Introduction

Volcanic eruptions are among the most destructive natural hazards and may disrupt life and human socio-economic activities even at a global scale. To reduce the potential impacts of volcanic eruptions, it is crucial to be able to anticipate them well in advance. It is generally found that volcanic eruptions are preceded by changes in the behavior of the volcano (i.e., volcanic unrest), including an increase in seismicity, surface deformation, and gas emissions1,2,3,4,5,6,7,8,9,10. These signals may be recorded by various monitoring instruments and, in many cases, they may also be felt by the populations leaving close to the volcano.

Every volcano has its own physical and chemical characteristics (internal structure, rock rheology, magma composition, etc.), so pre-eruptive unrests may show significant variations for the monitored geophysical and geochemical parameters from one volcano to another. A given volcano may show similar values for given parameter (e.g., RSAM) when preparing for eruption of similar size and characteristics (e.g., Merapi11) but may also show unrest episodes that differ from those from previous eruptions8,12. The situation is even more complex in the case of volcanoes that have been dormant for long periods and have not erupted in historical times. We have no record of monitoring data to use as a background, and the detailed knowledge about previous eruptions may also be lacking (e.g., Sinabung volcano, 2010, Indonesia13).

Establishing potential patterns in the evolution of volcanic unrest that could help identifying the outcome within a limited degree of uncertainty, requires the analysis of as many unrest episodes as possible14. The importance of having large datasets to use probabilistic analysis or other approaches in eruption forecasting15, stresses the need of a database of observational data freely available for being consulted. However, the available databases on volcanic unrest8,12,16,17,18,19, as well as the datasets stored in each volcanic observatory around the world, require to be open and containing data stored in a comparable format, so existing data can be reported and interpreted in the same way (i.e., using same formats and time scales)20 to be able to identify common or different behavior patterns among all volcano types and, thus, to establish effective methods to forecast volcanic eruptions.

However, eruption forecasting cannot be only based on the analysis of volcanic unrest. It also requires to identify which outcomes from such unrest (e.g.: no eruption, phreatic explosion, magmatic eruption) have the highest probabilities of occurrence and how they are associated with a given unrest pattern. This is necessary to implement the emergency plans according to the most probable outcomes. Therefore, eruption forecast also requires a hazard assessment that needs to be combined with the unrest analysis, in order to get a precise short-term hazard analysis that could identify how, when and where the next eruption will be, rather than only knowing whether the eruption will occur or not21,22. Otherwise, it may result in failed eruption forecasts either because eruption does not occur or because it does not correspond with the expected size (e.g.: La Sufriere Guadeloupe, 1976; Akutan, Alaska, 1996; Tungurahua, Ecuador, 1999; Paricutin, Mexico, 2006; Fourpeaked, Alaska, 2006; Mayon, Philippines, 2018).

Here I comment on some aspects related to how volcanic eruptions can be anticipated. I discuss the need to conduct long-term hazard assessment in addition to volcano monitoring, to be able to provide more robust eruption forecasting during volcanic crises. In this sense, I propose that a systematic integration of a detailed knowledge of the eruptive history of a volcano, based on the analysis of the geological (and historical) records, and monitoring data on its current state of activity, is the main way to forecast future eruptions.

Volcanic unrest

A volcanic eruption typically requires an overpressurised batch of magma to ascent to the earth surface or, conversely, decompression of the shallow magma systems due to mass unloading (e.g.: sector collapse) of the volcanic edifice, or rupture of the magma chamber caused by tectonics (e.g., earthquake). Magma may come from shallow or deep reservoirs (or chambers) where it has accumulated and differentiated, or even directly from the source region23. To reach the earth’s surface, magma needs to deform the surrounding rock, displacing it apart and opening new fractures, to create the necessary space and pathways to cross from deeper to shallower levels24,25. This will produce a series of changes in the vicinity of the magma that may be translated into surface deformation, seismicity, or other changes of potential fields that should be detected by ground based and remote geophysical monitoring systems1,2. Moreover, when magma approaches to the surface and pressure decreases, the dissolved gases may exsolve and separate from the liquid phase, thus giving rise to appearance of geochemical indicators of magma ascent26,27,28.

