Methodology

TubeTraveler is a live-camera discovery service. The World Live Events section exists to answer one question: something is happening somewhere — is there a public camera near it? This page explains exactly how that connection is made, and just as importantly, what it does not mean.

Where the event data comes from

Every event on this site originates from a public agency feed. We do not create events from social media posts, and we do not crawl general news. The current feeds are listed on the data sources page along with the attribution each agency asks for.

Feeds are read by a scheduled job on our servers and written into our own database. Visiting a page on TubeTraveler never triggers a call to an agency API — you are always reading our stored copy, with the timestamp of the last update shown on the page.

How duplicate events are merged

The same earthquake can be published by USGS and by GDACS within minutes of each other, with slightly different coordinates, magnitudes and titles. Showing both as separate events would be misleading, so incoming records are compared against events we already hold.

Two records are only merged when all of these hold:

• They are the same normalised event type.
• Their start times are within 36 hours of each other (12 hours for scheduled celebrations, where times are exact).
• Their coordinates are within 120 km for point events like earthquakes, or 350 km for events whose reported centre moves, such as tropical cyclones.

Passing those tests is necessary but not sufficient. We then score supporting signals — how close the coordinates and times actually are, how much the normalised titles overlap, whether the country matches, and whether reported magnitudes agree. Only records that clear that combined score are merged. When in doubt we keep two records rather than merge them, because a wrong merge silently deletes information and a duplicate does not.

Once merged, the coordinates and headline figures come from the highest-priority source, severity takes the most serious value any agency reported, and every contributing agency is kept and linked separately. Where agencies disagree on a number, we show both links rather than averaging them into a figure nobody published.

How cameras are matched to an event

We search our own directory of public live cameras for those with usable coordinates inside a radius chosen for that event type:

Earthquakes widen automatically with magnitude, up to 600 km for the largest events, because the area where shaking is felt grows with magnitude. When an agency publishes a track line or an affected-area outline, distance is measured to that shape rather than to a single centre point — otherwise a camera directly under a cyclone’s current path could be excluded because the storm’s reported centre had already moved past it. Our editors can also override the radius for an individual event.

The match score

Each candidate camera is scored out of 100:

• Distance — 45 points. Decays with the square root of the distance ratio, so the difference between 5 km and 40 km counts for much more than the difference between 100 km and 135 km.
• Camera category — 20 points. A harbour camera scores higher for a cyclone; a mountain camera scores higher for volcanic activity.
• Timing — 10 points. Higher while an event is active. For events where seeing anything depends on daylight, cameras where it is currently the middle of the night score lower.
• Live status — 15 points. A camera that is offline scores zero here, which makes it mathematically impossible for it to reach the threshold. Recency of our last automated check matters too.
• Location confidence — 5 points. Cameras whose coordinates are still flagged for review score lower.
• Curation — 5 points. A small bonus when the camera’s recorded city or country matches the event’s.

Cameras scoring below 65 are not attached at all. At most 6 cameras are shown per event, with the highest-scoring one presented as the nearest live camera.

What the distance means — and what it does not

Distance is measured from the coordinates the reporting agency published to the coordinates we hold for the camera. It is a statement about proximity. It is not a statement about line of sight.

A camera 12 km from an event may be:

• pointing in the opposite direction;
• separated from the event by hills, buildings or weather;
• delayed, buffering, or showing a static holding frame;
• showing an entirely ordinary scene in a place that is otherwise affected.

This is why every page says a camera may show general conditions near the area rather than that it shows the event. Unless one of our editors has explicitly confirmed what a camera is showing, we make no claim about its content. There is more detail in the guide to reading live cameras during a major event.

When an event ends

Events are closed automatically when the agencies that reported them stop publishing updates, on a timetable that suits the event type — 48 hours for an earthquake, up to two weeks for an ongoing volcanic episode. Events with an agency-published end time close at that time. Editors can close an event manually at any point.

Closed events move to an archived state after about three weeks. Archived events that still carry enough of our own reporting stay available and indexable; those that do not are removed from search results while their URL keeps working, so links shared at the time do not break.

Which pages appear in search results

Not every event we store deserves a page in Google. Each event gets a completeness score built from the length and presence of our own written sections, the number of attached live cameras, the summary, the location detail, and how many independent agencies reported it. An event page is excluded from search indexing when any of the following is true:

• No live camera could be matched to it.
• It has no summary, or no editorial sections.
• No source is attached, or the source is not one we recognise.
• Its completeness score is below 60 out of 100.
• It was merged into another event, or is still awaiting editorial review.

The same score gates advertising: pages that fall below the indexing threshold carry no ads, and neither do pages covering sensitive subject matter. The rules are in our editorial policy.

Automation and human review

Collection, deduplication, camera matching and the written sections on an event page are produced automatically from the structured fields the agencies publish. That automation never invents a figure: if an agency did not publish a depth, a wind speed or a casualty count, no sentence about it is written. Damage figures and casualty counts are never generated at all.

Our editors can override any of it — the title, the summary, the severity, the radius, which cameras are attached, which camera leads, whether the page is public, indexable, or carries ads. Once an editor has touched an event, the automated pipeline stops overwriting those fields and continues to update only the factual ones (coordinates, times, agency figures).

Reporting an error

If something here is wrong — a mismatched camera, a stale event, a location that does not make sense — please tell us. The corrections page explains what to include and how quickly we act.

TubeTraveler is not an emergency service. Camera footage may be delayed, unavailable, or unrelated to conditions outside the camera’s view. Follow instructions from local authorities and official emergency agencies.