Convective initiation and storm tracking from geostationary lightning mappers
Total lightning trends from geostationary optical mappers give forecasters a real-time proxy for convective updraft strength, often preceding severe weather by 10–20 minutes. This page covers the physics, the sensors, and the honest limits of the method.
Sensors
- GLM on GOES-16 / GOES-17 / GOES-18 (NOAA): Single-band optical sensor at 777.4 nm (oxygen triplet emission). Pixel footprint approximately 8 km at nadir, degrading toward the limb. Detects individual lightning events, groups and flashes at sub-millisecond temporal resolution. Full-disc coverage of the Americas and adjacent oceans with continuous, gapless imaging.
- LMI on FY-4A (CMA, China): Lightning Mapping Imager operating at the same 777.4 nm oxygen triplet. Spatial resolution approximately 7.8 km at nadir over East Asia and the western Pacific. Provides flash-rate data every millisecond, comparable in design philosophy to GLM though with independently characterised detection efficiency.
- MTG Lightning Imager on MTG-I1 (EUMETSAT, forthcoming operational): Four-camera design covering Europe, Africa and the eastern Atlantic. Designed for roughly 4.5 km pixel resolution at nadir, a meaningful improvement over GLM. First MTG-I satellite launched December 2022; LI commissioning and operational data delivery expected to follow qualification. Will close the persistent geostationary lightning gap over Europe and Africa.
- WWLLN (World Wide Lightning Location Network): Ground-based VLF network providing global cloud-to-ground stroke locations with median accuracy of roughly 10 km and detection efficiency that varies by region, typically 30–50% for strokes above 10 kA. Used operationally as independent ground-truth against GLM optical detections, particularly at the limb where GLM efficiency degrades.
Why total lightning is a better updraft proxy than radar alone
Radar reflectivity tells you where precipitation-sized particles are. It says less about what the updraft is doing right now. Lightning tells you something different: it records the rate at which ice crystals and graupel are being separated by collisions inside a vigorous updraft, a process that charges the cloud and discharges it electrically. The faster the updraft, the more collisions, the higher the flash rate. This relationship is imperfect and storm-dependent, but the directional signal is consistent enough to be operationally useful.
The practical consequence is the lightning jump. Studies using data from ground-based lightning mapping arrays, particularly the North Alabama Lightning Mapping Array, documented that a rapid increase in total flash rate, defined operationally as a two-sigma departure from recent trend, precedes severe weather reports (large hail, damaging winds, tornadoes) by a median of around 10 to 20 minutes. That lead time is not guaranteed and varies considerably by storm type and environment, but it is additional time that radar alone does not provide. GLM makes this signal available continuously across an entire hemisphere.
What the 8 km footprint and sub-millisecond timing actually mean in practice
GLM's pixel footprint of approximately 8 km at nadir is coarser than ground-based lightning mapping arrays, which can resolve individual discharge channels to within hundreds of metres. What GLM sacrifices in spatial precision it recovers in coverage: no ground network spans an ocean basin. The sub-millisecond event timing allows the sensor to distinguish individual optical pulses within a flash, which in turn allows classification into events, groups and flashes, each a progressively more integrated product.
For nowcasting, the operationally relevant product is flash rate per unit area per unit time, typically aggregated over one-minute windows and gridded to a resolution consistent with the pixel footprint. At that scale, the 8 km footprint is not the binding constraint; the binding constraint is usually the storm's own spatial extent relative to the grid. A compact, intense supercell may occupy only a handful of GLM pixels, making spatial gradients within the storm difficult to resolve. Broad mesoscale convective systems are better sampled.
Detection efficiency is the less-discussed limit. GLM detects daytime flashes less efficiently than night-time ones because solar background radiance competes with the 777.4 nm signal. Published characterisation work puts daytime detection efficiency roughly 10 to 20 percentage points below night-time values, though the exact figure depends on scene brightness and cloud optical depth. At the limb of the geostationary field of view, pixel area grows substantially and the signal path through the atmosphere lengthens, reducing both spatial resolution and detection efficiency further. For GOES-16, the limb degradation becomes significant beyond roughly 70 degrees from nadir.
Convective initiation: catching the storm before the radar echo
The earliest radar echo from a developing convective cell typically appears when precipitation particles have grown large enough to produce meaningful reflectivity, often 20 to 30 minutes after the first lightning. GLM can, in principle, detect lightning before a radar echo is present, because electrification begins as soon as the mixed-phase region develops sufficient ice. In practice, the first GLM flash from a newly initiated cell is often weak and isolated, and distinguishing genuine initiation from background noise requires careful thresholding.
Operational implementations at NOAA combine GLM flash-rate trends with GOES-ABI infrared cloud-top temperature to track rapid cooling, a proxy for overshooting tops, alongside the lightning signal. Neither channel alone is sufficient. The combination narrows false-alarm rates to operationally acceptable levels, though verification statistics vary by region and season. Buyers should treat convective initiation detection as a probabilistic guidance product, not a deterministic trigger.
