
Medium 4 2 9 2
Tall 4 3 4 12
The 36 days on which the "no echoes" category had the highest probability of
occurrence did not perform as well; only 12 (33%) were days of no echoes. The high count
of observed convection here was due to the large number of no echo occurrences in 1986-87
(Fig. 3). The TCCL/PB values had shown good potential convection but either the synoptic
conditions had changed or a mesoscale inhibitor (mid-level inversion, drying, etc.) existed to
115
prohibit echo development. In 1989, similar TCCL/PB values were associated predominantly
with moderate convection. Perhaps the addition of more years of data into the analysis will
improve the forecastability within this category.
Overall, considering that there were 4 categories of prediction, this technique
performed reasonably well, picking the correct category 44% of the time. It was most useful
on days that convection was expected, providing accurate indications of the height of
maximum echo top just over half the time.
e. Conclusions
Objective techniques for forecasting convection in Dlinois were developed using data
for 1986-87 and tested as part of the FY90 PACE research. Forecasts of the occurrence of
convective storm activity using the Modified K-Index from two close rawinsonde sites, which
were quite successful in predicting convection in earlier summers, were adversely affected
by a relocation of one of the sites. Results in 1989 from PIA, the closest site to the target,
paralleled its performance in the 1986-87 results. The combined-site format, used in
enhance the MK forecast in 1986-87 by taking advantage of two rawinsonde stations located
close to the target, failed with the substitution of PAH for SLO. The unrepresentativeness
of PAH was expected due to its distance from CMI. However, the combined-site technique
may prove useful to future research when planning the placement of supplemental
rawinsonde sites around even small target areas. The location of sites on the edges of a
target, both in upwind quadrants of predominant air flow and storm motion, may enhance
116
standard stability forecasts.
Objective forecasts of maximum radar echo tops, based on a combination of potential
buoyancy and the temperature of the convective condensation level, were successful in
predicting maximum tops in excess of 30,000 ft (9.1 km) 81% of the time, and tops < 20,000
ft (6.1 km), or no echoes 61% of the time. These forecasts were the basis of "Go"/"No-Go"
decisions for the experiment. An objective probable maximum height forecast succeeded
in predicting both tall and medium radar echo tops (out of 4 height categories) at rates for
each in excess of 50%.
The objective techniques presented here were successful in predicting the occurrence
of convection in the context of the PACE 1989 field experiment. Application of these
techniques to other geographical areas, however, will require an analogous assessment of the
thermodynamics and radar echo top relationships in each region.
f. References
Bidner, A., 1970: The Air Force Global Weather Central severe weather threat (SWEAT)
index - A preliminary report. Air Weather Service Aerospace Science Review, AWS
RP 105-2, No. 70-3, 2-5.
Braham, R.R., 1966: Project Whitetop. Dept. of Geophysical Sciences, Univ. of Chicago,
156 pp.
Charba, J.P., 1977: Operational system for predicting thunderstorms two to six hours in
advance. NOAA Tech. Memo. NWS TDL-64, 24 pp.
117
Colquhoun, J.R., 1987: A decision tree model of forecasting thunderstorms, severe
thunderstorms, and tornadoes. Wea. Forecasting. 2, 337-345.
Galway, J.G., 1956: The lifted index as a predictor of latent instability. Bull. Amer. Meteor.
Soc. 37, 528-529.
George, J.J., 1960: Weather Forecasting for Aeronautics. New York, Academic Press, 407-
415.
Johnson, D.B., 1982: Geographical differences in cloud-base temperature. Proceedings
Conf. on Cloud Physics, Chicago, II, 187-189.
Mather, G.K., B.J. Morrison, and G.M. Morgan, Jr., 1986: A preliminary assessment of the
importance of coalescence in convective clouds of the Eastern Transvaal. J. Climate
Appl. Meteor.. 25, 1780-1784.
Miller, R.C., 1967: Notes on analysis and severe storm forecasting procedures of the
Military Weather Warning Center. Tech. Rep. 200 (Rev), AWS, USAF.
[Headquarters AWS, Scott AFB, IL 62225.]
Miller, R.C., A. Bidner, and R.A. Maddox, 1971: The use of computer products in severe
weather forecasting (the SWEAT index). Preprints, Seventh Conf. Severe Local
Storms, Amer. Meteor. Soc, Boston, 1-6.
Peppier, R.A., 1988: A review of static stability indices and related thermodynamic
parameters. Illinois State Water Survey Misc. Pub. 104, Champaign, 87 pp.
Peppier, R.A. and P.J. Lamb, 1989: Tropospheric static stability and central North
118
American growing season rainfall. Mon. Wea. Rev.. 117, 1156-1180.
Schaefer, J.T., 1986: Severe thunderstorm forecasting: A historical perspective. Wea.
Forecasting. 1, 164-189.
Schultz, P., 1989: Relationships of several stability indices to convective weather events in
northeast Colorado. Wea. Forecasting. 4, 73-80.
Scott, R.W. and F.A. Huff, 1987: PACE 1986 forecasting program-design, operations and
assessment. Preprints of the 11th Conference on Weather Modification, Amer.
Meteor. Soc, Boston, 102-105, (ISWS Reprint 778).
Showalter, A.K., 1953: A stability index for thunderstorm forecasting. Bull. Amer. Meteor.
Soc. 34, 250-252.
Wilde, N.P., R.B. Stull, and E.W. Eloranta, 1985: The LCL zone and cumulus onset. J.
Clim. Appl. Meteor., 24, 640-657.
Woodley, W.L. and R.I. Sax, 1976: The Florida Area Cumulus Experiment: Rationale,
Design, Procedures, Results, and Future Course. NOAA Tech. Rep. ERL 354-
WMPO 6, Boulder, CO, 204 pp.
119
2. Operational Support for the University of North Dakota
PRECCIP provided assistance to the University of North Dakota (UND) during the
month of August 1990 in support of a project headquartered at CMI. The objectives of this
study was to track the movement of atmospheric trace gases within clouds over the Midwest
and monitor the way in which these gases were processed by clouds during their lifetime.
Our forecasting/nowcasting system was installed at CMI to assist UND in their daily
operations. We trained them in the use of the system and were available for assistance
when it failed or needed maintenance. However, we were not involved in the daily analysis
or archival of the data received.
The system we installed was essentially the same used by the PreCCIP staff in the
1986 and 1989 PACE field experiments. This included: (1) telephone communications
between the WSRC to CMI to accommodate printout of standard meteorological charts
(DIFAX) and access to a wide variety of digitized data (eg. surface, upper air, radar, etc.)
from NOAA; (2) a flatbed plotter used to draw various in-house products useful in daily
weather analysis such as thermodynamic diagrams, surface streamline and divergence charts
and upper air analyses; (3) a decwriter to record severe weather statements issued from
surrounding NWS sites, used primarily for aircraft safety; and (4) link to the University of
Wisconsin's MCIDAS system for transfer of satellite imagery data used to monitor cloud
development over the local area, and displayed on a graphics monitor we supplied.
120
3. Transfer of Forecasting Software and Data
All PRECCIP forecasting/nowcasting software programs and the archived
meteorological data from 1986 to the present were transferred from the former SWS
computer, a VAX 11-750, to our new system, a SUN-3. This was a major task involving the
transfer of at least 25 magnetic tapes of weather data and re-archived into a compatible
format with the new operating system. In the process, the 6250 bpi tapes were re-written
on much higher storing (2 GB) 8-mm cassette tapes. Some of the software programs used
to analyze data received from the forecasting/nowcasting system, including several library
routines, had been written with a large amount of VAX specific fortran; this conversion
required much re-coding for the programs to operate on the SUN.
Monitoring of the weather data from Zephyr was conducted on both machines for
several weeks to serve as a check of the new system.
121
HOT RADAR SYSTEM
A watertight structure to house a 3-phase, 60-cycle, 208 volt, 30-kilowatt rotary
regulator has been fabricated and attached to the rear underside of the radar van. This unit
and its associated controls isolates and regulates the voltage amplitude of the power supplied
to operate the radar equipment. The regulator and required cables have been installed in
the radar van and will be tested after the completion of a new airport power distribution
system.
Major improvements of the radar system are the installation of a Lassen SP-20, Sky
320, Microvax II, Adage 3000, Kennedy 9401, and software required for operation and data
storage.
The Lassen SP-20 is a flexible modular array processor capable of rapid computation
and fully programmable using assembler language. It has an input/output module that will
support 1 to 20 computational modules. The control/input module can accept rates
approaching 80 M bytes/second and each computational module has an output of similar
speed. The micro code for each module can run an independent program, and with inter
module communication, the system provides efficient sequential operations. The
programming feature substitutes software changes for hardwire modifications to alter
algorithms.
The Sky 320, a 16-bit integer signal processor, merges housekeeping data with the
data stream and can be used for data compression algorithms.
122
The MicroVax II, using operator directions, directs the appropriate control of the
radar system, the data displays, and manages the efficient recording of data.
The Adage 3000, a radar display computer, is a high speed micro-programmable
computer which is utilized to read and execute coordinate conversions of the data stream
as it is being archived for tape storage. The display buffer, a 1024 x 1024 array of 24 bit
pixels, utilizes crossbar switching and allows the display data to be subdivided for several
types of displays. Software has been implemented to permit operator interaction with the
display using a 3-button mouse.
The Kennedy 9400 tape unit is used for block recording at 6250 bits-per-inch density
with data rates approaching 200 kbytes per second. The tape machine is under computer
control and records data at the end of each processed group record.
123
WEATHER EFFECTS ON CROP YIELDS
1. Description of 1990 Experiment
The corn and soybean experiments in the rain shelters were repeated as part of our
FY90 research. The design was similar to the one in 1989. One shelter followed the
experimental design of 1987-1989, including both corn and soybeans. Two shelters were
used to repeat the 1989 corn planting date and plant population experiment. Establishment
of the i990 experiments was delayed due to a windy wet spring. The high winds prohibited
our covering the shelters until very late in May. Therefore, the first planting date of corn
was on 28 May and the second planting date 10 June. The first rain treatment in the
shelters was on 13 June.
2. Results from Previous PACE Experiments on Crop Yields
In past PACE experimentation, the response of corn yields to a 25% increase in
rainfall during typical dry, average, and wet summers in Illinois was evaluated over a 3-year
period in plots that could be covered by rain shelters. All natural rainfall from June through
August was excluded from the plots and water was applied at specified times to simulate the
typical dry, average, and wet rainfall summer patterns in Illinois. In contrast, most previous
studies on the effects of rainfall on corn (Zea mays L.) yields have been conducted using a
range of water applications from extremely dry to very well watered, but have not excluded
natural rainfall. Therefore, most previous studies have not controlled water applications as
strictly as the PreCCIP/PACE experiments.
124
A M017xB73 corn hybrid was planted on 28 May 1987, 12 May 1988, and 12 May
and 26 May 1989. Plant densities were 81,400 plants ha-1 in 1987 and 64,700 plants ha-1 in
1988; plots were established using both plant densities in 1989. Individual plant size was
found to be related to the plant density and rainfall, with the largest plants in the low density
high rainfall treatments. Final grain yield was related to individual plant size, rainfall, and
average temperature during the period from planting to tassel initiation. The average
temperature during this period explained 72% of the yield variation between the various
planting dates. The yield variation between the planting dates was greater than the rainfall
treatment variation within a planting date. Future PreCCIP research should include studies
to determine the physical/physiological reason for the affect of temperature during the early
vegetative stage on final grain yield.
3. Background
Efforts to estimate the effects of weather on corn (Zea mays L.) yield have been
undertaken using models (Thompson, 1969, 1986; Runge, 1968; Swanson and Nyankori,
1979; Offutt et al., 1987; Muchow et al., 1990), and field plot studies (Robbins and Domingo,
1953; Denmead and Shaw 1960; Benoit et al., 1965; Classen and Shaw, 1970; Herrero and
Johnson, 1981; Harder et al., 1982; Ouattar et al., 1987; and Grant et al., 1989). The focus
of these studies was on the effects of temperature and rainfall throughout the growing
season in the modeling studies, and during specific growth stages in plot or growth chamber
studies.
125
The modeling studies have used statistics from the USDA Statistical Reporting
Service to relate temperature and rainfall to corn yields over large areas (county size or
larger). In one of these studies, Thompson (1986) found that corn yields increased with
increasing July and August rainfall, decreased with increasing July and August temperature,
and decreased as both temperature and rainfall deviated from their normals in the 5 central
Corn Belt states. In a recent modeling study, Muchow (1990) reported that temperature
affected growth duration. Lower temperatures increased the time the crop would intercept
radiation, and thus as growth duration increased, yields increased.
Plot, greenhouse, and growth chamber studies provide understanding of the
mechanisms of crop response to temperature and water stresses during different growth
stages. For example, studies of the effects of temperature on corn growth during pollination
and early grain fill have shown that high air temperature during pollination (Herrero and
Johnson, 1980) and temperatures greater than 25 ° C or less than 15 ° C during the lag phase
of grain fill (Jones et al. 1981, 1984,1985), will reduce corn yields. The cause of these yield
losses were failure of pollen germination (Herrero and Johnson, 1980) and a reduction in
the number of endosperm cells and starch granules (Jones et al. 1985).
Eck (1986) studied the effects of water stress during various corn growth stages and
found that water deficits imposed 41 days after planting reduced leaf, stalk, and ear yields,
while those imposed 55 days after planting reduced only stalk and ear yields. Deficits during
vegetative growth reduced kernel numbers but had little effect on weight per kernel. Kernel
126
numbers were not affected by water deficits during grain filling unless severe deficits were
imposed early in the period. Grant et al. (1989) found the period most sensitive to a water
stress was from silking to 22 days after silking. Herrero and Johnson (1981) reported that
yield losses due to water stress during silking was due to reduced rate of silk elongation.
During the lag phase and grain fill periods, yield losses were due to a reduction in the
number of endosperm cells formed during the lag phase of kernel development, as well as
a reduction of the length of grain fill duration (Ouattar et al., 1987).
The plot studies discussed above use short term water stresses that usually do not
occur in the same manner as in a natural production setting, while the statistical modeling
studies rely on spatially and temporally averaged data to evaluate the effects of weather on
corn yields. Neither type of study lends itself to evaluating the response of corn yields to the
various rainfall patterns and moisture levels that occur in nature at a given site. Therefore,
our PreCCIP/PACE study was established in rain shelters where typical dry, average, and
wet summers in east central Illinois were simulated. The typical summer rainfall scenarios
were simulated by keeping natural rainfall out and applying rainfall amounts typical of the
summers. The objective of the study was to determine the yield changes that could be
expected by increasing each rainfall event by 25%. Rainfall treatments and production
practices were constant over the 3 years of the study. This paper presents the results of our
study conducted during the summers of 1987, 1988, and 1989.
