箱根は東京から西へ約100 km，車や鉄道を利用すると2時間程度で到着できる観光地である．大涌谷は，年間300万人の来客があるとされる箱根を代表する観光スポットであるとともに，箱根火山最大の噴気地帯で，2015年噴火はここで発生した．この写真で見えるいくつもの噴気は，この噴火で形成された噴気孔から噴出している．約3500年前に発生した最新のマグマ噴火後に何回か発生している大規模な水蒸気噴火もこの周辺で発生している．観光地が潜在的な噴火中心であることから，大涌谷における噴火予知は喫緊の課題である．写真は2015年噴火の噴出中心である谷地形としての大涌谷と（手前），観光地としての大涌谷（中程），およびをこの地のもうひとつの観光源である富士山（奥）を展望したものである．

（写真：宮下雄次　2021年4月26日撮影；説明：萬年一剛）

 A phreatic eruption is a phenomenon in which water near the surface expands rapidly due to magma-supplied heat, ejecting the surrounding rocks. Recent studies of conceptual models, subsurface structures, pre-eruption processes, and eruption processes of phreatic eruptions are reviewed. These eruptions often occur in volcanoes with well-developed hydrothermal systems, where a low electrical resistivity layer is found near the surface using magnetotelluric surveys. The low resistivity layer indicates a low-permeability structure that acts as a pressure-confining cap on the hydrothermal system. In the brittle-ductile transition zone above deep magma, a sealing structure associated with quartz crystallization develops. Volcanoes with open conduits that connect magma reservoir and surface crater also have the potential for phreatic eruptions. A low-permeable sealing structure in the shallow part of the conduit plays an important role in eruptions of this type of volcano. Phreatic eruptions are prepared by an imbalance in the hydrothermal system, which is caused by increases of heat, volcanic gases, and fluids from the deep magma reservoir, and are triggered by depressurization of the aquifer due to the breakdown of the cap/sealing structure. In recent years, eruptive processes have been modeled using data from broadband seismograms and tiltmeters near vents. At Ontake, Hakone, and Aso volcanoes, slow crustal movements or very low-frequency earthquakes were observed just prior to phreatic eruptions. These phenomena result from crack opening due to the rapid vaporization of liquid water. Incremental seismic activities, low-frequency earthquakes, and expansion of volcanic edifice, and geochemical changes in volcanic gases and hot springs are identified as long-term eruption precursors. These precursors reflect the supply of new magma, related changes in volcanic gases, and increased fluid pressure in shallow hydrothermal systems. Several new techniques for monitoring volcanoes to detect temporal changes in resistivity, crustal deformation, and chemical composition of hot springs and groundwater have been developed for forecasting eruptions.

 On August 3, 2014 and May 29, 2015, eruptions occurred at the Shindake summit crater of Kuchinoerabujima volcano in the Ryukyu Islands, southwestern Japan. The Japan Meteorological Agency (JMA) upgraded the Volcanic Alert Level (VAL) to 3 (warned zone within 2 km from the crater) after the onset of the 2014 eruption and to 5 (evacuation) after the onset of the 2015 eruption. The possibility of implementing early warnings for eruptions and forecasting the area most likely to suffer damage from volcanic eruptions are examined based on monitoring data and disaster-affected areas of historic eruptions. The onset of the 2014 eruption was preceded by a 15-year prolonged increase in volcanic activity that started in July 1999. Only a short-term tilt change was observed immediately before the eruption. The prolonged volcanic activity is characterized by: 1) repeated bursts of seismicity; 2) ground inflation events around the crater associated with increases in seismicity; 3) increases in geothermal activity and 4) appearance of fumarole. The short-term process consisted only of a tilt change of crater-side up one hour before the onset of the 2014 eruption. The phenomena prior to the 2015 eruption were more intense than those prior to the 2014 eruption, as demonstrated by seismicity, which included a felt earthquake six days before the eruption; larger ground deformation; higher rate of discharge of SO₂ gas; and, higher temperature. Despite more intense activity, VAL remained at 3. VAL was upgraded from 3 to 5 immediately after the 2015 eruption and then all of the residents were evacuated from the volcanic island by ferry boat. Decreases in seismicity, SO₂ gas discharge rate, and geothermal activity led to an initial reduction of the alert zone radius to less than 2.5 km in October 2015. A further reduction to less than 2 km (VAL 3) was later implemented based on deflation around the summit area in June 2016. Problems related to evacuation decision-making in May 2015 are as follows: 1) JMA had no clear judgment criteria for VAL 4 and 5; 2) Volcanic hazards had not been evaluated based on monitoring data in the precursory period; 3) JMA did not indicate a clear hazardous zone in the warning information of VAL 5 even after the eruption; 4) Disaster measures, which assumed a pre-eruption VAL rise, were not appropriately implemented due to the post-eruptive VAL 5 declaration; and, 5) Return of evacuees to their homes was delayed due to management failure by the decision-making board. It is recommended to develop a method for constantly evaluating volcanic hazards from monitoring data with the progress of volcanic activity.

