A volcano observatory has two jobs during a crisis that pull in different directions, because it has to describe what the instruments show honestly while knowing that the words it chooses will move airline schedules, hotel bookings, property values and possibly tens of thousands of people out of their houses. This piece covers how ground alert levels and the aviation color codes work and who receives them, why national schemes use different numbers for the same hazard, what an elevated alert does and does not predict, the pressure that builds when months pass with no eruption, and how the evacuation decision gets made when the science cannot give a date.

Two scales running at once: ground level and aviation color

In the United States the Geological Survey publishes two things about a volcano at the same time, and confusing them is the most common mistake I see in an operations center. The Volcano Alert Level has four steps, Normal, Advisory, Watch and Warning, and it describes ground-based hazard status for people who live, work and respond near the volcano. The Aviation Color Code has four steps of its own, Green, Yellow, Orange and Red, and it exists because the users are dispatchers, airline operations centers and air traffic managers who care about one specific product of an eruption, which is ash in the flight levels. The two are published as a pair, so a volcano can sit at Advisory and Yellow together, or at Watch and Orange, and the pairing is how the observatory says who should be paying attention.

The definitions matter more than the colors. Green and Normal mean the volcano is at its typical background state. Yellow and Advisory mean the observatory is seeing unrest above background, which can be earthquake swarms, ground deformation, gas emission or thermal change. Orange and Watch mean heightened or escalating unrest with an increased likelihood of eruption and no stated timeframe, or an eruption already underway with limited or no ash emission. Red and Warning mean a hazardous eruption is imminent, underway or suspected, with significant ash emission into the atmosphere either likely or already observed. Those last two definitions are the part people skip, because Red does not only mean before, it also covers during, and a volcano erupting lava with very little ash can be at Orange while it is actively erupting.

The Volcano Hazards Program runs five observatories covering US volcanoes, the Alaska, Cascades, Hawaiian, California and Yellowstone observatories, and each one sets the levels for the volcanoes in its area of responsibility. The Survey’s 2018 update of its National Volcanic Threat Assessment ranked 161 active and potentially active US volcanoes into five threat categories and put 18 of them in the very high threat class, which is a planning product rather than a status product and should not be read as a statement that anything is about to happen. Congress authorized a national volcano early warning and monitoring system in a public lands act signed in 2019, and the practical consequence for an emergency manager is that the monitoring density at your nearest volcano may have changed since the last time you read the plan, so ask the observatory what it currently has in the ground.

The notices themselves and who actually receives them

When a US observatory changes an aviation color code it issues a Volcano Observatory Notice for Aviation, the VONA, which is a short structured message with the volcano name, the location, the previous and current color code and alert level, the observed activity, the ash cloud height if there is one and the time of the observation. Separately it issues public-facing information statements, and for a volcano in an active crisis it issues daily or more frequent updates. The notices are pushed by subscription, and the single most useful administrative act available to a county emergency manager near a volcano is signing up the duty officer address and the dispatch center address for those notifications rather than relying on someone seeing a news story.

Ash forecasting is a separate function carried out by nine Volcanic Ash Advisory Centres designated under the International Civil Aviation Organization framework, at Anchorage, Washington, Montreal, London, Toulouse, Tokyo, Darwin, Wellington and Buenos Aires. Each one covers a defined area, takes satellite, observatory and pilot report input, runs dispersion models and issues volcanic ash advisories with forecast cloud positions at intervals out to eighteen hours. Those advisories feed the meteorological watch offices that issue SIGMETs to aircraft and feed the airlines’ own dispatch decisions, and the whole arrangement is coordinated through ICAO’s International Airways Volcano Watch. The observatory sets the color, the advisory centre forecasts the cloud, and the decision to cancel or reroute a flight belongs to the operator and the air navigation service provider.

The reason that division of labor exists is a pair of investigated incidents. On June 24, 1982 a British Airways Boeing 747 flew into ash from Galunggung in Indonesia and lost all four engines, which the crew restarted after descending, and on December 15, 1989 a KLM Boeing 747 flew into ash from Redoubt in Alaska and also lost all four engines before restarting them and landing at Anchorage. Both aircraft landed with everyone aboard alive, both suffered severe engine and airframe damage, and both events are documented in the investigation record rather than in anybody’s recollection. Read the official reports if you want the mechanism, because the short version, that silicate ash melts in the hot section and accretes on turbine components, is accurate but leaves out how little warning the crews had.

