Statistics | WxAlerts.org docs

Live counts of active alerts and GLM lightning, how fast each NWS alert feed delivers, and daily lightning totals in any time zone.

Base URL: https://api.wxalerts.org

The headline numbers on this site: how many alerts are in effect, how much lightning the GOES satellites are seeing, how quickly each alert feed delivers, and lightning per day.

GET /stats/summary
GET /stats/latency?hours=24
GET /v1/stats/lightning/daily?days=7&timezone=America/Chicago

summary and latency need no key. Browsers block calls from other websites (CORS allows only wxalerts.org), so call them from a server, script or app. They are limited per address as described under rate limits. lightning/daily needs an API key (alerts:read) and is served only under /v1; on the bare path it answers 401 with code api_key_required.

`GET /stats/summary`

Active alert counts and the GLM lightning window. No parameters. Cached for 15 seconds, which is about the data’s own cadence, so polling faster gets you the same answer.

{
  "server_time": "2026-11-09T19:07:18.638865Z",
  "alerts": { "active": 370, "with_geometry": 370 },
  "lightning": {
    "flashes": 1214878,
    "window_start": "2026-11-08T19:30:00.010682Z",
    "window_end": "2026-11-09T19:46:39.319450Z",
    "window_seconds": 87399,
    "rate_per_minute": 834.0,
    "retention_hours": 24.0,
    "covers_full_day": true
  }
}

field
`alerts.active`	alerts in effect now, the same set every alerts route counts
`alerts.with_geometry`	how many of those have a shape and can be drawn on a map
`lightning.flashes`	GLM flashes between `window_start` and `window_end`
`window_start`, `window_end`	the oldest and newest flash the live store holds
`window_seconds`	the span between them
`rate_per_minute`	`flashes` over that span, per minute
`retention_hours`	how long the live store is set to keep flashes
`covers_full_day`	whether the window spans at least 24 hours

The window is whatever the live store holds, measured, not assumed, so read its length from window_seconds rather than from retention_hours. Branch on covers_full_day before calling the count a day’s worth. Even when it is true, the window is rolling and never a calendar day: at 09:00 it reaches back into the day before. Do not label it “today”. For lightning per day, use lightning/daily.

`GET /stats/latency`

How fast each alert feed delivers, measured against the others. The ingest takes the same hazards from three feeds (nwws, NWWS-OI over XMPP; emwin, off the GOES broadcast; api, the api.weather.gov CAP feed) and records when each one first delivered each hazard.

parameter
`hours`	window to report over, 1 to 168. Default 24. Arrival records are pruned on a rolling schedule, so asking for more than is kept gives a shorter answer, not an error

{
  "server_time": "2026-11-09T19:07:19.262194Z",
  "window_hours": 24,
  "issuance_resolution_seconds": 60,
  "sources": [
    {
      "source": "nwws",
      "arrivals": 241, "contested": 241, "wins": 236, "win_share": 0.9793, "exclusive": 0,
      "median_behind_seconds": 0.0, "p95_behind_seconds": 0.0, "max_behind_seconds": 9229.67,
      "median_issuance_lag_seconds": 58.6
    },
    {
      "source": "api",
      "arrivals": 242, "contested": 241, "wins": 5, "win_share": 0.0207, "exclusive": 1,
      "median_behind_seconds": 30.43, "p95_behind_seconds": 106.25, "max_behind_seconds": 169.78,
      "median_issuance_lag_seconds": 94.3
    },
    "..."
  ]
}

Sources come most wins first. Two different measurements are in each row, and they are not interchangeable.

Relative, exact to the millisecond, and the fair way to rank the feeds:

field
`arrivals`	hazards this feed delivered in the window
`contested`	of those, how many at least one other feed also delivered. Only these can be raced
`wins`	contested hazards this feed delivered first
`win_share`	`wins / contested`, `0.0` when nothing was contested
`exclusive`	hazards no other feed carried. Coverage, not speed: the reason to keep a feed that never wins
`median_behind_seconds`, `p95_behind_seconds`, `max_behind_seconds`	how far behind the fastest feed this one arrived, over contested hazards. `null` when none were contested

Absolute, indicative only:

field
`median_issuance_lag_seconds`	median time from the NWS issuance stamp to arrival
`issuance_resolution_seconds`	`60`. The NWS stamps issuance to the minute, so every absolute lag can be off by up to this much. Treat it as an order of magnitude

max_behind_seconds catches outliers such as a feed that dropped and caught up later; the median and p95 are the figures to quote. Cached for 60 seconds.

`GET /v1/stats/lightning/daily`

GLM flash counts per day, with days cut in any time zone. Needs a key (alerts:read).

parameter
`days`	how far back to look, in days from now, 1 to 400. Default 30
`timezone`	IANA zone to cut days in, e.g. `America/Chicago`. Default `UTC`. An unknown zone is `422`, never a silent fall back to UTC

Pick the zone your readers live in. A UTC day ends in the evening across the Americas, so in UTC an evening storm lands on the next day.

{
  "server_time": "2026-11-09T19:07:13.249615Z",
  "timezone": "America/Chicago",
  "days": [
    {
      "day": "2026-11-09",
      "flashes": 238705,
      "peak_hourly_flashes": 40928,
      "hours_observed": 11,
      "mean_energy_j": 2.0169939528190287E-13,
      "first_flash": "2026-11-09T06:00:00.096895Z",
      "last_flash": "2026-11-09T16:59:59.906157Z"
    },
    {
      "day": "2026-11-08",
      "flashes": 1557001,
      "peak_hourly_flashes": 166906,
      "hours_observed": 24,
      "mean_energy_j": 1.5397250051263375E-13,
      "first_flash": "2026-11-08T06:00:00.114443Z",
      "last_flash": "2026-11-09T05:59:59.953460Z"
    },
    "..."
  ]
}

field
`day`	the local date in `timezone`
`flashes`	flashes that day
`peak_hourly_flashes`	the busiest single hour, a sign of how intense the day got that a total hides
`hours_observed`	hours with at least one flash. Under 24 on a quiet day, but also on a day with a gap in the data: check it before reading a low count as fair weather
`mean_energy_j`	mean optical energy per flash at cloud top, in joules. Not a measure of ground hazard
`first_flash`, `last_flash`	the first and last flash of that day, UTC

Days come newest first: today and the days - 1 whole days before it, cut at midnight in timezone, so days=7 is at most seven rows (a day with no flashes at all has no row). Today’s row is still filling in. Use hours_observed to tell a partial day from a complete one. Finished days never change; the response is cached for an hour.