127
4. Materials and Methods
The experimental plots were established on Drummer silty clay loam (fine-silty mixed
mesic Typic Haplaquolls) in an area where moveable rain shelters (Banwart, 1988) could be
positioned over the plots to exclude natural rainfall. The rain shelters were equipped with
sprinkler nozzles so that each 3x3 m plot could be treated with a different rainfall treatment.
Plots within the shelters were positioned in a grid 3 plots wide by 12 plots long with the long
side of the plot area along a north-south axis. The rain shelters covered the plots only when
it was raining or whenever rainfall treatments were being applied.
Climatological analyses of 88 years of historical data at Urbana, Illinois were used to
develop the characteristics of rainfall during typical dry, typical average, and typical wet
summers (Changnon, 1988). In addition to the daily rainfall totals, the time of day that the
rain would have fallen was specified and the rain treatments applied at those times. The
number of rain days and total rainfall applied during June, July, and August for each of the
typical summers is shown in Table 1. The 6 rain treatments used were 3 treatments
representing the typical dry, average, and wet summer rainfall, and 3 treatments representing
the typical rainfalls plus a 25% increase of each rain event.
A M017xB73 corn hybrid was used in the three years of the 1987, 1988, 1989
experiment. The dates of planting and plant densities used in each year are shown in Table
2. The two planting dates and two plant densities in 1989 were established to determine if
the differences in the 1987 and 1988 yields were due to the different date of planting and
128
Table 1. Number of rain days and total rainfall during June, July, and August representing typical dry, average, and wet summers in
East Central Illinois.
Summer Month
Days with Rainfall Total Rain Summer Month
≥ 0.254 & <6.35 mm
≥6.35 & ≥12.7 mm
≥12.7 & >22.8 mm ≥22.8 mm Days mm
Typical Dry Summer
June 6 2 1 1 10 71.1
July 4 2 1 0 7 43.2
August 5 1 0 1 7 45.2
Total 15 5 2 2 24 160.0
Typical Average Summer
June 5 2 2 1 10 101.6
July 4 2 1 1 8 86.4
August 5 1 1 1 8 91.4
Total 14 5 4 3 26 279.4
Typical Wet Summer
June 5 2 3 2 12 139.7
July 5 2 1 2 10 127.0
August 6 1 2 1 10 114.3
Total 16 5 6 5 32 381.0
129
_
plant density practices used in the 2 years. In 1987, the plant densities were those resulting
naturally from planting with a mechanical planter. The 1988 and 1989 plant densities were
established by over-seeding the populations and thinning to the desired density. At harvest,
the center 2.4 m of the 2 center rows of the 4 rows in each plot were harvested. The
number of plants, ears per plant, number of barren plants, grain yield (Mg ha-1), and
vegetative mass per plant were recorded. The number of kernel rows per ear, number of
kernels per row, and mass per kernel were determined after the ears had been oven dried.
The average daily maximum and minimum temperatures, and the average air
temperature in each growth stage was computed from a near by weather station. The
duration of each development stage was determined by computing the number of days
required to accumulate the number of growing degree days needed for the crop to reach the
next stage. Table 3 shows the assumed number of growing degree days required to reach
each stage of development from planting to silking (Hollinger, 1981). The lag phase, the
time when cell division of the seed endosperm is occurring, was assumed to require 200
growing degree days after silking (Jones et al., 1984). Table 4 shows the average daily
maximum and minimum temperatures and the average temperature between planting and
tassel initiation.
The grain yield and vegetative mass per plant data were analyzed using multiple
analysis of variance (MANOVA procedure SPSS-X, version 3.1, SPSS, Inc, 1988). Factors
included in the analysis were position of the plot in the rain shelter (west, central, east),
130
Table 2. Date of corn planting and plant densities at each planting.
Year Date of Planting
Plant Density (plants ha-1)
1987 May 28 81,400
1988 May 12 64,700
1989 May 12 64,700
81,400
1989 May 26 64,700
81,400
Table 3. Growing degree day accumulation required for the various stages of corn growth to be completed (Adapted from Hollinger, 1981).
Stage of Growth Degree Day Accumulation
Planting to Tassel Initiation 311
Tassel Initiation to Ear Initiation 81
Ear Initiation to End of Row Set 144
Tassel Initiation to Silking 433
131
Table 4. Average daily maximum and minimum temperatures, and average temperature and duration from planting to tassel initiation.
Year Date of Planting
Duration (days)
Mean Daily Temperature
Year Date of Planting
Duration (days)
Maximum (C)
Minimum (C)
Average (C)
1987 28 May 24 30.7 17.7 24.2
1988 12 May 31 29.3 11.9 20.6
1989 12 May 33 25.2 13.6 19.4
1989 26 May 28 27.6 15.6 21.6
plant density, rainfall treatments, average temperature from date of planting to tassel
initiation, average minimum daily temperature from date of planting to tassel initiation, and
the cross term average minimum daily temperature and the square of the average minimum
daily temperature from planting to tassel initiation. The effect of different plant densities
on final yield was tested within 1989 only because it was confounded with temperature in
1987 and 1988. The significant covariates found in the MANOVA were then used in a
linear regression analysis (Regression procedure, SPSS-X, version 3.1, SPSS, Inc, 1988) to
determine the slope and intercept of the relationship.
5. Results
The three years of the study were all significantly different. The first year, 1987, was
characterized by a hot dry spring and early summer followed by adequate July rainfall. On
31 July 1987, a 100 year storm dumped 102 mm of rain on the plot area in 4 hours. This
132
resulted in the plots being flooded at the time of pollination. The 1988 summer was dry
throughout the summer with above normal temperatures in July and August, and high
evaporative demand due to very low humidities. During the 1989 summer the temperatures
were cool throughout the spring and summer with near normal humidities. Because natural
rainfall was kept off the plots, the actual rainfall during each summer did not affect the
experiment except in 1987. During the winter, the plots were left uncovered so the soil
profile could be recharged.
Table 5 presents the means and standard deviations of the vegetative mass (gm
plant-1) and grain yields (Mg ha-1) for each of the 4 planting dates and the 6 rainfall
treatments. There was a large standard deviation of vegetative mass in the 1989 plantings
is because the means included the 2 different plant densities in 1989. The mean vegetative
mass for the 64,700 plant ha-1 density in 1989 was 101.1 g plant-1 and for the 81,400 plant
ha-1 density, 81.9 gm plant-1. Vegetative plant mass and grain yield increased with greater
rainfall for all planting dates.
Results of the analysis of variance are shown in Table 6 for both the vegetative mass
and grain yield. The significant independent variables for each dependent variable are
shown in the table. The plot position variable being significant indicates that there was a
border effect in the plots. The data show that the plots on the east and west sides of the
shelters had greater yields and plant masses than the plots in the center of the shelters.
133
Table 5. Vegetative mass and grain yields of the 4 planting dates and 6 water treatments.
Date of Planting
28 May 87 12 May 88 12 May 89 26 May 89
Treatment Mean SD Mean SD Mean SD Mean SD
Dry 71.5 2.5 84.5 9.9 85.2 12.5 77.9 10.5
D r y + 25% 76.7 5.8 83.4 6.3 87.5 13.2 82.8 12.3
Average 78.0 9.5 90.2 4.7 88.3 18.6 87.9 10.0
Average + 25% 79.6 9.1 95.8 10.5 93.8 17.2 94.5 15.0
Wet 88.1 14.7 97.1 12.8 101.4 17.6 104.3 12.8
Wet + 25% 89.3 7.8 109.8 3.8 113.9 16.4 99.6 11.2
Planting Date
Mean 80.5 93.5 95.0 91.2
Grain Yield (Mg ha"1)
Mean SD Mean SD Mean SD Mean SD
Dry 4.74 0.% 10.18 0.91 11.86 1.09 8.20 1.11
Dry + 25% 5.87 1.53 9.52 1.11 12.07 1.31 . 9.46 1.08
Average 5.28 0.63 10.41 0.79 12.27 2.19 9.08 1.17
Average + 25% 5.08 1.33 11.50 0.24 12.79 1.21 10.54 1.45
Wet 7.10 1.14 12.26 2.12 13.18 1.18 11.08 0.83
Wet + 25% 6.56 1.23 11.60 2.13 13.93 1.13 10.71 0.96
Planting Date
Mean 5.77 10.91 12.68 9.85
134
The regression equation relating vegetative mass to rainfall and plant population is
given by:
V = 131.5 - 0.0008724p + 0.0786r (1)
where V is the vegetative mass in gm plant-1, p is plant population in plants ha-1, and r is
total June, July and August rainfall in mm. The adjusted coefficient of variation of 0.581
was slightly less than that found by analysis of variance (Table 6). This difference occurred
because the plot position effect was included in the error term of the regression equation.
The relationship between the rainfall and vegetative mass (Fig. 1), showed that the effect
of plant density on vegetative mass was a result of a shift of the mean vegetative mass rather
than a response to rainfall. Separate regression analysis of vegetative mass vs rainfall as a
function of population showed that the slopes of the rainfall response were not different
between the two populations.
The average temperature between planting and tassel initiation explained more of the
variation in grain yield than rainfall throughout the summer. Plant density did not have an
effect on the final yield even though a plot of the response of grain yield to rainfall as a
function of population revealed two regression lines (Fig. 2). Even though there were 2
distinct regression lines shown, the slopes and intercepts were not significantly different.
This was due mainly to the larger variance in grain yield at the higher plant densities.
Temperature alone explained 72% of the yield variation between planting dates (Fig. 3).
Rainfall and temperature together explained 79.2% of the variation in a multiple regression.
135
Figure 1. Response of vegetative mass to the rainfall treatments. ( ) represents a plant density of 81,400 plants ha" , (——) represents a plant density of 64,700 plants ha-1.
Figure 2. Response of final grain yield to rainfall treatments Population 1 is 64,700 plants ha" (———), population 2 is 81,400 plants ha-i (---).
Figure 3. Response of final grain yield to average temperature between planting and tassel initiation.
This relationship is expressed as:
Y = 36.98 - 1.399T + 0.006935R (2)
where Y is grain yield in Mg ha-1, T is average temperature between planting and tassel
initiation in ° C, and R is total summer rainfall in mm. The difference between the variation
explained in the analysis of variance and the multiple regression was due to the omission of
the plot position term in the regression analysis. When grain yield was fitted separately by
simple linear regression to rainfall and temperature, the coefficients were the same as when
determined together with multiple regression. This indicates that in this study temperature
and rainfall are independent of each other.
6. Discussion
The independence of the rainfall and temperature in the analysis is expected because
rainfall was a controlled variable during the three years of experiment and temperature was
not. Under natural conditions temperature and rainfall are covariates, and any regression
analysis using natural rainfall and temperature would result in changes in the rainfall and
temperature coefficients, if fitted separately by simple linear regression and then together
by multiple regression.
Decoupling temperature from rainfall allowed the independent study of the effects
of temperature and rainfall on corn yield. The increase in corn yields with increasing rainfall
was expected. The increased yields were partially due to larger plants as shown by the
relationship between the size of individual plants and final grain yield (Figure 4). Individual
136
Table 6. Analysis of variance for vegetative mass and grain yield as functions of population, plot location, rainfall, and mean temperature
from planting to tassel initiation. For each analysis, only the significant independent variables are shown.
Variable Sum of Squares
Degrees Freedom
Mean Sum of Squares F
Vegetative Plant Mass
Rainfall 9168.6 1 9168.6 96.33
Plant Density 9745.3 1 9745.3 102.33
Plot Position 1645.0 2 822.5 8.64
Model 205558.9 4 5139.7 54.00
Residual 11516.9 121 95.2
Total 32075.8 125 256.6
Adjusted R2 0.629
Grain Yield
Average Temperature 674.3 1 674.3 524.26
Rainfall 69.6 1 69.6 54.10
Plot Position 35.3 2 17.7 13.74
Model 779.3 4 194.8 151.46
Residual 155.6 121 1.3
Total 934.9 125 7.5
Adjusted R2 0.828
137
Figure 4. Response of final grain yield to individual plant mass.
plants responded to both plant density and rainfall. The response to increased rainfall
indicates that rainfall was a limiting factor in determining final yields in this study. At the
higher plant density, plant competition for water further reduced the size of individual
plants. Even though individual plants were smaller at the higher plant densities, total
vegetative mass on an areal basis was greater at the higher plant densities. Reduction in
individual plant size with increasing plant density was also reported by Genter and Camper
(1972), Daynard and Muldoon (1981), and Tetio-Kagho and Gardner (1988a). The decrease
in yield at the higher plant densities differs from the findings of Tetio-Kagho and Gardner
(1988b) where grain yields increased with increasing plant density up to 100,000 plants ha-1.
The strong relationship between temperature during the early vegetative stages
(planting to tassel initiation) and final yield indicates the importance of conditions during the
early vegetative growth stages on the final yield. This result indicates that what happens
during the pure vegetative stage of growth, when leaves are being initiated and enlarged,
affects the subsequent development of the reproductive organs following tassel initiation.
Because plant mass was not determined at the time of tassel initiation, it is impossible to
evaluate the effects of temperature during the early vegetative stages on vegetative dry
matter accumulation.
Studies of the response of corn seedlings to soil and air temperature have indicated
that the optimum night temperature for growth is 15°C (Grzesiak et al., 1981). They
attributed this optimum temperature to a reduction in photosynthesis below 15 ° C and an
138
increase in dark respiration above 15°C. Barlow et al. (1977) found that when soil
temperatures decreased below 28°C the leaf elongation rate was decreased. This was
attributed to restricted water uptake which reduced plant water potential and to a decrease
in the shoot apical meristem temperature. Barlow et al. (1977) reported that leaf elongation
ceased at a plant water potential of -0.9 MPa, but that neither transpiration nor
photosynthesis were reduced significantly until plant water potential decreased to -1.2 to -1.3
MPa. Barlow et al. (1976) found that a greater reduction in leaf elongation without a
corresponding decrease in photosynthesis resulted in increased stores of soluble
carbohydrates in the plant.
These findings are supported by other studies which show different rates of leaf
elongation and dry matter accumulation are functions of temperature. Arnold (1975)
reported that the rate of sweet corn maturation was linearly related to temperature while
weight increase had a curvilinear relationship with an optimum temperature of 24° C. If
such a relationship exists during the early vegetative growth of corn, cooler temperatures
would result in larger plants at the time of tassel initiation. Brouwer et al. (1973) found that
low temperatures decreased leaf area development more than leaf differentiation. They
concluded that under favorable growing conditions, area growth and differentiation compete
for available carbohydrates leading to high arearweight ratios both in roots and leaves.
