Utilizing Low-Frequency Ground-Based Lightning Locating Networks to Simulate Optical Lightning Observations of Geostationary Satellites

The new generation of geostationary (GEO) satellites carries total lightning observing instruments, like the Geostationary Lightning Mapper (GLM) on the Geostationary Operational Environmental Satellite (GOES) satellites or the upcoming Lightning Imager (LI) on the Meteosat Third Generation (MTG) satellites. They see lightning as illuminated pixels in highly frequent (every 2 ms) optical images, referred to as events. GLM provides total lightning measurements over North- and South America and the adjacent oceans. MTG-LI will provide the same type of observations over wide parts of Europe, the Mediterranean Sea, Africa and the Atlantic ocean after its launch planned in 2022. Optical GEO lightning observations continuously cover large areas in support of numerical weather prediction (NWP) and early detection of severe storms.

This work aims at simulating MTG-LI observations over France to develop an assimilation technique using MTG-LI flash extent density (FED) prior to the satellite launch. Low-Frequency (LF) ground-based networks, i.e. the National Lightning Detection Network (NLDN) and Météorage in the US and France, respectively, provide total lightning observations. A proxy function is developed analyzing NLDN and GLM data. It is then applied to Météorage records to simulate MTG-LI data over France. Relations between the optical satellite data and the LF data are assumed to be similar in the US (GLM, NLDN) and in France (MTG-LI, Météorage).

The initial proxy function algorithm uses among others NLDN flash extent, duration and number of elements and produces synthetic GLM counterparts. Various statistical approaches ranging from simple regression methods to more advanced machine learning methods, such as random forest and multilayer perceptron, are tested to determine the relationships between characteristics of coincident NLDN and GLM flashes. The synthetic GLM characteristics are then validated against GLM observations to identify the best performing algorithm. Further processing of those synthetic characteristics deduces the locations and times of GLM events and consequently the FED product. This second part of the algorithm makes use of the locations and times of LF pulses/strokes, LF flash characteristics and random number generators.

The different lightning detection techniques will be first reminded and examples of concurrent observations will be shown. Then, the algorithms and their performances will be detailed. Finally, an example of a synthetic MTG-LI dataset including FED will be illustrated.