AMU Tasking Report
Science & Operations Officer
Applied Meteorology Unit / National Weather Service
Melbourne, FL
June 1997
2.1 Installation and Proof of Concept Test (PoCT)
2.2 Workstation Assessment
2.3 Algorithm Performance Assessment
2.3.1 Hail Detection Algorithm
2.3.2 Damaging Downburst Prediction and Detection Algorithm
2.3.3 Mesocyclone and Tornado Detection Algorithms
2.4 Integrations of Lightning (NLDN) Data
2.5 Supportive Activity
3.0 Summary
4.0 Considerations for the Potential Transition to Operations
5.0 Acknowledgments
6.0 Appendix
A Memorandum of Understanding (MOU) and subsequent professional
relationship has been formed between the National Severe Storms Laboratory
(NSSL) and the interagency weather community that supports space shuttle
launches,landings, and ground operations. The purpose of this relationship has been to evaluate the
utility of NSSL's Warning Decision Support System (WDSS) for potential
additive radar support to the nation's space program. Ultimately, and ideally,
output would be made available to operational personnel at the 45th Weather
Squadron (45th WS), the Space Flight Meteorology Group (SMG), and the
National Weather Service (NWS) in Melbourne (MLB).
A large percentage of WDSS tasking obligations (one-year) was given to
the Applied Meteorology Unit (AMU) at NWS MLB. The purpose of this
arrangement was mainly due to the collocation of the WSR-88D radar with the
office. The remainder of WDSS tasking obligations was given to the AMU at
the National Aeronautics and Space Administration (NASA). Installation of the
WDSS at NWS MLB took place during July 1996 followed by an intensive 30-day Proof of Concept Test (PoCT) during the month of August. NSSL
representatives remained on site during the PoCT. Hardware procurement
included a Sun-Sparc 20 (equivalent) and peripherals (external 8mm tape device,
external hard-drive). This workstation served as a central processor, receiving
base data from an additional wideband port (W3) and supplying post-processed
data to a pre-existing HP/UNIX network throughout the MLB operations area.
Informal reports previously delivered by the AMU/MLB include
incremental PoCT reports (08/13/96 and 08/21/96), progress reports and
correspondence (11/1/96, 11/13/96, 12/1/96, and 03/13/97), and a mid-point
tactical adjustment "Plan of Intent" (01/97) for the remainder of the contractual
year (internal - see Appendix A). During the year, successive NSSL reports
were also received (see NSSL Final Report - 1 May 1997) and subsequently
reviewed. A prioritized list of "Change Requests" and "Requested New
Features" was submitted to NSSL (see Appendix B) to be considered for
software upgrade. Several of these items have been included in a pending
software release which will be delivered by NSSL during the summer of 1997.
During the course of the one-year contractual arrangement, AMU/MLB
personnel evaluated the utility of the system within the east central Florida
environment. Impressions of workstation software design and output integrity
were addressed under various weather situations. The Hail Detection Algorithm
(HDA) and Damaging Downburst Prediction and Detection Algorithm (DDPDA)
were identified for NSSL optimization for the warm/wet season (May-Sept).
Discernment of large hail and damaging microbursts, during this time of year,
was determined by AMU/MLB to be the most challenging for area forecasters
with watch/warning/advisory responsibility and where artificial "decision
support" could provide the greatest benefit. The AMU/MLB also evaluated the
Mesocyclone Detection Algorithm (MDA) and Tornado Detection Algorithm
(TDA) for performance during high-shear situations (see Appendix C). During the year,
representatives from the 45th WS, SMG, and AMU/KSC visited MLB for WDSS
orientation and real-time experience.
As of the date of this publication, the integration of cloud-to-ground
lightning (NLDN) data has not successfully taken place. Technical problems
regarding hardware communications between the WDSS and the local NLDN
computer have hampered meaningful progress. MLB has strongly encouraged
NSSL to solve this problem by the end of the contractual period (end of June
1997). Therefore, the evaluation of such can not yet be addressed.
Finally, the AMU/MLB engaged in various local support activities and addressed concerns regarding the potential transition of WDSS into operations. Transitional issues were determined first for MLB due to collocation and then considered for potential remote implementation. Efforts in localized instruction, proficiency, and configuration optimization were made in the areas of end-to-end system recovery, system administration functions, data archiving and retrieval, POP-Ups and ALARMS, and map backgrounds. A list of transitional issues was also compiled for potential remote implementation. All decisions pertaining to potential remote implementation will be addressed by respective agency authorities.
The operational evaluation of the WDSS was tasked to the Applied
Meteorology Unit at NWS MLB. It was the intention and duty of each
AMU/MLB meteorologist to use the system and to document the overall utility
of WDSS in the east central Florida environment. The stability of the system
was assessed, the layout and usefulness of displayed data was considered, and
algorithm performance was examined. Direct interaction with NSSL
representatives was encouraged and occurred frequently. Matters regarding
system recovery, system administration, archiving, hardcopy generation,
operator training and proficiency, and site optimization were also addressed.
AMU/MLB meteorologists engaged in a 30-day Proof of Concept Test with on
site NSSL meteorologists during the month of August (1996) and then gained
solo experience with the WDSS during the remainder of the one-year
arrangement under various weather situations.
2.1 Installation and Proof of Concept Test (PoCT) - Since the WSR-88D
is collocated at NWS MLB, the WDSS hardware was installed at that
location during July 1996. The WDSS is comprised of three separate
parts:
1. RIDDS (Radar Interface and Data Distribution Streams) resides
on a Sun Sparc 5 (or equivalent) located in the Radar Data
Acquistion (RDA) shelter. RIDDS receives base data via
wideband 3, multiporting the throughput for as many as eight
external users.
Procurement of the RIDDS was not necessary for this project. NWS MLB had previously installed the RIDDS for projects
involving the Integrated Terminal Weather System and the Tropical Rainfall Measuring Mission.
2. RUDDS (Radar Utilities and Doppler Data Streams) resides
on a Sun Sparc 20 (or equivalent) located within the main-
frame computer room of the office. RUDDS receives the
real-time data stream from the RIDDS and then processes
the data through the various NSSL algorithms. Data from
the Rapid Update Cycle (RUC) is also ingested by RUDDS
and used for the near-storm environment algorithm for hail
detection. Changes to algorithm adaptable parameters and
pop-up boxes (alarms) and other user-adjustable parameters
are performed at this location.
