Data health

What this is, and what it is not

Where the data comes from, how the graphics are made, and the limits of what they can tell you.

MaverickWX takes the official forecasts published by government weather agencies and draws them. It is not a forecasting agency, it issues no warnings, and nothing on it should be the last thing you read before making a decision about your safety.

Do not use this for a life-safety decision

If a storm is heading for where you are, close this site and read the official one. In the United States that is weather.gov and the National Hurricane Center; elsewhere it is your own national meteorological service. Follow the instructions of your local emergency management authority over anything you read here.

Where the data comes from

Every storm on this site comes from one of two official warning centres, and each graphic says which:

  • NOAA/NHC — the United States National Hurricane Center, which warns for the Atlantic and the eastern and central Pacific.
  • JTWC — the Joint Typhoon Warning Center, a United States Navy and Air Force unit that warns for the western Pacific, the Indian Ocean and the Southern Hemisphere.

The NWS Guide section is about a third source — the National Weather Service's public API and the messages its Senior Duty Meteorologist posts about how the service itself is running. That section explains someone else's data too.

The U.S. Weather section reads the National Weather Service's warnings and watches and the forecasts of two of its national centres, the Storm Prediction Center and the Weather Prediction Center, with a few observations beside them. What it reads, and how, is its own section below.

We are not affiliated with any of them. They publish; we read what they publish, normalise it into one shape, and draw it.

How the satellite picture is made

The satellite window uses NOAA/NESDIS GOES-East imagery produced by CIRA and NOAA. It is a GeoColor day/night composite, not a raw visible-light photograph: after dark, the product combines infrared information so clouds and the surface remain readable. The source and its nominal observation time stay beside the image and in the published manifest.

The site crops that official full-disk product to a fixed North American view and publishes smaller WebP and PNG renditions. It does not recolor the weather values, infer clear sky from missing pixels, or present the development fixtures used by the test suite as current data. A new source image is expected about every ten minutes; when a check fails, the last verified frame may remain visible with a stale or unavailable label instead of being replaced by a blank or broken image.

What "derived" means here

Every graphic on this site is drawn by us from the agency's own numbers. The forecast track you see is the warning centre's forecast positions joined into a line; the intensity chart is their forecast winds. Nothing is smoothed, extrapolated or filled in. Where a source publishes no value, the graphic shows none rather than guessing one.

One rule is worth stating plainly because it is the easiest one to get wrong: the cone of uncertainty is official NHC GIS data and appears only on NHC storms. We never manufacture a cone for a JTWC storm by drawing a buffer around its track. A shape that looks like a cone but is not one would be the most dangerous thing this site could publish.

One place the word appears on the site is the Status column of a storm's forecast timeline. The two centres publish differently: NHC classifies most of its forecast points and leaves some blank, while JTWC gives a forecast wind at every step and no development stage at all. Where the centre stated a stage, that is what the column shows. Where it did not, the stage is worked out here from the centre's own forecast wind using the standard thresholds, and the row is marked derived — the positions and winds are theirs, the label is ours. Where the centre stated nothing and there is no wind to work from, the column shows an em dash and says so under the table. It never guesses, and it never prints a classification we cannot put into words.

The words in the Status column

A forecast point's stage, in the terms the warning centres use. These are development stages, not intensities: a storm's category comes from its wind, which is shown beside it.

Tropical depression
A closed circulation with maximum sustained winds below 34 kt (39 mph).
Tropical storm
34 kt up to 64 kt (39–73 mph). The stage at which a system is named.
Hurricane / typhoon
64 kt (74 mph) or more. The same storm by two names: hurricane in the Atlantic and eastern Pacific, typhoon in the western Pacific.
Super typhoon
A western-Pacific typhoon at 130 kt (150 mph) or more.
Subtropical depression / storm
A system with both tropical and mid-latitude characteristics — it draws energy from a temperature contrast as well as from warm water. Winds and hazards are real; the structure is not fully tropical.
Post-tropical cyclone
A former tropical cyclone that has lost its tropical structure. It can still be dangerous — Sandy was post-tropical at landfall — which is why the centre keeps naming it.
Extratropical
A cyclone driven by the temperature contrast between air masses rather than by warm ocean, with fronts rather than an eye. Ordinary winter storms are extratropical.
Remnant low
What is left when a tropical cyclone loses its organised thunderstorm activity: a low-pressure area with a circulation and little else, no longer meeting the definition of a tropical cyclone. The warning centres file it under the code LO, and a forecast that ends in one is a forecast that the system will dissipate. A remnant low can still carry rain, and it can occasionally redevelop.
Disturbance / tropical wave
An area of disturbed weather that may or may not become a cyclone. These usually appear as areas to watch rather than as forecast points.