However, a volcano may also change its state due to external factors not related to magma movement, such as changes in regional tectonics29 or in the conditions of the associated geothermal systems26,27,30, which will also result in a variation of the monitored parameters. Whether or not the observed activity at a given volcano is caused by changes in the magmatic system or by changes of the regional stress field or the geothermal system, and whether or not the unrest phase will end with an eruption, is the challenging question that needs to be answered with interpretation of monitoring data and good knowledge of the past behavior of the volcano5,8.

An increase of activity (seismicity, ground deformation, gas emissions) compared with a previous background level is what is generally known as volcanic unrest12 (Fig. 1). The background level of activity above which we may consider that volcanic unrest is occurring, needs to be defined for each volcano by experts who know about its present and past behavior. Establishing a background level of activity should be mainly based on the information provided by the monitoring networks, but also considering other volcanological aspects such as the past history of the volcano and historical chronicles. Comparison between volcanoes may sometimes help25,31, but volcanoes are complex natural systems, subjected to a large number of non-linear processes that make them easily depart from pre-assumed patterns. Even if eruptions of similar characteristics may occur at different volcanoes it is not necessarily true that unrest episodes preceding them have to be also similar (e.g., El Hierro 2011–2012 and La Palma 2021 eruptions in the Canary Islands). When establishing the background level for a particular volcano, it is important to keep in mind that the range of variation of monitoring parameters, as well as the time scales for these variations, may differ significantly from one volcano to another or even at the same volcano from one eruption to another32. So, establishing comparisons between volcanoes or even assuming the same degree of background activity for different volcanoes, even if they are similar in composition or behavior, is not a good solution. It is essential to thoroughly understand the specific volcano before determining its range of activity.

Fig. 1: Time evolution of volcanic activity during an unrest episode (not to scale).
figure 1

Volcanic unrest starts when there is an increase of activity (i.e., increase in the values of the monitored parameters) in the volcanic system with respect to a previous background level. In most cases (e.g.: www.WOVOdat.org) volcanic activity will increase progressively, with a clear acceleration at the last moment, until an event occurs. This may be an eruption, a phreatic explosion, or also a sudden decrease of activity due to the relaxation of the system (e.g., due to massive degassing), returning to its normal state without any eruption or explosion. The occurrence of an event marks the end (the outcome) of that particular unrest, even if volcanic activity increases again (new unrest episode). The relative stages in the evolution of the unrest using probabilistic terms to indicate how it may are also indicated, as well as the time window in which usually an alert should declared. Despite the shape of the curve represented here is similar in most volcanic unrests, the exact level of activity and duration of the unrest may be significantly different between volcanoes and between eruptions of the same volcano.

Full size image

The analysis of an unrest episode requires defining its time limits (when it starts and when it finishes) (Fig. 1). In some cases, the background level of activity may be not known because of lack of monitoring or the short time in which it has been operating. In such cases, the geological history of the volcano and historical records, when they exist, are the main tools to forecast future activity. In others, we may observe fluctuations (i.e., increases and decreases of activity) (e.g., Campi Flegrei caldera33) that may not help to define a clear tendency in the evolution of the unrest. Also, there are volcanic systems in permanent unrest since monitoring was implemented, so there is no confidence to establish a reference background level34. Moreover, in retrospective analysis, it would be interesting to consider time scales much longer than the monitored period or the unrest phenomena described in historical chronicles, but volcanic unrest is not registered in the geological records. Or, we may face the dilemma of whether or not significant increases of activity not ending with an eruption can be considered as unrest (Fig. 2).