 Volcanoes with shallow hydrothermal systems are often accompanied by background volcanic activity such as fumarolic activity, microseismicity, and ground deformation even in the non-eruptive phase. When elevated, they are said to be in a state of “unrest.” It is not difficult to imagine that such events of unrest reflect changes in the state of the shallow hydrothermal system beneath a volcano. However, there is currently no method by which these events can be used to quantitatively evaluate eruption imminency or predict eruption intensity based on physical and/or chemical models. A potentially useful application of such unrest events for probabilistically forecasting eruptions is discussed. First, the method proposed by Hashimoto et al. (2019) for compiling and evaluating the sources of unrest events, such as thermal demagnetization, is described. Then, the volcanic unrest index (VUI) of Potter et al. (2015a) is proposed as another key tool. Finally, a concept is proposed for integrating the VUI and the unrest data to make probabilistically forecasting eruptions feasible. Also described is a recent attempt to introduce the VUI for evaluating a volcano in Japan. Information on sources of unrest in the form of the scatter plot of Hashimoto et al. (2019) can be used as one of the rating criteria on the VUI worksheet. The key idea is to divide the source diagram into regions based on the probability of posterior eruptions given unrest events and to assign VUI scores to these regions. Such a procedure may augment the VUI's function, partially enabling probability-based eruption forecasting. Irrespective of whether the VUI is applied or not, it is essential to obtain temporally homogeneous monitoring data during both eruptive and non-eruptive periods for a quantitative evaluation of unrest events. Surveys and analyses carried out regularly over long time periods also play an equally important role. Therefore, to realize of probabilistic eruption forecasting, it is fundamentally important that monitoring networks are run properly and the data are shared appropriately.

 Earthquake swarms have occurred with volcanism repeatedly at Hakone volcano in Kanagawa prefecture, Japan. In 2015, a phreatic eruption took place about two months after the start of an earthquake swarm. Hakone volcano is a popular tourist destination. If it is possible to forecast at the early stages of an earthquake swarm whether or not an eruption will occur, the forecast could contribute to preventing disasters involving tourists. At Hakone volcano, increases in the ratio of components (CO₂/H₂S) contained in the volcanic gas from fumaroles were observed in synchronization with earthquake swarms and ground deformation in 2013 and 2015. Similar increases in CO₂/H₂S ratio were also observed in 2017 and 2019, although the increases in the CO₂/H₂S ratio in 2017 and 2019 were not as sharp as those in 2013 and 2015. Furthermore, the maximum values of the CO₂/H₂S ratio in 2017 and 2019 were lower than the values in 2013 and 2015. These differences in the CO₂/H₂S ratio may be related to the limited and smaller scale of volcanic activity in 2017 and 2019 relative to 2013 and 2015. Another explanation for the difference is a possible irreversible change in the underground structure of the Owakudani area, which is a geothermal area around Hakone volcano, because the phreatic eruption took place in the Owakudani area in 2015. During all four seismically active periods in 2013, 2015, 2017, and 2019, the CO₂/H₂S ratio decreased simultaneously with decreases in the number of volcanic earthquakes. The lower limit of CO₂/H₂S ratios after the peak of the CO₂/H₂S ratio time series was about 20 in all periods. This implies that subsequent unrest would not start until the CO₂/H₂S ratio drops to about 20. The CO₂/H₂S ratio might be an effective tool for forecasting activity at Hakone volcano. During the active periods in 2013, 2015, 2017, and 2019, extensions of the baseline across Hakone volcano were observed by GNSS with increases in the CO₂/H₂S ratio. A good correlation was found between the extensional velocity of the baseline length and the increasing rate of the CO₂/H₂S ratio. These variations could be examined at the early stage of unrest. The findings argue that the CO₂/H₂S ratio is a promising tool for predicting and evaluating volcanic activity at Hakone volcano.