How the Tropics page reads the products

The Tropics page lists every system the centres are warning on and, for each, the products the centre issued about it. A few rules decide what it says.

  • Systems are listed by where they are. A storm named in the East Pacific that crosses 140°W is listed under the Central Pacific, because that is the basin its current position is in and the centre warning on it (the Central Pacific Hurricane Center) is that basin's. Its id keeps the basin it was designated in.
  • "Forecast to strengthen" looks only ahead. The peak a row or a preview names is the strongest forecast point whose valid time is still ahead of the clock, and the window it looked in is stated. A past maximum is never an upcoming peak.
  • A warning badge is a verified count. It comes from the centre's own watch and warning list in the public advisory, read from the text and checked against the breakpoint geometry. A failed read is said to be a failed read; a centre that issues no civil watches (JTWC) is said to issue none; "no watches in effect" is said only when the advisory says so.
  • Every product has its own clocks. Issued is the centre's time on the product; retrieved is when this site read it; published is when the page's data was written. "Updated" on the page is the last of those, never the first.
  • Nothing on the map is derived. Wind extent is the centre's radii; arrival times are the centre's isochrones; probabilities are the centre's fields; surge and rainfall are the centre's and WPC's polygons. No cone is buffered, no arrival time is computed from distance and speed, no probability is read off a shape.
  • Satellite frames are the service's. The pipeline reads NASA GIBS's own list of available times each run, and the animation offers those times, labelled in UTC; when that list could not be read the page says the times are estimated.
  • Checking a location leaves nothing behind. The place list is a file of cities and coastal breakpoints served with the page (Natural Earth and NHC, both public domain); a coordinate, a map point or the device's position is compared against the published products in the browser. Nothing typed is sent anywhere and saved places live only in that browser.

Civil warnings for a place come from that place's national service. NHC and CPHC issue them for the United States and its territories; each storm's preview names the authority for the basin it is in and links to it.

How the Severe section reads its sources

The Severe section has four tabs: the Briefing, what is in force and what is forecast for the nation, a state or a place; the Live tracker, every warning in force on one map; Reports, the preliminary storm reports; and Analysis, what stations observe and what forecasters have analysed. Each agency's product is shown as that agency published it, and named. Two agencies' scales are never blended into one, and nothing on the page is a forecast of ours.

What it reads

  • NOAA/NWS: every warning, watch and advisory in force for weather on land, at the coast and of tsunamis, read every minute; the river, tide and lake gauges at action stage or above, with their flood stages and forecasts from the National Water Prediction Service; and each forecast office's Area Forecast Discussion and Hazardous Weather Outlook.
  • NOAA/SPC (the Storm Prediction Center): the convective outlooks for Days 1 to 8, their tornado, wind and hail probabilities and their written discussions, read every five minutes; mesoscale discussions; today's and yesterday's preliminary storm reports; and the hourly mesoanalysis pictures and observed soundings on the Analysis tab.
  • NOAA/WPC (the Weather Prediction Center): the Excessive Rainfall Outlook for Days 1 to 3 and its discussion, the rain forecast for each of those days, and the coded surface analysis: fronts, troughs and pressure centres.
  • NOAA MRMS, by way of nowCOAST: the radar estimate of the rain that fell in the past 24 and 72 hours.
  • NOAA/NWS Aviation Weather Center: every station's latest surface observation (METAR), for the Analysis tab's surface map.
  • NOAA/NCEP (the National Centers for Environmental Prediction): model guidance on the Analysis tab. The REFS ensemble's probabilities, read from NCEP's data server and drawn on our map, and the HRRR model's own pictures from NCEP's Model Analyses and Guidance site.

SPC's outlook and WPC's rainfall outlook are different products on different scales: SPC's five categories for severe storms, WPC's four for flooding rain. They are shown one at a time, each with its own key. A forecast office's discussion is quoted in its own words and dated; it never sets a risk, a time or a count.

Model guidance is labelled as guidance, as it is on the Tropics pages, and kept on the Analysis tab. Each picture or map says which model, which run, how many hours ahead and for what time. The REFS ensemble's maps show how many of its runs put a rotating or strong storm near each point: that share is how uncertain the forecast is, not how much of an area storms will cover. The HRRR is a single run of one model; its simulated reflectivity is the model's storms, drawn in radar's colours, and is labelled "Simulated" beside the picture, never shown as radar. Until NCEP makes REFS operational (planned for 3 November 2026), it comes from NCEP's pre-operational run, and the page says so. No model's numbers feed a risk, a time or a count anywhere else.

How a place is matched

A town or ZIP code is a point: the Census Bureau's own centre for it, from its 2024 Gazetteer of 32,333 places and 33,702 ZIP Code areas. The Briefing says what holds that point (a warning polygon, an outlook area) and never claims it for every address in the town. A state's page says the highest category anywhere in the state, from the state's outline, and says that it is not every point in it. A warning or a watch is counted for a state when a county or forecast zone it names is in the state, so one that crosses a state line is counted in both.