Under less favorable temperature conditions where area growth is inhibited, the area:weight
ratios would be reduced, accompanied by increased carbohydrate stores. With
139
differentiation progressing at a more "normal" rate, plants grown under cooler conditions
would result in more reserves being shunted to the developing ear when it begins to develop,
thus resulting in a higher potential yield under cooler conditions as observed in this study.
The above studies provide some understanding of the reasons for the strong
relationship between the temperature during the early vegetative stages of growth and final
yield. However, more detailed experiments need to be conducted to fully understand the
physical and physiological mechanisms causing the response.
Based on this study we conclude that a major determinate in the year to year corn
yield variability is the temperature during the early vegetative stage, especially in seasons
where water is not limiting. In those seasons where water is a limiting factor, both the
temperature during early vegetative growth and seasonal water availability interact to
determine the year-to-year variability, assuming constant management practices.
7. Economic Value of Precipitation Augmentation to Agriculture
An estimate of the value of added rainfall to agriculture in the East and Central crop
reporting districts in Illinois has been determined from the past 3 year's rain shelter
experiments and is shown in Table 7. The assumptions made are: (1) precipitation
augmentation will result in a 25% increase in rainfall in each rain event; (2) the price of
corn is $2.00 bu1 and the price of soybean is $4.00 bu"1; (3) the corn and soybean acreages
harvested are the same as those in 1987 in the Central and East crop reporting districts.
These values do not represent the total return for Illinois because only 2 of the 9 crop
140
reporting districts in Illinois are included. This study, however, shows a significant benefit
to Dlinois agriculture if precipitation augmentation is feasible.
Table 7. Estimate of the value precipitation augmentation to agriculture in the Central and East crop reporting districts in Illinois.
Crop Reporting District
East Central
Yield Change Crop Value Change Crop Value Change Total
Soybean bu
Corn bu
Soybean Corn Soybean Corn x 100
Dry 3.32 8.20 $21,480* $19,827 $19,355 $20,719 $81,383
Average -1.34 12.18 ($8,669) $29,451 ($7,812) $30,776 $43,745
Wet 5.75 0.73 $37,202 $1,765 $33,522 $1,844 $74,334
*All dollar amounts are thousands.
8. References
Arnold, C.Y., 1975: The relationship of inheritance and temperature to the yield of sweet
corn (Zea mays L.). J. Amer. Soc. Hort. Sci., 100:542-545.
Banwart, W.L., 1988: Field evaluation of an acid rain drought stress interaction. Environ.
Pollution. 53:123-133.
Barlow, E.W.R., L. Boersma, and J.L. Young, 1976: Root temperature and soil water
potential effects on growth and soluble carbohydrate concentration of corn seedlings.
Crop Sci., 16:59-62.
141
Barlow, E.W.R., L. Boersma, and J.L. Young, 1977: Photosynthesis, transpiration, and leaf
elongation in corn seedlings at suboptimal soil temperatures. Agron. J.. 69:95-100.
Benoit, G.R., A.L. Hatfield, J.L. Ragland, 1965: The growth and yield of corn. III. Soil
moisture and temperature effects. Agron. J., 57:223-226.
Brouwer, R., A. Kleinendorst, and J. Th. Locher, 1973: Growth responses of maize plants
to temperature. In: Plant Response to Climatic Factors. Proc. Uppsala Symp., 1970.
Unesco.
Changnon, S.A., 1988: Climate based representation of summer rainfall in Illinois. J.
Climate. 1:1041-1047.
Classen, M.M., and R.H. Shaw, 1970: Water deficits effects on corn. II. Grain components.
Agron. J.. 62:652-655.
Daynard, T.B., and J.F. Muldoon, 1981: Effects of plant density on the yield, maturity and
grain content of whole-plant maize. Can. J. Plant Sci.. 61:843-849.
Denmead, O.T., and R.H. Shaw, 1960: The effects of soil moisture stress at different stages
of growth on the development and yield of corn. Agron. J., 52:272-274.
Eck, H.V., 1986: Effects of water deficits on yield, yield components, and water use
efficiency of irrigated corn. Agron. J.. 78:1035-1040.
Genter, C.F., and H.M. Camper, Jr., 1972: Component plant part development in maize as
affected by hybrids and population density. Agron. J.. 65:669-671.
Grant, R.F., B.S. Jackson, J.R. Kiniry, and G.F. Arkin, 1989: Water deficit timing effects on
142
yield components in maize. Agron. J.. 81:61-65.
Grezesiak, S., S.B. Rood, S. Freyman, and DJ. Major, 1981: Growth of corn seedlings:
Effects of night temperature under optimum soil moisture and under drought
conditions. Can. J. Plant Sci., 61:871-877.
Harder, J.J., R.E. Carlson, and R.H. Shaw, 1982: Yield, yield components, and nutrient
content of corn grain as influenced by post-silking moisture stress. Agron. J.. 74:275-
278.
Herrero, M.P., and R.R. Johnson, 1980: High temperature stress and pollen viability of
Maize. Crop Sci.. 20:796-800.
Herrero, M.P., and R.R. Johnson, 1981: Drought stress and its effect on maize reproductive
systems. Crop Sci.. 21:105-110.
Hollinger, S.E., 1981: Environmental effects on corn ear morphology, planting to silking.
Ph.D. Thesis, Purdue University, West Lafayette, IN.
Jones, R.J., B.G. Gengenbach, and V.B. Cardwell, 1981: Temperature effects on in vitro
kernel development of maize. Crop Sci.. 25:761-766.
Jones, R.J., S. Ouattar, and R.K. Crookston, 1984: Thermal environment during endosperm
cell division and grain filling in maize: Effects on kernel growth and development in
vitro. Crop Sci., 24:133-137.
Jones, R.J., J. Roessler, and S Ouattar, 1985: Thermal environment during endosperm cell
division in maize: Effects on number of endosperm cells and starch granules. Crop
143
Sci.. 25:830-834.
Muchow, R.C., T.R. Sinclair, and J.M. Bennett, 1990: Temperature and solar radiation
effects on potential maize yield across locations. Agron. J.. 82:338-343.
Offutt, S.E., P. Garcia, and M. Pinar, 1987: Technological advance, weather, and crop yield
behavior. North Central J. of Agric. Econ., 9:49-46.
Ouattar, S., R.J. Jones, and R.K. Crookston, 1987a: Effect of water deficit during grain
filling on the pattern of maize kernel growth and development. Crop Sci., 27:726-
730.
Ouattar, S., R.J. Jones, R.K. Crookston, and M. Kayeiou, 1987b: Effect of drought on water
relations of developing maize kernels. Crop Sci.. 27:730-735.
Robbins, J.S., and C.E. Domingo, 1953: Some effects of severe soil moisture deficits at
specific growth stages in corn. Agron. J.. 45:618-621.
Runge, E.C.A., 1968: Effects of rainfall and temperature interactions during the growing
season on corn yield. Agron. J., 60:503-507.
SPSS, Inc., 1988: SPSS-X User's Guide. 3rd Edition. SPSS Inc., 444 N. Michigan Ave,
Chicago, IL.
Swanson, E.R. and J.C. Nyankori, 1979: Influence of weather and technology on corn and
soybean yield trends. Agric. Meteorol., 20:227-242.
Tetio-Kagho, F., and F.P. Gardner, 1988a: Responses of maize to plant population density.
I. Canopy development, light relationships, and vegetative growth. Agron. J., 80:930-
144
935.
Tetio-Kagho, F., and F.P. Gardner, 1988b: Responses of maize to plant population density.
II. Reproductive development, yield, and yield adjustments. Agron. J.. 80:935-940.
Thompson, L.M., 1969: Weather and technology in the production of corn in the U.S. Corn
Belt. Agron. J.. 61:453-456.
Thompson, L.M., 1986: Climatic change, weather variability, and corn production. Agron.
J., 78:649-653.
145
SUMMARY OF HEAVY RAINFALL RESEARCH
1. Time Distribution of Heavy Rainfall
A study was made of the time distribution of rainfall in heavy rainstorms. Analyses
were completed and a research report published during late spring 1990 (Huff, 1990). The
time distribution of heavy rainfall is an important factor in several fields, and it is useful for
post-storm assessment of individual storm events in weather modification operations. Time
distribution is very essential in the design and operation of water control systems, such as
urban storm sewer systems. It is also an important factor in the assessment of soil erosion.
The report provides the best available information on the time distribution
characteristics of storm rainfall and small basins encompassing areas up to 400 square miles.
Results are applicable not only to Illinois but to the Midwest in general, and to other areas
having similar heavy rainfall climates.
Information was presented in the form of families of curves derived for groups of
storms categorized according to whether the greatest percentages of storm rainfall occurs
in the first, second, third, or fourth quarter of a storm period. The time distributions were
expressed as cumulative percentages of storm rainfall and storm duration to enable
comparisons between storms. The individual curves for each storm type provide estimates
of the time-distribution characteristics at probability levels of 10% to 90% of the total storm
occurrences.
146
2. Spatial Distribution of Heavy Storm Rainfall
Under the precipitation research program of PreCCIP, a second study was initiated.
The purpose of this study was to develop spatial distribution characteristics in very extreme
rain events, defined as those occurring at a given location on the average of once in 100
years or longer. Requirements now exist for this information to address weather
modification assessment since urban-induced rainfall has been found to occur most often in
locally heavy rainfall situations.
Storm models are needed for these extreme weather events which take into
consideration the spatial characteristics with respect to storm intensity, the shape and
orientation of such storm systems, their movement, and the variation of the various storm
characteristics with storm duration. Models are also needed which apply to basins of various
sizes, ranging from urban drainage basins of a few hundred square miles to those
incorporating areas of several thousand square miles. Analyses are confined to Midwestern
storm systems for which much of the basic information for the model development has
become available from past hydrometeorological research carried out at the Illinois State
Water Survey. Approximately 250 storm systems are included in the data sample. These
span the period from 1901 to the present.
3. References
Huff, F.A., 1990: Time Distributions of Heavy Rainstorms in Illinois, ISWS Circular 13, 18
pp.
147
INADVERTENT WEATHER MODIFICATION RESEARCH
Studies of inadvertent, urban-produced changes in clouds and precipitation have been
initiated as a part of PreCCIP because of their relevance in understanding how man, either
purposefully or accidentally, can induce changes in the local and regional scales of clouds
and precipitation. In past years, Water Survey scientists, working in cooperation with
scientists of other universities, have investigated the general subject of urban effects on
clouds and precipitation with the primary focus on the summer conditions. These past
studies included a major project centered at St. Louis during 1971-75. During those years,
the Water Survey maintained a large dense recording raingage network in the St. Louis
region. These unique data became the basis for further studies in the non-summer seasons.
Analysis of precipitation events in the St. Louis area, based on pre-event low-level
windflow, was pursued to ascertain the presence of urban effects on fall, winter, and spring
precipitation. The data from the circular dense raingage network were used to define
quadrant (NW, NE, SE, SW) average precipitation. Winds before each event (443 events
during 1971-1975) were used to define the urban plume and to identify which quadrant was
"downwind" of the city.
Results for fall revealed a 17% increase in precipitation downwind of St. Louis, and
a 13% increase in events with their peak rainfall occurring downwind. Both outcomes were
found to be statistically significant at the 1% level. Further, the downwind enhancement was
greatest when pre-event winds were from the southeast, and when average precipitation in
148
the quadrant with the maximum rainfall value was either quite light (<5.1 mm) or quite
heavy (>17.9 mm). The fall results agree with earlier findings for summer rainfall which had
revealed a 25% rainfall increase due to enhancement either in isolated light airmass showers
or during heavier, well-organized convective systems.
Winter precipitation results indicated little precipitation change downwind of St.
Louis. However, when southwest pre-event winds existed (a flow often associated with
winter convection), there was a statistically significant downwind increase in winter
precipitation. However, when pre-event winds were from the southeast or northwest (flows
frequently associated with stratiform precipitation), downwind decreases in precipitation
occurred.
The number of spring precipitation events maximized downwind of St. Louis was
significantly greater than expected by chance, particularly in light (<5.1 mm) precipitation
events. However, total spring rainfall downwind of St. Louis was increased by only 4%.
There was no suggestion of decreased precipitation in spring or fall seasons. The urban
influences that enhance precipitation in these three seasons appear to be related to
precipitation conditions with convective processes. Urban influences on the more stratiform
precipitation situations was negligible. These results were summarized into a scientific paper
that will be published in the Journal of Applied Meteorology. Further studies of inadvertent
atmospheric modification are being planned including studies of potential urban effects on
nocturnal rainfall events during the summer season.
149
PROJECT PLANNING AND ASSESSMENT
During the 12-month period of this project, an in-depth assessment was made of the
PACE project. Several factors were considered in this analysis, including project personnel,
project leadership, financial support for the program as it relates to future needs and long-
term stability, in-state and regional support for weather modification research, the
dimensions of the current program and other opportunities for investigations, and various
scientific issues that were and are affecting the scientific interest and government support
of weather modification research.
Consideration of these various issues and factors affecting PACE, coupled with the
change in the NOAA program to one of "atmospheric modification," led us to refocus the
effort and, in essence, broaden it. To this end, the program previously known as PACE was
renamed during this past year, as the "Precipitation-Cloud Changes and Impacts Project,"
or PreCCIP.
The goal of PreCCIP has been established to study and understand cloud and
precipitation processes in Illinois that: 1) occur naturally, 2) occur as a result of human
changes in the environment, and 3) could be produced by purposeful cloud seeding. The
program includes studies of the impacts of these changes on the environment and economy
of Illinois and the Midwest, and the resulting implications for policy development.
This new programmatic dimension allows us to address a wide range of relevant
atmospheric issues such as the local and regional influences on the atmosphere and related
150
climate change, changed extremes of rainfall, alterations in cloud cover, causes of droughts,
and lake effect storms. All the research areas of PreCCIP have been key elements of past
Illinois State Water Survey scientific programs. The programmatic dimensions of PreCCIP
fall in three broad areas. These are as follows.
1. Studies of Inadvertent Changes in Clouds and Precipitation
These studies are seen as critically important to understanding atmospheric
modification, the climate change issue, and the economic environmental effects. The
envisioned research includes:
1. further studies of existing historical project databases from St. Louis and
Chicago to better understand the physical effects of urban-industrial areas on
clouds and precipitation;
2. potential field studies to gather data on cloud and precipitation processes
affected by human-enhanced changes in the atmosphere;
3. investigations of regional changes in clouds and/or precipitation due to
effluents from large urban areas, aircraft contrails, and other sources; and
4. studies of how nature affects clouds and precipitation, including atmospheric
effects due to Lake Michigan and major topographic features of Illinois and
the Midwest.