Different countries, different numbers, same mountain

There is no global volcano alert scale, and the World Organization of Volcano Observatories has compiled the national systems precisely because they differ in the number of steps, the trigger criteria and the meaning of the top step. The Philippine Institute of Volcanology and Seismology uses Alert Level 0 through 5, with the higher levels tied to defined danger zone radii that are announced with the level. Indonesia’s volcanology agency uses four levels, Normal, Waspada, Siaga and Awas, and publishes them with recommended exclusion radii. Japan Meteorological Agency eruption warning levels run 1 through 5 and are written in terms of the action expected of the public, from awareness of an active volcano at the bottom to evacuation at the top, and JMA has revised the wording of the upper levels to align with the language used for flood and landslide warnings, so check the agency’s current phrasing rather than an older translation.

GNS Science in New Zealand uses Volcanic Alert Levels 0 through 5 and states plainly in its own material that the level describes current volcanic status rather than forecasting what happens next, which is the clearest official statement of that distinction I know of. Mexico’s national civil protection center publishes the Popocatepetl traffic light, the semaforo de alerta volcanica, with green, yellow and red each subdivided into phases, and it has held that volcano in the yellow band for years at a time, raising it to Yellow Phase 3 in May 2023 and returning it to Phase 2 the following month. The Servicio Geologico Colombiano uses a four-color scale for Nevado del Ruiz and others, raised Ruiz to orange in 2023 after a long period at yellow and later returned it to yellow, and its bulletins are where the dates and current level should be read rather than here.

Iceland’s Meteorological Office runs the aviation color code for Icelandic volcanoes alongside published hazard assessment maps that divide an area into zones with the dominant hazard named for each, which is a different communication product with a different purpose. For a mutual aid or airline planner the practical consequence of all this variety is that a number by itself transmits nothing across a border. A report that a volcano is at level 3 is only meaningful when the country and the issuing agency are attached to it, because level 3 of 5 in Manila, level 3 of 4 in Jakarta and level 3 of 5 in Tokyo are three different statements about three different situations.

What an alert level is not telling you

An alert level compresses a continuous, messy, multi-parameter judgment into one of four or five boxes, and the compression throws away exactly the information a decision maker wants. It does not carry a probability, it does not carry a time window, and it does not tell you the size of the eruption that the observatory has in mind. Two volcanoes at the same level can be in entirely different situations, one with a shallow intrusion and rapid deformation and one with elevated gas and nothing else, which is why the text of the bulletin matters more than the color band and why anyone who reads only the level is working from a headline.

The step changes also create a discrete problem out of a continuous one, which researchers who study these systems have written about at length, including the volcano crisis communication literature edited by Carina Fearnley and colleagues. Unrest does not climb smoothly through neat thresholds, so an observatory sometimes has to decide whether a signal that is halfway between two definitions justifies a change that will trigger evacuations, and if it changes the level on ambiguous data and nothing happens it spends credibility, while if it waits for unambiguous data it may have taken the time available to move people. That trade is unavoidable in a stepped system, and it is the reason good observatories publish narrative bulletins alongside the level instead of only the level.

The other thing the level does not tell you is the base rate, which is that most volcanic unrest does not end in an eruption. Christopher Newhall and Daniel Dzurisin’s 1988 USGS compilation of historical unrest at large calderas of the world established that pattern for calderas specifically, documenting many episodes of seismicity and deformation that subsided without magma reaching the surface. Live with that fact rather than arguing with it, because it means an elevated alert is an honest statement about elevated hazard and not a promise, and because a jurisdiction that treats every elevated alert as a countdown will burn out its public and its own staff before the eruption that actually comes.

Elevated for months: four episodes worth knowing

La Soufriere on Guadeloupe produced phreatic explosions and intense seismicity in 1976, and French authorities evacuated the southern part of Basse-Terre in August of that year, with the population moved commonly cited at around 70,000 people and the evacuation lasting into the autumn. No magmatic eruption followed. The episode also produced a public scientific dispute between Haroun Tazieff and scientists of the Institut de Physique du Globe de Paris over whether the hazard justified the measure, conducted partly in the press while people were out of their homes, and the economic damage to the island was severe by every account I have read. The published figures for cost and for the number evacuated vary by source, so treat any single number you see as one estimate among several.

In May 1982 the USGS issued a notice of potential volcanic hazard for the Long Valley caldera in California after a sequence that included four magnitude 6 earthquakes in May 1980 and measured uplift of the resurgent dome. The reaction in Mammoth Lakes was anger, because the notice landed on a town whose economy is built on visitors, and business owners and local officials said publicly that the announcement had done them real harm. No eruption occurred. The Survey subsequently revised how it issued and worded such notices, and that episode is one of the reasons US volcano communication now uses defined levels with published criteria instead of one-off hazard notices.

