REU 2002 Student Research Topics

Nonmeteorological Factors in Warning Verification

Student: Kathryn Carroll
Mentor: John Ferree (WDTB), Jim LaDue (WDTB), Andy Wood (CIMMS)
Abstract:

The purpose of this study is to increase the understanding of decision-makers on how nonmeteorological factors may impact performance improvement goals for individual forecast offices. The impacts of population density and distance to the nearest WSR-88D upon the National Weather Service warning performance measures are examined. This study focuses on warning verification for the St. Louis, Missouri County Warning Area (CWA).

No significant correlations between the performance scores and population density were found. Although no relationship was found, performance measures are likely impacted more by the density and veracity of spotter networks than by the amount of residents. There seems to be a slight correlation with radar distance and two performance measures: False Alarm Ratio and Critical Success Index. Furthermore, a collection of counties with similar population densities at varying distances were grouped together for comparison purposes with the performance measures. Results of this collection showed a more pronounced relationship for all measures. There is also a moderate correlation between the normalized number of warnings and events with both population density and distance to the nearest radar.

Modeling Oklahoma City Rainfall Occurrence Using a First-Order Markov Chain

Student: Erik Crosman
Mentor: Harold Brooks (NSSL), Matt Wandishin (UofAZ)
Abstract:

Rainfall occurrence for Oklahoma City is modeled using first and second-order Markov chains, which generate long time series of daily rainfall occurrence data. Occurrence frequencies of wet and dry spells are derived from the models. Model confidence is high for extended wet periods and short dry spells, but drops sharply for dry spells exceeding thirty days. This indicates that the processes underlying the persistence of long wet and dry runs may be significantly different, and that a better understanding of regional dry spell-climatology is needed.

The Effect of Population Grown on Killer Tornadoes

Student: Somer Erickson
Mentor: Mark Shafer (OCS)
Abstract:

Killer tornado data and population data have been studied in the region east of the Rocky Mountains in the United States, in order to look for any increase or decrease in the number of killer tornadoes when compared with population density. The data that is used is from 1950-2000, because these are the most accurate data years available. Before then records are not as accurate or reliable. These records are broken down by state and county. The population is then adjusted for natural change and then categorized into one of five categories: rural, small town, average, suburban, and urban. After much comparison it was found that there was no real change in the number of killer tornadoes in urban counties. However, it was determined that there was some increase in the number of killer tornado events in suburban counties. These results are useful to us so that we may better prepare for such occurrences and to create better ways of warning at-risk populations.

Lightning Prediction and Verification Within the IHOP Domain

Student: Kevin Goebbert
Mentor: Phillip Bothwell (SPC)
Abstract:

The prediction of cloud-to-ground lightning and the verification of the lightning forecast probability equations are examined. Lightning is predicted for three days for the 00-03 UTC time period from the 12, 9, 6 and 3-hr forecasts and the 00-hr analysis. Lightning prediction improves as the forecasts approach the analysis time. Of the three forecast days two days are scored for verification. A forecast Bias near 1 was obtained for the two days with a POD > 0.55, FAR < 0.50 and CSI > 0.40 for the 10% probability. Strengths and weaknesses of the forecast equations are discussed with possible solutions for improvement of the predictive scheme.

A Composite Study of Warm-Season Severe-Weather Episodes in the Florida Peninsula

Student: Michael Hardiman
Mentor: Pete Banacos (SPC), David Manning (NWSFO-TUL)
Abstract:

Archived Storm Prediction Center (SPC) Outlooks and National Lightning Data Network (NLDN) cloud-to-ground lightning strike data over the Florida Peninsula for the warm season (April-September) 1994-2000 are examined to create a limited-scope climatological perspective of the nature of Florida Peninsula convection and SPC outlook skill during the warm season. Upper-air rawinsonde data is used to create composite soundings in numerous stratifications based on the nature of the daily surface features, deep-layer shear values, and months of the year in an attempt to find a thermodynamic or kinematic signal that delineates days with non-severe and severe convection.

Results show that SPC skill decreases during the months of June and July, as scattered thunderstorms in a weak shear environment develop on a near-daily basis over the Florida Peninsula.

Results of the composite rawinsondes indicate a distinct reoccurrence of a midlevel dry air layer on most large severe weather days in the peninsula. Other factors such as an increasingly steep 850-500 mb and 700-500 Mb lapse rate are also apparent on days exhibiting severe thunderstorms.

Case studies demonstrate the appearance of the aforementioned mid-level dry layer on many severe weather days, with a few outliers possibly indicating the existence of more than one typical profile for severe weather in the Florida Panhandle.

