VulcanWatch AI
Support desk · volcanoes & earthquakes near you
This is a reading room, not a siren. VulcanWatch AI repeats what monitoring agencies have already published and explains their wording. It issues no warnings, forecasts no eruption and predicts no aftershock. Anything urgent reaches you through your local civil protection authority, and that is the source to act on.
What the marks on the map are
A cone — a volcano from the Smithsonian catalogue. Its outline is filled with the alert level an agency has published for it, and stays grey when nobody publishes one. Tapping it opens the catalogue entry: type, elevation, coordinates, geological summary, eruption record, and the current bulletin if one exists.
A circle — the epicentre of a recorded earthquake. The circle grows with magnitude and fades as the event ages out of your time window. The card behind it carries depth, both timestamps, the reporting agency, felt reports, instrumental shaking, the PAGER level and how the solution was computed.
Distance is measured from one point and one point only. Either your position or a point you long-press on the map. Dragging the map to look elsewhere moves the search area, never the ruler — a volcano in Iceland keeps the distance it has from where you actually are.
Alert levels, and the grey that is not one
Alongside the level a volcano may carry an aviation colour code — green, yellow, orange or red. That one describes the hazard to aircraft from ash, it is issued for pilots, and it can differ from the ground-level rating on the same mountain on the same day. The card shows whichever of the two the observatory published, with the date and the observatory's name.
Where the model writes, and where it does not
The model has three jobs and no fourth: turning the fields of a card into a few plain sentences, translating an official bulletin into your app language, and answering questions about the one object whose card is open. It is shown the same fields you are, and it is instructed never to estimate, round, forecast or fill a gap from general knowledge.
Every generated paragraph carries a caption saying so. Text under that caption can be wrong — a mistranslation, a dropped qualifier, a misread field — so the original bulletin is one tap away on each card, and it is the text that counts. Magnitudes, depths, coordinates, times and alert badges carry no such caption, because they never pass through the model at all: they are drawn from the agency answer exactly as published.
Questions that reach this desk
The map around me is empty.
Usually the filters rather than the planet: the radius, the time window and the minimum magnitude all cut. Widen the radius from the drawer, or switch to Nearest, which ignores radius entirely and returns the closest objects wherever they happen to be.
A volcano near me shows no colour.
Only the volcanoes a monitoring agency actually rates carry a level — 69 of them in the USGS list against 1,214 entries in the catalogue. The rest are grey. Grey means unrated, and we will not paint it green, because green would read as checked and calm.
A magnitude changed between two openings.
Agencies revise solutions as more stations report. The card carries the revision time, the station count and the azimuthal gap so a changed number can be told apart from a different event.
Why is the geological summary in English?
The Smithsonian publishes those summaries in English only, and we show them as written with a line saying so. Machine translation happens in a separate, captioned block — never silently in place of a source text.
Nothing loads at all.
With no connection the app falls back to a catalogue extract built into it: names, coordinates, elevations. No live alert levels and no earthquakes, since both are readings with a publication time and a stored copy would be a stale number wearing a live one's clothes. The header says what you are looking at; pull down to refresh once you are back.
Must I give my location?
No. Decline it and long-press the map to set a point instead; everything is then measured from that point. The prompt can be answered later from iOS Settings if you change your mind.
What do the notifications send?
Only what you asked for on objects you added to Watching: an alert level that changed, a newly published bulletin, or a quake above a magnitude you set inside a radius you set. Profile carries one master switch over all of them, and declining the system prompt breaks nothing else in the app.
Does any of this cost money?
Nothing at all. The app carries no store, no subscription, no paid tier; nothing on any screen is held back until someone pays, and no advert is shown inside it. The agency services behind the numbers are public and free to read.
What was that tracking prompt for?
It asks whether we may attribute your install to the advert it came from. Answer either way: the map, the agency data, the alert levels and the explanations behave identically, and the answer can be changed in iOS Settings later.
Can it tell me an eruption is coming?
No, and nothing in it tries. It reports levels and events that have already been published by an agency. Forecasting is not something we do, not something the model is allowed to do, and not something to wait for from this app.
What does Reset data remove?
Your watched objects, your saved preferences and the cached agency answers. It happens on the device, it cannot be undone, and the next launch starts from onboarding.
Can I change the app language?
Yes — Profile, App language. The interface and the bulletin translations both follow it immediately, without restarting the app.
What travels and what stays
- Sent when you ask for an explanation
- The agency fields already on screen — name, coordinates, alert level, magnitude, depth, times — the official bulletin if there is one, and your question. Your coordinates are not among them: the distance is computed on the device and only that finished number travels.
- Who handles it
- A language model is reached through the Runware API: it relays the request, then hands back whatever the model wrote. None of that runs on your phone — it is their infrastructure, under their terms. Nothing leaves the device until you accept the data-processing notice, and the map, the distances and the watchlist are all fully usable with that notice declined.
- Kept on the device
- The last agency answers, so the map opens offline; your watched objects and your preferences. Clearing them is one tap in Profile.
- Policy and deletion
- The full text sits at vulcanwatch.observer/info/privacy-terms. For a deletion request, write from the address the request concerns and say what you want removed; we reply on the same schedule as anything else at this desk.
Writing to us
One address, read by a person:
hi@miawhitehi.clickA report is far quicker to act on with these in it:
- iPhone model and iOS version;
- the app version, printed at the bottom of Profile;
- the object it concerns — the volcano's name, or the magnitude and time of the quake;
- what the screen showed against what you expected, and a screenshot if one exists;
- for a wrong number: whether the original bulletin, opened from the card, said the same.