Procurement of the RUDDS hardware (and peripherals) was necessary for this project.
3. RADS (Radar Analysis and Display System) resides on an
HP755 and is networked to the remaining HP workstations within the office. RADS allows the users to view,
manipulate, and analyze radar products and data tables. RADS is the operator-level interface with WDSS.
Since the HP network at NWS MLB was pre-existing, procurement of RADS hardware was not necessary for this project.
Operators gained experience during the PoCT, utilizing algorithm output
during various severe weather events. They were quickly impressed with the
amount of available data, especially individual storm parameter trend
information and tabular Cell/Mesocyclone/Tornado information. Storms were
also automatically "ranked" according to their prioritized potential severity.
Consistently, WDSS added value to current operational radar display
information obtained from the Principal User Processor (PUP) of the WSR-88D.
WDSS output was considered more helpful in certain situations given the
presence of NSSL's Damaging Downburst Prediction and Detection Algorithm
(DDPDA). Although not yet optimized for the east central Florida environment,
the DDPDA did provide clues to the threat of damaging microburst winds. The
PoCT encompassed several days of severe weather over east central Florida with
damaging winds specifically noted at PAFB on 01 August and again on 13
August 1997. The 13 August event prompted an internal case-study
investigation due to the amount of damage at PAFB. Performance of the WDSS
during this event was included in the report (see related internal AMU
Memorandum). (click here for assorted WDSS images for 13 August 1997)
Generally, NSSL meteorologists remained on site during the PoCT while
NWS operators gained experience. Incremental impressions were supplied by
both NSSL and the NWS during the 30-day exercise. NSSL also complied with
a request for multiple short-seminars regarding "algorithm-thinking" for a deeper
understanding of WDSS output.
2.2 Workstation Assessment - The WDSS was evaluated from an
operational workstation perspective. Primarily, this was a function of
data display and manipulation at the RADS level and the effectiveness of
having non-competitive direct access (more timely output) to a more
powerful central processor (RUDDS). Operational meteorologists were
tasked to evaluate the system during various severe weather situations for
total "decision support" and value-added aspects over the current display
system (PUP).
In general, the WDSS was found to be superior to the current
operational data display system. Since the output was supplied to MLB's
HP/UNIX network, radar data was available throughout the operations
area and not confined to just one workstation (PUP). Output can be
displayed within multiple opened "windows" on a single monitor.
Unfortunately, however, these windows can not be re-sized (at times
tremendous amounts of data may be squeezed into a smaller than desired
display window). On the other hand, simple hardware improvements,
such as screen resolution and processing power, add to the effectiveness
of WDSS. During the course of the year, operators continually identified
two features on WDSS with high operational value. Firstly, the WDSS
will display reflectivity data, even when the radar itself is in precipitation
mode, which is less than 5 dBZ. This is essential for finding/tracking
subtle features such as fine lines, cloud elements, and boundaries. SMG
has a high interest in identifying areas of convection initiation as well as
cloud tracking. For this reason, the display format is more desirable than
the current system. Secondly, the WDSS will keep track of the historical
trends of a host of individual thunderstorm related parameters (for
examples...see Appendix C). This function would be extremely difficult,
if not impossible, for a human operator. Parameter information with
respect to individual cells, mesocyclones, and tornadoes is available
within both trend sets and tables. Storms are "ranked" according to their
potential severity to focus the operators attention on certain storms.
These two features offered the forecasters a high degree of operational
satisfaction.
Other positive features which added value to the task of evaluating
radar echo return include multiple image looping (more than one loop at
a time), faster looping capability, improved panel display (more than 4
panels), access to raw data values, and hodographs.
Several negative aspects of WDSS were also encountered. The
RADS display software was found to be intensive with respect to
workstation resources. It was very difficult to run other screen intensive
programs simultaneously. This was thought to be related to color
allotment which would thus require a statement of minimum hardware
specifications (i.e. graphics card). The current WDSS does not offer
cross-sectional capabilities or constant altitude plan position indicator-type (CAPPI) products. These products are considered operationally
necessary to satisfy the diverse needs of the WSR-88D/KMLB user
community for space operations support. Also, a plan-view of Vertically
Integrated Liquid (VIL) would be desirable. The RADS auto-update
feature for receiving real-time data from the RUDDS should be defaulted
to the "ON" position and should catch the user's attention whenever it is
switched to the "OFF" position to avoid delay/cancellation of critical
information.
2.3 Algorithm Performance Assessment - It should be noted that although
the WDSS is fitted with improved algorithms, it can not overcome the
physical limitations inherent with radar systems. Limitations such as
beam sampling, cone-of-silence, and anomalous propagation/clutter still
have adverse impacts on algorithm performance. Nonetheless, an
attractive aspect of the WDSS was access to the latest NSSL-research
algorithms. Improved Mesocyclone and Tornado Detection Algorithms
(MDA & TDA respectively) were found to be more robust than the
current operational algorithms. These algorithms show better
performance and future promise, especially for the detection of
mesocyclones of smaller physical dimensions (i.e. associated with tropical
cyclone outer rainbands) and tornadoes which are not associated with
supercells (i.e. landspouts and waterspouts). The highest operational
interest was with the Hail Detection Algorithm (HDA) and the Damaging
Downburst Prediction and Detection Algorithm (DDPDA). Currently, a
microburst/downburst algorithm is not available among the suite of
operational WSR-88D algorithms (1997).
2.3.1 Hail Detection Algorithm (HDA) - During the PoCT, it was
theorized that the HDA may be "over-warning" for summer time
thunderstorm situations. This impression was based upon the lack
of ground-truth verification during times when hail size (HAIL)
was determined to be 1.0" or larger and the chance for severe hail
(SVRH) was greater than 50 percent. From an operational
perspective, a decision support system for large hail that possessed
a high probability of detection but along with a high false alarm
rate could have negative impacts. With much of the HDA
displayable output based on "percentage-likelihood", operators
were forced to weigh the data (the "odds") to make
warning/advisory decisions. An identified storm with hail,
determined by the HDA, to be larger (and at times much larger)
than 1.0" (with better than even odds) would be highly regarded by
the operator as a severe storm candidate. Since severe hail is
threshold defined (3/4"), an incorrect determination of hail size
(over-sizing) would likely affect the probability of severe hail and
thus lead to the issuance of unverified weather warnings/advisories.