151
2. Studies of Planned Precipitation Modification in the Midwest
These studies will include:
1. in-depth analysis of existing databases from PACE field experiments in 1986
and 1989;
2. scientific-based evaluations of future operational cloud seeding projects
conducted in Illinois;
3. study and monitoring of precipitation modification results from other sources
relevant to Midwestern conditions; and
4. field efforts for exploring issues raised by results from the 1989 field project.
3. Investigations of the Physical Effects of Altered Clouds and Precipitation on the Environment and Economy
Studies in this area will include:
1. research concerning the impacts of changed cloud and precipitation conditions
on agriculture, transportation, and other weather-sensitive sectors;
2. studies of potential changes in the hydrologic cycle;
3. studies of other environmental effects including water quality; and
4. interdisciplinary studies concerning how these effects and impacts relate to the
development of state and national policies.
This assessment of the PACE project and the resulting planning effort for PreCCTP
have been concluded. A major new direction for the program has resulted.
152
FY90 PUBLICATIONS FROM ILLINOIS PreCCIP
1. Formal Publications
Changnon, S.A. and S.E. Hollinger, 1990: Crop Yield Results from Simulated Rain Applications to Agricultural Plots in Illinois in Journal of Weather Modification. Vol. 22, 1990. pp 58-62.
Abstract
Ten different levels of rainfall were applied (during 1987, 1988, and 1989) to agricultural plots in central Illinois to discern effects on corn and soybean yields. Increases in rainfall during a hot dry summer (June-August 1988) revealed sizable yield gains. For one inch of added rainfall, the yields increased 10 bu/acre for corn and 4 bu/acre for soybeans. In a summer of near average rain (1989), the increases were less, about 5 bu/acre for corn and 3 bu/acre for soybeans. When summer rainfall exceeded 14 inches, yields of both crops were decreased. The various rainfall tests revealed that rain increases done only on days when natural rainfall was <0.1 inch provided no detectable yield increases, whereas a 40% increase on all rain days (the largest increase tested) produced the greatest crop yield increase (up to the 14-inch optimum). Corn yields reacted very favorably to added rains on days with ≥1.0 inch of rain.
Garcia, P., S.A. Changnon, and M. Pinar, 1990: Economic Effects of Precipitation Enhancement in the Corn Belt in Journal of Applied Meteorology. Vol. 29, No. 1, January 1990. pp 63-75.
Abstract
Policy formulation in weather modification requires an understanding of the economic effects from altered weather. The focus of this study is to provide insight into the beneficiaries of a functioning weather modification technology when applied at various spatial and temporal levels. An econometric model which links the corn/soybean production to U.S. cattle, hog and poultry sectors is used to determine the effects of precipitation enhancement in the U.S. Corn Belt, a humid climatic region. A regional supply formulation permits assessment of weather modification on production, prices, revenues to producers, and savings in consumers expenditures on meat. The results provide insight into the distribution of economic effects, emphasize the importance of careful planning in the use of
153
_
weather modification technology, and provide useful information on the roles of local, state, and federal governments in the support of weather modification.
Westcott, N.E., 1990: Radar Results of the 1986 Exploratory Field Program Relating to the Design and Evaluation of PACE, in Journal of Weather Modification. Vol. 22,1990. pp 1-17.
Abstract
The initial phase of the Precipitation Augmentation for Crops Experiment (PACE), directed at enhancing rainfall was conducted in Illinois during the summer of 1986. This first experiment resulted in a limited sample, (19 clouds and 3 experimental units), but one sufficient to provide information pertinent to the design and evaluation of future efforts.
In particular, it was determined that differences in meteorological conditions may mask any seeding signature present, requiring that the experimental units be stratified or normalized. It was found that for these clouds, the height of the echo at first detection and the age of the echo at treatment (Agl or placebo) have an important bearing on the expected growth of the echo. Additionally, the area and reflectivity of the echo at 6 km at the time of treatment seem to be related to the maximum size attained by the echo cores. That is, the larger the echo at treatment, the larger the echo can be expected to grow.
However, the growth of the echo core in terms of reflectivity, height and area appeared to slow as the echo cores matured. This suggests that explosive growth sometimes expected from cloud seeding may not be the rule in this area of the country and that a comparison of before and after treatment growth rates may not be a good evaluation tool. Rather, the post treatment growth, with the experimental units stratified by the age of the echo at treatment and by the height of formation may be more useful in discriminating seeding effects.
Radar derived predictor variables were examined to assess echo behavior based on the ambient weather conditions, and response variables were developed which may be useful in detecting potential seeding effects. Inferences were made with regard to stratification of the data, experimental unit definition, and cloud behavior. This work emphasized, as others have found, the need for predictor variables and a large sample.
154
2. Informal Publications and Presentations
Czys, R.R., 1990: Observed Versus Diagnosed Ice Production Rates in Warm-Based Midwestern Cumuli. Conference on Cloud Physics, July 23-27, 1990, San Francisco, CA. 5 pp.
DeGrand, J.Q., A.M. Carleton, and P.J. Lamb, 1990: A Mid-Season Climatology of Jet condensation Trails from High Resolution Satellite Data. Conference on Radiation, San Francisco, CA. pp 309-311.
Scott, R.W. and R.R. Czys, 1990: Convection Intensity and Seeding Suitability Forecasts in a Dynamic Seeding Experiment in Illinois. 12th Conference on Weather Analysis and Forecasting, Oct. 2-6, 1989, Monterey, CA. pp 107-110.
Scott, R.W., and R.R. Czys, 1990: An Objective Process to Forecast Convective Precipitation and Seedability of Clouds in the Midwestern United States. Submitted to European Geophysical Society, XV General Assembly, Copenhagen, April 23-27, 1990.
Huff, F.A., 1990: Time Distributions of Heavy Rainstorms in Illinois, ISWS Circular 173, 18 pp.
155
APPENDIX A: SELECTED PREPRINTS
156
CROP YIELD RESULTS FROM SIMULATED RAIN APPLICATIONS TO AGRICULTURAL PLOTS IN ILLINOIS
by Stanley A. Changnon and Steven E. Hollinger
Illinois State Water Survey Champaign, Illinois 61820
ABSTRACT
Ten different levels of rainfall were applied (during 1987, 1988, and 1989) to agricultural plots in central Illinois to discern effects on corn and soybean yields. Increases in rainfall during a hot dry summer (June-August 1988) revealed sizable yield gains. For one inch of added rainfall, the yields increased 10 bu/acre for corn and 4 bu/acre for soybeans. In a summer of near average rain (1989), the increases were less, about 5 bu/acre for corn and 3 bu/acre for soybeans. When summer rainfall exceeded 14 inches, yields of both crops were decreased. The various rainfall tests revealed that rain increases done only on days when natural rainfall was ≤0.1 inch provided no detectable yield increases, whereas a 40% increase on all rain days (the largest increase tested) produced the greatest crop yield increase (up to the 14-inch optimum). Corn yields reacted very favorably to added rains on days with ≥1.0 inch of rain.
1. INTRODUCTION Weather modification research in Illinois since
1970 has included efforts to estimate the effects of additional rainfall produced by weather modification on two major U.S. crops, corn and soybeans (Changnon, 1986). Several studies utilized weather-crop yield regression models to estimate the yield changes relating to use of differing levels of rainfall during various parts of the growing season. Phenological models have also been used to more accurately assess weather effects on crop yields. However, like all models. crop-weather models depend upon considerable information about applications that are not widely available and thus must include key assumptions. The most recent modeling involved a new regression model utilizing the 1960-1986 data to estimate yield effects from changed rainfall under current levels of farm practices and related technology (Garcia et al., 1986). How the model's yield predictions for 1987 and 1988 related to yields from agricultural plots where additional rainfall was applied has been investigated (Changnon et al., 1989).
The principal current investigation to assess the effects of altered summer rainfall on yields in central Illinois has involved use of agricultural plots at the University of Illinois experimental farms. One set of these plots allows for ten tests, with one including the actual summer (June-August) rainfall. The other nine tests involve application of a range of potential rainfall modification capabilities that can be envisioned for convective rainfall in the Midwest. These tests are itemized in Table 1.
This paper focuses on the yield results obtained from the nine rainfall simulations on these plots in 1987, 1988, and 1989. In each of the three years, the same varieties of corn (a Mo17 x B73 Cross, and a Williams variety of soybeans) were planted. Planting dates varied somewhat being 28 May 1987, 16 May 1988, and 14 May 1989. Different planting dates do have an effect on crop yields and the weather relationships, but these subtle effects could not be separated out in this
3-year sample. The results reveal that any effects of the different planting dates appear minimal. Problems related to wind damage to the corn test fields in 1987 unfortunately kept these data from being adequate for analysis. The plots utilized for corn and for soybeans each contained 30, 10'x10' plots. Each plot had its own sprinkler which was connected to a central pumping system where any level of water application can be made for any plot. For each treatment, there were 3 replications randomly chosen within the larger 30-plot area. Water increases for the 9 treatment models were applied as soon as each rain ended. The natural rainfall amount for a distinct rain event was used to calculate the amount of additional rain to be applied, and the different amounts. Thus, if a rain of 0.3 inch occurred, the 3 plots designed to receive a 10% increase received 0.03 inch from the pumping system immediately after the rain day ended. Further details about the techniques, the facility, and methods of planting are given in an earlier paper (Changnon and Hollinger. 1988). It should be noted that the fertilizer treatments and planting densities of both crops remained the same in all three years. Thus, the major between-year difference was the weather conditions.
2. RESULTS The summer weather conditions of the three
years sampled fortunately provided a wide range of differing conditions, particularly as it pertained to total rainfall. The 3-summer values appear at the bottom of Table 1, along with their departures from the 3-year normals for 1951-80. The rainfall values show an extremely wet 1987 with 17.8 inches, an extremely dry 1988 with only 5.2 inches, and a near normal 1989 with 11.0 inches. Summer temperatures were very near normal in 1987 and 1989, but extremely above normal in 1988 which was a part of a major Midwestern drought.
The 10 rainfall applications to the plot included the actual rainfall and 9 rainfall enhancements of the actual rainfall. Treatments 2, 3, and 4 involved 10%,
58
Reprinted from the Journal of Weather Modification, Vol. 22, 1990.
Table 1. Crop Yields and Summer Weather Conditions in 1987, 1988. 1989.
Corn Yields, bu/acre
Rainfall Applications 1987 1988 1989
1. Natural rainfall - 48.3 219 2. Increase all rain days by 10% - 55.9 225 3. Increase all rain days by 25% - 62.7 232 4. Increase all rain days by 40% - 69.3 226 5. Increases to 0.1 to 1.0 inch rain days by 10% - 65.1 221 6. Increases to 0.1 to 1.0 inch rain days by 25% - 54.5 220 7. Increases to 0.1 to 1.0 inch rains days by 40% - 66.7 231 8. Increase all >1.0 inch rain days by 10% - 51.8 234 9. Increase all > 1.0 inch rain days by 40% - 67.0 231 10. Increase all ≤0.1 inch rain days by 40% - 49.3 216
Soybean Yields, bu/acre
1. Natural rainfall 32.7 15.1 45.1 2. Increase all rains by 10% 26.1 16.9 48.4 3. Increase all rains by 25% 30.7 18.7 54.6 4. Increase all rains by 40% 23.7 20.8 52.8 5. Increases to 0.1 to 1.0 inch rain days by 10% 32.1 17.1 45.7 6. Increases to 0.1 to 1.0 inch rains days by 25% 27.6 16.7 48.2 7. Increases to 0.1 to 1.0 inch rains by 40% 27.1 18.9 49.0 8. Increase all > 1.0 inch rain days by 10% 29.6 17.1 48.2 9. Increase all > 1.0 inch rain days by 40% 26.6 17.9 55.0 10. Increase all <0.1 inch rain days by 40% 34.4 16.3 43.8
Summer rainfall/ 17.84 5.24 11.07 departure (in.) (+6.9)(-5.70)(+0.16)
Summer temperature/ 74.3 76.3°73.33° departure (°F) (+0.9°)(+3.1°) (-0.1 )
25%, and 40% increases, respectively, to all rain days during the summer. Treatments 5, 6, and 7 involved identical levels of increases but applied only to daily rains in the 0.1- to 1.0-inch category. Treatments 8 and 9 were increases of 10% and 40% to only rain days of 1 inch or heavier, and treatment 10 was an increase of 40% to all daily rains of 0.1 inch or less. These were designed to bracket a range of capabilities seen as potentially likely with a future Midwestern rain enhancement technology.
Results for Corn. In general, the lowest yields came with natural rainfall in both years (1988 and 1989) or with the natural rainfall plus a 40% increase in the very light, 0.1 inch or less, rains. The yield values and rainfall values are shown in Table 1 and figure 1. The highest yields in 1988 (a severe drought year) came with the maximum rainfall applications including: 1) an increase of 40% to all rains (69.3 bushels per acre) as top ranked; 2) an increase of 40% to the 1 inch rains which ranked second; and 3) an increase of 40% to the rains of 0.1 to 1.0 level, which ranked third. The results from 1988 show a good linear relationship of increased yields to increased rainfall (Fig. 1). The primary non-linear anomalies in 1988 were the two applications of additional rainfall only to the 1-inch or heavier rains. These two values are denoted on figure 1. showing they represent a relatively greater increase
in yields than the amount of added rainfall received would indicate, as based on the other 8 outcomes.
Examination of the 1989 corn yield results (Fig. 1) when the natural rainfall was near average and all rain additions thus ran above average, reveals a different outcome. The highest values came with the additional water applied to 1 -inch or nearer values, as shown in figure 1, and with the increase of 25% to all rains which ranks as the second highest yield outcome (Table 1). The results for corn yields and rainfall in 1989 suggest that the application of 15.5 inches of rain, the value for the natural rainfall plus 40% more water, was too much rainfall. It has long been recognized that the prairie soils of the Midwest can receive too much rain for optimum corn growth, and the data for 1989 suggests that this maximum was somewhere around 14 inches, recognizing that the temperatures of 1989 were near normal, and the effect of the distribution of the daily rainfall events during June-August.
In general, the results from the two years of rain simulations indicate a moderately strong relationship of corn yields to the summer rainfall, with a seemingly linear increase up to approximately 14 inches of rain during the June-August period. Undoubtedly the distribution of the daily rainfall events had an effect, but the results do allow one to estimate the value of a rainfall increase on corn yields. For example, in the hot
59
_
dry summer of 1988, an inch of rainfall, say from 6 to 7 inches, appeared to increase corn yields by about 10 bu/acre. In 1989, the addition of 1 inch of rainfall, say from 11 to 12 inches, appeared to provide approximately 5 additional bu/acre.
