|
Listed first are particularly notable research results as
well as the accomplishments
of all participants in the current grant (2011-2015). Skip down
to the same information for the 2007-2010
participants.
Last Updated: December 21, 2011
Special Research Nuggets
The items here are especially significant outcomes of REU projects.
2011 Participants:
Rapid Identification of Severe Surface-Level Winds
REU student Adam Taylor quantified the impact of having low-level, dual-Doppler
radar data available for detection of severe surface winds. Mr. Taylor
found that operational forecasters would be able to identify areas of severe
winds much faster, and with much greater accuracy, if overlapping low-level
radar coverage (e.g., CASA data) was widespread. Furthermore, Adam found
that even a simple tool that corrected radar-derived wind speeds for height
above ground (applying a wind profile correction) could aid forecasters in
estimating surface wind speeds.
Anticipating Tornado Casualties for Emergency Planning
Amber Cannon used a GIS analysis to compare rates of incidence of
fatality, by population density, for the Alabama portion of the
27 April 2011 and 3–4 April 1974 tornado outbreaks. If replicated
and done over many geographic areas, this research could be combined
with the distribution (number, intensity, size) of tornadoes expected
with convective outlooks to help FEMA anticipate rates of incidence
of fatality/injury hours to days ahead of time, enabling them to
leverage resources for immediate response.
Assimilation of AQUA Data Improves Track Forecast of
Hurricane Danielle
Although results are preliminary, Travis Elless's research highlighted
the importance of studying the impact of the data and data assimilation
methods on tropical cyclone forecasts. His work will be continued
in Dr. Xuguang Wang's research group.
Simplifying Microphysics Parameterization to Achieve
Better Forecasts of Convection
Diversity in the physical parameterizations used in forecast ensembles is already
known to provide robust variance amongst the ensemble members in mesoscale forecasts
(resolution of 10 to 30 km). Sam Lillo’s research made a first systematic
look at some possible means to achieve physics diversity within a single advanced
microphysics parameterization for convection-resolving forecasts (resolution
~1km). The work has implications for the Warn-on-Forecast initiative, which
aims to assimilate radar data to provide short-term forecasts of severe weather. The
density of radar data can drastically reduce ensemble spread, and this research
considered sensitivities that affected warm-rain physics, precipitation efficiency,
and large ice hydrometeor characteristics that may help maintain storm-scale
ensemble efficiency.
Rapid-Scan Dual-Polarimetric Radar
Alex Lyakhov used the RaXpol,
a state-of-the-art dual-polarimetric mobile radar to scan
a supercell and weak tornado. His research documented rapid changes
in tornado and mesocyclone evolution during tornadogenesis and tornado
dissipation and their relationship to polarimetric supercell signatures.
Highlights of Student
Research Accomplishments
Rebekah Banas considered operational forecasts
of heavy precipitation events along the Sierra Nevada mountains of
California. She found that the precipitation amounts
are consistently under predicted and that forecast quality lessens
as elevation is increased.
Eric Beamesderfer compared storm motion estimates to observed storm
motions for different storm modes and environments. Eric found
that storm motion estimates were fairly inaccurate, with deviations
up to 20 m s-1 very common. He also found that storm-relative
helicity influenced storm motion the most. However, since SRH
is heavily tied to storm mode, it was hypothesized that storm mode
might be the most important predictor of motion.
Amber Cannon successfully completed a GIS analysis
that helped her to compare rates of incidence of fatality, by population
density, for two tornado outbreaks affecting Alabama. She successfully
manipulated datasets from several different sources, and included
raster, vector, and photographic information in her analysis.
Tracey Dorian examined daily MODIS imagery to
estimate whether our cloud detection algorithm was producing accurate
estimates of mean cloudiness over different regions of the globe. She
determined that our original procedure significantly underestimated
stratus estimates, and recommended new threshold values.
Travis Elless successfully run the operational global data
assimilation and forecast system to study the impact of Aqua data on hurricane
track forecast of Danielle (2010). He found the assimilation of AQUA data improved
the intensity track of Hurricane Danielle.
Sam Lillo took the first systematic look at supercell storm
forecast variability resulting from sensitivity to parameters within a multimoment
bulk microphysics scheme. The goal was to identify some parameters or
parameterizations that could be diversified provide smooth (as opposed to multi-modal)
spread in forecast ensembles. Focus was put on large ice (graupel and
hail) characteristics, for example fall speed, rime density, and upon warm-rain
physics response to cloud condensation nuclei.
Alex Lyakhov used the RaXpol, a state-of-the-art dual-polarimetric mobile
radar to investigate a supercell and weak tornado. His research
documented rapid changes in tornado and mesocyclone evolution during tornadogenesis
and tornado dissipation and their relationship to polarimetric supercell signatures.
Using high-resolution SHAVE hail reports, Sarah Mustered investigated radar
and environmental parameters to determine hail size at the surface. Sarah
found that more novel matching techniques will be necessary given the high-resolution
reports might match different parameters to wide distribution of hail sizes. She
also found that while combinations of radar and environmental parameters did
not stratify hail size well, there were certain parameter spaces more favorable
to large hail production.
Highlights of Adam Taylor's work include a comparison
of Oklahoma Mesonet surface wind speeds with radar-derived estimates
from WSR-88D and CASA. Adam tested the impact of dual-Doppler
and wind profile corrections to radar-derived estimates when comparing
against surface wind observations.