As a result, MLB recommended that the HDA be optimized by
NSSL for the warm/wet season. Previously archived base data
could be used. It was felt that the HDA statistical storm database
might be biased towards Southern Plains-type events. A storm
database unique with peninsular Florida severe hail events during
the warm/wet season would be desired. Also, relationships
between stone size and the height of the melting level during the
Florida summer should be examined.
An area of further intrigue involved the automatic updating
of specific adaptable parameters for use by the HDA. Information
regarding the height of the 0o C and -20o C levels was
automatically updated every three hours according to output from
the Rapid Update Cycle (RUC). Since the same algorithm was
also implemented on the operational system (RPG/PUP),
forecasters were able to compare output based on manual
parameter input from the latest Cape Canaveral sounding vs. the
automated parameter input from the RUC. No obvious differences
were observed. Automated algorithm parameter maintenance
schemes should be expanded within this algorithm and with the
other main algorithms.
2.3.2 Damaging Downburst Prediction and Detection Algorithm
(DDPDA) - Also during the PoCT, the Damaging Downburst
Prediction and Detection Algorithm (DDPDA) was identified for
optimization for the central Florida warm/wet season. Currently,
there is no operational version of a microburst algorithm among the
suite of WSR-88D algorithms. Output was deemed useful for
identifying areas of predicted/detected moderate or severe
downbursts, especially in terms of potential. However, it was
theorized that the DDPDA may also be excessive in its indication
of downburst activity. Interestingly, problems associated with
clutter suppression were revealed by NSSL during this exercise.
Seemingly, there is a difference in the way clutter suppression is
applied for elevations 2.4o and higher (batch cuts). Clutter
suppression regions (CSR) applied by NWS MLB operators may
have further aggravated the problem in some cases. NWS MLB
has an aggressive CSR policy for invoking suppression and is
taking a closer look at an equally aggressive policy for CSR
removal.
2.3.3 Mesocyclone (MDA) and Tornado (TDA) Detection
Algorithms - MLB also evaluated the performance of the MDA
and TDA algorithms during cool-season / higher-shear situations.
The MDA was found to be superior to the current mesocyclone
algorithm and very good in overall performance. However, it did
tend to be oversensitive when identifying cyclonic shear features.
The TDA was also considered superior to the current tornado
algorithm since it allowed for high shear tornadic-scale features to
be identified without being associated with a
mesocyclone/supercell. However, the TDA still needs
improvement as it tended to give false alarms apparently due to
transient "noise" in the velocity field. Quality control checks need
to be improved. Operators had higher confidence in the
MESOTVS than the TVS.
The specific study can be found in Appendix C. It concerns two separate severe weather events (29 March and 23 April 1997) and a null case (22 April 1997).
2.4 Integration of Lightning (NLDN) Data - As of the date of this
publication, the integration of cloud-to-ground lightning (NLDN) data has
not successfully taken place. Technical problems regarding hardware
communications between the WDSS and the local NLDN computer have
hampered meaningful progress. MLB has strongly encouraged NSSL to
solve this problem by the end of the contractual period (end of June
1997). Therefore, the evaluation of such can not yet be addressed.
2.5 Supportive Activity - The AMU/MLB has also engaged in peripheral
supportive activity. This activity involved end-to-end assessment for
complete operation of the WDSS. In particular, this activity included:
The WDSS has proven to be superior to the current radar
processor/display system. Operators have identified two features which offer a
testament to its worth. The first feature is its skill in individual thunderstorm
parameter accounting for identifying potential severity. The second is the ability
to display reflectivity data below 5 dBZ when in precipitation mode.
The WDSS provides access to NSSL's latest research algorithms. These
algorithms have been proven to be more robust than current operational
algorithms (RPG/PUP) which will not be updated (or ported) for several years.
While robust, they are still not yet ideal. Currently, there is not a
microburst/downburst algorithm available on the operational WSR-88D system.
Interestingly, this was a weather hazard that was identified as difficult for
forecasters to issue timely watches/warnings/advisories, especially during the
summer. Having access to such an algorithm through WDSS is beneficial.
WDSS also offers the ability to site/situation optimize algorithms to the east
central Florida environment according to specific agency/user requirements
(however, since the source of the base data remains constant, inherent physical
radar limitations do not change). The WDSS boasts a more powerful central
processor, resulting in timely weather products to its dedicated/associated users.
The WDSS promotes the concept of local control of such matters without being
tied to network radar obligations.
NSSL has accepted a list of "Change Requests" and "Requested New
Features" and has responded with a reasonable (itemized) reply for integration
into the next version of WDSS software to be delivered in June 1997. Certain
upgrades include cross-sections and an "Auto Update" flag.
Finally, NSSL is confident that they will be able to integrate NLDN
lightning data into the local WDSS by the end of the contractual one-year period
(June 1997). The fruits of such integrated data sets remain to be determined.
The AMU/MLB is in favor of implementing the WDSS into 45th WS
operations and SMG operations for additive radar support. However, since the
WDSS is considered Research & Development software, it would only be
supported by NSSL as such. Therefore, it could not replace the current WSR-88D display (PUP), but rather complement it. The suggested physical location
of the WDSS would be adjacent to the PUP for that reason.
If operational transition were to take place at one or both locations, the
following items must be addressed:
a) Large-volume base data connection from the WSR-88D/KMLB radar to
either SMG or the 45th WS (or both).
b) Data Display platform for RADS display (SMG/45th WS) and RUDDS.
c) Unit Radar Committee meeting to decide local WDSS support activity (as well as wideband 3 & RIDDS policy).
d) Available technical support from NSSL during implementation phase.
e) Operator Training with respect to both system-related aspects and meteorologically-related aspects.
An alternative longer-term consideration might involve the AMU (KSC
& MLB) engaging in the identification of software requirements for the USAF open-architecture PUP (OPUP) through WDSS experiences in accordance with 45th WS
operational needs.
