A thunderstorm warning system is only as good as the data that triggers it. Before promising an alarm, you need to know whether your lightning source sees the storms that actually happen. So we ran a test that is simple to explain and heavy to compute: count every lightning flash for a year around 81 Latin American sites, and compare the result with a measurement made by a different instrument, based on different physics.
What we measured
We used NOAA's geostationary optical lightning sensor, which watches the continent and publishes a file every 20 seconds. It detects total lightning: cloud-to-ground and in-cloud. We read all 1,576,283 files from 4 May 2025 to 3 May 2026, one by one, with zero read failures. Across the whole field of view we processed 422.5 million flashes.
For each flash we computed the distance to 81 sites in 13 countries and kept those within 25 km. We also counted those within 16 and 8 km, the monitoring and surrounding areas of the IEC 62793 standard.
The comparison
For 37 of those sites there is an independent measurement from a global ground network that picks up the radio waves of lightning (VLF), over the same 12 months and the same radius. A satellite flash and a network stroke are not the same unit, so we did not compare exact counts. We measured three things: whether both rank the sites the same way, whether they grow together, and how many network strokes there are per satellite flash.
The result is clear. The rank correlation is 0.95, on a scale where 1 is the exact same order. On log scales the correlation is 0.97 and it holds across five orders of magnitude, from sites with a handful of strikes a year to sites with tens of thousands. The network counts a median of 1.45 strokes and pulses per satellite flash, which is expected since one flash usually carries several strokes. 70% of sites fall within a factor of 2 of that ratio. Dropping the sites with fewer than 100 flashes a year, the correlation is still 0.93.
Where they disagree
The comparison also shows where the sources part ways, which is as useful as the agreement. In southern Peru and northern Chile the ground network counts 2.5 to 4.3 strokes per satellite flash, more than twice the usual ratio. One hypothesis consistent with the literature is that the satellite is less sensitive to short, dim flashes, typical of small high-altitude storms. In the Atacama Desert the opposite happens, but with so few events the ratio is unstable: at one site, a single day, 31 July 2025, accounts for 88% of the year's flashes. Neither hypothesis has been proven yet.
When lightning strikes
The census leaves a second practical finding. In the Andes only 7% of flashes happen at night; the peak comes between 14:00 and 17:00. In Colombia, the Caribbean and Central America 29% happen at night, and at one site in northern Colombia more than half. A warning that only works by day does not protect everyone equally.
Exposure isn't what you might expect, either. The five most exposed sites on the list are in northern Colombia, but a high-altitude site in southern Peru had lightning within 16 km on 193 days of the year.
What this does not say
Some limits are worth stating before anyone asks. The two sources measure different units, so the comparison holds for order and magnitude, not for matching counts. Neither is ground truth: the satellite detects 70 to 90% of flashes and locates each one to within about 4 km, which matters near the 8 and 16 km rings. It is a single year, and lightning activity shifts with El Niño and La Niña. Each site is a single point. And this census validates the input data, not the alarm: detection, false alarms and lead time are measured separately, event by event, under IEC 62793.
Why it matters
Two instruments with different physics, one seeing the light of lightning from 36,000 km and one picking up its radio waves thousands of kilometres away, share no sources of error. When they agree, it is unlikely both are wrong in the same way. That is the evidence a meteorologist asks for before trusting an alarm, and it is the foundation we built Pulse on.
References
Goodman, S. J. et al. (2013). The GOES-R Geostationary Lightning Mapper (GLM). Atmospheric Research 125–126, 34–49. · Rudlosky, S. D. and Virts, K. S. (2021). Dynamic GLM detection efficiency. Monthly Weather Review 149, 979–998. · Said, R. K., Inan, U. S. and Cummins, K. L. (2010). Long-range lightning geolocation using a VLF radio atmospheric waveform bank. Journal of Geophysical Research 115, D23108. · Albrecht, R. I. et al. (2016). Where are the lightning hotspots on Earth? Bulletin of the American Meteorological Society 97, 2051–2068.