Ask a roofer whether hail maps are accurate and you will get one of two answers, both wrong. Either the map is gospel — the square is red, therefore the roof is toast — or the map is marketing, because somebody once knocked a red street and found nothing.
The truthful answer is narrower and more useful than either: a hail map is an excellent instrument for choosing where to spend your crew's day, and a poor instrument for proving anything about one roof. Everything below is an attempt to say exactly where that line sits, using what the published research actually found rather than what a vendor would like you to believe.
We make a hail map. Read this accordingly. We have tried to make it the kind of page that is still useful if you buy somebody else's.
"Accurate" is three different questions
Most arguments about hail map accuracy are people answering different questions at each other. There are three, and they have very different answers.
1. Did hail occur over this address on this date? Radar is genuinely good at this. Detecting that a storm was producing hail is a much easier problem than sizing it, and the swaths modern products draw are reliable at the event level. If a map shows a significant hail event over a neighborhood, a hail event almost certainly happened.
2. How big were the stones? Meaningfully harder. This is where the error bars live, and where the honest answer is "within about a quarter inch, most of the time, in the middle of the range."
3. Did that hail damage this roof? The map cannot answer this at all, and no vendor's map can. Damage depends on the age and type of the covering, the slope and exposure, the wind direction during the event, and prior condition. A map that claimed to answer question three would be lying.
Almost every disappointment with hail maps comes from expecting a question-three answer out of a question-two instrument.
What the number is actually made of
Briefly, because we covered it in detail in how to read a hail map.
NOAA's Multi-Radar Multi-Sensor system stitches the national radar network onto a grid roughly a kilometer across and updates it every couple of minutes. From the radar returns above the freezing level it computes a Severe Hail Index, and from that index — combined with temperature profile data from a weather model — it produces MESH, the Maximum Estimated Size of Hail.
Two details in that sentence matter enormously for accuracy, and they are the two nobody mentions.
"Maximum." MESH estimates the largest stone the storm could produce over that square. Most of what fell was smaller. When a homeowner says "it was mostly marbles with a few big ones," they are describing exactly what MESH is designed to report the top of.
"Combined with temperature profile data from a weather model." The hail estimate inherits the errors of the model that supplied the temperature profile. NOAA's own training material is explicit that the product is subject to the biases and deficiencies of the mesoscale model used to derive that profile. This is a chain of estimates, not a measurement.
What the research found
This is the part the marketing pages skip.
The bias runs in both directions. Murillo and Homeyer, in work published in 2019, evaluated the original MESH equation against the distribution of reported hail sizes and found it underestimates smaller hail sizes and overestimates larger ones. That is why they published recalibrated versions — one fit to the 75th percentile of the sampled hail size distribution (MESH75) and one to the 95th (MESH95) — and found the two performed best at different thresholds, MESH75 around the severe-hail threshold and MESH95 for significant hail.
The practical translation: the middle of the range is where you should trust the number most. A square reading 1.25 to 1.75 inches is telling you something fairly dependable. A square reading 3 inches is telling you "very large hail happened here," and you should read it as a category rather than a measurement.
Specific storm structures fool it, and they fool it low. NOAA's Warning Decision Training Division lists the known cases where MESH underestimates:
- Storms tilted over in strong, deep-layer vertical wind shear
- Left-moving supercells
- Supercells with a very large bounded weak echo region — which can show up on the map as a hole of low values sitting exactly where the worst hail was
- Storms producing low-density, dry hailstones
That last category is worth remembering at the door. And the bounded-weak-echo case is worth remembering when a swath looks like it has a bite taken out of the middle: that is sometimes a real gap and sometimes the algorithm losing the plot precisely where the storm was strongest.
The beam rises with distance. Radar beams travel in a straight line while the earth curves away beneath them, so the further you are from a radar site, the higher in the storm the beam is sampling. A storm 120 miles from the nearest radar is being observed very differently from one 20 miles away. This is a property of radar, not of any vendor's product, and it is one reason coverage quality is not uniform across a state.
The dirty secret: verifying a hail map is nearly impossible
Here is the problem that makes precise accuracy claims suspicious no matter who is making them.
To know whether a hail map was right, you need to know what actually fell. The only national record of that is human reports — trained spotters, law enforcement, emergency management, and the public calling the National Weather Service. Those reports are the best ground truth available and they are badly non-random:
- They cluster where people are. A two-inch stone in an empty wheat field generates no report. The same stone in a subdivision generates twenty.
- They cluster near roads and daylight. Hail at 3am in rural country is systematically under-reported.
- They are estimated, not measured. Most reports are a person comparing a stone to a coin or a ball from memory, sometimes after it has partially melted. Reported sizes bunch onto familiar reference objects — quarter, golf ball, baseball — rather than distributing smoothly.
- They are one point. A report is one yard at one moment. The grid square around it covers ten to fifteen lots.
So when anybody — including us — tells you a hail product is accurate to within a specific figure, understand that the yardstick itself is a rubber ruler. The honest framing is that radar estimates and ground reports agree well enough, often enough, for the estimates to be operationally useful, which is exactly how the National Weather Service itself uses them.