The author would like to thank each of the AMU/MLB Meteorologists (Scott Spratt, Steve Hodanish, Dan Petersen, Randy Lascody, Peggy Glitto, and Tony Cristaldi) for their thorough day-to-day operational evaluations. Appreciation is extended to both Roger Willis and Jim Lane for their local systems support. Insightful comments were often provided by Mark Wheeler and Winnie Lambert of the AMU/KSC and their collaborative spirit is also appreciated. Bill Conway and Pam MacKeen of NSSL are recognized for WDSS field support. A special thanks goes to Bart Hagemeyer and the remainder of NWS/MLB, Tom Adang and the 45th WS, Frank Brody and SMG, Frank Merceret and the AMU/KSC, and John Madura and Jack Ernst of NASA, for their overall vision and endorsement of this project.
Appendix A - Internal Plan of Intent (mid-term adjustment)
Appendix B - Submitted Change Requests & Requested New Features
Appendix C - Cool Season Analysis of Algorithms
Internal Plan of Intent
This plan has been drafted to ensure the effective evaluation of the WDSS workstation by the
AMU at NWS MLB for the remainder of the contractual year (thru June '97) with the National
Severe Storms Laboratory. The WDSS is being evaluated for the purpose of potential transfer into
the operations at the 45th Weather Squadron and Space Flight Meteorology Group in support of the
nation's space program.
Local Project Manager: Dave Sharp/SOO
The SOO will continue comprehensive oversight of the project, locally. He will remain the
primary Point of Contact and liaison with NASA, 45th WS, SMG, and NSSL. It will be his
responsibility to purpose the efforts of the AMU at MLB to obtain the most effective evaluation of
the entire WDSS system on behalf of all agencies. It will be his responsibility to document any
progress, problems, and results as necessary. The SOO will consider the larger issues involved with
transitioning the WDSS into each agency's operations while ensuring the fulfillment of the MOU.
Local Project Oversight: Scott Spratt/AMU Focal Point
The AMU Focal Point will assume local responsibility for the workstation(s). He will ensure
that the complete system is well maintained and evaluated according to the purposed plan. He will
offer input to the SOO concerning progress and results and serve as advocate for proper WDSS
exploitation at NWS MLB. He will also serve as an additional AMU MLB liaison.
Local Project Evaluators: AMU Meteorologists-NWS MLB
The AMU shift will be responsible for the routine execution of WDSS evaluation and
operation according to the purposed plan. Progress and results are to be noted and passed along to
the SOO and/or FP. Each AMU meteorologist will also be responsible for certain "areas of focus"
to expedite an efficient evaluation through the end of the contractual period.
Areas of Focus
SYSTEM/WORKSTATION EVALUATION:
Generally speaking evaluation of the WDSS system, including the RADS display workstation,
has been completed (Aug-Dec. 96). Many suggestions were prioritized and submitted to
NSSL. They have, in turn, responded to each suggestion (either "change request" of "new
feature request") and have provided item-by-item comments on the feasibility of incorporating
them into the next WDSS software release. The next software issuance is scheduled for June
1997 in which several of our suggestions are to be included.
System/Workstation evaluation issues should now be confined to items which impact the
coherent operation of WDSS.
TRANSITION WDSS INTO NWS MLB OPERATIONS:
Proficiency of METs with all RADS display features.
Proficiency of METs in system recovery of WDSS system.
Proficiency of METs in handling tape archiving.
Optimize POP-Ups & ALARMS.
Sanitize and mature Map Backgrounds.
Consider alternate color schemes (if necessary).
(Accomplishment of most elements will entail the creation of local instructions.)
SYSTEM ADMINISTRATION:
System level adjustments to the WDSS will be addressed to ensure the integrity of the
SPARC 20 and its networked relationship with each of the HPs.
ALGORITHM EVALUATION:
Along with the routine operation of the WDSS, the AMU shift will also be tasked to evaluate
the performance of each of the meteorological algorithms in the ECFL environment. Each
AMU meteorologist will consider the performance of all algorithms and make notes
accordingly.
i. Collect DDPDA cases - microburst algorithm.
1. Spring cases
2. Wet season cases - May/June
ii. Collect HDA cases - hail algorithm.
1. Spring cases
2. Wet season cases - May
iii. Evaluate performance of MDA and TDA (mesocyclone and tornado) algorithms
during high shear situations.
iv. Take advantage of unique opportunity to test various adaptable parameters regarding
the Precipitation Accumulation algorithms in real time.
v. Facilitate adjustments to algorithm adaptable parameters.
INTEGRATION OF LIGHTNING DATA:
As part of the MOU, NSSL will attempt the integration of another data set into WDSS. Lightning data has been targeted as that additional data set. To minimize time and effort, while maximizing value added, it has been determined that NLDN data would be the best choice. AMU MLB will work with NSSL to incorporate NLDN data. Once incorporated, the AMU will then evaluate its value.
The following list was compiled after considerable consultation with AMU/KSC.
CHANGE REQUESTS: (focus on top 10)
* FYI
REQUESTED NEW FEATURES: (focus on top 5)
* FYI
Scott M. Spratt, Stephen J. Hodanish, Daniel K. Petersen, and David W. Sharp
Applied Meteorology Unit - Nexrad Weather Service Office - Melbourne, Florida
1. Introduction
The following serves as an operational evaluation of the performance of the National Severe Storms Laboratory (NSSL) Warning Decision and Support System (WDSS) at Melbourne Florida (KMLB) during the 1996-97 "cool season" (November-April). This report is part of a comprehensive operational evaluation of WDSS in accordance with a NASA/NSSL contract (August 1996-May 1997).
Three individual convective events were reviewed which occurred within east-central
Florida (ECFL; Fig. 1) during late March and April. The first event consisted of a cluster
of strong to severe storms which produced a macroburst at the Kennedy Space Center
(KSC). The second case involved a small non-severe thunderstorm which developed late
at night between KMLB and Patrick Air Force Base (PAFB), and the third examines a
thunderstorm which produced a microburst together with another cell which produced
several tornadoes to the west and north of KMLB and KSC (Orange/Seminole counties
and Volusia county, respectively).
In real-time, products upon both the WSR-88D Principal User Processor (PUP) and the
WDSS Radar Analysis and Display System (RADS) were examined. Contributions to this
evaluation were derived from documented real-time WDSS algorithm assessments, as well
as a post-analysis of archived data surrounding the event. The focus of the review will be
upon output from four of the primary WDSS algorithms: the Hail Detection Algorithm
(HDA), the Mesocyclone Detection Algorithm (MDA), the Tornado Detection Algorithm
(TDA), and the Damaging Downburst Prediction and Detection Algorithm (DDPDA). In
addition to examining the algorithm output associated with the cells identified above,
output associated with adjacent cells were also inspected for possible "false alarms".
