data
NOAA and NWS Storm Data Sources Explained
The separate NOAA and NWS storm data products people treat as one: what each covers, what it refuses to answer, and when to use each.
Ask where NOAA storm data comes from and you get several answers that are each partly right. There is a live API for current conditions and warnings. There is a daily storm reports page. There is a long research file of reports going back decades. There is an official archive publication with its own searchable database. And there is a radar-derived inventory. These are separate products with separate purposes, and they disagree with each other by design. Treating them as one source is the quickest way to answer the wrong question with confidence.
Here is each one, what it actually covers, and what it refuses to do.
The live API at api.weather.gov
The public NWS web API sits at api.weather.gov. It is free, it serves JSON-LD, and no API key is required. The documentation states that keys will be introduced later. What it requires now is a User-Agent header that identifies your application, with a contact website or email recommended. A library default is not an identification.
Rate limits are not publicly documented. The same page describes them as generous for typical use and says a request over the limit returns an error that can be retried once the limit resets, usually within about five seconds. It also notes that proxies are more likely to hit the limit than direct clients. Plan around that detail. If every request leaves through one shared address, you are sharing a budget with everything else behind it.
The endpoints cover points, gridpoints, active alerts and stations. Radar is the first thing people expect and do not get: only radar status data is available here, and display imagery comes from separate services. History is the second. The API does not provide a long-term archive, and the alerts endpoint covers only the past seven days. NCEI hosts the official archive of NWS alerts.
That is a coherent product, not a broken one. It answers questions about current and imminent conditions. It is not a research dataset and does not present itself as one.
SPC daily storm reports
The Storm Prediction Center's daily storm reports are drawn from NWS Local Storm Reports, usually sent in near real time. The day is not a calendar day: reports are grouped on a 1200 UTC to 1159 UTC day. Today's page updates every 10 minutes starting at 6 AM CST or 7 AM CDT, and a separate page covers just the last three hours. Archived days can be retrieved by date, and SPC offers KML files and a REST service.
Two limits are stamped on the product itself. The first is status. SPC labels these reports preliminary and shows them as is, and the reports index says so in plain capitals: All Reports Are Considered Preliminary. The second is scope. SPC does not decode non-thunderstorm Local Storm Reports, such as hurricane-related wind, so a tropical system can produce a pile of local reports that never surface in these pages. For post-storm summaries, SPC sends users to NWS Storm Data by way of the NCEI Storm Events database.
Use this to see what came in while the storm was happening. Do not use it as the citation of record.
The SPC severe weather database
SPC also publishes a severe weather database compiled from NWS Storm Data. Tornado reports go back to 1950. Hail and damaging wind reports go back to 1955. The current year's reports are preliminary, which means the most recent part of the file is the least settled part of it.
One discontinuity dominates any long-period analysis. The NWS raised the severe hail threshold from 0.75 inch to 1.00 inch in 2010, and for legacy purposes the 0.75 inch reports remain in the annual hail files, so users may need to filter. Count rows straight through the period of record and you have mixed two definitions of severe, with the looser one applying to the earlier decades.
The same page documents its own biases, which saves everyone an argument. Most wind reports rest on a fallen tree rather than a measured gust, which pulls them toward places with more trees to fall. Reports cluster around population centers. The database exists to verify NWS warnings and may not capture every storm event. And monetary loss figures in the database are described as highly suspect, which reads as a direct instruction not to build a damage estimate on them.
Storm Data and the NCEI Storm Events Database
This is the official version of the same events, and it is slow on purpose. NWS guidance states that both the Storm Data publication and the reports in the Storm Events Database are available 90 to 120 days after the event. Coverage runs tornadoes since 1950, severe thunderstorms since 1955, and all other weather events since 1996. That last line matters more than it looks: anything outside tornado, hail and thunderstorm wind has a much shorter record than people assume.
Local NWS offices build these entries from sources such as damage surveys, emergency managers and SKYWARN spotters. If an event is missing from both sources, the guidance says it was not reported to the National Weather Service. That is a statement about reporting, not about weather.
One procedural detail decides whether your work survives scrutiny. Certified, official copies must be requested from NCEI, not from local NWS offices. A saved page from a reports site is a different artifact with a different standing.
The Severe Weather Data Inventory
The Severe Weather Data Inventory at NCEI is what people mean when they say they want radar data without decoding radar. Its datasets are NEXRAD Level-III Storm Structure, meaning storm cells, offered in a filtered version with maximum reflectivity at or above 45 dBZ and in an all-cell version; NEXRAD Level-III Hail signatures, offered in a filtered version with maximum size greater than zero and probability 100 percent and in an all-signature version; NEXRAD Level-III Mesocyclone signatures; the Digital Mesocyclone Detection Algorithm; NEXRAD Level-III Tornado signatures; and lightning strikes from the Vaisala NLDN.
You can reach it three ways: an interactive map, a bulk download of the entire database as CSV over HTTP, and web services in the form of a REST API. Output formats are Shapefile, KMZ, CSV and XML. That combination is why it shows up in so many pipelines. The bulk path and the service path serve different needs, and neither one asks you to process radar yourself.
Read its stated caveats before you trust a row. The page states that SWDI adds no quality control beyond archival processing, that missing data does not mean no severe weather occurred, and that much of the automatically derived data is radar-based and represents probable rather than confirmed conditions. Probable rather than confirmed is the phrase to carry forward. A hail signature is a radar inference, not an observation of a stone on the ground. The same page also notes that NCEI is migrating data to the cloud, which may result in temporary data access delays, and that order fulfillment is intermittently degraded. Write retries rather than assuming a failed pull means there was nothing there.
Sort by the question you are asking
Is it happening now? That is the API at api.weather.gov, with active alerts and the seven-day alert window, plus the SPC last-three-hours and ten-minute report pages for what people are calling in. Nothing in this group is an archive, and the alerts window closes behind you.
What happened last week? That is the SPC daily reports archive, retrieved by date, with the preliminary label kept attached to every number you quote from it. Storm Data for last week does not exist yet, because it runs 90 to 120 days behind the event.
What can I certify? That is Storm Data through NCEI, with a certified copy requested from NCEI rather than captured from a local office page. Expect to wait months, and expect the certified record to differ in places from the preliminary version you watched come in.
A fourth question sits underneath all of them: what probably happened where nobody was standing? The Severe Weather Data Inventory answers that one, in its own words, with probable rather than confirmed conditions. It is the right tool for coverage and the wrong tool for proof, and most bad storm analysis comes from using it as the second thing.