LTN · Observational
Lightning
NASA LIS/OTD/ISS flash density × ECDF normalization · v2.0 · 2026-09
Lake Maracaibo, Venezuela — densest lightning on the planet
Observed total flash density from NASA satellite optical detection (LIS/OTD/ISS), 1995–2023.
What ships
Surfaces — 0–1 susceptibility derived from annual TOTAL flash density (intra-cloud and cloud-to-ground, flashes / km² / yr) on the NASA LIS/OTD/ISS satellite record. v2.0 changes the measured quantity, not just the calibration: v1.x counted cloud-to-ground only, so absolute densities and class boundaries both shift and v1.x values must be re-derived, not rescaled. A score of 0.000 means no flash was detected there across the record — measured, not missing.
Every release: 0–1 score · 5-class band · per-pixel uncertainty · input-coverage mask · STAC item · methodology doc.
Inputs
NASA LIS/OTD/ISS satellite flash climatology
Version history
What changed between releases — the same notes that ship with the data. Newest first.
v2.0
2026-09
current
- Rebuilt on a new observation record. Lightning is now derived from the NASA LIS/OTD/ISS satellite flash climatology (1995–2023) instead of the WGLC ground network used through v1.1.1. The satellite record is longer (29 years against 11), global rather than network-limited, and in the public domain.
- The measured quantity changed: total lightning, not cloud-to-ground. The satellite instruments are optical and record intra-cloud flashes as well as cloud-to-ground. Intra-cloud typically outnumbers cloud-to-ground by 2–3 to 1 and the ratio varies geographically, so absolute flash densities and the physical class boundaries both shift. Scores differ from v1.1.1 — the 0–1 susceptibility is a percentile rank and absorbs the level change, but it is re-referenced against a new distribution. Refresh any downstream threshold or cached score keyed to a v1.x vintage. If your workflow specifically needs cloud-to-ground rates, this layer is no longer that number.
- Correction for uneven satellite observing time. A satellite in low Earth orbit sees each location for only part of the record, and a poorly-observed place that lightnings rarely will often record exactly zero flashes — 31% of land in the raw climatology, and 82% of land above 60°N. Those places fall off the bottom of a log scale and score 0 regardless of their real hazard. v2.0 applies a Gamma-Poisson credibility estimate, weighting each cell's own observations against its 1.1° neighbourhood in proportion to how long it was actually watched. Land scoring zero falls from 31% to 7% over the land the satellites observed, and agreement with the independent ground network rises from 0.823 to 0.878 rank correlation — with the improvement concentrated in the least-observed land (0.665 → 0.839) and small where observation was already good (0.792 → 0.810).
- A limit worth stating: 8% of the delivered layer's land is tied at exactly zero (7% of observed land, plus the 1.5% polar cap that is filled with zeros). Those neighbourhoods recorded no detected flash across the whole record, so the layer cannot rank the least-electrified land on Earth — Iceland, Tierra del Fuego and most land above 75°N score 0 and are indistinguishable from each other. This is not new: v1.1.1 reads 0 at the same places and floors a slightly larger share of land (10.3%). A wider smoothing radius would manufacture an ordering by importing lightning from hundreds of kilometres away; it was tested, scored worse out of sample, and was not adopted.
- The coverage band is now meaningful, and worth reading. No instrument in the record observed above roughly 80°N, which is 1.5% of land. Rather than return NoData there, v2.0 fills those pixels from the nearest observed cell and sets coverage to 0.5 to mark them as extrapolated. Coverage is 1.0 everywhere else. Treat a high-Arctic score as a floor value.
- The raster covers more latitude, which matters only if you decimate and index. v1.1.1 was 432000 x 192000 (70°S–90°N); v2.0 is 432000 x 216000 (90°S–90°N). Both sit on the *same* grid — identical pixel size (0.000833333333°), identical origin (−180, +90), identical phase — so row *i* is the same latitude in both and any georeferenced read is unaffected, whether by coordinate, by window-from-bounds, or through the API. The scored footprint is also unchanged (84°N–60°S, the rest NoData). The one thing that breaks: reading both into a common decimated shape. 216000 → 1080 rows is 0.1667°/row while 192000 → 1080 is 0.1481°/row, so the correspondence drifts by up to 20° down the image while looking entirely valid. Read by bounds, not by array index.
- Product footprint is unchanged at 84°N–60°S; Antarctica excluded by standard.
v1.1.1
2026-07
- Full-resolution coastline (GSHHG-f). All four layers are re-clipped against the canonical GSHHG-f shoreline (~100 m native), so barrier islands and narrow coastal strips that previously read as NoData now carry a score.
- Recalibrated percentile scale (representative ECDF). Unlike the other v1.1.1 surfaces, lightning's calibration was corrected this vintage. The global percentile reference is now sampled representatively across every land tile rather than a tile subset, so the 0–1 susceptibility is spatially unbiased. Scores differ from v1.1: the spatial pattern of relative lightning risk is preserved, but absolute percentile values are re-referenced (better-distributed, with less high-end saturation). Any downstream thresholds or cached scores keyed to v1.1 lightning should be refreshed against this vintage.
- Product footprint is 84°N–60°S; Antarctica excluded by standard.
v1.1
2026-04
- Compact uint16 encoding. The surfaces were re-encoded from float32 to compact integer COGs with a documented scale (physical = raw × scale, nodata 65535) — smaller and self-describing, with no loss of precision at the delivered resolution. Method and values are unchanged from v1.0.
v1.0
2026-04
- Initial release — observed flash-density climatology. Susceptibility derived from the WGLC gridded lightning climatology (annual flash density, flashes/km²/yr), ranked to a 0–1 percentile scale. Because flash density spans several orders of magnitude (from ~0.001 over Patagonia to 50+ around Lake Maracaibo), the tail is log-compressed before ranking so the densest hotspots don't flatten everywhere else. A five-class map (< 0.5 → > 15 fl/km²/yr) ships alongside per-pixel uncertainty and coverage.
- Land-only, ocean = nodata. Ranking and output are restricted to land, so ocean cells don't drag the whole land distribution toward the high end.
Other hazards
LandslideAvalancheWildfireEarthquakeWindstormExtreme heatExtreme coldFrost / freezeSoil erosionHurricaneDroughtVolcanic EruptionTsunamiStorm SurgeHail