What you type into the place search is sent to this site and nowhere else, to find the place; it is not kept, and our server's log leaves it out. "Use my location" is answered inside your browser: your position is not sent to us, stored or put in the address. It offers the nearest listed place instead.

The section's clocks

Issued
When the agency published the product, read from the product itself. Each outlook day has its own: SPC's Day 1, Day 2 and Day 3 outlooks are issued on separate schedules, so one can be newer than another, and each says so.
Valid
The period a forecast is for. SPC's and WPC's day runs from 12:00 UTC to 12:00 UTC the next morning (7 AM to 7 AM Central Daylight Time), so before 12:00 UTC "today" in SPC's terms is still the day that began yesterday morning. A warning is valid until the end its issuing office set.
Checked
When we last read the source successfully, as in "NWS alerts checked 5:05 PM". It is the clock that says whether what you see is current. Each source has its own, under "Every source's clock".

Times are in your browser's time zone with UTC beside them, or in UTC alone if you choose; the choice is kept in this browser. Without scripting, times are in UTC.

What a storm report is

SPC's storm reports are preliminary: what spotters, emergency managers and the public told a forecast office, as it was told, for SPC's convective day. A report is not a confirmed tornado or a damage survey. One storm can be reported many times, a row can be listed twice, and a later row can correct an earlier one. No count on the page is a count of storms. A magnitude SPC was not given is "not given", never zero, and SPC's file does not say whether a wind gust was measured or estimated.

When a source is down

Each source is current, down or not reported, judged against how often it is read. "No warnings in effect" is said only when the warnings list is current and empty. If the warnings cannot be read, the page says it cannot confirm them, and the last list it held is shown with its time, never as current. One source failing never stops another from being shown.

Model guidance is not the forecast

Some storms carry a model-guidance graphic — a bundle of individual model runs, sometimes called a spaghetti plot. It is there because it is genuinely informative about how confident the forecast is. It is labelled as non-official everywhere it appears, and it always sits below the official track, because a striking picture is easy to mistake for an authoritative one.

Four clocks, kept apart

A weather page that shows one timestamp is hiding three. Every storm's full briefing shows all four, under Where this came from; the Tropics workspace keeps the three that are about publishing under Data details:

Forecast cycle
When the forecast itself was initialized. The National Hurricane Center runs its forecast on the synoptic hour — 00, 06, 12 or 18 UTC — and issues the advisory built on it up to six hours later, so this is normally earlier than the advisory below it. It is the time the track graphic names, and the time each forecast hour counts from.
Advisory issued
When the warning centre published the advisory. This is the one that matters.
Source retrieved
When we fetched it. Occasionally a few minutes before the advisory's own stamp: NHC posts a product to its servers slightly ahead of the nominal hour printed on it. Both times are shown as measured rather than tidied into an order that looks tidier.
Site published
When we finished drawing the graphics for this storm.

The gap between the advisory and the last is our latency. The age of the advisory is the age of the forecast, and it is the one shown beside each storm.

When something is missing

Sources go down, and an outage looks exactly like quiet weather if a site is careless about it. This one is not. If we cannot reach a source, the page says so; if data is older than the issuing centre's own schedule would predict, the storm it belongs to says so. An empty list means an empty ocean only when we can confirm it.

Malformed or incomplete data never replaces a good previous state. When an update cannot be trusted, the last coherent one is kept and labelled as retained rather than thrown away — a stale forecast that says it is stale is more useful than a blank page.

You can see the current condition of every feed on the data health page, which is also what the indicator in the top bar reads from.

Counting readers, without tracking them

We keep a rough count of which pages get read, so we know what is worth the effort. It is first-party — the count is ours, sent to our own server, and shared with nobody. It uses no cookies and stores no identifier of any kind: not your IP address, not a device fingerprint, nothing that could tie one page view to another or to you. Each view is a single, standalone tally.

What a page view records is deliberately coarse: the path you looked at, the host of the site you arrived from (never the full link, which stays in your browser), a rough size band for your window, whether you are in light or dark mode, and a two-letter country that our network edge derives without us ever seeing your address. Those counts are rolled up into a daily total and nothing finer is kept.

If your browser sends a Do Not Track or Global Privacy Control signal, we check it before sending anything — so nothing about your visit leaves your browser at all. The count is also best-effort: if the machine that receives it is down, that view is simply lost. We do not queue it, and for a page-view tally that is a fine trade.

For developers

Everything the site draws is available as JSON. The Storm API is a small set of static objects fetched with a plain GET — no key, no signup — and each graphic ships a provenance sidecar tracing it back to the exact agency issuance it came from.