Where hail maps are most and least reliable
| Situation | Confidence | Why |
|---|---|---|
| Middle of a well-defined swath, 1″–2″ band | High | Best-calibrated part of the size range, strongest radar signal, usually corroborated by spotter reports |
| Near a confirmed spotter report | High | An actual human measured a stone nearby |
| Close to a radar site | Higher | The beam is sampling lower in the storm |
| The outer edge of a swath | Low | The boundary is modeled, not observed. Expect a band of uncertainty either side |
| Very large estimates (2.5″+) | Moderate | Known to over-read at the top of the range. Read it as a category |
| Tilted storms, left-movers, big BWER | Low, biased low | Documented underestimation cases — the map may be hiding the worst of it |
| Far from any radar, overnight, rural | Low | High beam sampling plus almost no ground truth to check against |
| Any claim about a specific roof | None | The map does not know the roof exists |
What this means at the door
The accuracy conversation matters most in the ninety seconds you have on someone's porch, and it is where most contractors quietly damage their own credibility.
Do not say "the radar shows your roof has hail damage." It does not, it cannot, and the homeowner who calls their agent and repeats that line to an adjuster has just made your inspection harder.
Do say what is actually true, which is more persuasive anyway: "NOAA radar recorded a hail event over your address on May 14th with stones estimated around an inch and a half, and there was a confirmed report of golf-ball hail about three miles from here. That doesn't tell me what happened to your roof — I'd like to get up there and find out."
That sentence does four things a claim about damage cannot. It cites a source. It names a date. It offers corroboration. And it makes the inspection the point, which is where you were trying to get anyway.
Expect people to misremember size, and do not argue about it. Someone who says "golf balls" when the map says an inch is not lying. Reported hail sizes bunch onto familiar objects, and memory inflates. It changes nothing about whether the roof is worth inspecting.
There is more on the conversation itself in our door knocking guide.
What this means for the claim
Two things, and both cost money when they go wrong.
The date of loss has to match a real event. A wrong date is one of the more common reasons a legitimate claim comes back denied or partially paid, and it is entirely avoidable. If your market took hail on the 14th and again on the 22nd, the map is what tells you which one you are filing on. This is also where a standalone map subscription starts costing you: it can show you the date, but it cannot carry that date into the job record, so it gets retyped by a human, and humans retype dates wrong. In HailMate the storm date comes off the swath the job sits inside, because the map and the job are the same system.
Carriers use weather data too. Third-party weather verification is a normal part of claims handling, and an adjuster may cite it to argue no significant hail occurred at an address on the claimed date. You want the radar evidence attached to your file rather than only to theirs — the estimated size, the event date, the nearby spotter reports. It does not win the argument by itself. It stops you from losing it by default.
What actually settles the question is the inspection: test squares, bruised mats, granule loss, dented soft metals, photographs with a date and a location. That is evidence about the roof. Everything on the map is evidence about the cloud.
A sixty-second sanity check before you send a crew
Do this on any swath before you commit a day to it.
- Look for a spotter report near your target streets. A human-confirmed stone size within a few miles is the strongest single corroboration you can get for free.
- Check the probability layer, if your tool has one. A hot square with low probability of severe hail, sitting alone at the fringe of an event, is the classic signature of radar reading something that was not hail.
- Look at the shape. Real hail swaths are elongated along the storm's direction of travel. A blotchy, disconnected pattern with no axis is worth a second look.
- Check the age of the housing stock. A 2-inch swath over a subdivision built last year is worth less than a 1.25-inch swath over roofs from 2009. The map does not know this and neither does your competition's map.
- Check whether somebody already worked it. A three-week-old red swath in a metro has been picked over. A one-week-old moderate swath forty minutes out frequently has not.
Does paying more buy more accuracy?
Sometimes, and not in the way the pricing implies.
Human-reviewed maps — HailTrace is the best-known example — put meteorologists on the radar to draw and check swaths. That genuinely helps with the odd storm structures that fool automated algorithms, and it is the right product if you need a report that will survive cross-examination in court. It does not change the underlying physics: a person looking at radar is still inferring hail size from radar.
Deeper archives — Interactive Hail Maps advertises history back to January 2011 — buy you history, not precision. A fifteen-year archive is genuinely valuable for reconstructing an old date of loss and worth nothing at all for deciding where to knock this week.
Automated street-level swaths are what most storm crews actually need, and the accuracy difference between vendors here is smaller than the price difference, because they are all processing the same public NOAA radar data. What differs is resolution, how the bands are drawn, what corroborating layers sit alongside, and whether the map is attached to anything you can run a business on.
If you are comparing tools, we put every published price side by side, sourced and linked, in our comparison of the five hail apps roofers actually use — including the ones that cost more than we do.
The bottom line
A hail map is a prospecting instrument of genuinely high quality and a claims instrument of strictly limited scope.
Trust it to tell you a hail event occurred and roughly how severe. Trust it most in the middle of a well-defined swath, near a confirmed report, in the one-to-two-inch range. Trust it least at the fringes, at the extremes, and in the specific storm structures NOAA has documented as fooling it.
Never ask it what happened to a particular roof. Go look.
You can see the underlying data on any US address with the free hail tracker — no account, no card — or put a whole state on the full map with canvassing on it for $79 a month.