Satellite, radar or ground network: which storm evidence holds up?

Every contested storm claim eventually turns into a contest of weather records. One side brings a radar map, the other a spotter report, someone mentions a weather station twenty miles away. Each of these sources measures something real, and each has blind spots. This guide walks through the main kinds of storm evidence, what each one actually measures, where it is strong and where it is weak, and ends with a checklist you can use on any weather report, including ours.

Why no single source settles a storm claim

Severe weather is small and fast. Hail falls in swaths that can be a few kilometers wide, and lightning strikes are points. A claim, meanwhile, is about one address and one evening. Every observing system makes a trade between how precisely it measures and how much ground it covers. Stations are precise but sparse. Radar covers wide areas but estimates rather than measures at the ground. Satellites see everything at once, but in kilometer-sized pixels. The strongest files combine sources whose weaknesses do not overlap.

Human storm reports

In the US, National Weather Service offices collect local storm reports from spotters, emergency managers and the public, and NOAA's Storm Prediction Center publishes them daily. They often include the most persuasive detail a claim can have: a measured hailstone, a time, a street. Their weakness is coverage. The SPC itself describes its list as preliminary and likely missing reports, and a storm that crosses an empty field at night may produce no report at all. The official, final record is NCEI's Storm Events Database, which takes a few months to compile.

Weather stations

An official station measures wind, rain, temperature and pressure to a high standard, which is why stations carry weight in court. But a station measures its own location. Unless the loss is close to it, station data describes a different place, and with hail, a few kilometers can be the difference between a direct hit and nothing.

Radar hail estimates

Weather radar sees inside the storm. NOAA's Multi-Radar Multi-Sensor system merges radars into a national grid and estimates hail size with the MESH product. It is excellent for mapping the swath of a storm across neighbourhoods. Two cautions apply. MESH is an estimate built from reflectivity aloft, not a measurement at the ground, and NSSL researchers describe it as representing roughly the 75th percentile of hail diameters rather than the largest stone. And radar quality depends on the beam's height and distance from the nearest site, which varies widely from place to place.

Ground lightning networks

Networks of ground sensors triangulate lightning strokes. Vaisala's US National Lightning Detection Network reports 84 meter median location accuracy and 95 % detection of cloud-to-ground strokes. For the narrow question of how close a strike came to a building, this is the most precise evidence there is. It is commercial data, and its quality depends on the sensor network in the region.

Geostationary satellites

GOES-19 and GOES-18 watch the Americas continuously. Their imager measures cloud-top temperature and ice-particle signatures every few minutes, and their Geostationary Lightning Mapper records in-cloud and cloud-to-ground lightning with 70 to 90 % flash detection at about 8 km resolution. Over Europe, EUMETSAT's Meteosat Third Generation plays the same role. The strength is uniform coverage: the same measurement over a city, a farm in Santa Fe and an empty highway, with no spotter needed. The weakness is resolution: a satellite says how severe the storm was over an area, not what landed on a single roof.

Model data, which is not an observation

Many weather reports include wind from a numerical model such as NOAA's GFS. Models are useful context, but they are a computed estimate on a grid, not something an instrument measured at the site. A report should label model fields as such. If a wind speed in a report has no named instrument behind it, ask where it came from.

• Spotter reportsStrongest detail, patchy coverage, preliminary until the official record is compiled.
• StationsHighest measurement quality, only at the station's location.
• RadarBest for mapping a hail swath; estimates, with quality that varies by distance from the radar.
• Ground lightning networksMost precise strike locations; commercial, quality tied to sensor density.
• SatellitesUniform coverage and a full timeline of the storm; kilometer-scale resolution.
• ModelsContext only; never a measurement at the site.

How the sources line up in a real case

Our US sample report for an address in Kankakee, Illinois, on March 10, 2026 shows four sources reading one storm. A preliminary SPC report described a 6.14 inch hailstone found 0.2 km from the address. MRMS MESH peaked at 54.6 mm over the 1 km analysis circle. GOES-19 measured cloud tops as cold as −75.4 °C and a 77.5 % hail probability on the point, and its lightning mapper logged 2 flashes inside the circle. None of the numbers match each other, and none should: each instrument measures a different thing. Together they leave little room to argue about the date and place.

Agreement between independent sources is worth more than precision from any one of them.

A checklist for judging any weather report

• Is every source named?Satellite, radar, network, station, model. Unnamed data cannot be checked.
• Is the time window stated in local time?And does it actually contain the reported loss? UTC dates trip up many reports.
• How far is the data from the site?A distance for every observation, or a radius for every map.
• Is the resolution stated?Meters for ground networks, kilometers for satellites. The conclusion should not be finer than the data.
• Are preliminary data labelled?SPC reports, for example, are preliminary by NOAA's own description.
• Are model fields labelled as models?Wind from a forecast model is not a measured gust.
• Does it state its limits?A report that can never say not conclusive is not doing analysis.
• Can you verify it is unaltered?A code or certificate that lets a third party open the issued original.

Hold us to the same list. Our published Argentine case for Bouquet, Santa Fe, on March 28, 2026 concluded hail was probable, with a 63 % peak probability and 30 minutes of persistence, and says plainly that the satellite signal is compatible but not conclusive on its own. Every Skybridge report names its satellite, radar and report sources, states the window in the site's local time, treats model wind as model output rather than a measurement, and carries a code anyone can use to open the original on our site.

For the claim-side view of the same problem, read Hail claim denied? How to prove the date of loss and Lightning strike report for an insurance claim.

Questions to ask about the other side's weather report

• Which instrument, and how far away?A station 30 km away that saw no hail says little about an address in the swath.
• Which time zone?If the window is in UTC, check whether the loss falls inside it in local time.
• Measured or modelled?Ask for the instrument behind every wind speed and hail size.
• Preliminary or final?Preliminary storm report lists are incomplete by design; their silence is weak evidence.