2. KSC Macroburst Event - 29 March 1997
During the afternoon of 29 March 1997, a complex of strong thunderstorms developed over
interior central Florida and moved east to the coast. As the storms entered Brevard
county, the KMLB radar indicated additional intensification and organization as maximum
reflectivities increased to greater than 60 dBZ and the dominant echo assumed a
"spearhead" configuration. Based on these observations and earlier reports of pea-sized
hail and wind gusts to near 45 kt in Seminole and Orange counties, NWS MLB issued a
severe thunderstorm warning (SVR) for Brevard county at 1915 UTC. Although the storms
did produce small hail and minor wind damage over Orange and Seminole counties, the
majority of the severe weather, including widespread damaging winds (macroburst),
occurred in the Titusville/Kennedy Space Center (KSC) areas of northern Brevard county
between 1920 and 1950 UTC.
2a. Rawinsonde Analysis
Modification of the upstream 1200 UTC TBW (Tampa Bay) sounding for the actual surface
temperature and dewpoint recorded at the Shuttle Landing Facility (SLF/TTS) prior to the
event (88o / 72o F) produced moderate values of Convective Available Potential Energy
(CAPE; positive buoyancy). Assuming a lift of the mean low-level parcel, over 1700 J Kg-1
were realized, while the surface parcel lift would produce a CAPE of nearly 3700 J Kg-1.
The (best) Lifted Index (LI) was -9o C and a dry adiabatic lapse rate was observed from the
surface to near 800 mb. Additionally, the surface to 700 mb Theta-e difference of 36o K
revealed a high potential for vigorous downdrafts and translated strong surface wind gusts.
A relatively low wet-bulb zero (WBZ) height of 10,200 ft also indicated the threat for large
hail to reach the surface.
A westerly flow prevailed at all levels of the atmosphere, with a 80 kt wind maxima at 250
mb . Winds at and above 600 mb were greater than 40 kt, while closer to the surface,
winds were weaker, averaging 10 kt. Storm-relative helicity was considered negligible
(using the actual storm motion).
Morning observations from the Tampa Bay (KTBW) radar indicated scattered strong
convection over the northeastern Gulf of Mexico. This convection was associated with a
large scale outflow boundary which was propagating east-southeast towards the west
central Florida peninsula. Extrapolation of the boundary indicated it would reach east
central Florida during the time of maximum afternoon heating.
2b. Hail Detection Algorithm (HDA)
Several strong thunderstorm cells formed slightly ahead of an outflow boundary over Lake,
Orange and Seminole counties between 1730 and 1800 UTC. As this complex of storms
moved generally east at 35 kt into the coastal waters offshore Cape Canaveral by 2000
UTC, several significant cell mergers occurred. The storms which were responsible for the
severe weather were identified by the Storm Cell Identification and Tracking (SCIT)
algorithm as cells 24, 26, 39, and 64. Of these cells, cell 64 was the most ominous and
persistent (Fig. 2). It was identified and tracked by SCIT from 1841 to 1951 UTC.
Between 1800 and 1841 UTC, several individual cells were identified over Lake, Orange,
and Seminole counties with probabilities of hail between 70 and 100%. However, the
probabilities for severe hail remained at or below 40%, and hail sizes remained less than
1 inch. Spotter reports indicated that pea-sized (1/4 inch) hail fell across portions of
Seminole county around 1840 UTC in association with these storms, but later a delayed
report indicated that nickel-sized (3/4 inch) hail also occurred in the county at about the
same time.
From the time cell 64 was initially identified by SCIT until 1921 UTC, a steady increasing
trend was evident in hail probability, severe hail probability, hail size, maximum reflectivity,
and vertically integrated liquid (VIL) (Fig. 3). Hail probabilities remained at 100% from
1846 to 1936 UTC. From 1906 until 1926 UTC, the probability of severe hail remained
between 70 and 100% and the HDA estimated sizes between 1.5 and 2.25 inches.
Throughout much of this time, the cell was moving across sparsely populated areas of
eastern Seminole and northwestern Brevard counties. By 1920 UTC, the cell had reached
the city of Titusville and several reports of 0.75 inch hail were received. Delayed reports
indicated that additional hail, up to 0.75 inches, occurred through 1940 UTC as the cell
moved east across the KSC. As cell 64 began to rapidly weaken after 1936 UTC (hail
probability decreased from 100% to 0% in a single volume scan), additional strong cells
intensified along the immediate coast and just offshore (cells 18 and 39; Fig. 4).
Numerous other cells were identified across east central Florida throughout the afternoon
with hail probabilities of 60% or greater (e.g. 5 cells during the 1836 UTC volume scan).
However, it is important to note that no cells (over the east-central Florida landmass) other
than the strong/severe storm discussed above (cell 64) indicated severe hail probabilities
greater than 60% or hail sizes greater than 1 inch. Interestingly, the HDA did indicate
severe hail probabilities greater than 60% and hail sizes between 1.25 and 2.0 inches in
association with three cells which developed rapidly just offshore Cape Canaveral after
1930 UTC.
2c. Mesocyclone and Tornado Detection Algorithms (MDA and TDA)
The new paradigm employed by NSSL for mesocyclone detection is to detect all storm-scale circulations of various dimensions and strengths (not only those meeting
"mesocyclone" criteria), and then to manually diagnose whether they may be associated
with severe weather or tornadoes on the ground (see WDSS Algorithm Documentation).
This philosophy was very apparent as numerous circulations were detected by the MDA
during the 6 consecutive volume scans between 1901 and 1926 UTC. In all, the MDA
detected a total of 20 circulations associated with cell 64 and a total of 27 circulations
associated with the collective cluster (15 n mi radius). This large number of circulation
detections might seem overwhelming to the radar operator at first. However, keeping the
paradigm in mind and further examining the MDA output, the probability of detection is
therefore high and operators are alerted to areas upon which to focus their attention. For
cell 64, the Strength Rank of all MDA-associated circulations were in the range of 1 to 6
(defined as very weak to moderate mesocyclones), Furthermore, the Mesocyclone
Strength Index (MSI) values remained below 3600 (weak to moderate) on the majority of
identified circulations. The highest MSI value was 5145 as the cluster was moving from
Orange county into Brevard county (1911 UTC) just prior to dime size hail and wind
damage reports. However, three other cells did acquire MSI values between 4000 and
4700 with no significant weather reported. A review of the mesocyclone trend tables
indicated that although many of the circulations were deep (>10,000 ft), their strongest
rotational velocities generally remained in the mid-levels.
