Preliminary investigation into lightning hazard prediction from high resolution model output
Stuart D. Miller Jr., University of Missouri, Columbia, MO; and G. W. Carbin, J. S. Kain, E. W. McCaul Jr., C. J. Melick, and A. R. Dean
The Lightning Threat 3 parameter (McCaul et al. 2009), hereafter referred to as Flash Rate Density (FRD), blends simulated graupel flux at the -15°C level with simulated vertically integrated ice content (VII) in a column to explicitly predict the occurrence of lightning. Both graupel flux and VII were implemented in a version of the Weather Research and Forecasting (WRF) model in Northern Alabama and were calibrated, weighted, and blended based on verification with the North Alabama Lightning Mapping Array (LMA). FRD was integrated into the National Severe Storms Laboratory (NSSL) 4 km Advanced Research WRF (ARW) model in early 2010. Since then, it has run daily (initialized at 0000 UTC) yielding maximum hourly values to 36 hours. This research seeks to ascertain the validity of FRD on the CONUS scale by verification with National Lightning Detection Network (NLDN) data mapped to the ARW grid. In order to do this, gridded FRD data from 12 March 2010 through 11 June 2010 underwent a binary conversion based upon an empirically derived threshold value so as to evaluate lightning/no lightning. First, the hourly percent domain coverage for FRD was compared to that of the NLDN data. Next, a Gaussian probability distribution function was applied to the FRD data to provide an estimate of lightning density, which can also be interpreted as a probability of lightning occurrence. The reliability and Relative Operating Characteristic (ROC) curve for the data were plotted for FRD forecast hours 13 – 36 over the period. Then, the raw data were mapped to the 212 grid used by the National Centers for Environmental Prediction (NCEP) Short Range Ensemble Forecast (SREF) system by flagging the nearest 40 km grid point to any active grid point in the native ARW grid. The 212 grids were summed over a four hour period each day (2000 UTC to 0000 UTC) to be consistent with the Storm Prediction Center's (SPC) Experimental Enhanced Resolution Thunderstorm (ENHT) outlooks. Finally, The 40 km, 4 hourly data underwent the Gaussian smoothing function. The hourly NLDN data were gridded to the SREF 212 grid as above, without smoothing. Once this was done, the reliability and ROC curve for FRD were evaluated using 0-10%, 10-40%, 40-70%, and 70-100% probability bins and compared to the reliability and ROC curve for SPC's ENHT outlooks (which forecast 10%, 40%, and 70% probabilities of lightning). This was performed over the data set each day valid from 2000 UTC to 0000 UTC. In the same way, the FRD probability fields were compared to the calibrated cloud physics thunder parameter (CPTP) (Bright et al. 2005) which is currently in operations in the SREF 212 and is the primary tool used by the SPC to issue ENHT outlooks.
Extended Abstract (820K)
Session 4B, Forecasting Techniques and Warning Decision Making: Short-Range Forecasting I
Monday, 11 October 2010, 4:30 PM-6:00 PM, Grand Mesa Ballroom D
Browse or search entire meeting
AMS Home Page