Assimilation of Geostationary Lightning Mapper Observations into the NOAA GSL Rapid Refresh Forecast System

Lightning activity, similar to high radar reflectivity, is a useful proxy for the location and intensity of convective storms. Unlike radar reflectivity, lightning observations are now available over a very large and spatially contiguous area: The GOES-R series satellites carrying the Geostationary Lightning Mapper (GLM) instrument provide continuous detection of total (IC+CG) lightning activity over much of Earth’s western hemisphere. These observations provide an indicator of convective activity that covers a much larger area than existing weather radar networks, including over oceanic regions, in mountainous regions where radar beams are often blocked by terrain, and in other regions where coverage gaps exist in current radar networks. Since lightning activity is co-located with thunderstorms and since the intensity of total lightning activity is correlated with microphysical and kinematic quantities within thunderstorms, lightning observation data such as the Flash Extent Density (FED) product provided by the GLM have the potential to lessen initial condition errors in both the location and intensity of thunderstorms within convection-resolving numerical weather prediction models. Improvements in model initial states are in turn expected to improve the accuracy of model forecasts.

Around a decade ago, before the first GOES-R launch, proof-of-concept efforts were underway to demonstrate the effective ensemble assimilation of simulated FED observations and pseudo FED observations into a storm-scale research model poorly suited for forecasting. More recently, assimilation techniques built upon that earlier work have matured to almost full operational readiness following the addition of the ability to assimilate GLM FED observations into NOAA GSL’s Rapid Refresh Forecast System (RRFS). The RRFS is a near-operational convection resolving forecast model with a domain covering all of North America and nearby locations and with the hybrid ensemble-variational (EnVar) Regional Data Assimilation System that allows for the use of static, flow-dependent, or blended background error covariances. Two implementations of GLM FED data assimilation for the RRFS have been developed. The first uses the Gridpoint Statistical Interpolation (GSI) data assimilation framework and is intended for use with RRFSv1, and the second uses the newer Joint Effort for Data assimilation Integration (JEDI) framework and is eventually intended for use with RRFSv2. In both versions, the lightning data is assimilated hourly and is linked to model physics using an observation operator that relates FED to the graupel mass mixing ratio in the model.

Retrospective experiments have been completed using the GSI-based implementation and the experiments show benefits to short-range forecasts of convection when GLM FED data is assimilated. This is especially the case if radar reflectivity data is not available for assimilation, a situation where experiments indicate that the assimilation of GLM FED data improves short-range forecasts of precipitation across a range of storm intensity and precipitation accumulation thresholds. Results from an additional retrospective experiment using the JEDI implementation will also be presented and compared with the results from the GSI-based experiments.