Another MDA parameter called Probability of Tornado (POT) is available as output in the
mesocyclone table with an example shown in Figure 5. This parameter employs a Neural
Network (NN) which diagnoses the POT for each circulation detected, and was "trained"
using 22 storm days, including some Florida storms (see WDSS Algorithm
Documentation). At 1911 UTC, a circulation associated with cell 64 achieved a POT of
98%. However, the POT remained below or significantly below 60% at all other times.
Although no tornadoes occurred, the associated parent thunderstorm did produce hail
greater than 0.75 inches and damaging downburst winds near and shortly after the time
of maximum POT.
The TDA detected a total of 5 tornadic vortex signatures (TVS's), associated with 3
different cells. One cell, accounting for 2 of the TVS's, was located out over the Atlantic,
10 miles east of the KMLB radar. Due to the location of the cell, ground truth was not
attainable. The three remaining TVS's occurred in association with cells 24 and 64, over
Seminole and Brevard counties, near the times of reported hail and wind damage.
Although all TDA detections were associated with strong to severe cells, no tornadoes
were visually observed or tornado damage reported.
2d. Damaging Downburst Prediction and Detection Algorithm (DDPDA)
Output of the DDPDA is in the form of "severe" detections (SEVDET) and predictions
(SEVPRD) for cells capable of producing surface wind speeds in excess of 50 kt, and
"moderate" detections (MODDET) and predictions (MODPRD) for cells capable of
producing surface wind speeds of 30-50 kt. Elevated cores (HICORE) and areas of severe
convergence (SEVCNV) are also indicated as early signatures of cells with microburst
potential. All predictions are issued for 2 volume scans, while detections are only issued
for the volume scan in which the low-altitude downburst signature was detected (see
WDSS Algorithm Documentation).
During the event, DDPDA predictions and detections were frequent. This is not surprising
since vigorous downdrafts were expected. In fact, during the 24 volume scans between
1800 and 2000 UTC, 109 such occurrences were displayed. Unfortunately, nearly half of
the microburst predictions/detections were indicated as "severe" and with just twelve
percent categorized as "detections". A SEVPRD was associated with cell 64 at 1841 UTC
and lasted until 1911 UTC, then with a SEVDET through 1921 UTC. A SEVPRD again
occurred from 1926 until 1951 UTC. Figure 6 shows certain DDPDA parameter trends
associated with cell 64 during this time period. A wind gust to 42 kt was recorded at
Titusville (TIX) at 1920 UTC. Another wind gust was measured to 58 kt at the Shuttle
Landing Facility (1936 UTC) and 85 kt at the 200 ft level between launch pads 40 and 41
at KSC (1940 UTC). However, according to surface observations and spotter reports, the
remaining severe detections and predictions were not associated with 50 kt surface gusts.
In summary, the DDPDA performed well for the particular analyzed severe storm but also
tended to over-emphasize the severe threat for surrounding storms.
return to the top of appendix
3. Isolated Non-Severe Event - 22 April 1997
A weak mid-level short-wave trough moved across ECFL late on 22 April 1997. The
disturbance triggered the development of several isolated showers and thunderstorms
during the late night and early morning hours. Although most of the convection occurred
across north Florida, a small multi-cell cluster of thunderstorms affected ECFL. The
isolated convection initiated 23 n mi west of KMLB at 0450 UTC and quickly strengthened
to 60 dBZ within 15 minutes while moving east at 30 kt. The strongest cells reached the
coast near PAFB between 0535 and 0540 UTC (Fig. 7), then moved offshore and
dissipated by 0615 UTC. There was no damage experienced, or severe weather reported,
with this event. Interestingly, the WDSS highlighted numerous mesocyclonic circulations,
microbursts, and tornadic signatures which indicated an extremely high likelihood of
severe weather during the period that the storms were in vicinity of PAFB.
3a. Rawinsonde Analysis
Modification of the 0000 UTC TBW sounding for actual surface conditions recorded at
MLB prior to the storm (72o / 64o F) produced little in the way of surface-based positive
buoyancy. In fact, lifting the surface parcel only resulted in a CAPE of 274 J Kg-1, with a
negative area of 21 J Kg-1 which had to be overcome. The lifted index was -2o C while the
surface-700 mb theta-e difference was only 17o K. It is hypothesized that the approaching
short-wave trough teamed with low-level moisture advection to facilitate isentropic lift over
the area and subsequent localized low-topped convection near the coast. The VAD winds
from the KMLB radar between 0500 and 0600 UTC showed southerly winds at 30 kt at 1
kft and above while veering slightly to the southwest and eventually west with height.
Speeds reached to near 90 kt at 30 kft. Given the presence of the isolated convection and
the VAD winds, a potential for marginal severe weather did exist.
3b. HDA
Due to the observed low tops and low VIL characteristics, large hail was not a significant
concern.
3c. MDA and TDA
At 0515 UTC the first circulation was identified as a "low top" circulation but with a suspect
maximum rotational velocity of 64 kt. Manual assessment of the lowest-level SRM data did
not show appreciable cyclonic shear or significantly high velocities to support the extreme
values of maximum rotational velocity(mxrotv-64), maximum shear (mxshr-37), maximum
gate-to-gate (mxgtg-101) and predicted 54 percent probability of severe weather. A broad
set of outbound gates did indicate 25 to 31 kt of flow (with an azran of 298o/14 north of the
"meso symbol") with a few gates of 0 to 2 kt inbound flow (with an azran of 280o/13). The
WDSS subtracted direction/speed on the SRM was listed as 258o/18, which would become
a critical factor when compared to the cell table speed/direction. Similar MDA tabular
output values persisted during the next several scans, where seemingly "speed shear"
accounted for the algorithm-identified mesocyclone. The algorithm table direction/speed
increased to a remarkable 268o/74. Continuity suggests that although the algorithm
claimed to be following the same meso, manual analysis indicated that the algorithms
(MDA & TDA) were perplexed by the shear at the leading (eastern) portion of the shear
zone and an apparent "noisy" base velocity field within the 2.4o slice. The MDA identified
eight low top circulations at 0526 UTC, again focusing on the shear at the eastern edge
of the storm where the area of outbound velocities were interrupted by the cell cluster
approaching central Brevard county. The cell direction/speed was listed in the algorithm
table as 274o/64, with no direction/speed listed for any of the other algorithm-identified low
top circulations. Three separate cells were identified as having TVS's (all within a
geographic area of about 4 miles) and with mxgtg values from 124 to 130.