A third interesting aspect of the outcomes of both years relates to the yield values obtained with the rainfall increases (10% and 40%) applied only to the one inch and heavier rain days. In both years, the results indicate relatively great yield increases, as shown in figure 1. Tests 8 and 9 resulted in relatively higher corn yields than the data on total rain for the other 8 trials would have predicted. This suggests that the capability to increase rainfall on days when the rain is relatively heavy would be of considerable value. In 1989 there were two such days of heavy rains over an inch, one on June 23 and another one on June 26. In 1988. the drought year, there was only one day, on July 19. Rain in late June and July are very critical to increased corn yields.
Soybeans. The soybean yields associated with the natural rainfall and the 9 tests of increased rain are shown for the three years in Table 1. In 1987. an extremely wet year, the highest yields came with the actual rainfall and the 40% increase only to the light rains (less than 0.1 inch). As the rainfall was increased, the yields decreased such that the maximum possible rainfall increase, 40% to all rain, which produced a total of nearly 25 inches of rain, resulted in the lowest soybean yield attained, 23.7 bushels per acre. Any rainfall increase applied in a summer as wet as 1987 is
not desirable for soybeans. There is no explanation available for the odd outcome achieved with the 25% increase in 1989.
Inspection of the 1988 soybean results reveal a very different outcome. In this hot, dry summer, the peak soybean yields of 20.8 bu/acre, came with the heaviest summer rainfall. 40% of water added to all summer rain days. The second greatest yield increase came with 40% added to all rain days of 0.1 to 1.0 inch per day. As shown in figure 2, the soybean yield outcomes were a strong linear relationship with the amount of summer rainfall, ranging from a low of 15 bu/acre with 5.2 inches of rain, up to nearly 21 bu/acre with 7 inches. The apparent increase in soybean yields with one inch of rain in such a summer was approximately 4 bu/acre.
The soybean yields in 1989, when there was near average precipitation and temperatures, also varied considerably, but were much higher than in 1987 (too wet) or 1988 (too dry). The peak yield in 1989 (55 bu/acre) came with the 40% increase to only the 1 inch rain days, with the next two highest yields achieved with the 25% and 40% increase to all rains. The distribution of the yield outcomes, as revealed in figure 2, suggests a curvilinear relationship yields with the summer rainfall. Drawn on figure 2 is a curve drawn to approximately fit the data points for the three years. The data points suggest that at about 14 inches of rainfall, a reversal occurred with a decrease in yields as the summer rains became heavier; note that the total rain with 40% increases, 15.5 inches, had a lower yield
Figure 1. The relationship of total summer rainfall in 1988 and 1989 with com yields at agricultural plots in central Illinois. The 9 rain increases
were based on 9 simulated rainfall increases.
60
Figure 2. The relationship of total summer rainfall in 1987, 1988, and 1989 with soybean yields at agricultural plots in central Illinois. The 9
rain increases were based on 9 simulated rainfall increases.
than did the 14 inches based on a 25% increase in the summer rain. The 1989 data points for summer amounts less than 14 inches suggests that a 1-inch rain increase, say from 11 to 12 inches, resulted in about 3 bu/acre increase in yields.
3. SUMMARY Inspection of the corn and soybean outcomes
with the simulated rain increases in three summers reveals certain common findings. First, increases of rainfall on only days when light rains occurred 0.1 inch or less, produced no perceptible changes in yield in either crop. Thus, a modification capability limited to such days has no agricultural value. Second, given the limited data sample and given the temperatures and conditions in these three years, there is an indication that summer rainfall amounts in excess of 14 inches leads to decreases in both corn and soybean yields. Prior research established that the prairie soils of central Illinois can contain only so much water and thereafter ponding and saturated soils produce detrimental effects to both crops. A third finding for both crops (given this limited sample) is that given that natural rainfalls are less than 14 inches, a 40% increase
in all rain events was the best possible outcome tested. That is, a 40% increase to all rain events increases the yields of both crops more than any other type of simulated increase tested. Although not unexpected, this outcome confirms past modeling studies.
The results for both crops suggest that the relative value of 1 inch of rain (or any amount of rainfall) is much greater when the summer rainfall is low. For example, 1 inch of rain when summer totals is between 5 and 7 inches produced a 10 bu/acre increase in corn, whereas with summer total rainfall between 11 and 13 inches, the impact of an additional inch of rain was approximately 5 bu/acre to corn. Soybean responses to an inch of rain (given the summer total was <14 inches) were less than with corn. In the hot, dry summer of 1988, an additional inch of rainfall produced approximately 4 additional bu/acre of soybeans, whereas in the wetter near normal summer of 1989, the increase was about 3 bu/acre. Every additional inch of rain above a summer total of 17.8 inches (1987) produced about a 3 to 4 bu/acre decrease to 20 inches; thereafter, the decrease in yields with additional rainfall became much greater.
61
The results for both corn and soybeans suggest the following guidelines relative to weather modification planning. First, increases in rainfall on days when a 0.1 inch or less is produced appear to be of negligible value to crop yields. Summer rain events that produce <0.1 inch are typically of short duration, 15 minutes or less at a point, and are often followed by clearing and a rapid return to pre-rain temperatures. Hence, we conclude that most of this light rainfall is evapotranspirated and very little enters the root zone of corn and soybeans. However, increases in rain on days when rainfall is more than 0.1 inch show positive yield benefits, and the greater the increase, the greater the crop yield benefit, at least up to totals that are too much for soils to handle, apparently about 14 inches for the summer (with near normal temperatures). Furthermore, for the corn production, results suggest that increases of 10% or more to days when 1 inch or more of rain occurs are particularly beneficial to crop yields.
Acknowledgements. This work was conducted under the PACE program. NOAA grant COMM NA89 RAH09086. The work of Gene Ziegler in conducting the operation of the agricultural plots and in the collection of crop data is deeply appreciated.
REFERENCES
Changnon, S.A., 1986: Illinois weather modification program: PACE. Proceedings Tenth Conference on Weather Modification. AMS. Boston, 315-319.
Changnon, S.A., and S.E. Hollinger. 1988: Use of unique field facilities to simulate effects of enhanced rainfall on crop production. J. Wea. Mod.. 20. 60-66.
Changnon, S.A., Hollinger, S.A., and P. Garcia. 1989: Analyzing the effects of additional rainfall on corn and soybean yields. Proceedings Sixth Conference on Applied Climatology. AMS, Boston, 6 pp.
Garcia. P.. Offutt. S.. and M. Pinar. 1986: Technological Advance. Weather, and the Potential Economic Benefits of Weather Modification. Proc. Tenth Conf. on Wea. Mod.. AMS, Boston, 285-290.
62
Conf. on Cloud Physics July 23-27, 1990 San Francisco, CA
OBSERVED VERSUS DIAGNOSED ICE PRODUCTION RATES
IN WARM-BASED MIDWESTERN CUMULI
Robert R. Czys
Illinois State Water Survey Champaign, Illinois
1. INTRODUCTION
Attention to the fact that "first" ice particles are found in clouds at temperatures warmer and concentrations higher than expected from heterogeneous nucleation alone dates back many years (Palmer, 1949; Brewer and Palmer, 1949). First ice particle concentrations in excess of that expected from ice nuclei have been reported for clouds from all over the world; cumulus in England (Murgatroyd and Garrod 1960), Australia (Mossop et al. 1972), South Africa (Krauss et al. 1987), and in the United States in cumuli over Missouri (Koenig 1963), Florida (Hallett et al. 1978), West Texas (Jurica and Scro 1985), and Central Indiana (Czys 1989).
One mechanism often indicated to explain the discrepancy between first ice and ice nuclei is that of splinter production during riming. This mechanism was demonstrated in laboratory experiments that showed the ejection of secondary ice crystals when supercooled droplets froze onto an ice substrate (Hallett and Mossop 1974). Subsequent laboratory work indicated that certain physical criteria must be met for the process to operate. For example, the process is temperature dependent in that ice splinters are produced at temperatures between -3 and -8°C, with peak production at -5°C (Hallett and Mossop 1974). Droplets larger than 25µm (Mossop and Hallett 1974) and droplets smaller than approximately 13µm (Mossop 1978) must be present. The rate of ice splinter production has been found to be proportional to the rate of sweep out of droplets larger than 25µm (Mossop 1976) and has been found to be sensitive to the velocity of the riming body (Mossop 1976). Finally, the rate of splinter production has been found to be influenced by the surface temperature of the riming body (Heymsfield and Mossop 1984; Foster and Hallett 1982).
Mossop (1978) has provided a relationship for calculating ice production rates. This expression is, P - aNLNs
m, where P is the ice splinter production rate, NL is the accretion rate for droplets > 25µm , Ns the accretion rate for droplets < 13µm and a and m are constants relating experiment conditions such as temperature and riming rate. However, because this equation is inconvenient to use with aircraft data, the practice has developed to suspect that a rime-splintering process has operated in a cloud whenever higher than expected ice concentrations are found and the in-cloud conditions meet the criteria for rime-splintering. Recently Harris-Hobbs and Cooper (1987) introduced an expression for diagnosing ice production rates by rime-splintering that not only embodies many of the criteria for rime-splintering, but is also fairly convenient to use with microphysical measurements.
This paper reports on the results of a comparison between observed and diagnosed ice production rates for a small sample of updrafts in warm-based Midwestern cumulus congestus.
2. DATA
Data used in this study were collected as part of the 1986 field program of the Precipitation Augmentation for Crops Experiment (PACE). The overall goals and objectives of PACE have been described by Changnon (1986). In-cloud measurements Were made using a light twin engine airplane (Beechcraft Baron) equipped to measure standard meteorological state parameters, such as, temperature, pressure and dew point as well as microphysical instrumentation such as a FSSP, and 2DC and 2DP probes. Vertical air velocity was estimated using a method discussed by Lawson (1979).
Data were collected on 8 July 1986 near the tops of eight cumulus congestus clouds feeding a main storm system located just southwest of Indianapolis, Indiana. Penetrations were made near the -10°C level directly at the center of cloud with no more than 5000 and no less than 1000 ft of cloud top above the level of cloud entry. Data used in this study are only from the updraft portions of each penetration. In all, 27 updrafts were identified, where an updraft was defined as any period of in-cloud measurement having at least 3 consecutive seconds of positive velocity in excess of 1 m s-1. Thus, the data are largely representative of the result of adiabatic and collection processes in vertical ascent with other influences minimized.
3. DIAGNOSIS OF ICE PRODUCTION RATES
Secondary ice production rates were diagnosed using an expression developed by Harris-Hobbs and Cooper (1987). The equation has the form:
(1)
where R g , N g ( R g ) , and v(Rg) are equivalent radius, size spectrum, and fall speed for graupel, respectively. The variables r, n(r), and v(r) are respectively, cloud droplet size, size spectrum, and fall speed. Fall speed for graupel and droplets were calculated using relationships provided by Heymsfield (1978) and Beard (1976) , respectively. The collection efficiency E(Rg,r) is that for collisions between graupel and cloud droplets and was taken as unity in the calculations for Midwestern clouds presented here. The constant C was given as 0.16 in Harris-Hobbs and Cooper (1987) and was determined by fitting Eq. 1 to the laboratory data of Mossop (1978).
Examination of Eq. 1 clearly reveals that it takes into account many of the sensitivities and dependencies of the rime-splintering process. For example, temperature dependence is accounted for by the function f(T) which is unity at -5°C and linearly decreases to zero at -3 and -8°C, respectively. Droplet size dependence is taken into account by the function g(r) which is defined as:
(2)
which account for the large and small droplet dependence in addition to the sensitivity of the process to the fraction of the graupel surface covered by droplets less than 13µm (G<13) to the total cloud droplet population (Ga11). Finally, the dependencies of sweep out rate and velocity of the riming body is also evident in Eq. 1.
4. RESULTS
Figure 1 is a plot of observed ice production rates versus that diagnosed for rime-splintering process using Eq. 1. Ice production rates (IPR's) for Midwestern clouds are denoted by solid circles. Horizontal and vertical lines passing through the solid circles are representative 70% confidence intervals based on the Poisson statistics discussed by Cornford (1967). Confidence intervals are based solely on the uncertainty of measuring ice concentration and do not take into account measurement uncertainties associated with, for example, measuring distributions of supercooled cloud droplets with an FSSP in the presence of ice (Gardiner and Hallett, 1985). To avoid cluttering Fig. 1 approximately half of the IPR's were chosen randomly to have their confidence intervals plotted. Diagonal lines indicate IPR ratios of 1:1, 1:10, and 1:100 in favor of diagnosed ratios.
For comparison, diagnosed and observed IPR's for Florida cumuli as reported by Harris-Hobbs and Cooper (1987) are plotted in Fig. 1 as open circles. Confidnece intervals for these data were not provided in their paper. To spread out the cluster of data over the largest possible area three of the Florida data do not appear in Fig. 1. These data points are (2.2,.002), (1.6,.002), and (.15,.002).
As can be seen in Fig. 1, observed and diagnosed IPR's for Midwestern and Florida cumuli scatter over the same general area. This level of agreement is not surprising since the characteristics of clouds from either region have many similarities and because the method of calculation used on the Midwestern data is very similar to the Florida data thanks to private communication with Harris-Hobbs. An important similarity is that the distribution of convective cloud-base temperatures for the Midwest is close to that for Florida (Johnson, 1982). Furthermore, the range of main updraft velocities is also similar (Czys, 1988; Sax and Keller, 1980). Consequently, clouds from both regions have active coalescence processes that tend to produce drizzle and rain drops before cloud top
Figure 1. Observed versus diagnosed ice production races.
reaches the 0°C Isotherm (Hallett et al. 1978; Braham, 1964). Cloud droplet populations tend to be greater in Florida than in Midwestern cumulus, but clouds in both regions generally possess many droplets less than 13µm, as well as, droplets greater than 24pm diameter. Most importantly, clouds from either region develop ice in excess of that expected from the action of ice nuclei alone. For example, the observed ice production rates plotted in Fig. 1 for Midwestern cumuli correspond to initial graupel concentrations which range from 1 to about 15 roughly 10 to 100 times the activity of ice nuclei at -10°C (Fletcher, 1962).