FY-4A LMI and the emerging global picture
GOES-16 and GOES-18 together cover the Americas. FY-4A LMI covers East Asia and the western Pacific. The forthcoming MTG Lightning Imager will cover Europe and Africa. When all three systems are operational, geostationary lightning monitoring will span most of the populated tropics and mid-latitudes continuously. The remaining gaps are the Indian Ocean basin and the central and eastern Pacific west of the GOES-18 limb, where WWLLN and spaceborne low-Earth-orbit sensors such as the ISS-mounted ASIM instrument provide partial, lower-temporal-resolution coverage.
FY-4A LMI data are distributed by the China Meteorological Administration and are available to international users under data-sharing arrangements, though access pathways are less standardised than for GOES GLM data distributed via NOAA's open data programme. Intercalibration between GLM and LMI remains an active research area; flash-rate comparisons in the overlap region near the date line show broadly consistent behaviour but non-trivial detection efficiency differences that complicate direct quantitative comparison.
Operational products and their honest limits
GLM Level 2 products are publicly available from NOAA with latency typically under five minutes from observation time, making them usable for real-time nowcasting. The primary derived products are gridded flash-extent density and flash-rate density, available at one-minute temporal resolution. These are the inputs to lightning-jump algorithms implemented in operational forecasting tools such as NOAA's WarnGen environment.
The method has documented skill against severe weather reports over the continental United States, where verification datasets are dense. Skill over oceanic and data-sparse continental regions is less thoroughly verified, partly because ground-truth severe weather reports are sparse there. Flash-rate thresholds calibrated for Great Plains supercells may not transfer directly to tropical oceanic convection or to high-shear, low-instability environments where storms produce severe weather with modest lightning rates. Any operational deployment should include regional recalibration against local verification data.
Satellize ingests GLM Level 2 data streams as part of its atmospheric analytics suite and can configure lightning-jump alerting and flash-rate trend monitoring for client-defined domains. The same pipeline that supports crop-condition monitoring for the Kingdom of Tonga programme can be adapted for convective hazard products over maritime or island domains where ground-based radar coverage is absent.
Typical figures
| Pixel footprint at nadir (GLM / LMI) | ~8 km (GLM), ~7.8 km (LMI); grows significantly toward limb |
| Temporal resolution (event detection) | Sub-millisecond event timing; Level 2 gridded products at 1-minute aggregation |
| Spectral band | 777.4 nm (oxygen triplet lightning emission line), all current geostationary mappers |
| Coverage (GLM) | Full disc: Americas, adjacent Atlantic and Pacific; continuous, no scan gaps |
| Data latency (GLM Level 2, NOAA open data) | Typically under 5 minutes from observation |
| Daytime detection efficiency penalty (GLM) | Approximately 10–20 percentage points below night-time, solar-background dependent |
| Limb detection efficiency (GLM) | Degrades materially beyond ~70° from nadir; pixel area and atmospheric path both increase |
| Archive depth (GLM on GOES-16) | GOES-16 operational from late 2017; GLM data available from early 2018 via NOAA |
| MTG Lightning Imager design resolution | ~4.5 km at nadir (four-camera mosaic); operational delivery pending commissioning |
| WWLLN stroke location accuracy | Median ~10 km; detection efficiency ~30–50% for strokes above 10 kA, region-dependent |
Analytics Satellize can run
| Lightning jump alerting | Two-sigma flash-rate departure algorithm applied to 1-minute GLM flash-rate density time series over client-defined polygons or storm-tracked cells | Real-time alert feed (JSON or webhook) with storm cell ID, jump magnitude, and time-to-threshold; configurable lead-time and false-alarm-rate trade-off |
| Convective cell tracking and lifecycle analysis | Object-based tracking of flash-rate density clusters combined with GOES-ABI infrared cloud-top temperature; cell birth, merger, split and dissipation logged | GIS layer (GeoJSON) updated every 5 minutes showing tracked cell polygons, flash rate, cloud-top temperature and trend arrow |
| Storm severity probability nowcast | Logistic regression or gradient-boosted classifier trained on GLM flash-rate trend plus ABI overshooting-top proxy; outputs calibrated probability of severe weather within 30 minutes | Gridded probability raster at 10 km resolution, updated every 5 minutes, delivered via WMS tile endpoint or flat GeoTIFF |
| Maritime convective hazard bulletin | GLM flash-rate density aggregated over user-defined maritime corridors or exclusive economic zones; WWLLN used to cross-check limb-affected areas | Automated PDF bulletin and structured JSON feed, one per hour or on-demand for active events; suitable for vessel routing and offshore operations |
| Historical flash-rate climatology for site risk assessment | Aggregation of GLM Level 2 archive (2018 to present) over client site footprint; monthly and diurnal flash-rate distributions computed with detection-efficiency correction applied | Site risk report with flash-rate percentile tables, diurnal cycle charts and seasonal breakdown; delivered as PDF with underlying CSV |
| Multi-sensor lightning detection cross-validation | Colocation of GLM events with WWLLN strokes in time and space windows; detection efficiency and false-alarm ratio estimated per region and season | Calibration report quantifying GLM performance over client domain; used to set operationally appropriate detection thresholds |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.