Previous Grant, 2007-2010
Listed first are particularly notable research results from REU participants'
work. Skip down to accomplishments
of all participants in
the 2007-2010 grant.
Special Research Nuggets
The items here are especially significant outcomes of REU projects.
2010 Participants:
Deficiencies in boundary layer parameterizations may hurt model
forecasts of shallow cold air.
William Leatham investigated eleven model forecasts for events that
included arctic air arriving in the southern Plains in advance of
winter storms. The average error for position of the freezing
line was quite large and increased with time (up to 135 km at 24
hours). A preliminary investigation revealed that radiative
schemes may be a strong contributing source to the error. Precipitation
was often well forecast in space and time, and model soundings became
saturated to the wet bulb temperature. But the models allowed
too much heating of the boundary layer so that the sounding was too
warm prior to the onset of precipitation. In at least one instance,
the GFS and NAM models forecast the freezing line to retreat more
than 30 km northward during the peak of the solar cycle, while the
observed freezing line progressed an equal amount in the opposite
direction (southward).
Defining Spatial Vulnerability From Tornadoes Based
on Fujita Scale
Eric Hout's research defined the idea of spatial
vulnerability of counties based on the standardized tornado
fatalities for individual counties over time. Previous studies
on tornado vulnerability have provided insight on how individual
factors influence overall social and spatial vulnerability.
However, few studies have been conducted to evaluate the aggregated
effect on vulnerability when these factors coincide. Additionally,
a definition of vulnerability has been absent from the meteorological
literature. Thus, to provide a more comprehensive view of vulnerability,
this study proposes a mathematical definition for spatial vulnerability,
and then uses tornado casualty data from 1950 through 2009
to calculate vulnerability on a county level for seven time
periods. Hout analyzed trends of spatial vulnerability for
each county and interpreted the spatial patterns among counties
with increasing or decreasing trends of spatial vulnerability.
Some patterns could be attributed to regional and others to
local effects, which suggest regional and local influences
on social responses to tornadoes of different damage (Fujita)
scales.
Incorporating Societal Impacts into Development of
Warn-on-Forecast
The National Severe Storms Laboratory has begun research
to move towards a Warn-on-Forecast (WoF) system. WoF will include
probabilistic information from ensemble model forecasts and forecaster
input with much greater lead time than today's current warnings. Sarah
Stalker's research was the first to address some of the
societal impacts of moving to a WoF system. She interviewed six individuals
in the Norman, OK region who were affected by the 10 May 2010 tornado
outbreak. Subjects noted that seeing a projected path of the
storm, similar to that provided by graphical hurricane outlooks,
was more important than the additional lead time to them. Further
research will develop upon these results and help tune WoF products
throughout their development.
2009 Participants:
Advanced modeling techniques help forecasters stay
in tune with snow band prediction.
Banded snow is one of the greatest winter weather forecasting challenges
faced operationally with large economic and human safety consequences. Numerical
models frequently fail to provide forecasters with adequate guidance
to anticipate banded snow. Using techniques normally applied to springtime
convective events, Astrid Suarez Gonzalez demonstrated
that a combination of high-resolution, data assimilation and ensemble
forecasts can greatly aid in the forecast decision-making process.
Low-level Scanning Could Be Key to Reducing NWS Tornado
False Alarms
Hannah Barnes investigated NWS tornado warning statistics
for marginal storm events. Hannah quantified the false alarm
rate (FAR) and probability of detection (POD) for three scenarios:
i) days with no reported tornadoes; ii) days with only one reported
tornado; and iii) outbreak days with ten or more reported tornadoes
within a Weather Forecast Office county warning area. Hannah
identified three key results. First, the large-scale environment
differed little between zero and one tornado days, but both differed
significantly from large outbreak events. Thus, there were
no large-scale signatures differentiating between zero and one tornado
days. Second, the circulation intensity of false alarms at
the lowest height , as scanned by WSR-88D radars, was notably weaker
than those associated with confirmed tornado warnings. Third,
the presence of obscured velocity data (marked by ‘purple haze’)
led to a 15% increase in the false alarm rate. Hannah’s
findings are critical first steps in understanding how to reduce
the number of tornado false alarms. These findings may determine
how future radar systems are deployed and how optimal scan strategies
could be utilized.
One to two hour tornado warning lead-time may not
be necessary for general public
Stephanie Hoekstra's research provided insight into
whether a 1-2 hour tornado warning lead-time (also known as warn
on forecast) is currently demanded by the general public. On average,
participants stated needing a minimum lead-time of 13.8 minutes but
would like 33.5 minutes in an ideal situation. Her work is significant
because while longer lead-times are often the focus of meteorological
research, little to no research has been published regarding how
the public would respond to such a warning. Stephanie and her mentors
are aiming to publish her results in a peer reviewed journal.
NWS Lead Time for Severe and Damaging Hail Adequate
for Preventive Measures
Lauren Potter quantified NWS warning lead times
for reported severe hail and damaging hail events. Lauren compared
two years (1999 and 2000) of severe hail reports and ten years (1999-2008)
of damaging hail reports from Oklahoma, Colorado, Massachusetts and
South Carolina. Interestingly, she found no significant differences
among those states in the lead time of reported severe hail or damaging
hail. The mean lead time for severe hail was 18-22 minutes,
with a lead time ranging from 19 to 29 minutes for damaging hail
across the four states. Overall, Lauren found that about 72%
of reported hail occurs during a Severe Thunderstorm warning and
another 7% occurs during Tornado Warnings. Such a relatively
long lead time and warning rate provides the general public and emergency
and government services with the opportunity to take preventive cautions
and thereby mitigate at least some property damage from hail.