Close inspection of Fig. 1 reveals a tendency for rates diagnosed for Midwestern updrafts to be 5 to 10 times higher than observed, and in a few instances diagnosed rates are about 100 times larger than observed. Of the 23 Midwestern updrafts 13 (=60%) have diagnosed rates at least five times greater than observed. Furthermore, the occurrence of diagnosed rates in excess of that observed appears to have occurred by more than can be explained by measurement uncertainty, even in light of the fact that confidence intervals are generally not as small
as desirable. It is also interesting to note that both Midwestern and Florida IPR's show a few updrafts with very low observed rates and very high diagnosed rates, about 1:100. This size of discrepancy suggests that some updrafts can have very favorable conditions for rime-splintering, but do not necessarily have the highest observed rates of ice production.
Thus, even though the ice in Midwestern clouds tends to originate in concentrations greater than expected from ice nuclei, the results presented in Fig. 1 suggests that in many cases observed rates of production are much less than expected than if a rime-splintering process was operating at full potential. In support of this finding, mean observed and diagnosed IPR's for updrafts in Midwest cumuli are 35 and 101 m-3 s-1, respectively. The discrepancy is slightly larger for the Florida data with 23 and 179 m-3 s-1 for mean observed and diagnosed rates, respectively (for the Florida data means include the three data points not plotted in Fig. 1). Thus, rates diagnosed for the Midwest sample of updrafts are about 3 time larger than observed while diagnosed rates for the Florida sample are about 8 times larger than observed.
5. DISCUSSION
The results of this study suggests that even though initial ice concentrations in Midwestern updrafts are larger than expected from ice nuclei, the rate of ice production in many cases is substantially less than expected from a rime-splintering process. This is an unexpected result because in most updrafts, the criteria for a rime-splintering process are met: sufficient numbers of droplets smaller than 13µm and larger than 24µm exist, updraft velocities in comparison to droplet and graupel fall speed are about right, and supercooled drizzle and rain drops are usually present. It is unclear as to why a rime-splintering process either did not operate or operated below potential in many of the updrafts in this sample. The discrepancy between observed and diagnosed IPR's can not be completely explained away in terms of measurement uncertainty even though the uncertainty is larger than desirable. Furthermore, the diagnostic expression used appears to take into account most of the physical criteria for the rime-splintering process. The expression was said to produce good agreement between laboratory data, and was even shown to indicate observed rates in excess of diagnosed for measurements taken in clouds over Montana (Harris-Hobbs and Cooper, 1987).
The fact that results for Midwestern updrafts (and perhaps those for Florida) differ from those for clouds in Montana might rest in a fundamental microphysical difference between the clouds; Midwestern (and Florida) clouds tend to contain supercooled drizzle and rain drops in their updrafts while the cold-based clouds of Montana generally do not. If the rime-splintering process is indeed sensitive to the surface temperature of the riming body, then the occasional collection of supercooled drizzle or rain drop might be suspected of resulting in the suppression of splinter production by increasing the surface temperature from the release of latent heat as the liquid of the collected drop freezes. Furthermore, when rime-splintering operates in the absence of supercooled drizzle and rain drops, the rate of splinter production is proportional to the fraction of small to large droplets that cover the graupel surface. When the process operates in the presence of supercooled drizzle and rain drops, the structure of frail ice on the graupel surface may be upset when a supercooled drizzle or rain drop is intercepted and coats the surface. Therefore, the presence of supercooled drizzle and rain drops may not necessary be advantageous to a rime-splintering process although the presence of these hydrometeors are closely associated with the onset of ice (Brahara, 1964).
conditions at -10°C appear to be favorable. Another factor that will lead to over estimating rime-splintering rates is that graupel are less numerous and smaller in the ice multiplication zone than they are when observed at -10°C. Lastly, the results apply to measurements taken on the same day for a relatively small sample of clouds. The question thus arises as to whether similar results would be obtained in other samples.
Acknowledgments. This research was supported by the Precipitation Augmentation for Crops Experiment under NOAA cooperative agreements COMM NA87RAH07077, COMM NA88RAH08107 and COMM NA89RAH09086 and by the National Science Foundation under Grant NSF ATM-8819751. The author wishes to thank Norm Ostrander (Pilot) and Don Stone (Flight Engineer) for their support in the 1986 PACE field program.
REFERENCES
Beard, K.V., 1976: Terminal velocity and shape of cloud and precipitation drops aloft. J. Atmos. Sci., 33, 851-864.
Braham, R.R., 1964: What is the role of ice in summer rain-showers? J. Atmos. Scl., 21, 640-645.
Brewer, A.W., and H.P. Palmer, 1949: Condensation processes at low temperatures and the production of new sublimation nuclei by the splintering of ice. Nature, 220, 687-689.
Changnon, S.A., 1986: Illinois weather modification program: PACE. Preprints Conf. on Weather Modification, Arlington, VA, 315-319.
Cornford, S.G., 1967: Sampling errors in measurements of raindrop and cloud droplet concentrations. Met. Mag., London, 96, 271-282.
Czys, R.R., 1989: Ice initiation by collisional forcing in warm-based cumuli. J. Appl. Met., 28, 1098-1104.
Finally, caution must be taken in generally applying these findings. First, because it was necessary to assume that the conditions at -10°C were representative of the conditions for rime-splintering in the updrafts as they passed through the ice multiplication zone. Clearly, subtle differences between the population of supercooled cloud droplets observed at -10*C and the population that existed in the ice multiplication zone is one factor that can lead to lesser rime-splintering rates even though
Czys, R.R., 1988: Microphysical characteristics of warm-based cumuli: Observations at -10°C. Proceeding of the 10th International Cloud Physics Conference, Bad Homburg, Federal Republic of Germany.
Foster, T., and J. Hallett, 1982: A laboratory investigation of the influence of liquid water content on the temperature dependence of secondary ice crystal production during soft hail growth. Preprints, Cloud Phys. Conf., Chicago, III., 123-126.
Hallett, J., R.I. Sax, D. Lamb, and A.S. Ramachandra Murty, 1978: Aircraft measurements of ice in Florida cumuli. Quart. J. R. Met. Soc, 104, 631-651.
Hallett, J., and S.C. Mossop, 1974: Production of secondary ice crystals during the riming process. Nature, 249, 26-28.
Harris-Hobbs. R.L., and W.A. Cooper. 1987: Field evidence supporting quantitative predictions of secondary ice production rates. J. ACmos. Sci., 44, 1071-1082.
Heymsfield, A.J., 1978: The characteristics of graupel particles in northeastern Colorado cumulus congestus clouds. J. Atoms. Sci., 35, 284-295.
Heymsfield, A.J., and S.C. Mossop, 1984: Temperature dependence of secondary ice crystal production during soft hail growth by riming. Quart. J. R. Met. Soc, 110, 765-770.
Johnson, D.B. , 1982: Geographical variations in cloud-base temperature. Preprints AMS Conf. on Cloud Physics, Chicago, IL, 187-189.
Jurica, G.M., and K.D. Scro, 1985: Microphysical properties of cumulus congestus clouds observed during Texas HIPLEX. Pro. Fourth WHO Conf. on Weather Modification, Honolulu, World Meteor. Org., 363-368.
Koenig, L. R., 1963: The glaciating behavior of small cumulonimbus clouds. J. Atmos Sci., 20, 29-47.
Krauss, T.W., R.T. Bruintjes, and J. Verlinde, 1987: Microphysical and radar observations of seeded and nonseeded continental cumulus clouds. J. Climate Appl. Meteor., 26, 585-606
Lawson, R.P., 1979: A system for airborne measurement of vertical air velocity. J. Appl. Met., 18, 1363-1368.
Mossop, S.C, 1978: The influence of drop size distribution on the production of secondary ice particles during graupel growth. Quart. J. R. Met. Soc, 104, 323-330.
Mossop, .S.C, 1976: Production of secondary ice particles during the growth of graupel by riming. Quart. J. R. Met. Soc, 102, 45-57.
Mossop, S.C, and J. Hallett, 1974: Ice crystal concentrations in cumulus clouds: Influence of drop spectrum. Science, 186, 632-634.
Mossop, S.C, R.E. Cottis, and B.M. Bartlett, 1972: Ice crystal concentrations in cumulus and stratocumulus clouds. Quart. J. Roy. Meteor. Soc, 98, 105-123.
Murgatroyd, R.J., and M.P. Garrod, 1960: Observations of precipitation elements in cumulus clouds. Quart. J. Roy. Meteor. Soc, 86, 167-175.
Palmer, H.P., 1949: Natural ice-particle nuclei. Quart. J. Roy. Meteor. Soc, 75, 15-22.
Sax, R.I., and V.W. Keller, 1980: Water-ice and Water-Updraft relationships near -10°C within populations of Florida cumuli. J. Appl. Met., 19, 505-514.
1 6 t h Conf . on S e v e r e Loca S t o r m s , O c t . 2 2 - 2 6 , 1990 K a n a n a s k i s , P r o v i n c i a l P a A l t a . , Canada
SINGLE DOPPLER RADAR OBSERVATIONS OF A MINI-TORNADO
Patrick C Kennedy, Nancy E. Westcott, and Robert W. Scott
Illinois State Water Survey Champaign, IL 61820
1. INTRODUCTION
A primary motivation for the installation of a national network of Doppler weather radars (NEXRAD) is to improve tornado detection capabilities and to increase warning lead times. The realization of these benefits is crucially dependent upon the correct recognition of tornado - related rotation signatures in the radial velocity data. Thunderstorm research results from the southern Plains indicate that the development of a mid-level mesocyclone and its subsequent descent often precede the touchdown of a major tornado (JDOP Staff, 1979). However, important variations to this genesis pattern exist. Wakimoto and Wilson (1989) present evidence that the interaction of a developing thunderstorm with pre-existing shear instability in the boundary layer can produce a tornado circulation that grows upwards. In addition, outflow from mature storms has been found to be capable of generating tornadoes along the edges of gust fronts and downbursts (Forbes and Wakimoto, 1983). Thus, the task of radar recognition of tornadic storms is made difficult by the variety of velocity patterns "that these storms may present. The purpose of this paper is to expand the spectrum of Doppler radar - observed tornadic thunderstorms through the documentation of an unusually small tornadic thunderstorm and the synoptic setting in which it occurred.
2. CASE DESCRIPTION
2.1 The Synoptic Setting
Several periods of thunderstorms, some of which were severe, occurred in Illinois beginning in the late morning hours of 19 May 1989. The tornado described in this paper touched down at 2007 (all times arc CDT) near the town of Bement, 30 km southwest of the CHILL radar site at the University of Illinois Willard Airport, Champaign (CMI). At 1900 a general southerly surface wind flow
prevailed in Illinois ahead of a moderate cold front which was advancing across the central Plains. A wind shift occurred in west central Illinois due to a trough line embedded in the large scale surface flow (Fig la). Aloft, the trough line also was well defined at the 850 and 500 mb levels (Fig lb and c). Based on the 1900 NGM 500 mb initialization (not shown), the absolute vorticity along this trough line reached 20xl0-5 s-1 at a center just northwest of CMI. A pronounced mesoscale comma pattern associated with this vorticity maximum was present in the 1931 visible satellite image (Fig 2), with deep convection in an arc from western Indiana to southern Illinois. Despite this highly organized cloud distribution, the constant pressure analyses showed no significant cold air regions aloft along the trough at either 850 or 500 mb.
The horizontal distribution of thermodynamic instability at 1900 was related to the satellite - observed cloud pattern. Near the arc of deep convection in southern Indiana moderate instability was present in the National Weather Service (NWS) Paducah, Kentucky (PAH) sounding (not shown). The lifted index for PAH was -3 and the K index was 36. In the tornado environment behind the convective line, less instability was present in the Peoria, Illinois (PIA) sounding where the lifted and K indices were 0 and 30 respectively (Fig. 3). Except in the vicinity of a stable layer near 550 mb, the PIA sounding was moist. The wind in this sounding veered through 100 degrees between the surface and 500 mb, but the speeds in this layer averaged less than 10 ms-1. The average radar • determined storm motion was from 260 degrees at 8.4 ms-1.
2.2 The Tornado
Based on NWS reports, the tornado was preceded by a funnel cloud sighted at 1912 near the north side of Decatur (DEC). The tornado touched down at 2007 just south of Bement and apparently dissipated approximately 25 minutes later. All eyewitness accounts indicate that both the tornado's diameter and forward speed remained small throughout it's life. Virtually all of the 15 km path crossed
Figure 1. Regional analyses at 1900 CDT on 5/19/89. a) Surface data; solid contours are sea level pressures in mb. At station locations, temperature is plotted above dew point (both in degrees F); missing values are zeroes. Wind speeds are in knots. Geographical locations referred to in the text are marked. b) 850 mb data; solid contours are heights in dm, dashed contours are temperatures (C). At station locations temperatures (C) are plotted and wind speeds are in knots. c) 500 mb data; as in (b).
Figure 2. GOES 1 km resolution visible image at 1931 CDT 5/19/89. Tornadic storm cloud is marked with an arrow; the range ring is 150 km from the CHILL radar.
open farmland; some power lines were downed and F0 to near Fl damage was done to the few structures that were affected.
3. DOPPLER RADAR OBSERVATIONS
Throughout the afternoon and evening hours of May 19 the Illinois State Water Survey / National Science Foundation CHILL radar was collecting data in support of the PACE 89 research project (Changnon et al., 1990). The operating characteristics of the CHILL are summarized in Table 1. During the tornado event, the radar was conducting a series of sequential 360 degree PPI scans at one degree elevation steps up to the echo top. The time required to complete a volume scan increased from 2 to 6 minutes as the storm approached the radar.
Figure 3. Skew - T log P plot of NWS PIA 1900 CDT 5/19/89 sounding. Only single moist adiabat and saturation mixing ratio lines are shown. Wind speeds are in knots.
Table 1 : CHILL Radar Operating Characteristics on 5/19/89
Wavelength (cm) 10.7 Half Power Beamwidth (Deg) 0.96 Peak Transmitted Power (dBM) 88.07 Min Detectable Signal (dBM) -110.29 Range Resolution (m) 300 Unambiguous Velocity (ms-3) 26.42 Pulses per Integration Cycle 50
The size of the tornadic echo remained remarkably small in all of the PPI data. The low level area enclosed by the 40 dBZ contour averaged only some 6 km in diameter (Fig. 4a). In most of the scans the horizontal reflectivity gradient maximized around the western end of the echo. During some sweeps when the tornado was in progress, a small hook pattern was evident in the higher reflectivity contours (see Fig. 4a). Maximum echo top height as defined by the 15 dBZ reflectivity level stayed in the 5 - 6 km AGL range (see upper portion Fig. 5). This height was in the vicinity of the stable layer in the sounding (see Fig. 3). The azimuthal variation of radial velocity in the PPI data showed evidence of a persistent cyclonic circulation in the parent echo (Fig. 4b). The velocity change between adjacent - beam gates at the same range increased from 15 to 20 ms-1 as the storm approached the radar and the tornado developed. An overview of the vortex history is shown in a time - height cross section of the maximum azimuthal radial velocity shear normalized to 25 km range (Fig. 5). Rotation developed near the midlevel of the echo and descended with time. The DEC funnel cloud sighting at 1912 occurred when an early shear maximum appeared aloft; no reports were found relating to the second period
of shear intensification aloft around 1930. The reported tornado activity occurred in association with the descent of the 40xl0-3 s-1 shear envelope to near - surface heights that began just after 2000.
a) Reflectivity field; lowest contour is 20 dBZ with a 10 dBZ interval.
b) Individual range gate radial velocities (ms-1). An estimated environmental flow from 240 degrees at 10 ms-1
has been subtracted; negative velocities are towards the radar. Gate spacing is 300 m in range and .88 degrees in azimuth. Figure 4. CHILL radar data from 2.4 degree elevation PPI sweep at 2022 CDT on 5/19/89. In both panels the heavy contours in the tornadic echo are 50 dBZ, and the range ring increment is 5 km.
