Research

May 19th, 2010

GRADUATE WORK

UNDERGRADUATE WORK

Thesis Research: Evaluating WSR-88D Methods to Predict Warm Season Convective Wind Events at Cape Canaveral Air Force Station and Kennedy Space Center

Forecasting downbursts is challenging since it is difficult to tell which convective cells (e.g. thunderstorms) will generate downbursts, when they will occur, and what the expected maximum gust will be. Nonetheless the 45thWeather Squadron (45 WS) is responsible for predicting wind gusts that exceed 35 kt, as well as 50 kt for Cape Canaveral Air Force Station (CCAFS) and Kennedy Space Center (KSC). Recently, an attempt has been made to use information from the WSR-88D’s Storm Cell Identification and Tracking (SCIT) algorithm. Previous work has shown that attributes from the SCIT algorithm could be used as a precursor in nowcasting a convective wind event. Additionally, results showed an intriguing relationship between the reported wind gust and height of the freezing level. However due to the small sample size, continued evaluation is needed. Also, the findings were based on the volume scan at or just before the time of the maximum reported peak wind gust. It would have been more useful if this information was correlated with previous volume scans, in order to provide a longer lead-time for forecasting these events.

Using WSR-88D storm structure data of warm season convective events from 2003-2007, predicted wind gusts derived from previous methods are tested against actual peak wind gusts reported from one of the 36 weather towers located on the CCAFS/KSC complex. Scatter plots, root mean square error (RMSE), and mean absolute error (MAE) were used to evaluate these methods. In addition, several new techniques to predict wind gusts were developed and assessed in order to improve forecast skill. Statistical methods, including logistic regression and classification and regression trees (CART), were used to develop a forecast technique with maximum skill, i.e. maximizing the True Skill Statistic (TSS) to find the optimum compromise between the Probability of Detection (POD) and Probability of False Alarm (POFA). The performance was verified using independent data from 2008-2009 not used to develop the techniques.

Results show that the previous methods do not perform as well as earlier results indicated. For a much larger sample size, high RMSE and MAE values, as well as low correlation coefficients were determined for the time of onset, with consistent performance for earlier volume scans. In addition, it appears that the relationship between the peak wind gust and the height of the maximum reflectivity in comparison to the freezing level has less validity than previously thought.

The new statistical methods introduced in this thesis appear to show promise in nowcasting convective winds on the CCAFS/KSC complex. Most of the models show high performance when tested against the independent dataset, with relatively high POD’s and low POFA’s. TSS values varied, however the promising methods had values above 30 percent. Values of TSS between 30 and 50 percent are considered marginally useful for operations at the 45 WS.

Radar Meteorology Project: Validation of the Radar Gust Equation Used at the Kennedy Space Center and Cape Canaveral Air Force Station

Forecasting for convective winds can be a rigorous task due to their small spatial resolution. Nonetheless the 45th Weather Squadron is responsible for predicting wind gusts that exceed 35 kts for the Kennedy Space Center and Cape Canaveral Air Force Station. While many resources are available, research has been performed in attempt to use WSR-88D technology. An equation developed by Loconto (2006) implements information from the storm structure alphanumeric product to predict the speed of a gust from a convective storm. This research will attempt to validate this equation outside the Cape Canaveral area. Scatter plots of reported wind gusts versus predicted wind gusts for three different areas (Continental Interior, Florida, Northeast Seacoast) show that Loconto’s equation may not be useful in all areas. In attempt to create a threshold for each area, multiple linear regression models are run in order to generate a new location specific equation to predict convectively driven gusts.

Advanced Meteorology Statistics Project: Forecasting for Convectively Driven Wind Events at the Kennedy Space Center: Statistical Methods

Forecasting for convective winds can be a rigorous task due to their small size. However many different methods, including those suggested by Loconto (2006) have been tested and put into operation. Unfortunately the probability of detection and false alarm ratios are not values we would like them to be. Using statistical methods described in class, a 13 year data set is used that contains 45 different variables as possible predictors. Different models are run via the R statistical program and they include logistic regression, Fisher’s discrimination, classification trees, and bootstrapping. Once the model is created, a leave one out cross validation is performed along with calculating the misclassification rate, probability of detection, and false alarm ratio. Results show that all of the methods are better than the ones currently in use. Additionally, relevant variables such as the average relative humidity and K-Index are deemed significant in many of the models.

Satellite Meteorology Project:A Statistical Analysis of the GOES-12 Sounder Derived Lifted Index

In 1994, Derived Product Imagery was implemented on the GOES-8 sounder to provide a representation of the atmosphere in areas where radiosonde data is limited. While numerous research has been done to test its validity, no recent research has been performed on the lifted index. This research tests the correlation between lifted index values on the GOES-12 Sounder with radiosonde measurements over January, March, June, and September of 2007 for four Florida stations. McIDAS was used to extract lifted index values from the GOES-12 Sounder using a 15 X 15 pixel grid and performing a Barnes analysis to calculate the average lifted index for each station. Linear regression, bias and RMSE are calculated to look at the correlation and averaged differences between the two LI values. While Miami had the best correlation, the convective months of June and September had poor results. This is possibly due to the assumptions, biases, and problems with the derived product imagery as discussed in this paper.

Advanced Synoptic Meteorology Project: Synoptic Analysis of Heavy Snowfalls over Central New England, 1996-2007

Forecasting snowfall totals for central New England can be a difficult task as slight synoptic and topographical changes can enhance or inhibit precipitation across mesoscale distances. The goal of this research project is to determine differences and similarities between the synoptic setup of storms that produce significant snowfall in Plymouth, NH, and the setup of systems that do not. Using archived climatological, METAR, and surface data, cases were selected and snowfall initialization times were utilized for NCEP/NCAR composites. The composites were split into three different sets: 44 cases where one of three selected sites (Plymouth NH, Concord NH, St. Johnsbury VT) received more than the local NWS-defined winter storm warning criteria of 7 or more inches of snowfall, of which 25 cases where Plymouth received warning criteria, and 19 cases where Plymouth received less than the warning criteria. Subjective analysis shows that while there are multiple similarities, subtle differences can be the deciding factor in whether Plymouth does or does not receive substantial snowfall.

Hollings Scholarship Research: Analyzing Conventional and Emerging Radar Technologies for the May 8th, 2007 Central Oklahoma Tornado Case

Since the implementation of the NEXRAD WSR-88D Doppler radar, forecasters have been able to better detect small-scale phenomena within storms. Nevertheless, some phenomena, such as tornadic vortex signatures, still remain difficult to detect due to the time required to complete volumetric scanning and the effects of increasing range from the radar on sampling. Over the past few years new radar technologies have been developed in an attempt to address these issues. This research, performed at the National Weather Center in Norman, Oklahoma, looks at both radars in operation (Oklahoma City’s WSR-88D and FAA’s Terminal Doppler Weather Radar) and radars used primarily for research (NSSL’s Phased Array Radar and a network of CASA radars).

Using the WDSS-II graphic user interface and data from a mesoscale convective vortex on the evening of May 8th, 2007, small scale circulations were identified and horizontal shear was calculated to compare the radars ability to see the storm. Eight circulations were analyzed and results showed that the radars each had an advantage and disadvantage. The Terminal Doppler Weather Radar calculated shear well, but its range was small compared to the other radars (~90km). While the Phased Array and CASA radars had high resolution data every 30-60 seconds, relatively small aerial coverage and limited operational periods of these experimental radars prevented a complete evaluation of these systems for the May 8, 2007 case.

The Human METAR is published under the Attribution-Noncommercial-No Derivative Works 3.0 Unported license.

Comments are closed.