At 0530 UTC, five low top circulations and one TVS were identified as the small cluster
moved towards PAFB. The associated cell table at this time (Fig. 8) highlighted the
presence of a tornadic mesocyclone (TVSMESO) and a severe microburst (SEVDET). The
TVS was apparently in response to a couple of gates containing 64 kt of inbound velocities
centered at an azran of 319o/12. This was evident at both the 0.5o and 1.5o scans.
Random ("noisy") inbound/outbound velocities continued to contaminate the higher level
slices. The mesocyclone table output listed direction/speed values ranging from 351/6 to
287/82 all within this same small area. The next scan (0535 UTC) brought the
identification of six low top circulations and one TVS (Fig. 9). Then by 0540 UTC, only two
low top circulations were identified.
Given the meteorological situation, this small cluster of thunderstorms was capable of
producing marginal severe weather. If nothing else, environmental winds of 30 to 40 kt
(near 1 - 2 kft) could have been easily advected downward to the surface in downdrafts.
Erroneous indications of multiple mesocyclones and TVS's under such circumstances
could mis-lead warning meteorologists. The likely culprit for this case was the velocity
"noise" in the batch cuts (2.4o and above).
3d. DDPDA
During the event, multiple microburst predictions and detections were erroneously
indicated as well. Since the DDPDA, MDA, and TDA all had problems resolving the event,
a common mis-interpretation of velocity base data is surmised. This, again, is believed to
be linked to "noise" in the batch cuts (2.4o and above).
return to the top of appendix
4. Widespread Severe Event - 23 April 1997
Throughout the morning and early afternoon of 23 April 1997, a line of strong to severe
thunderstorms moved rapidly east across all of ECFL. An approaching mid level short
wave trough and strong upper level divergence allowed the storms to remain organized
and strong. Ahead of the line, several isolated cells developed and rapidly became severe.
Although severe weather was reported in six ECFL counties, only two events will be
evaluated in this section. The first event produced microburst wind damage in Orange and
Seminole counties and a gust to 50 kt at the Orlando Executive Airport (ORL) between
1445 and 1450 UTC. NWS MLB issued Tornado warnings for both counties in advance
of the damage and recorded severe gust. The second event produced several short-tracked F0 tornadoes in Volusia county. NWS MLB issued a tornado warning well in
advance of the damage.
4a. Rawinsonde Analysis
The 1200 UTC modified sounding for TBW indicated a deep layer of moisture, with a dry
layer aloft. Lifting the near surface layer produced an unstable environment with a CAPE
of nearly 3000 J Kg-1 and an LI of -6o C. Additionally, the surface to 700 mb theta-e
difference was 34o K, indicative of vigorous downdraft potential and microbursts.
Although the vertical wind profile was nearly unidirectional, speeds increased quickly just
above the surface. In fact, speeds were greater than 40 kt at and above 1 kft. The resultant
storm-relative helicity was 252 m2s-2 revealing a shear profile conducive to rotating storms.
The WBZ height was only 9,400 ft.
4b. HDA
Throughout the evaluation period of the severe thunderstorm which moved across Lake,
Orange, and Seminole counties (1407-1502 UTC), the HDA indicated hail probabilities of
80% or greater, except for at 1457 UTC, when 50% was output. Severe hail probabilities
steadily increased to 70% by 1447-1452 UTC (Fig. 10). Hail sizes were forecast to be 1
inch or less, except for at the times of the 70% severe hail determination, when a forecast
of 1.25 was shown. The high severe hail probabilities and sizes likely resulted, in part, due
to a peak of maximum reflectivities of 58-60 dBZ at a height of 23000-24000 feet. Although
no reports of hail were received in association with this event, the time and location of
maximum several hail probabilities and sizes where coincident with the wind damage
described earlier.
For the Volusia county complex of storms, SCIT identified several individual cells between
1443 and 1502 UTC with hail probabilities of 100% or greater, severe hail probabilities of
80% or greater, and hail sizes of 1.25-1.50 inches. The times of the maximum severe hail
probabilities and sizes corresponded with numerous reports of 0.75-0.88 inch hail. One
of the two cells which was associated with the large hail detection, was also responsible
for the two tornado occurrences.
Other than the Lake/Orange/Seminole and Volusia county cells described above, only one
other cell (over land) was labeled with a probability of severe hail greater than 60% and
a hail size greater than 1 inch during the one hour period. This cell only produced the
significant hail parameters for a single volume scan and was located over a sparsely
populated area of northeast Lake county at the time. Therefore, even though no reports
of hail were received, hail occurrence can not be ruled out.
4c. MDA and TDA
Prior to 1400 UTC, an isolated cell developed over southern Lake county, 10-15 n mi
ahead of a solid line of storms with an embedded bow echo. The isolated cell, identified
by SCIT as cell 3 (Fig. 11), moved northeast at 40 kt and began to acquire rotation.
Continuously from 1407 to 1437 UTC, the MDA associated a "weak circulation" with the
cell. However, the algorithm re-identified the circulation feature 3 times (circulation 17, 44,
and 85). Hereafter, circulation 85 continued to be identified until its demise after 1500
UTC. Up until 1442 UTC, all identified circulations associated with the cell were ranked as
"2-3" (weak), but during the following 2 volume scans, the circulation intensified to a rank
of "6", defined as a moderate mesocyclone. The rapid intensification was likely a result of
increased convergence as a strong outflow associated with the bowing cell located to the
west collided with isolated supercell. The mesocyclone trend data revealed that as
maximum low-level velocities reached 50 kt and the rotational base decreased to below
6 kft. Resultant MSI values increased to slightly above 4000, indicative of a strong
mesocyclone (Fig. 12). Coincident with the time and location of maximum rotation, wind
damage occurred as numerous large trees were downed in Seminole and Orange counties
and a wind gust to 50 kt was recorded at ORL. Although the wind damage appeared to
have been caused by a macroburst and straight-line winds, it is possible that a short-lived
F0 tornado may have occurred as well. After the brief period in which the mesocyclone
achieved a "strong" classification, the circulation gradually weakened. The weakening
occurred as the line of storms to the west merged with the echo and the strong outflow
winds pushed ahead of the complex.