Exploring Viability of Social Networking to Communicate
Weather Information
In order to begin developing an understanding of social networking
as a means for communicating weather information to the general public,
particularly with regard to time-critical information about severe
weather, Justin Wittrock developed and distributed
a web-based survey designed to address fundamental questions about
this issue. In contrast to many REU projects, which represent
student participation in an ongoing research program established
previously by the mentor, Justin’s project was self-initiated. He
learned how to develop an effective survey and in particular, how
to pose questions in a neutral manner. He also gained experience
in the Institutional Review Board process, learned how to identify
communities and sample them in appropriate ways, and how to apply
statistical analysis techniques. Most importantly, he was shown
how to explain findings, rather than simply present them, and pose
questions for further study based upon them.
2008 Participants:
Tornado Warning Performance Dramatically Improves
Inside a Tornado Watch
Kelly Keene's research showed that having a tornado watch in place
prior to the issuance of tornado warnings vastly improves the performance
measures associated with tornado warnings, particularly the critical
measure of probability of detection (POD). Specifically, when a tornado
watch is in place, the average POD for tornadoes over the last 10
years is around 0.85, while when no watch is in place the POD drops
significantly to 0.50. This drop in POD suggests that not having
a watch in place prior to a tornado warning equates to where the
warning performance was some 20 years ago just prior to the implementation
of the Doppler radar network.
Model of New York Harbor Improving through Ensemble
Data Assimilation
The New York Harbor Observing and Prediction System (NYHOPS) is being
upgraded (funded by the Navy's Small Business Innovation Research
(SBIR) program) to make better use of observations routinely collected
in New York Harbor. Jon Poterjoy's project on ensemble Kalman filter
localization accomplished, by hand, what will become an automated
procedure that maximizes the impact of a wide variety of ocean observation
systems deployed in New York Harbor.
2007 Participants:
Student Discovers Error in Data
As Doug Crauder was scoring his velocity products looking for velocity
dealiasing errors, he came across a situation where there were
noisy velocities in a meteorologically benign area near the radar. Doug
realized these noisy regions had the classic teardrop shape associated
with range folded echoes. A closer look identified strong
storms in the fourth Doppler trip were causing the problem. Normally
the range folded data should be shown as "purple haze." In
this case the new range aliasing mitigation technology developed
by Sachidananda and Zrnic (SZ-2) is not correctly sorting the data. National
Severe Storm Laboratory scientists agree they will need to tweak
threshold parameters to clean up the data.
Type of Weather Watch Affects Warning Performance
Jessica Ram's research showed that the type of watch issued by the
Storm Prediction Center affects warning performance at local National
Weather Service Forecast Offices. Specifically, tornado warning
performance was highest for Particularly Dangerous Situation (PDS)
Tornado watches, and lowest when no watch of any kind was in effect. The
study also showed that 93% of all tornado events occurred inside
some type of watch, with 3/4 occurring either in a PDS tornado
watch or a tornado watch. An interesting result of the forecaster
survey is that watch type seems to influence an individual warning
threshold, such that it is lowest for a PDS tornado watch.
Long-Term Changes in Atmospheric Instability Could
be Related to Increasing Temperatures
Victor Gensini found that there are long-term changes in the frequency
of high instability in the atmosphere in the US, with high values
occurring at the beginning and end of the analysis period associated
primarily with increased low-level moisture. To first order,
trend resembles the US annual temperature record, implying that it
could occur more frequently in a greenhouse-enhanced climate. In
South America, on the other hand, instability decreased throughout
the period, dominated by drying conditions.
Students' Dataset Forms Basis of Competition
The storm classification data set created as part
of Eric Guillot's REU project is being used in a competition sponsored
by the Artificial Intelligence Committee of the American Meteorological
Society at their 2008 annual meeting.
Highlights of Student Research Accomplishments
2010 Participants:
With current and anticipated climate change, there arise questions
related to how the hydrologic cycle may be affected in a region.
Using a GCM ensemble Christopher Bednarczyk studied
potential changes to the Blue River Basin of Oklahoma. Depending
on the emissions scenario, streamflow is projected to decrease 10
to 30%. This is important because several area communities get water
from this river, and there has also been talk of outside communities
pumping water to supplement their own future water supplies.
Jeffery Deppa studied WRF model forecasts of the low level jet (LLJ)
over a wind farm in southwest Oklahoma. As part of his investigation
of mountain wave dynamics he studied parameters such as the Froude
number and used IDV to visualize the forecasts. The WRF forecasts
indicated that the strongest winds at turbine height might actually
occur a few kilometers downstream of the ridgetop wind farm.
Todd Ferebee investigated the use of multi-radar,
multi-sensor severe weather products in determining where different
hail categories did or did not fall. He found several products did
fairly well in depicting where no hail, non-severe hail, severe
hail and significant-severe hail fell. Several other products
showed delineation between just two categories, such as non-severe
hail vs. severe hail or significant-severe hail vs. all other categories.
Todd gained experience with the R statistical program and the Warning
Decision Support System--Integrated Information command line utilities.
Stacey Hitchcock learned the importance of programming
knowledge, writing skills,
and data visualization in meteorological research. She also
learned several different forms of forecast verification, including
the use of Performance Diagrams (Roebber Diagrams) to convey large
amounts of information succinctly in a single figure.