Figure 5. Time • height plot of maximum radial velocity shear between adjacent • beam range gates. Units: inverse seconds x 1000 normalized to 25 km range. Areas where shear exceeds 40 x 10-3) inverse seconds are stippled. Echo top heights are indicated by asterisks.
4. DISCUSSION AND CONCLUSIONS
From a forecasting perspective, the synoptic scale patterns associated with the Bement tornado were not those of a classic spring season midwest tornado day. The PIA sounding showed only weak instability and both the strength and directional curvature of the wind profile were modest. The NWS severe weather forecasters recognized the marginal nature of the situation. During the afternoon hours Illinois was placed in a slight risk area for severe thunderstorms, due to the interaction of the passing 500 mb vorticity lobe and diurnal heating. At 1650 the Severe Local Storms Mesoscale Discussion was directed towards the developing line of thunderstorms along the Illinois -Indiana border; no tornado watch was in effect when the Bement tornado occurred.
The synoptic setting was also dissimilar from those noted in cases of "cold air funnels". Coolcy (1978) reported that these funnels generally occurred several hundred kilometers to the rear of surface cold fronts beneath areas of significantly cold air aloft. Neither of these characteristics applied to this storm environment.
The most prominent map feature associated with the tornado was the well developed 500 mb vorticity center located over north central Illinois at 1900. In this regard, the environmental combination of "strong dynamics and weak thermodynamics" was similar to that found in another marginal severe weather case described by Moller and Ely (1985).
In addition to the somewhat unusual synoptic setting, some of the radar characteristics of the Bement storm were also noteworthy. The parent echo was quite small in size; both the height of the maximum echo top and the diameter of the low level 40 dBZ contour were approximately 6 km. Due to this limited size, the echo appeared insignificant when the full range (150 km) of the CHILL was displayed.
Within this small parent echo the general pattern of radial velocity shear development was similar to that observed in the southern Plains (JDOP staff, 1979) in that the shear first intensified aloft and then approached the surface prior to the tornado. However, in keeping with the limited stature of the echo, the "mid-level" circulation in this case intensified near 3 km AGL in contrast to the 5 - 6 km mid-tropospheric heights noted in the southern Plains storms. The shear magnitudes were approximately half those found in strong southern Plains tornadoes; the 40 - 55 x10-3 s-1 adjacent gate shears were more similar the intensities observed in Colorado events (Wakimoto and Wilson, 1989).
While far from being a major event, this tornado illustrates some of the challenges involved in efforts to issue warnings for all tornadoes based on automatic detection. Neither the synoptic setting nor the full range surveillance view of the parent echo appeared threatening. Due to the limited depth of the echo, Doppler detection of the circulation pattern at long ranges could easily be compromised by beam overshooting. Even if properly resolved, the small vertical extent of the echo might cause some mesocyclone identification algorithms to reject such a feature. Thus, full utilization of spotter reports and magnified radar display capabilities will continue to be important even after the implementation of automated Doppler - based severe weather detection algorithms.
Acknowledgments:
Dr. E. A. Mueller of the Illinois State Water Survey provided consultation in the analysis of the CHILL radar data. Information shared by personnel from the University of Illinois Department of Atmospheric Science is also acknowledged: Bill Chapman recounted his tornado intercept observations and Harold Brooks summarized the results of his damage survey. John Brother Jr. provided assistance in the drafting of the figures. This research was supported by the CHILL radar NSF grant ATM 83-20095 and by the NOAA Precipitation Augmentation for Crops Experiment (PACE) grant COMM NA89RAH07077.
References:
Changnon, S. A.. R. R. Czys. R. W. Scott, and N. E. Westcott, 1990: The Illinois Precipitation Modification Program. Submitted to the Bult of the Amer. Meteor. Soc.
Coolcy, J. R.. 1978: Cold air funnel clouds. Mon. Wca. Rev.. 106. 1368-1372.
Forbes, G.S. & R.M. Wakimoto, 1983: A concentrated outbreak of tornadoes, downbursts and microbursts, and implications re-garding vortex classification., Mon. Wca. Review. 111. 220-235.
JDOP Staff, 1979: Final report on the Joint Doppler Operational Project (JDOP) 1926-1978. NOAA Tech. Memo. ERL NSSL-86. Norman, Oklahoma, 84pp.
Moller, A. and G. Ely, 1985: On the operational problem of warning for the anomalous severe weather event: the Dallas county tornado of December 13. 1984. Preprints, 14 AMS Conference on Severe Local Storms, Boston, p346-349.
Wakimoto, R. M. and J. W. Wilson, 1989: Non-supercetl tornadoes. Mon. Wca. Rev., 117. 1113-1140.
16th Conf. on Severe Local Storms, Oct. 22-26, 1990 Kananaskls P r o v i n c i a l Pa rk , A l t a . , Canada
ARMORED AIRCRAFT OBSERVATIONS OF A SEVERE HAILSTORM IN ILLINOIS
Dennis J. Musil and Paul L. Smith Nancy E. Wescott
Institute of Atmospheric Sciences I l l inois State Water Survey South Dakota School of Mines and Technology 2204 Griffith Drive
501 East St. Joseph Street Champaign, I l l inois 61820 Rapid City, South Dakota 57701-3995
1. INTRODUCTION Penetrations were made with the armored
T-28 research aircraft (Johnson and Smith, 1980; Detwiler, 1990) into a developing severe Illinois hailstorm that occurred on 25 May 1989. These penetrations were made as part of an investigation into the evolution of convective storms during the Precipitation Augmentation for Crops Experiment (PACE) conducted by the Illinois State Water Survey (ISWS) during the spring of 1989 (Changnon et al., 1989). The study area for the project includes approximately a 100-mile radius from Champaign, Illinois (Fig. 1).
The purpose of the T-28 investigations in PACE was to obtain a data set from penetrations of cloud regions not normally accessible to conventional research aircraft. Aircraft data emphasizing microphysical, thermodynamic, kinematic, and electrification measurements were obtained by the T-28 on 25 May. The T-28 data are being analyzed in conjunction with ISWS/NSF CHILL 10-cm radar data, as well as radiosonde and standard weather data. The results provide valuable information about the internal features of this storm, as well as the applicable precipitation mechanism. Furthermore, the aircraft observations were the first ever made inside an Illinois severe storm as a result of a planned penetration.
2. DESCRIPTION OF THE T-28 SYSTEM The aircraft is equipped for measurement
of a variety of cloud microphysics and state variables, as well as navigation and aircraft performance variables. Most of the instrumentation is discussed in Johnson and Smith (1980). Included in the instrumentation configuration for the first time in 1989 were four electric field mills borrowed from the New Mexico Institute of Mining and Technology and the National Center for Atmospheric Research.
A new PC-based data acquisition system was installed early in 1989. The basic recording rate is once per second, but the option exists to sample selected variables at higher frequencies. The pilot can enter event codes and has the capability to re-boot the data system in flight should a failure occur.
3. STORM DESCRIPTION On 25 May, three mesocyclones developed and
passed through east central Illinois, leading to
Fig. 1: Project study area for PACE.
numerous reports of damaging strong winds and large hail. The storms formed in a moist atmosphere; dewpoints in the project area were about 18°C in advance of a cold front moving through the area. Aloft, a fast-moving trough was approaching the area and estimates of the instability showed a lifted index of about -6°C. Cloud base height and temperature were estimated from a computation of the CCL from a CLASS sounding at 1224 CDT as ~2 km MSL and 14°C, respectively,
3.1 T-28 Penetrations The storm penetrated by the T-28 was just
reaching severe levels, but had radar reflectivities that exceeded the usual T-28 penetration limit of 55 dBz (Fig. 2a). At the time of the T-28 penetrations, it was oriented southwest-northeast and was moving from about 240° at about 20 m s-l.
Two penetrations of the storm were made. They were accomplished at an altitude near 4.1 km MSL (~3.9 km AGL), which corresponded to a temperature around 0°C. During the first penetration (not shown here), the aircraft was traveling
Fig. 2: Radar data from ISWS/NSF CHILL 10-cm radar on 25 May 1989. (a) CAPPI display at 4 km showing the track of the-T-28 from 154806 (N) -155047 (X) - 155347 (T). (b) Vertical cross section along north-south line indicated in 2a. Contours at 10-dB intervals labeled in dBz. Scan volume times are given in 2a.
southwest several kilometers in advance of the line, while the second penetration was in a northeasterly direction through reflectivities near 60 dBz (Fig. 2a). Several equipment problems developed during this penetration, including questionable position keeping. Thus, a decision was made to return to base for safety reasons.
The location of the aircraft was obtained using the Research Aircraft Tracking System (RATS). Transponder data from an FAA radar were telemetered to the CHILL for real-time display and recording Test flights in early May indicated that the RATS positions were within 1-2 km of the true aircraft position. 3.2 Radar Summary
The storm penetrated by the T-28 formed to the south of, and was connected to, another SW/NE oriented line (Fig. 2a). The motion of the entire storm complex was toward the east at about 20 m s-1. The penetrated storm sustained a large area of reflectivities greater than 55 dBz for more than 30 min and was moving at about 20 m s-1 toward the ENE. During this period, the storm traveled over a rural area where no surface reports were made. Approximately 30 min after the southern line was penetrated, a new system which | spawned damaging hail and winds formed within this same broad area of echo.
The high-reflectivity region was strongly sloped with height toward the south-southeast, especially in the lowest 3 km, and extended to about 6 km AGL; reflectivities > 40 dBz extended to about 8 km (Fig. 2b). The single-Doppler data indicated a vertically-oriented updraft (to -8 km)
to be co-located with the high-reflectivity area above about 2 km. 4. T-28 OBSERVATIONS
The first penetration was about 10 km east of the most intense part of the storm and exhibited only weak microphyslcal and dynamical characteristics. The most interesting data were found in the second penetration; the track passed through the highest reflectivity present at flight level. Figure 3 shows plots of several cloud variables along the entire second penetration (nearly 50 km in length), while various characteristics of the precipitation-sized hydrometeors are given in Fig. 4. The latter were derived from the foil impactor carried on the T-28 and analyzed laccording to a numerical method described by Peterson et al. (1990).
The penetration encountered one rather well-developed updraft region (1549-1551 in the plots) with moderate speeds of about 10 m s-1; it was indicated to be co-located with the high-reflectivity region shown in Fig. 2. Since the aircraft travels at a speed of about 100 m s-1, this updraft region was about 12 km long. Another weakly organized updraft region was located further to the northeast (~1553) where the aircraft passed ~10 km east of a second, weaker reflectivity maximum in the storm.
The cloud droplet concentrations were about 400 cm-3 in the strongest updraft and about 500 cm-3 in the secondary updraft, although the cloud liquid water concentration (LWC) as determined from a Forward Scattering Spectrometer Probe (FSSP) was substantially higher (maximum about 1.5 g m-3) in the stronger updraft. Values of LWC from a Johnson-Williams hot-wire device were about 30% higher in this same region until a hailstone knocked the wire loose just after 1550 COT.
Fig; 3: Time history of selected variables for Penetration 2 on 25 May 1989: (A) cloud droplet concentration from PMS FSSP in cm-3; (B) reverse flow temperature in °C; (C) LWC from FSSP in g m-3 (D) vertical winds from T-28 measurements in m s-1 (numbers represent the regions of sample 2D-C images in Fig. 5): and (E) turbulence intensity (c1/3) in cm2/3 s-1.
Fig. 4: Time history of precipitation particle characteristics from foil data for Penetration 2 on 25 May 1989. Upper panel: mass concentration, in g m-3; middle panel: maximum diameter in mm; lower panel: number concentration of particles > 1 mm (bold) and number concentration of particles > 5 mm (thin).
Direct measurement of cloud base conditions was not made on the cloud penetrated. However, a parcel ascent based on the CCL leads to an adia-batic LWC of ~3.5 g m-3 at the penetration level. , This indicates that the T-28 observations of cloud LWC. in the main updraft were about 40% of adiabatic. That is quite similar to other T-28 measurements in the southeastern part of the U.S. (Musil and Smith, 1989) and in the High Plains (Musil et al., 1990). It also agrees with values found by Ackerman et al. (1979) in actively growing cumuli penetrated near 0°C. As they indicated, the observations suggest that a substantial amount of the cloud water, even at this elevation, had been converted into precipitation-sized particles through coalescence. An interesting feature of the T-28 observations here is that such values were found over a large region (~6-8 km) of the primary updraft. The reduced LWC values in the eastern part of the updraft, where substantial amounts of precipitation and large particles were , also found (see below), are reminiscent of the structure of hailstorm updrafts in other regions ' (e.g., Musil et al., 1990).
The turbulence encountered was only moderate by T-28 standards, despite the presence of such high radar reflectivities and hail. The peak values of turbulence intensity near the southwest edge of each updraft region were about 10 cm2/3 s-1. The primary updraft showed turbulence peaks near both edges. The values in the interior regions of the primary updraft were rela-tively small compared to the edges, a phenomenon often found in T-28 penetrations of storms in the High Plains (Musil et al., 1986). Other secondary peaks were found throughout the penetration.
Particles in precipitation sizes, as determined from foil impactor measurements, revealed high precipitation mass and number concentrations on either edge of the strongest updraft region (Fig. 4). A broad region of high concentrations was located in the downwind portion of that updraft, and narrow maxima occurred on
the upwind edge (Fig. 4 ) . There is a suggestion that the hydrometeor data would better match the radar data if the aircraft track was displaced ~2 km further SE. This would help explain the relative minimum of large particles between 1549-1550. These uncertainties are still being examined.
