Applications of the GOES-R Geostationary Lightning Mapper for Severe and Local Storms.

Abstract

The recently deployed GOES-R series Geostationary Lightning Mapper (GLM) pro- vides forecasters with a new, rapidly-updating lightning data source to diagnose, fore- cast, and monitor atmospheric convection. Illuminated pixels (events) from the GLM in each 2 ms frame are joined into groups, and these groups are then clustered spatially and temporally into flashes. Lightning observations from the GOES-East and GOES-West GLMs cover a majority of the western hemisphere, with a considerable region of sensor overlap across the United States. Coincident lightning observations from both GLM sensors have revealed variations in flash detection efficiency and their observed characteristics within the region of sensor overlap. Additional studies leveraging dual-GLM observations of individual lightning flashes, combined with information from available ground-based lightning location systems, are needed to further validate and apply these data within the operational and severe storms research communities. Gridded GLM products have been developed to improve operational forecast applications, with variables including Flash Extent Density (FED), Minimum Flash Area (MFA), and Total Optical Energy (TOE). While these gridded products have been demonstrated and are now used by the U.S. National Weather Service, there is a continual need to integrate these products with other datasets available to forecasters such as radar, satellite imagery, and ground-based lightning networks. Data from the Advanced Baseline Imager (ABI), Multi-Radar Multi-Sensor (MRMS) system, and one ground-based lightning network were compared against gridded GLM imagery from GOES-East and GOES-West in case studies of two supercell thunderstorms, along with a bulk study from 13 April through 31 May 2019, to provide further validation and applications of gridded GLM products from a data fusion perspective. Increasing FED and decreasing MFA corresponded with increasing thunderstorm intensity from the perspective of ABI infrared imagery and MRMS vertically integrated reflectivity products, and was apparent for more robust and severe convection. Flash areas were also observed to maximize between clean-IR brightness temperatures of 210 to 230 K, and isothermal reflectivity at -10 ◦C of 20 to 30 dBZ. TOE observations from both GLMs provided additional context of local GLM flash rates in each case study, due to their differing perspectives of convective updrafts. One of the most notable flashes within the life cycle of a thunderstorm is the appear- ance of its first lightning flash (i.e. ’first flash’), which can indicate that deep, moist convection with a substantial updraft has initiated. Additionally, the first lightning flash can be the most hazardous to the public due to the limited time to take protective action. GOES-East/-West GLM first flashes were identified and investigated over the continental United States in 2022 for comparison with ground-based lightning location systems, radar data, and satellite imagery. First, individual first flashes from the GLM were manually identified and studied from multiple lightning location systems. Over three-thousand candidate first flashes from seven cases of semi-discrete convection were then manually analyzed to select an objective identification criteria for first flashes. Over 170-thousand first flashes from both GLMs were then collected across the CONUS for all of 2022, and studied by region to provide insight into the factors that influence thunderstorm development and storm electrification. First flashes most frequently coincided with periods of peak convective instability daily and seasonally, and occurred in regions with -10◦C isothermal reflectivity values between 35 and 45 dBZ, and ABI clean-IR brightness temperatures between 220 and 240 K. When matched with lightning flashes from a ground based lightning network and verified using a machine learning model, most first flashes were found to be intra-cloud flashes with exception to the southwestern United States. Machine learning methods have become more widely adopted when diagnosing light- ning activity and thunderstorm intensity, with operational and research applications. Based on coinciding satellite and radar observations from the two previously described studies, a two-tired machine learning problem was posed to identify (TRF1) and then classify convection (TRF2) using GLM observations as its truth. Two random forests were trained on four radar and satellite variables from seven cases of convection. Between the satellite and radar variables, isothermal reflectivity at -10◦C demonstrated the greatest importance for both models. When identifying convection clean-IR bright- ness temperatures less than 220 K and -10◦C reflectivity greater than 23 dBZ were frequently associated with lightning activity, even at low probabilistic thresholds from the first tier random forest model. When identifying convection capable of producing robust flash rates, the second tier random forest model demonstrated that -10◦C reflec- tivity greater than 30 dBZ were frequently associated with more intense convection. Lastly, a case study from 14 May 2022 demonstrated the capability of both tiers of the model to provide diagnostic information for thunderstorm activity and intensity.