Eric Hout's research defined the idea of spatial
vulnerability of counties based on the standardized tornado fatalities
for individual counties over time. Previous studies on tornado vulnerability
have provided insight on how individual factors influence overall
social and spatial vulnerability. However, few studies have been
conducted to evaluate the aggregated effect on vulnerability when
these factors coincide. Additionally, a definition of vulnerability
has been absent from the meteorological literature. Thus, to provide
a more comprehensive view of vulnerability, this study proposes a
mathematical definition for spatial vulnerability, and then uses
tornado casualty data from 1950 through 2009 to calculate vulnerability
on a county level for seven time periods. Hout analyzed trends of
spatial vulnerability for each county and interpreted the spatial
patterns among counties with increasing or decreasing trends of spatial
vulnerability. Some patterns could be attributed to regional and
others to local effects, which suggest regional and local influences
on social responses to tornadoes of different damage (Fujita) scales.
Eric learned the process of developing a research
project with an emphasis in GIS and spatial vulnerability of tornadoes.
He experienced the entire research procedure: defining a research
question, literature review, data acquisition, analysis, and interpretation.
He has learned GIS skills in data integration and spatial analysis.
Christopher Kerr learned to process and analyze CASA X-band polarimetric
radar data and made comparisons between the radar measurements and
calculated radar variables from disdrometer data. He calculated mean
biases and errors of the measurements for the radar data with and
without attenuation correction. The biases and errors are significantly
reduced with attenuation correction, but substantial residual errors
exist even after the correction. The residual errors vary depending
on the location of the storm and the propagation path through the
storm. This indicates that the attenuation effects have not been
fully accounted for and further study is required.
Major ice storm events have become a seemingly routine component
of southern U.S. winters during the past decade. In order to determine
where and how frequently ice storms have occurred during the past
decade, Carly Kovacik conducted a climatological
analysis of ice storm events across the southern United States. Her
research accomplishments included the development of a 10-year dataset
of ice storm events across the southern U.S. (specifically KS, MO,
OK, AR, TN, TX, LA, MS), an analysis of events to determine spatial
and temporal characteristics, and a preliminary investigation into
atmospheric mechanisms potentially responsible for changes in ice
storm frequency observed during the past decade. Carly quantified
an ice storm maxima within a region stretching from far southwest
Oklahoma northeastward through Missouri. One particularly important
result of this research revealed significant geographical inconsistencies
in ice storm frequencies across National Weather Service Forecast
Office boundaries. Although limitations in the National Climatic
Data Center’s Storm Storm (and Storm Events) archives are known,
this result emphasized the lack of a universal definition for ice
storms nation-wide. Through this project Carly gained skills in building
datasets, analyzing phenomena spatially, and effectively communicating
results orally and in writing. The work she accomplished is already
contributing to continuing efforts to study southern U.S. ice storms
at the Oklahoma Climatological Survey.
Forecasters have long observed that operational models are too slow with the
southward progression of shallow, arctic air across the sloping terrain immediately
east of the Rocky Mountains. This deficiency affects forecasters’ confidence
in forecasting precipitation type and issuing winter storm warnings at lead
times of only 12 to 24 hours. William Leatham set out
to quantify the problem. Inspecting eleven model runs related to four
winter storms, Bill found an average error of about 60 km on the location of
the cold front and, more importantly, 107 km on the location of the surface
freezing line. Freezing line error increased from east to west, and,
as expected, the model error was toward the north. Twenty-one hours into
one forecast, much of the freezing line from Oklahoma City to the New Mexico
border was observed farther south than any of the twenty-two members of the
Short Range Ensemble Forecast (SREF) had forecast. With a robust ensemble,
this result should be rare, and finding this result with a case sample size
of one is troubling. Preliminary inspection found that diabatic influences
may play a large role in creating model error with respect to the freezing
line. Building a larger data set will help confirm whether a strong model
bias exists, and testing sensitivity to diabatic processes may provide insight
into potential causes.
Sarah Stalker investigated public actions and reactions
to the 10 May 2010 Norman, Oklahoma tornado. Sarah showed great poise
and enthusiasm throughout the summer and acted as a self-starter
in order to get her research accomplished in the limited time-period
of the REU program. She learned how to work through the Internal
Review Board (IRB) process, which included a detailed training and
developing a description of her research process and goals. Interviews
were completed with individuals living in the path of the tornado
in order to gain perspective on what they choose to do and why during
a tornado. A qualitative research technique (thematic analysis) was
used to analyze her data and associate with past work via a conceptual
model. Participants did not feel any direct threat during early storm
development and advisories and waited until the final moments to
take shelter, though all subjects later believed they should have
taken action sooner. Participants also stated that information on
the projected storm-track, similar to that provided by graphical
hurricane outlooks, was more important to them than longer lead-times.
Sarah continued this project after REU, adding participants in Minnesota
following the tornado outbreak that occurred in that region on 17
June 2010. She will present her work at the Sixth Symposium on Policy
and Socio-Economic Research at the upcoming 2011 AMS Annual Meeting.
Joshua Turner investigated whether the urban heat
island affects storm trajectories. He analyzed output from
a cloud tracking algorithm that provided locations and size of storms
throughout their lifetime. The goal was to see if changes
in storm velocity (both direction and speed) were correlated to urbanized
centers. A threshold filter was applied to eliminate spurious
results from different storms joined together. There was no
clear relationship of changes in velocity to urbanization, but signals
may have been swamped by noise due to the time limitations in developing
the thresholding filter. Josh’s
MatLab programming techniques were greatly expanded during this project.