Peak concentrations of particles > 1 mm (N1) and > 5 mm (N5) exceeded about 500 m-3 and 60 m"3, respectively. The corresponding peak mass concentrations were about- 10 g m - 3
; These valueS are similar to those found by Musil and Smith (1989) in storms in the southeastern part of the United States. The relative amounts of the large particles in ice and liquid phase are unknown, because it is not possible to determine phase conclusively with foil impactor data.
The ~40-s period of high number and mass concentrations between 1550-1551 in Fig. 4 appears to be related to the diminished values of cloud LWC (due to accretional growth?) and vertical •velocity (due to precipitation loading?) in the downwind region of the updraft (Fig. 3). Reduced values of LWC in regions of moderate updraft are fairly common in T-28 observations of hailstorms (Musil et al., 1990).
Maximum precipitation particle sizes according to the foil device were of the order of several mm, with overall maximum sizes about 10 mm. Past observations with a hall sensor (which was inoperative on this day) have shown that, due to sample volume considerations, particles about twice as large as those found on the foil are likely to be present. The occurrence of such larger particles would be consistent with the radar observations and was also indicated by some of the dents in unprotected parts of the aircraft that were noted following the flight.
The Particle Measuring Systems 2D-C probe carried on the T-28 provides information about the sizes and habits of hydrometeors < 1 mm or so. Even though it was not functioning properly during the flight on 25 May, enough good images were present to provide an idea of the particle habits (Fig. 5). In the main updraft region, which also included the region of highest LWC, broken images (indicative of high concentrations of liquid) and some raindrops were the rule (Panel 2). Small ice particles in the form of graupel and irregular
Fig. 5: Sample 2D-C images near the regions numbered in Panel D of Fig. 3.
shapes were evident near the edges of the main updraft region (prior to 154830 and after 155050, Panels 1 and 3 of the figure), where the 2D-C happened to be most reliable. The ice particles on the northeast side of the updraft tended to be larger than in other regions (see also Fig. 4). The secondary updraft region further to the northeast was characterized primarily by a variety of ice particles, including graupel, broken, and irregular particles (Panel 4). Note that some of these were small enough to be carried upward in the updraft, suggesting the likelihood of a recirculation process in this region.
Particle size and mass distributions from . the 2D-C data are still under investigation. However, the malfunctioning 20-C probe may not provide an adequate sample of hydrometeors for a detailed analysis. 5. DISCUSSION
The presence of a relatively broad region of moderate updrafts containing high concentrations of large hydrometeors in conjunction with relatively large amounts of cloud liquid water suggests an efficient development of precipitation. The high concentrations of precipitation-sized hydrometeors and sub-adiabatic cloud LWC suggest that the cloud water is rapidly converting to rainwater (and probably hail at higher levels) faster than cloud water is formed in the updrafts.
The observations in this storm are similar in some respects to those obtained in storms in the southeastern part of the United States (Musil and Smith, 198g), where a coalescence process was found to be active (Tuttle et al., 1989). Raindrops were found in the main updraft at the 0°C level. Thus, it appears that coalescence is an important growth mechanism in the Illinois hailstorm as well. Precipitation appears to develop directly in the updraft, which at least partly overlapped the high-reflectivity region of this storm. There was at least 2 km thickness of warm cloud where coalescence could proceed before reaching the 0°C level. The observed moderate updrafts seemingly allow a sufficient amount of time for the process to occur.
The diminished values of LWC in the regions of greatest concentration of precipitation par-ticles in the downwind portion of the main updraft also suggest that these large particles play a strong role in depletion of the cloud water. In addition to growth by coalescence, it is likely that accretion is occurring by loading of water onto graupel. That is particularly likely along the updraft edges, where larger graupel particles are falling through the weaker updrafts. Additional depletion of cloud water by mixing (as indicated by the stronger turbulence) is also occurring there.
This storm clearly produced hail of at least moderate sizes. The T-28 observations suggest the downwind side of the main updraft to be a likely region Tor major hail growth. This characteristic shows up in some storms of the High Plains (e.g.. Musil et al., 1990). However, unlike the usual High Plains storms, this Illinois storm shows some precipitation developing in the middle of the updraft. This indicates that embryo
sources and other details of the hail growth mechanisms in the two regions may be different.
Cloud bases are usually much higher in High Plains storms (Fankhauser and Wade, 1982), typically at a temperature of ~5-10°C. Stronger updrafts are also common in the High Plains storms. The strong updrafts tend to limit the conversion from cloud water to precipitation 1n the main updrafts of those storms. Therefore, one usually has to look elsewhere for plausible sources of hailstone embryos.
The sources of hailstone embryos are a major uncertainty concerning hailstorm processes. The main updrafts of High Plains storms provide little opportunity for embryo development. In this Illinois storm, however, potential embryos were found right in the main updraft. Whether those particles in fact participated as embryos probably cannot be determined from the available data. Further study is needed to deal with such questions, and clearly one should not attempt to generalize from the observations on a single storm, however interesting.
Acknowledgments. This research was supported under the NOAA Federal/State Cooperative Program in Atmospheric Modification Research, under Purchase Order No. BPO 190,423 from the University of Nevada Desert Research Institute. The T-28 facility operation is supported under NSF Cooperative Agreement No. ATM-8620145.
REFERENCES Ackerman, B., R. G. Grosh and R. Y. Sun, 1979: Assessing
midwest cloud characteristics for weather modification. ISWS Contract Report No. 216. 170 pp.
Changnon. S. A.. R. R. Czys, F. A. Huff. E. A. Mueller. J. B. Nespor, R. W. Scott and N. E. Wescott, 1989: The Precipitation Augmentation for Crops Experiment - 1989: Phase II: Exploratory Phase. Operations Report prepared by Illinois State Water Survey. 54 pp.
Detwller, A. G.. 1990: Armored T-28 Research Aircraft Facility. Bulletin 90-1. Institute of Atmospheric Sciences, S.O. School of Nines and Technology, Rapid City, SO. 22 pp.
Fankhauser, J. C, and C. G. Wade, 1982: Hailstorms of the central High Plains, (ed., C. A. Knight and P. Squires), Vol. I, The National Hall Research Experiment, 5-33.
Johnson, G. N., and P. L. Smith, Jr., 1980: Meteorological instrumentation system on the T-28 thunderstorm research aircraft. Bull. Amer. Heteor. Soc. 61. 972-979.
Musil, D. J., and P. L. Smith, 1989: Interior characteristics at mid-levels of thunderstorms in the southeastern United States. Atmos. Res., 24_, 149-167. , A. J. Heymsfield and P. I. Smith, 1986:
Microphysical characteristics of a well-developed weak echo region in an intense High Plains thunderstorm. J. Climate Appl. Meteor., 25, 1037-1051. , S. A. Christopher, R. A. Deola and P. L. Smith,
1990: Some interior observations of southeastern Montana hailstorms. [Submitted to J. Appl. Meteor.]
Peterson, B. A., D. J. Musil and P. L. Smith. 1990: Computerized reduction of airborne foil impactor data from C0HMEX thunderstorms. Preprints Conf. Cloud Physics, 23-27 Jul 1990, San Francisco, CA, Amer. Heteor. Soc, —.
Tuttle, J. D., V. N. Bringi, H. D. Orvllle and F. J. Kopp. 1989: Multiparameter radar study of a microburst: Comparison with model results. J. Atmos. Sci., 46, 601-620. i
I2th Conference on Weather Analysis and Forecasting, Oct. 2-6, 1989, Monterey, CA
Pl.4
CONVECTION INTENSITY AND SEEDING SUITABILITY FORECASTS IN A DYNAMIC SEEDING EXPERIMENT IN ILLINOIS
Robert W. Scott and Robert R. Czye Climate and Meteorology Section
Illinois State Hater Survey Champaign, IL 61820
1. INTRODUCTION Objective forecasting procedures
were developed to assist In the prediction of convectlve cloud elements considered suitable for treating during a dynamic seeding field experiment. The effort stems from post-experimental analyses of data collected during the initial field stage of the Precipitation Augmentation for Crops Experiment (PACE) conducted across east-central Illinois during the summer of 1986 (Changnon, 1986). A wide variety of forecasting tools were on hand (ochs and Kidder, 1988) including: 1) satellite transmission of standard meteorological charts (NAFAX) from the National Meteorological Center and substantial digital and worded data via the FAA604 and Public Products circuits, 2) half-hourly visible GOES satellite imagery at 1-km resolution obtained via a telephone hookup to the Man-Computer Interactive Data Access System, re-mapped on an IBM-AT and displayed on an IBM Professional Graphics Display monitor (Kidder and Ochs, 1987), and 3) local rawinsonde releases using the Cross-chain Loran Atmospheric Sounding System, a semi-automated sounding facility of the National Center for Atmospheric Research.
Due to the speed of data acquisition and reduction, items such as stability index calculations, a modified version of the 1-D Great Plains Cumulus Model (obtained from the Bureau of Reclamation; Hirsch, 1971) and computer-generated plots such as thermodynamic diagrams and hourly surface charts of sea-level pressure, streamlines and divergence over the Midwest were produced in near real-time to assist In both the forecasting and nowcasting duties.
Post-experimental research objectives were to improve forecasts of convectlve activity from studies in the following areas: (1) stability index prediction and analysis, (2) suitability of the environmental airmass for a favorable dynamic response to seeding,
and (3) suitability of ln-cloud conditions for dynamic seeding to be effective.
2. STABILITY INDICES Results reported by Scott and Huff
(1987) from the July - August 1986 PACE field project indicate that at least 5 stability indices calculated from 1200 GMT data produce positive indications of precipitation occurrence between 1700 -0200 GMT from 70* - 80* of the time. Slightly greater success was observed in the prediction of no rain. In addition, on days when all 5 indices gave indication of precipitation, the maximum radar echo tops seen from local NWS radars generally exceeded 9.1 km (30,000 ft) with half of these occurrences exceeding 15.2 km (50,000 ft). Similarly, when no or just one index Indicated rainfall, tops were usually under 6.1 km (20,000 ft) or, more often than not, rain did not occur.
3. SUITABILITY OF THE AMBIENT AIRMASS Environmental airmass suitability
studies developed an experimental procedure for use in future weather modification field programs which may objectively predict the operational status of each day (Go, No-Go and Stand-By) by relating certain thermodynamic parameters aloft to maximum radar echo tops. A scatter diagram was constructed of the potential buoyancy of the atmosphere (PB, defined as the difference between the environmental temperature at 500 mb and that of a parcel raised moist adlabatlcally from the convectlve condensation level, CCL) verses the temperature of cloud base (taken as the CCL temperature, CCLT) obtained from the local NWS 1200 GMT rawinsonde soundings during the summer months of 1986 and 1987. In addition, each point on the plot was categorized by values of the daily maximum radar echo top observed
107
from nearby NWS radar sites during the 1700 - 0200 GMT period. When viewed in this way, the organization of the data suggest that days with cloud base temperatures ranging between 14° - 22° C and potential buoyancy between 2° - 9° C are also days which tend to produce the tallest clouds (Fig. 1). Coding on the figure is as follows: 0 = no echoes, 1 = echoes with tops below 6.1 km, 2 = tops of 6.1 - 9.1 km, 3 = tops of 9.1 - 12.2 km, 4 = tops of 12.2 - 15.2 km, and 5 = echoes with tops exceeding 15.2 km.
Probability forecasts on the occurrence of no-echo, small (height < 6.7 km), medium (6.7 km < height < 12.8 km), and large (height > 12.8 km) echoes were produced by discreetizing values of maximum echo height within the CCLT and PB domain (Fig. 2). Consider first, days with precipitation. As both CCLT and PB increase, so do the probabilities for the medium to large maximum echo heights. Days with no echoes predominate on days with relatively cold CCLT and negative PB, however, moderately nigh probabilities were found with much warmer CCL_ and positive PB. It appears that this is due, in part, to the lack of an adequate triggering mechanism or a change in airmass between the morning sounding and afternoon radar observation period on those days.
4. IN-CL0UD SUITABILITY A relationship originally
developed by Mather et al. (1986) is that CCLT and PB can be used to discriminate between clouds which have an active coalescence process (production of supercooled drops > 300 urn by the time clouds reach the -10 C level) and those that do not. The relationship is:
L = 8.6 - CCLT + 1.7 PB where L is termed the drop coalescence discriminator. When L is negative, the production of large drops is expected. L is shown graphically as the diagonal line in Fig. 1.
The relationship was developed over the eastern Transvaal of South Africa and it is unknown if the same relationship exists in the midwestern U.S. However, because the distribution of cloud base temperatures for the Midwest is similar to that for South Africa, we believe that the equation is a good first estimate to test in the field. Furthermore, from three days during PACE 1986 for which we have cloud physics data, it appears that L is a good discriminator. The more negative L was, the generally more abundant and larger the size of supercooled water drops that were observed. This is important for operational programs In that we may now have a way, which needs testing, to
Fig. 1. Thermodynamic parameters stratified by maximum radar echo height.
108
Fig. 2. Probabilitie* of occurrence in percent of 4 categoriei of convection, a) none, b) light, c) moderate, and d) heavy.
anticipate in-cloud properties, and objectively distinguish between days that are and are not suitable for dynamic seeding. Field testing of all of the above forecasting techniques will be conducted during the summer of 1989 in the second field program of PACE.
5. ACKNOWLEDGEMENTS This research was supported by
grant COMM-NABBRAH08107 via the Federal Grants to the States Program and administered by NOAA under a cooperative agreement.
6. REFERENCES Changnon, S.A., 1987: Precipitation
Augmentation for Crops Experiment: Phase II, Exploratory Research, Year I. Contract Report 430, Illinois State Water Survey, Champaign, IL.
Hirsch, J., 1971: Computer Modeling of Cumulus Clouds during Project Cloud Catcher. Inst. Atmos. Scl., S. Dak. Sch. of Mines Rep. No. 71-7, 61 pp. [NTIS PB-281028/AS].
Kidder, S.Q. and H.T. Ochs, III, 1987: A low-cost system for the acquisition and display of GOES satellite Images. B. Amer. Meteorol. Soc, 41, 8 pp.
Mather, G.K., B.J. Morrison, and G.M. Morgan, Jr., 1986: A preliminary assessment of the Importance of coalescence in convectlve clouds of eastern Transvaal. J. Clim. Appl. Meteor., 25. 1780-1784.
Ochs, H.T., III and S.Q. Kidder, 1989: A low-cost forecasting/nowcasting system for remote field locations. J. Atmos. and Ocean Technol., 6, 218-221.
Scott, R.W. and F.A. Huff, 1987: PACE 1986 forecasting program - design, operations and assessment. Proceedings of the 11th AMS Conf. on Wea. Mod., Edmonton, Alta., Canada.
110