The Trustworthiness of VVP Wind Estimates

Student: Karly Klein
Mentor: Alan Shapiro (SOM), Paul Robinson (CAPS)
Abstract:

Velocity Volume Processing (VVP) is a simple wind retrieval technique which uses radial velocity data from one Doppler radar to estimate the full wind field. The technique assumes that the wind is constant among x and y, but varies linearly with z (within user-specified volumes). After running a single-Doppler code, the results are checked against an estimate of the winds obtained from a dual-Doppler wind analysis. We hypothesize that the VVP procedure itself can be used to estimate the regions of the wind flow where the technique can be trusted. Tests performed with gust front data suggest that it is possible to quantify the trustworthiness of VVP winds using VVP estimates. These trusted winds can be used as a data source in numerical weather prediction to improve the forecast of weather.

Significant Severe Thunderstorm Proximity Soundings

Student: Stephanie Nordin
Mentor: Jeff Craven (SPC), Harold Brooks (NSSL)
Abstract:

Four hundred and sixty-eight 0000 UTC proximity soundings were examined in an attempt to find parameters that may discriminate between significant tornadic and significant non-tornadic environments. Significant severe weather is defined as a storm having an F2 or greater tornado, 2.00" hail or greater, and/or 65 knot wind speeds or greater. The data set was constructed between the dates of September of 1993 through December of 1996. In this study, proximity is defined as a significant severe weather event that occurred within 100 nm of a United States rawinsonde site, and between 2100 UTC and 0300 UTC (six hour time period centered on 0000 UTC launch). It was shown that low-level shear and mean layer lifted condensation level heights were the best discriminators in identifying these environments.

A Surface Wind Climatology for Central Oklahoma in July Using Oklahoma Mesonet Data

Student: Andrew Philpott
Mentor: Jeff Basara (OCS)
Abstract:

Wind direction and the relationship between wind speed 2 meters above ground and wind speed 10 meters above ground were studied for six Oklahoma Mesonet stations near Oklahoma City during 1994-2001, with an emphasis on applications to surface airflow experiments. Given any 10 meter wind speed, the 2 meter wind speed was within some 1 m/s range. In different years, 55-84% of winds were southerly and 7.5-31 % were easterly, but on average only 4% of winds were westerly or northwesterly. Wind direction also depended on time of day, with maximum winds at 7-10 pm being 150-160 degrees and maximum winds at 10 am-1 PM being 200-210 degrees. These conclusions can assist prediction of wind behavior near the surface in July.

A Climatology of Drizzle for North America

Student: Addison Sears-Collins
Mentor: Bob Johns (CIMMS), Dave Schultz (CIMMS)
Abstract:

Drizzle is a significant forecast challenge. Although most recent efforts to understand drizzle have been associated with marine stratocumulus and stratus clouds, few studies have examined the spatial and temporal distribution of drizzle. This paper investigates those subjects through a climatology created using surface observations from stations across North America between 1976 and 1990.

A monthly distribution of maximum drizzle occurrence reveals that 40% of the stations in North America have a drizzle maximum from November to January. These stations cover much of the United States. In Canada, the temporal variability is greater, and maximum drizzle occurs from September to December.

An analysis of the hour of most frequent occurrence of drizzle for North America reveals that east coast stations of the United States tend to report maximum drizzle earlier in the day (~10-11 UTC) than West coast stations (~14-15 UTC). This result suggests that drizzle is affected by the diurnal cycle because the increase in mixing of the boundary layer at sunrise produces instability, resulting in turbulence that increases opportunities for collision-coalescence to produce drizzle. Freezing drizzle also may be related to the diurnal cycle. The result of a recent study on freezing drizzle will be used for comparison.

An examination of the average number of hours of drizzle at each hour per year reveals that the temporal distribution at these high drizzle frequency locations in North America is varied. In addition, due to either local effects or a low number of drizzle observations over the 15-year period, a small percentage (13%) of stations dispersed around the continent report drizzle maxima from June to August.

The Influence of Initial Conditions on the SPC 2001 Ensemble Cloud Model Forecasts

Student: Shanna Sampson
Mentor: Kim Elmore (CIMMS)
Abstract:

As an extension to the Storm Prediction Center's 2001 Spring Experiment, a cloud model ensemble is used to determine its utility as a forecasting tool. Based on the Day 2 outlook, soundings are extracted from three mesoscale models over a 160 x 160 km area. These soundings serve as initial conditions for two cloudscale ensemble models. One ensemble contains a mix of soundings from the Operational Eta and a locally run version of the Operational ETA using the Kain-Fritsch convective parameterization, and the other ensemble contains a mix of soundings from the same locally run version of the Operational ETA and the Rapid Update Cycle model. Previous work has only examined the results of all the models combined. This paper addresses the question of how much each mesoscale model influences the verification results. Ensemble model results are characterized by five storm-length categories: no storms, short-lived storms, medium-lived storms, long-lived storms, and supercells. Except for long-lived storms and supercells, combining initial conditions from all mesoscale models yields the best results. For long-lived storms and supercells, verification results are heavily influenced by initial conditions from the Operational ETA