Available datasets of global wetlands, water bodies, and seasonally
inundated areas do not meet the needs of greenhouse gas flux models,
which are used to estimate the flow of trace gases such as methane
from the land surface into the atmosphere. Kevin Van Leer contributed
to efforts to develop advanced water mapping techniques by investigating
the effect that pixel scale has when flooded area is determined from
satellite remote sensing imagery. Kevin's research showed that classified
coarse-resolution imagery from the Moderate Resolution Imaging Spectroradiometer
(MODIS, 250- to 2000-m) can significantly underestimate total inundation
area compared with fine-resolution Landsat imagery (30 m). Furthermore,
he showed that 250-m MODIS imagery did not improve the inundation
area estimate compared with coarser-resolution imagery. Through this
project, Kevin developed skills in spectral analysis, image classification,
and spatial analysis of satellite imagery. He also gained experience
in the Matlab programming environment and in the handling of large
remote sensing datasets.
The Storm Prediction Center (SPC) has begun to develop a measured
severe thunderstorm gust dataset that is partially independent of
the National Weather Service's severe weather report database. Andrew
Winters played an integral role in the development of the
measured wind dataset by parsing surface observation data and then
subsequently analyzing spatial relationships in a Geographic Information
System framework and analyzing near-storm environmental variables
linked to each individual gust. Through his work, the SPC is
in the early stages of obtaining an objective, non-damage biased
severe thunderstorm wind gust climatology. Preliminary results
show measured severe wind gusts to be most frequent in the Plains
and portions of the Midwest. This is in contrast to the severe
thunderstorm wind database showing the maximum in severe thunderstorm
wind gust/damage frequency over the Appalachians. It was found
that a much lower frequency of measured wind gusts exists there compared
to the Plains. Through this work Andrew increased his skills in manuscript
writing, oral presentations, statistical methods, and Geographic
Information System analysis techniques.
2009 Participants:
Hannah Barnes investigated NWS tornado warning
statistics for marginal storm events. Hannah quantified the
false alarm rate (FAR) and probability of detection (POD) for three
scenarios: i) days with no reported tornadoes; ii) days with only
one reported tornado; and iii) outbreak days with ten or more reported
tornadoes within a Weather Forecast Office county warning area. Hannah
identified three key results. First, the large-scale environment
differed little between zero and one tornado days, but both differed
significantly from large outbreak events. Thus, there were
no large-scale signatures differentiating between zero and one tornado
days. Second, the circulation intensity of false alarms at
the lowest height , as scanned by WSR-88D radars, was notably weaker
than those associated with confirmed tornado warnings. Third,
the presence of obscured velocity data (marked by ‘purple haze’)
led to a 15% increase in the false alarm rate. Hannah’s
findings are critical first steps in understanding how to reduce
the number of tornado false alarms.
Wind ramp events – abrupt changes in wind power output due
to variations in wind speed – are a growing concern in the
wind power industry. Kristen Bradford examined the
climatology of wind ramp events at 34 METAR sites in the Southern
Plains during June-July 2009. The observations were used to validate
Weather Research and Forecasting (WRF) model forecasts on a 3-km
grid. Owing to the paucity of instrumented tower data, 10-m winds
were used for the study. The WRF model performed well during frontal
passages but did not capture the temporal variability of the observations.
Similarly, although there little overall bias in the forecast wind
speeds, many more ramps were noted in the observations owing to the
temporal variability.
David Gorree collected 20 years of 1-km resolution,
biweekly maximum-value composite normalized-difference vegetation
index (NDVI) data from polar orbiting satellites over the contiguous
United States and converted the data to vegetation fraction for periods
centered near 1 April and 1 May. Analyzed these data to produce
mean, maximum, minimum, and standard deviation fields and explore
interannual vegetation variability. Developed improved skills
in programming languages and visualization tools.
Stephanie Hoekstra got a taste of what it is like
to integrate social science into meteorology. She looked at how the
public perceives severe weather as well as whether tornado warning
lead-times longer than the current average lead-time (about 13 minutes)
are in demand. She surveyed National Weather Center (NWC) visitors
ranging from 18 to 65+ years of age. Many social science studies
only sample undergraduate students, so the broad range in ages is
noteworthy. Stephanie found that her sample population perceived
weather risks and fatalities fairly accurately, but interesting patterns
were emerging for the way different age groups or people from different
areas perceived and ranked these risks. She also found that the participants
would prefer a tornado warning lead-time of at least 13.8 minutes,
with an ideal lead time of 33.5 minutes. Stephanie learned about
creating surveys and the difficulties that can accompany that process.
Additionally, she learned some introductory methods for analyzing
survey data, as well as ways to correlate the perceptions of those
surveyed to the climatology of severe events.
Alex Kowaleski evaluated lightning and severe thunderstorm
forecasts from the European Storm Forecast Experiment (ESTOFEX).
He found that between-forecaster variability is about the same or
less than between-season variability, suggesting that the ESTOFEX
forecasters put out products that look like they come from a single
unit rather than individual forecasters. By utilizing new approaches
to visualization of forecast performance, Alex was able to show the
progression of forecast performance through the year and between
different years. Such techniques will be applied in the future to
forecasts from the NWS Storm Prediction Center.
