Developing Empirical Lightning Cessation Forecast Guidance for the Kennedy Space Center

The Kennedy Space Center in east Central Florida is one of the few locations in the country that issues lightning advisories. These forecasts are vital to the daily operations of the Space Center and take on even greater significance during launch operations. The U.S. Air Force's 45th Weather Squadron (45WS), who provides forecasts for the Space Center, has a good record of forecasting the initiation of lightning near their locations of special concern. However, the remaining problem is knowing when to cancel a lightning advisory. Without specific scientific guidelines detailing cessation activity, the Weather Squadron must keep advisories in place longer than necessary to ensure the safety of personnel and equipment. This unnecessary advisory time costs the Space Center millions of dollars in lost manpower each year. This research presents storm and environmental characteristics associated with lightning cessation that then are utilized to create lightning cessation guidelines for isolated thunderstorms for use by the 45WS during the warm season months of May through September. The research uses data from the Lightning Detection and Ranging (LDAR) network at the Kennedy Space Center, which can observe intra-cloud and portions of cloud-to-ground lightning strikes. Supporting data from the Cloud-to-Ground Lightning Surveillance System (CGLSS), radar observations from the Melbourne WSR- 88D, and Cape Canaveral morning radiosonde launches also are included. Characteristics of 116 thunderstorms comprising our dataset are presented. Most of these characteristics are based on LDAR-derived spark and flash data and have not been described previously. In particular, the first lightning activity is quantified as either cloud-to-ground (CG) or intra-cloud (IC). Only 10% of the storms in this research are found to initiate with a CG strike. Conversely, only 16% of the storms end with a CG strike. Another characteristic is the average horizontal extent of all the flashes comprising a storm. Our average is 12-14 km, while the greatest flash extends 26 km. Comparisons between the starting altitude of the median and last flashes of a storm are analyzed, with only 37% of the storms having a higher last flash initiating altitude. Additional observations are made of the total lightning flash rate, percentage of CG to IC lightning, trends of individual flash initiation altitudes versus the average initiation altitude, the average inter-flash time distribution, and time series of inter-flash times. Five schemes to forecast lightning cessation are developed and evaluated. 100 of the 116 storms were randomly selected as the dependent sample, while the remaining 16 storms were used for verification. The schemes included a correlation and regression tree analysis, multiple linear regression, trends of storm duration, trend of the altitude of the greatest reflectivity to the time of the final flash, and a percentile scheme. Surprisingly, the percentile method was found to be the most effective technique and the simplest. The inclusion of real time storm parameters is found to have little effect on the results, suggesting that different forecast predictors, such as microphysical data from polarimetric radar, will be necessary to produce improved skill. When the percentile method used a confidence level of 99.5%, it successfully maintained lightning advisories for all 16 independent storms on which the schemes were tested. Since the computed wait time was 25 min, compared to the 45WS' most conservative and accurate wait time of 30 min, the percentile method saves 5 min for each advisory. This 5 min of savings safely shortens the Weather Squadron's advisories and saves money. Additionally, these results are the first to evaluate the 30/30 rule that is used commonly. The success of the percentile method is surprising since it out performs more complex procedures involving correlation and regression tree analysis and regression schemes. These more sophisticated statistical analyses were expected to perform better since they include more predictors in the forecasts. However, with the predictors available to us, this was not the case. While not the expected result, the percentile method succeeds in creating a safe and expedited forecast.