Rabaul caldera in Papua New Guinea entered a seismic and deformation crisis in the early 1980s, the observatory raised its alert stage, evacuation planning was carried out, and the activity then declined without an eruption. The eruption came on September 19, 1994, from Tavurvur and Vulcan, and the population’s own decision to leave in advance of it, informed by the planning done during the earlier crisis, is generally credited with holding the death toll to a handful. Campi Flegrei in Italy offers the slow version of the same problem, with ground uplift and seismicity that forced the relocation of tens of thousands of residents from central Pozzuoli in the 1982 to 1984 crisis and construction of a new district to house them, and the Italian civil protection department moved the caldera’s base alert from green to yellow in December 2012, where matters have continued to develop since. Check the department’s current statements for status, because that one is genuinely live.

Money, tourism and the politics of a long alert

An elevated alert has costs that begin immediately and fall on people who cannot see the hazard. Hotel cancellations, closed trails and cable cars, deferred construction, insurance repricing, mortgage difficulty and falling property values all start with the announcement rather than with the eruption, and the businesses carrying those costs are local while the alert is set by a national scientific agency. That asymmetry is the engine of almost every political fight over a volcano alert, and it explains why pressure to downgrade tends to arrive through elected officials and business associations rather than through scientific channels.

Aviation’s exposure is larger and faster. The April 2010 eruption of Eyjafjallajokull in Iceland closed much of European airspace for roughly a week, and the International Air Transport Association estimated the airline industry’s lost revenue at about 1.7 billion US dollars, an industry estimate rather than an audited figure, while European air traffic figures published at the time put cancelled flights around the 100,000 mark. The regulatory response was to move away from a blanket avoidance approach toward ash concentration thresholds and operator safety risk assessments, developed through ICAO, the European aviation authorities and the UK Civil Aviation Authority, and the current thresholds and procedures should be read from the aviation authorities rather than from a summary like this one.

The property side has recently been tested in Iceland. Grindavik was evacuated on November 10 and 11, 2023 during an intense seismic swarm associated with a magma intrusion beneath the town, an eruption began north of the town in December 2023, and repeated eruptive episodes followed along the same fissure system. Rather than leave several thousand residents holding unsaleable houses inside a hazard zone, the Icelandic parliament legislated in 2024 to have the state purchase residential property in the town, which is the clearest recent example of a government treating a long hazard as a property and fiscal problem instead of only a warning problem. Iceland’s Meteorological Office publishes the hazard assessments and eruption updates for that area, and it is the place to look for what has happened since.

The evacuation call when the science will not give a date

The observatory does not order evacuations anywhere that I am aware of, and USGS is explicit that it provides hazard information while decisions on protective actions belong to civil authorities. That division is correct and it is also uncomfortable, because the official with the legal authority to move people is being asked to act on a judgment they cannot independently verify, on a timeline nobody will state, with the local economy and their own political survival on the other side of the ledger. The Armero disaster in Colombia in November 1985, which I have written about separately on this site, is the case study in what happens when the hazard information is good and the decision chain does not act on it, and it is worth reading alongside this piece rather than summarized here.

The mechanics that make the decision tractable are zones and triggers set in advance. Mount Pinatubo in 1991 is the standard example, where PHIVOLCS and USGS built a five-level alert scheme during the crisis, tied each level to a danger zone radius, and expanded the radius in steps to 10, then 20, then 30 kilometers as the unrest escalated, with Clark Air Base evacuated on June 10 and the climactic eruption on June 15 arriving alongside a typhoon that turned airfall ash into a load heavy enough to collapse roofs. Published death tolls differ depending on whether they count only the eruption period or include the later lahars and illness in evacuation camps, so give the range and the source rather than a single figure, and USGS accounts credit the warnings and evacuations with saving thousands of lives and large amounts of property while noting that those savings are estimates.

Two structural tools are worth copying. The Philippines uses permanent danger zones around its most active volcanoes, which are areas where settlement is restricted at all times, so that an evacuation at an elevated alert becomes an expansion of an existing line rather than an argument about whether a line should exist. Italy’s Vesuvius plan divides the surrounding area into zones by hazard, with the red zone holding a population commonly put in the hundreds of thousands, and assigns each red zone municipality a partner region elsewhere in the country that will receive its evacuees, which turns the destination question into a standing agreement instead of a scramble. Verify the current zone boundaries and populations with the Italian civil protection department, because the plan has been revised more than once.