Lauren Potter quantified NWS warning lead times
for reported severe hail and damaging hail events. Lauren compared
two years (1999 and 2000) of severe hail reports and ten years (1999-2008)
of damaging hail reports from Oklahoma, Colorado, Massachusetts and
South Carolina. Interestingly, she found no significant differences
among those states in the lead time of reported severe hail or damaging
hail. The mean lead time for severe hail was 18-22 minutes,
with a lead time ranging from 19 to 29 minutes for damaging hail
across the four states. Overall, Lauren found that about 72%
of reported hail occurs during a Severe Thunderstorm warning and
another 7% occurs during Tornado Warnings. Such a relatively
long lead time and warning rate provides the general public and emergency
and government services with the opportunity to take preventive cautions
and thereby mitigate at least some property damage from hail.
Astrid Suarez Gonzalez considered numerical forecasts
of banded snow. Banded snow is one of the greatest winter weather
forecasting challenges faced operationally with large economic and
human safety consequences. Her work focused specifically on
using techniques that have been successfully used to improve forecasts
of springtime convective systems: namely, convection-permitting forecasts,
ensemble forecasts, and data assimilation. The work she did
is quite unique: Some people argue that winter phenomena may not
be as sensitive to these techniques because the flow is usually forced
by larger-scale processes. However, Astrid demonstrated that these
techniques can greatly improve a forecast. Astrid came to REU
with a great interest in anything related to numerical modeling. She
had never heard of data assimilation before but was really enthusiastic
to learn and try and is now very interested in this research area.
Cristal Sampson (funded through CUNY's REU) conducted
evaluation research on existing user feedback surveys from the ongoing
research and development of the National Weather Radar Testbed's
Phased Array Radar (PAR). The PAR is a research radar that is under
consideration to replace the Weather Surveillance Radar-1988 Doppler
(WSR-88D). As a new technology it is important to provide user insight
into the development stage to ensure intended users have the most
usable tool upon deployment and not only understand the operational
utility of PAR. Results from experiments held in 2008 and 2009 have
already assisted researchers developing PAR. The participants of
these experiments evaluated real-time and archived cases; after each
evaluation questionnaires were completed. The responses to two archived
cases were analyzed in this paper using a data-driven method. The
results show how high-temporal resolution data of PAR impacted the
participating forecasters in a simulated warning environment. Suggestions
are made to improve future research and development.
Jeff Viel performed a robust statistical analysis
of temperature time series data of US cities for which there are
weather contracts that trade on the Chicago Mercantile Exchange. Jeff
was able to utilize Fourier decompositions of the data to remove
the seasonal signal in the first and second statistical moments,
leaving a distribution of historical residuals. Jeff demonstrated
that the residual distributions are, in most cases, not drawn from
a normal population. This finding is very important, for it has implications
on the way in which options on weather futures contracts could be
priced. Then, using the statistics, he developed a stochastic
model that attempts to simulate realistic temperature paths for the
eventual purpose of incorporation into a pricing model. The
preliminary results of Jeff's model seem very promising.
Travis Visco used a new least, linear squares derivatives
(LLSD) technique (developed at NSSL) to derive azimuthal shear and
radial convergence fields. He compared the shear and convergence
fields to tornado tracks to form a distribution. His study represents
the first such effort to obtain these distributions of the derived
LLSD fields. Travis also separated out the first tornadoes from storms
in our database. By doing this, he could isolate (i.e., avoid interference
from ongoing tornadoes) trends in the fields prior to tornado touchdown.
These trends showed a general increase in azimuthal shear prior to
tornado touchdown, especially in the leading 5 to 10 minutes. Travis'
most significant find was the large spread in the distribution of
azimuthal shear values.
Jonathan Vogel conducted a survey of NWS meteorologists
to assess the impacts of super-resolution radar data on signature
recognition and warning-decision making. The majority of meteorologists
surveyed indicated that they did see an improvement in signature
recognition for various signatures noted in the literature (i.e.
gust fronts/boundaries). When it came to warning-decision making,
the meteorologists were a little more reserved in their comments
because they wanted to give super-resolution data a little more time
before making a judgment. Jonathan gained valuable experience in
developing and conducting human surveys. He also gained experience
in analyzing radar data.
In order to begin developing an understanding of social networking
as a means for communicating weather information to the general public,
particularly with regard to time-critical information about severe
weather, Justin Wittrock developed and distributed
a web-based survey designed to address fundamental questions about
this issue. In contrast to many REU projects, which represent student
participation in an ongoing research program established previously
by the mentor, Justin’s project was self-initiated. He learned
how to develop an effective survey and in particular, how to pose
questions in a neutral manner. He also gained experience in the Institutional
Review Board process, learned how to identify communities and sample
them in appropriate ways, and how to apply statistical analysis techniques.
Most importantly, he was shown how to explain findings, rather than
simply present them, and pose questions for further study based upon
them.
2008 Participants:
Blake Allen investigated the effects of enabling
prediction of a second microphysical moment, number concentration,
for each hydrometeor category in a mixed phase, bulk microphysics
scheme in a cloud-resolving numerical prediction model. The electrification
of the storm was also found to be quite sensitive to changes in the
microphysics complexity, due at least in part to variations in cloud
ice and graupel production in the different model runs. Along the
way, he learned about scientific computing in a UNIX environment,
using 3D visualization tools, and helped uncover errors in the model
code as it was being developed.