Fig. 2: Time scale evolution of long- and short-term hazard analyses and variation of the degree of detail, considered in eruption forecast.
figure 2

Long-term hazard assessment: long-term assessment is based on historical and geological data, as well as on simulation models of possible hazards, and refers to the available time window before an unrest episode occurs in a volcanic system that currently shows no signs of unrest. Short-term hazard assessment: refers to the unrest phase, when complementary information resulting from the combination of long-term analysis and real-time monitoring data is used to update the status of the volcanic hazard and to forecast a possible eruption. Unrest: any variation with respect to the background level or, in other words, any change in the state or dynamics of the volcanic system, recorded by monitoring networks and/or perceived by the nearby populations, which correlates with a volcanic event (outcome in the terms of hazard), being this an eruption or no eruption. In the case that the unrest does not end up with an eruption, the return of the geophysical and geochemical indicators to a background level will coincide with the end of the unrest episode. Outcome: end of the unrest associated with a hazard occurrence. No outcome: end of unrest without associated hazard. Historical: time period that goes from the appearance of written records to present. Monitoring: time period that covers the registered instrumentally volcanic activity. Geological record: time period that covers all geological registers from a specific volcano or volcanic system.

Full size image

Any unrest episode will have different stages and forecasting possible outcomes should be more reliable as time passes and more information is gathered. Figure 3 shows a flow chart-type diagram detailing the sequential way in which we could approach the different steps of a volcanic unrest analysis. First, it is necessary to characterize the volcanic system in terms of eruption dynamics, eruption frequency, and magma composition. In addition, we need to gather information on the geodynamic setting and degree of deformation (seismicity, strain, etc.) that the volcanoes may be currently experiencing at a regional or local level. This allows discriminating between local deformation that may be attributed to the volcano from that related to the activity of plate boundaries or mantle instabilities (e.g., mantle plume upwelling). Another important step is to identify groups of unrest indicators that will better describe the evolution of unrest in a volcano8. These will surely include seismicity, surface deformation, potential fields, gases, and may be other that we could consider in each particular case and depending on monitoring facilities available35. For example, in open vent volcanoes degassing and seismicity seem to be better indicators of unrest36 than deformation, whereas currently, the dome complex of Laguna del Maule is experiencing a huge deformation and significant seismicity, although limited or no degassing37. The next step is to specify which particular parameters we will consider (e.g.: for seismicity: seismic energy released, total number of VT events, presence of LP events, etc.; for gases: total gas flux, presence of SO2, etc.), and which ranges of variation may be assumed to consider a change in activity significant or representative of the evolution of the volcanic unrest18.

Fig. 3: Volcanic unrest analysis.
figure 3

Schematic representation of the stages included in the volcanic unrest analysis, indicating the different aspects or parameters to be considered, and the possible uses of such analysis (see text for more explanation).

Full size image

Furthermore, to enhance our understanding of the volcanic system’s behavior, it is essential to gather monitoring data during any eventual eruptions. This will reflect possible variations of monitoring parameters related to variations in eruption dynamics, which could indicate variations in the plumbing system. This analysis needs to be complemented with the petrological and rheological characterization of erupted products in near real time, as variations in their composition and physical properties may explain the changes observed in the monitoring signals38,39. Such systematization of unrest indicators, and combination with other geological data, should contribute to better understanding the physical meaning of the various levels of unrest, thus improving volcano forecasting.

Eruption forecasting, long- and short-term hazard assessments

In general terms, forecasting is defined as the process of making predictions of the future based on past and present data and analysis of trends40,41. In volcanology, forecasting basically intends to predict the occurrence of future eruptions (in space and time) based mainly on the analysis of unrest episodes2,21,24,35,42,43,44,45,46. This means that eruption forecasting concentrates on using real-time monitoring data to determine whether or not an eruption will occur, and if so, when it will happen. However, such forecasting is incomplete if we are not able to anticipate the kind of eruption and its potential extend (size) and impacts, for which it is necessary to also consider the past history of the volcano22,47. This task may be achieved by combining long-term hazard assessment with volcano monitoring, thus obtaining the short-term assessment and, finally, the eruption forecast15,21.

Long-term hazard assessment is based on historical and geological data, and refers to how the volcano has behaved in the past, before a new unrest episode occurs21,47,48,49,50,51. In this case, fundamental geology is essential to establish reliable basis on which to build the hazard assessment structure, to determine the time constraints of the volcanic processes and eruption frequencies, and to characterize the products (i.e., hazards) from past eruptions47. Based on this information, long-term hazard assessment should identify all possible eruptive scenarios that have occurred in the past and, from them, the most probable that may occur in the future (Fig. 4). Long-term hazard assessment is essential for territorial planning and elaboration of emergency plans, as well as to implement educational programs addressed to inform general public on volcanic hazards.