Interestingly, the POT parameter associated with the circulation remained below 30% for
all volume scans except one. During the 1447 UTC scan, the time in which the wind
damage and severe gust occurred, the POT jumped to 98%.
At the same time the cell discussed above was occurring, another severe cell was moving
across Volusia county. This cell, identified by SCIT as cell 6 developed along the
Lake/Volusia county line prior to 1400 UTC. As the cell moved northeast at 45 kt, it too
was over-taken and merged with a line of storms from the west around 1430 UTC. By
1447 UTC (Fig. 13a & Fig. 13b), a new cell began to form on the south flank of cell 6. Also
on this volume scan, 2 TVSMeso's (mesocyclones associated with TVS detections) were
identified along a shear-zone apparent within the velocity data, and along the leading edge
of the low-level 50 dBZ echo. Manual assessment indicated a third, but weaker circulation
along the shear-line, just south of the MDA identified features. By the next volume scan,
both TVS features continued to be identified and had reached the immediate coast, with
the weaker circulation still manually recognizable to the southwest. At 1457, the TVS
features had moved offshore and become weak circulations, and the manually detected
circulation was along the immediate coast and obscured by range folding at the lowest
elevation, however a 29 kt shear couplet was seen within the 1.5o slice. On the following
scan, the feature which had been manually assessed earlier, had moved offshore and
strengthened, and was identified as a TVSMESO.
Visual spotter reports and a post-storm damage survey indicated that 2 short-track F0
tornadoes occurred in association with the manually identified mesocyclone. The first
tornado occurred inland from the coast at approximately 1452 UTC, and the second
occurred along the immediate coast at about 1457 UTC. A thorough ground survey of the
area affected by the MDA-identified TVSMESO features revealed that no damage occurred
their vicinity (POT associated with each cell exceeded 90% for at least one scan).
Although the MDA appeared to miss detection of the circulation which produced tornado
damage, the algorithm consistently identified the 2 much stronger circulations which
occurred along the shear-line, just north of the weaker feature. While manual techniques
did detect the weaker feature, attention was focused slightly to the north on the relatively
stronger circulations.
Due to the large number of strong circulations identified by the MDA during the time of the
Seminole/Orange county and Volusia county events, some of the circulations of interest
were ranked low in the mesocyclone table or were not included on the table at all. For
example, the mesocyclone table for the 1452 UTC volume scan contained 4 couplets, 5
meso's, a TVSMESO, and 5 weak circulations, most of which were located offshore.
However, two TVS features over Orange county (one associated with the damage
described above) were not listed on the table due to the relatively higher ranking of other
circulations. Many other volume scans during this event were similar to 1452 UTC scan.
4d. DDPDA
The DDPDA initially identified a MODPRD with the Lake/Orange/Seminole county cell at
1417 UTC, then upgraded it to SVRPRD on the following volume scan. This prediction
continued though the remainder of the evaluation period, except for during the 1437 and
1442 UTC scans when no microbursts and a MODPRD was identified, respectively. The
reason for the decreased prediction/detection on these two scans was not apparent in the
microburst trend table (Fig. 14). Although the MMDPA did identify a MODPRD/SVRPRD
at the time of the damaging winds (1445-1450 UTC), the decreased potential indicated on
the previous two scans provided contradicting information.
The complex of cells which eventually produced the tornadoes along the Volusia county
coast at 1452-1457 UTC were initially identified by the DDPDA prior to 1430 UTC. A
SEVPRD was displayed with the tornado-producing cell up through the time of the first
tornado. During the volume scan in which the cell reached the immediate coast and
produced the second tornado, no microburst predictions/detections were indicated. On
the following scan, the cell was no longer identified by SCIT.
Throughout the period of the two events, numerous other microburst predictions occurred
in associated with cells which did not produce strong surface gusts. The majority of these
predictions occurred along the gust front which propagated away from the line of
convection. Interestingly, the DDPDA never produced any microburst detections
(MODDET or SEVDET).
return to the top of appendix
5. Summary and Conclusions
for Cool Season Analysis
5a. Summary
Hail algorithm: Useful, given the general guidance mentioned below. Sizes too large and
too many "high" hail probabilities, but severe hail probabilities good. Subjectively speaking,
it appears that cells which have a severe hail probability of at least 60% (usually 80% or
more) AND a hail size of 1.25" or more equates to hail of 0.75" or more at the surface.
Mesocyclone algorithm: Good. Lots of detections in short periods of time. However, many detections aloft and "behind" gust front. Detection of meso's (i.e. rating of 5 of more) and MSI values greater than 3600 appear to correlate reasonably well with wind damage and tornadic potential.
Tornado algorithm: Needs further work. Several false alarms apparently due to "noise" in
the velocity field. Need improved quality control. TVSMESO gives greater confidence than
TVS. Although TDA missed the 2 tornadoes in Volusia county (4/23/97), two TVS's were
identified with stronger circulations in adjacent cells. Since non-supercell tornadoes are
also searched for, the TDA may be oversensitive.
Microburst algorithm: Needs further work. Too many false alarms. No way to differentiate
between "real" microburst potential and false alarms.
5b. Conclusions
WDSS is very helpful to account for potential cell severity. Trend tables may allow a 1-2
volume scan lead time on warnings over PUP cell evaluations. However, too many
hail/meso/burst detections can "distract/confuse" the user. On the other hand, after
sufficient use...the user can discern which parameters are "important" (i.e. MSI and meso
rank...severe hail probability combined with large hail sizes) and what can be discounted
for a given event. Products must be examined rather than blindly following algorithm
output (i.e. many false alarms). Users must be thoroughly familiar with mesoscale
meteorology to make informed decisions using WDSS output and not overwarn.
Responsibility still belongs to the operator.
Although strong mesocyclone and large hail indicators did not always correlate with
tornadoes and hail respectively, they often were associated with locations of wind damage
(i.e. severe weather predictor).