Severe weather watches are a part of a series of products issued
by the Storm Prediction Center (SPC) that are used to alert forecasters,
emergency managers, the media, and the public of the likelihood of
the occurrence of severe weather. What makes severe weather watches
important is their ability to help improve public safety and help
save lives as they make people aware of the potential danger of severe
weather within their area in the hours immediately following the
issuance of the Watch. Becky Belobraydich surveyed
the college students at Northern Illinois University and The University
of Oklahoma. The students' responses were then analyzed to see what
they knew about Watches and how they responded to them. The responses
were right in line with what we thought they would be, and they point
to the fact that the majority knew there county but did not know
the counties next to them. More research and public education is
warranted to get the SPC Watches to the full effect.
Tim Bonin looked into the notion that we have experienced
more ice storms in the southern plains in recent decades, compared
to prior history. His research combined two large datasets: the climatological
record of winter precipitation and upper-air data that informs it.
He did a solid job of working around, and reasoning through, some
limitations of the data. His results showed that the precursors for
icing events remained largely unchanged, but the scenarios that potentially
support high-end events may have increased in the last decade.
Madison Burnett focused on evaluating the amount
of tornado activity that occurred during the autumn and winter months
of 2007-08 and comparing that activity with the historical record
of tornadoes and tornado-related fatalities. She discovered a substantial
upswing in all tornado reports over the past 50 years but very little
change, to a slight downward trend, in the strongest and most violent
tornadoes reported during the cool season months over this same period
of time. Further analysis revealed that cool seasons with a large
number of tornado-related fatalities have appeared in the record
about once a decade over the past 50 years. In order to conduct this
work, Madison had to become proficient not only in the format of
the NWS/SPC tornado database but also in the use of the structured
query language used in evaluating tornado data contained within the
database. In a side project comparing the SPC and NCDC tornado databases,
Madison uncovered a peculiar inconsistency where extra counties existed
in the SPC database for tornadoes tracking up to and perhaps crossing
a county border. This inconsistency was most noticeable for a period
of the 1970s and may have been due to different tornado coding standards
used at the time. Further investigation into this issue is needed.
Brad Hegyi applied lake-effect snow forecasting
parameters to lake-effect snow cases on the west side of Lake Michigan
to see if those parameters were helpful in forecasting those relatively
rare lake effect events. He found that northeast and north-northeast
winds at 850 and 925 mb were common to western Lake Michigan lake-effect
snow events, in addition to a minimum of a 13°C temperature difference
between 850 mb and the lake surface.
Christina Holt investigated the physical characteristics
of a tornado producing mini-supercell that occurred over Oklahoma
during tropical storm Erin. The mini-supercell had a shallow circulation
only 4.5 km in diameter that extended through a depth of only 3 km.
These physical attributes are consistent with previous studies of
similar storms. Christina's study was unique in that she was able
to use data from the Multifunction Phased Array Radar (MPAR), operating
in Norman, OK, to sample the rapid intensification to a tornadic
phase. This transition took only 3 minutes. The case serves as an
example of how higher temporal sampling might improve hurricane-spawned
tornadoes and improve forecasts and warnings of them.
Kelly Keene examined tornado warning performance
in relation to watches for 1997-2007. Her database consisted of over
30,000 tornado warnings and over 15,000 tornadic events. She found
that the issuance of any watch improves the overall tornado warning
performance. She found that the Probability of Detection (POD) of
tornado warnings increases by .327 when a tornado watch is in effect,
as opposed to when no watch is in effect. Lead time from tornado
warning occurrence is improved by an average of five to six minutes
when a tornado watch is in effect, as opposed to no watch. Finally,
when a tornado watch is in effect, there is a slight decrease (in
the amount of 0.81) in False Alarm Ratio (FAR) compared to when no
watch is in effect.
Jennifer Newman analyzed data from nine interviews
with meteorologists from two key stakeholder groups in the Southern
Plains, NWS forecasters and TV broadcasters, to attain specific information
about current radar capabilities and how those capabilities helped
or hindered participants' ability to fulfill their roles. Her analysis
revealed that the problems participants spoke of fell into four basic
needs. First, meteorologists clearly conveyed the need for reliable,
clean, and accurate radar data. Second, several stories involved
weather situations that evolved more rapidly and on smaller spatial
scales than WSR-88D can sample. Third, both groups told stories illustrating
advantages of high-resolution and low-altitude station or TDWR radar
data, and how the lack of that information in other areas hampered
their awareness of the weather that was occurring. Finally, size,
distribution and type of hydrometeors in both warm and cold season
events were critical information participants could only partially
infer in data from current radar systems.
Jonathan Poterjoy's project addressed a fundamental
question about using observations to improve ocean models: what area/volume
within an ocean model should a single observation influence? His
work shows that the answer varies widely depending on the local bathymetry,
depth and by variable, and importantly, that there are high correlations
at great distances from the observation that are spurious and must
be trimmed (i.e., localization). His work also shows that some cross
variable correlations (i.e., salinity/temperature) are significant.
With the knowledge generated by Jonathan's project, there is a now
benchmark for devising automated methods to calculate localization
distances.
Christopher Wilson successfully used a high-resolution
hail verification dataset to evaluate several hail diagnosis techniques.
Chris's project may also be one of the first projects to use lead
time, in a meaningful way, in algorithm performance evaluation. Chris
tracked observed hail sizes and storm attributes at discrete times
(i.e., a radar volume); this is in contrast to other studies which
arbitrarily relate storm attributes to hail sizes by using a time
window (e.g., +/- 20 min). While different diagnosis techniques had
high probability of detection and low false alarm ratio scores, which
hint at good performance, relatively high probability of false detection
scores hampered overall performance (determined by Heidke Skill Score).