Fig. 4: Long-term hazard assessment.
figure 4

Schematic representation of the sequential steps included in long-term hazard assessment, indicating the input parameters, possible outputs of each phase of such analysis, and the corresponding actions (see text for more explanation).

Full size image

Long-term volcanic hazard assessment is an essential part of any volcanic risk reduction program47. It can be considered as a sequential process aimed at implementing several actions necessary to determine the level of hazard of a particular area and, therefore, to contribute to reduce its potential impact. These actions include spatial analysis, temporal analysis, simulation of the eruptive scenarios, and elaboration of partial and total hazard maps. At each step it corresponds a variety of input data, such as historical, geographical, and geological data, theoretical models, and multiple outputs (Fig. 4). The results are highly dependent on the data used, and thus the completeness and quality of data sources is crucial for a proper long-term hazard evaluation. The spatial analysis consists on identifying from where past eruptions have been sourced and, consequently, will help to infer “where” the next eruption can take place, i.e., the spatial probability of occurrence of a new eruptive vent (volcanic susceptibility52). This analysis needs to be completed with the temporal probability, which informs on when past eruptions have occurred and on which scenarios characterize each past eruption. The temporal analysis consists of identifying all the possible eruptions recorded in the historical and geological records of the volcano and establishing, from this information, its eruptive recurrence, with which we can answer the question of “when” the new eruption will be. Complementing the information obtained from the temporal analysis, we have to proceed with the identification of all possible eruption scenarios occurred in the past and, from all them, which have occurred with a higher frequency (i.e., the most probable scenarios), and their potential characteristics and extend. This requires simulations to reproduce the possible eruptive scenarios and identifying the areas that may be affected by each of them. The last step in the long-term assessment is the elaboration of partial (for each hazard in particular or combinations of some of them) and total (considering all potential hazards) maps47. All the results from the long-term hazard assessment are expressed in probabilities, which implies the use of different methodologies (event trees, Bayesian Inference, Belief networks, expert assessment, etc.) to assign them, but I am not going to explain them here since it exceeds the purpose of this Perspective Contribution (see refs. 15,49,50,51,52,53,54,55)

During a volcanic unrest, such information will constitute the basis to identify the most probable outcome by conducting a short-term hazard assessment (Fig. 5). This combines the long-term analysis and real-time monitoring data to update the status of the volcanic hazard15,21,32,48,53. Short-term evaluation will constraint how, where and when the eruption will take place, although a considerable degree of uncertainty may still remain depending on how well we know our volcanic system56,57,58. Short-term hazard assessment (Fig. 5) permits to update the previous susceptibility map, to infer the most probably outcome (e.g.: eruptive scenarios) of the unrest phase, and, consequently, to identify the exposed elements and the actions to be undertaken to protect them. Therefore, results from the short-term analysis will be addressed to refine the emergency preparedness and to respond in the most effective way to the event.

Fig. 5: Eruption forecasting steps.
figure 5

Schematic representation of actions to be undertaken as a function of the time available (see text for more explanation).

Full size image

Discussion and conclusions

The best way to reduce volcanic risk is by anticipating to volcanic eruptions. This may be achieved by undertaking a series of actions both in the long term, when the volcano is not showing signs of activity above its characteristic background level, and in the short term, when the volcano enters into a new unrest period. Among others, these actions should include hazard assessment, implementation of monitoring systems, effective land planning, development and implementation of emergency plans, promoting educational programs, and implementing mitigation measures to reduce vulnerability. From the scientific side, hazard assessment and volcano monitoring are the most important contributions to reduce volcanic risk, together with educational actions addressed to improve population knowledge on volcanic hazards.

