A Geostationary Lightning Pseudo-Observation Generator Utilizing Low-Frequency Ground-Based Lightning Observations

Abstract Coincident Geostationary Lightning Mapper (GLM) and National Detection Network (NLDN) observations are used to build a generator of realistic lightning optical signal in the perspective simulate Imager (LI) from European NLDN-like observations. Characteristics GLM NLDN flashes train different machine-learning (ML) models, which predict simulated pseudo-GLM flash extent, duration, event number per (targets) several characteristics. Comparing statistics observed targets targets, most suitable ML-based target generators identified. The then further processed obtain events flash-scale products. In data assimilation, extent density (FED) is derived both data. best accumulated hourly FED sums with bias 2% observation while cumulated absolute differences remain about 22%. A visual comparison reveals that features local maxima at similar geolocations as However, often exceeds regions convective cores high rates. area > 0 5 km × pixel by some differs only 7%–8% values. recommended uses linear support vector regressor (linSVR) create FED. It provides balance between simulation, sum, electrified area.