Coming back down, and the word false alarm

Downgrading an alert is harder than raising one, because the evidence for decline is the absence of things rather than the presence of them, and an observatory that lowers a level and then sees unrest resume has a credibility problem that is worse than the one it was trying to avoid. Most agencies handle this by requiring a sustained period of declining seismicity, deformation and gas emission before a step down, by describing the reduction in the bulletin narrative rather than only changing the color, and by lowering one step at a time. When you brief the public on a downgrade, say what was observed to change and over what period, because a level that drops with no stated reason teaches people that the levels are political.

The phrase false alarm deserves to be retired from the way agencies talk about this, and the reason is technical rather than rhetorical. An alert level describes hazard status at the time it is issued, so an elevated alert that is followed by no eruption was a correct statement about a real state of unrest, and the volcanoes at Long Valley, Guadeloupe and Rabaul in the episodes above were genuinely doing something measurable at the time. Say that out loud in the after-action briefing, because the alternative framing, in which the scientists got it wrong, is the framing that gets the next warning ignored. Aberfan and Vajont, both covered elsewhere on this site, show what ignoring a monitored and reported hazard costs.

The residue of a long alert is the thing to manage deliberately. People who spent months displaced, businesses that lost two seasons, and elected officials who took the political damage all carry that experience into the next crisis, and their behavior next time will be shaped by whether anyone acknowledged the cost and whether any compensation or assistance mechanism existed. That argues for building the economic side of a volcanic contingency plan at the same time as the evacuation side, naming which state or national programs could apply, what documentation a business would need, and who at the county level owns that conversation, because those questions get asked in week three and there is no good answer available by then if nobody looked before.

What to do at your agency

• Have your duty officer subscribe the agency’s shared duty email address and your dispatch center to the notification service of the volcano observatory responsible for your area this week, then send a test message through your own notification tree so you know the notice reaches a shift supervisor at three in the morning.
• Ask the observatory’s scientist in charge, by name and in writing, for the current alert level criteria and the hazard zone maps that apply to your jurisdiction, and file the response with the date it was received in the hazard annex you already maintain.
• Put one item on the agenda of your next local emergency planning committee meeting asking who in your county has the legal authority to order an evacuation for a volcanic hazard, and get the answer written into the plan with the statute or ordinance identified by your county attorney rather than by you.
• Have your planner write one paragraph into the existing evacuation annex stating which zones are evacuated at each alert level, so the trigger is decided before an alert exists rather than during one.
• Ask your airport manager or the nearest air traffic facility which Volcanic Ash Advisory Centre covers your airspace and how ash advisories reach airfield operations, and record the answer, because in an ashfall event your airport becomes both a closed facility and a possible staging area.
• Brief your fire and EMS shift commanders on ashfall specifics at a shift meeting that is already scheduled, covering respiratory protection for crews, engine air filter consumption on apparatus, roof load on structures with wet ash, and the fact that ash does not behave like snow when it is wet.
• Have your public information officer draft and get approved now the message you would issue for an alert level increase with no eruption, including the sentence explaining that an elevated level is a statement of current unrest rather than a prediction of when an eruption will occur.

Takeaways

• US volcanoes carry two published statuses at once, a four-step Volcano Alert Level for ground hazards and a four-step Aviation Color Code for ash in the flight levels, and the pairing tells you which audience the change is aimed at.
• Red and Warning cover an eruption that is underway with significant ash as well as one that is imminent, and a volcano actively erupting lava with little ash can legitimately sit at Orange and Watch.
• Observatories set levels and issue Volcano Observatory Notices for Aviation, nine ICAO-designated Volcanic Ash Advisory Centres forecast the ash clouds, and the decision to cancel or reroute a flight belongs to the operator and the air navigation service provider.
• National alert schemes differ in the number of steps and in what the top step means, so a reported level is meaningless without the country and the issuing agency attached to it.
• An alert level is a compressed statement of current status that carries no probability, no timeframe and no eruption size, which is why the narrative bulletin matters more than the color.
• Most volcanic unrest ends without an eruption, a pattern documented for large calderas in Newhall and Dzurisin’s 1988 USGS compilation, and an elevated alert that ends quietly was still a correct statement about real unrest rather than a false alarm.
• The costs of an elevated alert land immediately on local tourism, construction, property values and aviation, which is where the political pressure to downgrade comes from, and Iceland’s 2024 legislation to buy residential property in Grindavik is a recent example of a government treating that exposure as a fiscal problem.
• Observatories do not order evacuations, so pre-agreed hazard zones tied to alert levels, as used at Pinatubo in 1991 and in the Philippine permanent danger zones, are what make the civil authority’s decision possible under uncertainty.
• Write the conditions for a partial return and for daytime access before the evacuation order is signed, because the return argument arrives months later and improvised return policy sets precedent you cannot take back.