When a volcano enters into a state of anomalous activity or unrest, monitoring data should provide the necessary information to infer the causes of such unrest and its potential evolution. The evolution of an unrest episode will depend on the causes of the unrest (magmatic, tectonic, or geothermal), which may give different outcomes (magmatic eruption, phreatic explosion, sector failure, or others) in a range of locations with different possible eruption magnitudes, products, scope, etc.32,50. Each particular scenario is expected to result from a particular pattern in precursory activity. However, there are factors in each scenario that cannot be anticipated merely by studying monitoring data but which can be identified by examining the past eruptive history of the volcano. In an ideal situation, the short-term hazard assessment should be constructed on the long-term assessment, so knowing well the past eruption history of the volcano. However, when this is not known and an unrest starts, the short-term has to rely on an incomplete knowledge of the volcano and/or on comparisons with other potentially similar volcanoes. A good example of this situation was El Hierro eruption in 2011–2012, in the Canary Islands. In the case of El Hierro, no previous hazard assessment existed, so the most probable scenario, a submarine eruption, as it was shown by a subsequent study59, was not anticipated. Consequently, scientific advisors and decision-makers considered possible eruptive scenarios that had much lower probabilities of occurrence, which implied making decisions with a higher cost than necessary54.

The tremendous scientific and technological advance experienced by volcanology in recent years has resulted in the deterministic forecast of the eruptive behavior of volcanoes with a high eruption frequency and a good knowledge of their eruptive past, as is the case of the Stromboli volcanoes60,61,62,63,64,65 and also of some more infrequent eruptions24,66. Moreover, the 2021 Fagradalsfjall eruption, Iceland, is a good example on how application of short-term eruption forecasting based on a previous long-term hazard assessment may provide accurate predictions (about 90% of coincidence in this case) even in places of long eruption recurrences (about 800 years since last eruption)67.

However, despite these advances, real-time analysis of pre-eruptive activity in many volcanoes still cannot provide exact numbers (threshold values) to precisely predict when an eruption will begin. In fact, we are still far from acquiring the necessary amount of observational data able to guarantee the efficient application of statistical methods capable to identify discriminating patterns in the pre-eruptive behavior of volcanoes21. The fact that all volcanic systems, even those theoretically pertaining to the same volcano type, present different characteristics for what concerns internal structure, plumbing system, state of strain and stress, or magma rheology, does not facilitate the use of specific values as markers or thresholds that may be applied to forecast the future of a volcano. However, comparison between volcanic unrests among different volcanoes has occasionally provided, for example, coincidence in the total accumulated seismic energy values reached before eruptions68, but in the same way these values have been clearly exceeded by orders of magnitude in other unrest episodes without leaving to an eruption69.

Fortunately, the amount of available information is progressively increasing with the deployment and implementation of new monitoring networks, organization of new volcano observatories, and the development of new research on volcanic eruptions. Initiatives such as WOVOdat19, complemented with other databases (Smithsonian GVP17; LAMEVE70; VOGRIPA, https://www2.bgs.ac.uk/vogripa/; DomeHaz71; CCCB72; Volcanic Unrest12; VUI18; VOLCANS31, etc.), containing information on different aspects of volcanic phenomena (volcanoes, unrest, large eruptions, calderas, domes, etc.), should be welcome and fully supported, as they offer good data sources to identify pre-eruptive and eruptive behaviors among different volcanoes. However, the existing data are still too disperse, stored in a wide variety of formats and not always available. Fixing this gap requires an open collaboration between all scientists and the definition of common formats and agreement in their use in order to ensure that available data are compatible and comparable.

With this contribution, I only pretend to offer a systematic approach to what should be done to forecast future eruptions, by combining long-term hazard assessment (i.e., the geological and historical records of the volcano) with real-time monitoring data. To identify unrest patterns at the same volcano and from volcanoes with similar characteristics, it is crucial to obtain as many retrospective analyses as possible. This would help to categorize volcanoes according to unrest types or levels of unrest. Moreover, it would facilitate establishing predictive models based on the identification of patterns that repeated in the past and that may represent a particular relationship between unrest evolution, magma and rock properties, and potential eruption dynamics. In a similar way, the study of the past eruptive history of volcanoes will provide the clues to define their future potential hazard level. And is the combination of both, long-term hazard assessment and real-time monitoring, which will result in the most accurate forecasting for future eruptions.