 Steam-blast eruptions are classified into three categories: (1) hydrothermal eruption caused solely by a phase change of hydrothermal water within a hydrothermal system; (2) phreatic eruption caused by a new thermal input derived from a magma body in a sub-volcanic aquifer; and, (3) ultravulcanian eruption (gas eruption), a type of vulcanian eruption, which is caused by gas degassed from magma accumulating under a lava plug. It is proposed that these can be classified from a petrological analysis of eruption products based mainly on the authors' previous contributions. Volcanic ash from hydrothermal eruptions is characterized by abundant altered lithics. At some composite volcanoes, altered lithics exhibit a wide variety of alteration types including siliceous, advanced argillic, phyllic, and potassic alterations, which are considered to originate from alteration zones of composite volcanoes. The association of alteration zones are correlated with those around porphyry copper deposits. The products of phreatic eruptions are composed mainly of strongly acid altered rocks, but may also contain fresh volcanic rock fragments. The rocks are derived from selectively/partially altered rocks under the crater. Ultravulcanian eruptions mainly release fresh lithic fragments and may also emit sulfur compound minerals (mainly sulfate), but the products contain no alteration minerals indicating hydrothermal acid leaching.

 Since a phreatic eruption is caused by ruptures in hydrothermal systems beneath volcanoes, detecting and monitoring a hydrothermal system can play an important role in predicting such an eruption. Interferometric Synthetic Aperture Radar (InSAR), which detects ground deformations over a large area, may be a key technology for use in various fields, as shown from the exponential growth of recent studies in terms of number and quality. The present contribution reviews surface deformations caused by the hydrothermal system of Hakone volcano, as detected by InSAR before, during, and after the 2015 eruption. The opening of the NW–SE-trending crack and localized uplift in the Owakudani fumarole area were captured by InSAR analyses during the 2015 unrest at Hakone volcano. Moreover, an InSAR time series analysis showed steady subsidence on the west side of the Owakudani fumarole area. Based on models explaining these surface displacements, the shallow hydrothermal system of Hakone volcano is characterized by NW–SE to WNW–ESE-trending crack-shaped fluid supply paths and pocket-shaped fluid reservoirs. During the 2015 and previous phreatic eruptions, it is probable that fluid was supplied using the same crack-like path, implying that fluid was repeatedly supplied using the same structure. Therefore, in order to predict the occurrence of phreatic eruptions at Hakone volcano, it is necessary to monitor volcanic activity by taking into account these structures. The activity of Hakone volcano, including formations of these NW–SE to WNW–ESE-trending cracks, is dominated by a regional stress field. This stress field is caused by shear deformation due to plate motion occurring in this region; that is, the subducting Philippine Sea Plate, and the colliding Izu Peninsula.

 Hakone volcano has been in an active phase since 2001, as implied by frequent volcanic unrest every 2-5 years, with each accompanied by deep inflation (6-10 km), increase of deep low-frequency events (DLFEs) at a depth of ∼20 km, increase of CO₂/H₂S ratio in fumarole gas, and surge of volcano tectonic earthquakes (VT; < 6 km deep). A series of episodes of volcanic unrest culminated in a small phreatic eruption (erupted volume; ∼100 m³) in 2015; however, lesser unrest in terms of seismic activity occurred in 2017 and 2019. Recent studies on crustal structures based on seismic tomography indicate a magma chamber 10-20 km beneath the volcano, which might be connected to a large magma chamber beneath Fuji volcano, approximately 30 km NW of Hakone. Interestingly, the DLFEs beneath Hakone volcano seem to take place in a high attenuation zone that connects the magma chambers. Deep inflation beneath Hakone volcano, however, is clearly located at a shallower location than the magma chamber of Hakone. The increases of CO₂ and He within the fumarole of Hakone during its unrest may indicate degassing of magma at depth. The maximum fumarole temperature after the eruption and constraints on subsurface temperature (∼200°C at 400 m deep indicated by the mineral assemblage and ∼370°C at 4 km below sea level where is the lower depth limit of VT) imply a vapor-dominated hydrothermal system in the volcano from the bottom of the cap structure (∼100 m deep) to a depth of possibly 2-4 km. Such a vapor-dominated system may allow rapid transfers of magmatic gases and their emission from the fumarole area in the very early phase of volcanic unrest, as was observed. Hakone lacks long period events (LF) and volcanic tremors, which are common at many active volcanoes. Because such events are considered to be related to fluid migration, the vapor-dominated system can be attributed to their absence in Hakone. An estimation of the water mass balance implies that the amount and rate of inflation in the hydrothermal system are comparable to those emitted from the fumarole area in pre-eruptive calm periods. Thus, continuous inflation at depth can be explained by crystal depositions from the hydrothermal fluid. The high temperature of steam emitted in the fumarole area after the eruption indicates destruction of the container of the hydrothermal system, which also caused the lower VT activity and CO₂/H₂S ratio during post-eruptive unrest.