Finally, Chris showed that for lead times greater than 10 min, all
evaluated hail diagnosis techniques showed poor skill.
Jeff Zuczek investigated the climatology of when
weather would likely be favorable for prescribed burning. Such burns
are critical for controlling invasive plants, including the Eastern
Red Cedar, which is notorious for breaking up pastureland and wildlife
habitats. Wind climatology using Oklahoma Mesonet data from January
1994 through May 2008 was analyzed to determine the number of days
each year meeting all criteria for prescribed burns, using a consensus
definition for "favorable burn day" from Oklahoma's 11
burn associations. Jeff's results showed good burning conditions
were more likely earlier, rather than later, in the burn season.
2007 Participants:
Rachel Butterworth took the opportunity to work
on developing a proposal that, if funded, could lead to hear master's
degree. Whether or not that might come to pass, she learned a great
deal about how to research and develop a proposal. She took her initial
idea of addressing the gaps in scientific literacy among the general
public with the anticipated capabilities of new radar technologies
such as the CASA concept and developed an education and communication
plan to that would help people take advantage of weather technology
in their daily activities.
Douglas Crauder successfully demonstrated the feasibility
of using two rather than three Doppler scans with the Multiple PRF
Dealiasing Algorithm when one of the two Doppler scans uses the phase
coding logic developed by Sachidananda and Zrnic to mitigate range
aliasing. The significance of using two rather than three Doppler
scans is that nearly thirty seconds can be removed from the time
required to complete a volume scan of data which is normally between
five and six minutes. During severe weather a reduction of
thirty seconds is important to operational forecasters who want a
rapid update for assessing storms. The Applications Branch
expects to submit change requests to the WSR-88D system to add new
volume coverage patterns based on his findings.
Victor Gensini analyzed time series from 1958-1999
of high values of atmospheric variables important for severe thunderstorms
in regions of North and South America, based off of NCAR/NCEP reanalysis
data. He learned how to look at cumulative distribution functions
of very large datasets in an efficient manner, so that comparisons
between different periods and locations could be assessed.
Eric Guillot found that the amount of forecast
skill involved when issuing tornado and severe thunderstorm warnings
is closely related to the type of storm that causes the severe weather.
It was found that, for a sample of over 3000 warnings, both tornado
warnings and severe thunderstorm warnings issued for isolated supercells
and convective line storms have better skill scores than those issued
for pulse and non-organized storms. Lead times were consistently
longer for supercell and line storms, while usually very short for
pulse and non-organized storms. We concluded that measures of forecast
skill are particularly sensitive to the type of storm. Thus, any
measurement of forecast skill, such as the year-over-year skill measure
of an individual forecast office, has to take into account the types
of storms in that office’s warning area in the time period
considered. This project focused on the analysis of multi-radar,
multi-sensor data from convective storms, statistics, and severe
weather verification techniques.
Stephanie Henry helped to develop a procedure for
determining cloud forests in Central America using MODIS imagery
with 250m spatial resolution. Cloudiness was extracted from
the visible images via an algorithm and this cloudiness was further
stratified by its annual and diurnal variations. Together these
allowed the mapping of regions of differing cloudiness, which could
then be related to cloud forests estimated via other means. The
procedure can be applied globally to map different vegetation regimes
based on the satellite observed cloudiness.
Luke McGuire developed a tool that allows us simulate
satellite orbit tracks and sensor field-of-view configurations and
project this track onto the globe. Then, using a cloud-cover
database, he developed a method to determine for each satellite footprint
the probability that the sensor will encounter clouds – and
if it does, the expected cloud altitude. This tool allows us to investigate
the impact of cloud-cover on satellite configurations. The
results can be used to assess the utility of new sensors by allowing
for a robust simulation of clouds when we simulate the satellite
measurements. Luke wrote an excellent summary of his work and
we are in the process of expanding this into a refereed journal paper.
Scott Powell learned a number of social science
statistical approaches and how to properly manipulate variables involved
in how people make decisions based on weather information. He was
very quick to learn a new statistical software package (SPSS), effectively
incorporated the valuable tips his mentor, a professor in the Department
of Communications, gave him on presentation skills, and wrote a well-formatted,
organized, and developed final paper. Scott found that individuals'
responses to weather information vary demographically, especially
by geography, age, and gender. Californians, for example, reported
less planning, readiness, and trust in weather information, no matter
the source. Over one third of the sample population did not know
the difference between a severe weather watch and a warning.
Jessica Ram successfully quality controlled and
organized thousands of storm reports and national weather service
issued warnings for over 250 watches from the first few months of
2006. She learned all about 2x2 contingency tables and statistics
related to warning and watch performance like false alarm rate, probability
of detection, and critical success index. She also received
over 40 completed surveys from NWS forecasters all across the country.
Bo Tan worked on developing a strategy to relate
satellite imagery to tropical wave positions identified with special
radiosonde data collected over West Africa during the NAMMA - 2006
field program. The procedures that Bo explored will be expanded
to develop multi-year climatologies based on geostationary satellite
imagery that distinguish rapidly developing tropical waves from those
that develop slowly, or do not develop. This ongoing study
should benefit longer-range hurricane forecasting over the Atlantic.
|
