[S-VC32] Mitigation of volcanic disasters - basic and applied researches

[SVC32-02] Probabilistic eruption forecasting based on statistical analysis for ground deformation data of Vulcanian explosion at Sakurajima volcano

*Kyoka Ishii¹, Masato Iguchi¹ (1.Disaster Prevention Research Institute, Kyoto University)

Keywords:Probabilistic eruption forecasting, Sakurajima volcano, Vulcanian explosions, Ground deformation

Assessment of probability in the size and timing of eruptions is important for the evaluation of ash fall hazards. By combining the probability of eruption size and timing with ash fall amount and area predicted by numerical simulations, a quantitative assessment can be made which could include the uncertainties of eruptions. Sakurajima volcano is characterized by frequent Vulcanian explosions more than several hundred times a year, so ash fall is the most familiar hazard to nearby residents. Such frequent eruptions can provide a huge database of eruptions and enable the evaluation of eruption probability using statistical approaches. Therefore, this study aims to obtain the probability distribution of eruption size and timing based on the eruption database of Sakurajima and to perform short-term prediction of eruptions. In Sakurajima volcano, inflation and deflation of magma chamber are typically observed before and after the occurrence of a Vulcanian explosion, respectively. Therefore, we investigated the statistical features of ground deformations due to explosive events that occurred at the Showa and Minami-dake craters in 2009-2020. First, we focused on duration of inflation, inflation volume, deflation volume, and deflation rate (= deflation volume / inflation volume), and made databases of these four indicators. We used strain data (radial and tangential components) recorded at an underground observation tunnel located 2.4 km from the summit crater. We assumed a single spherical pressure source beneath the crater and estimated the volume change at the pressure source using the ratio of the two strain components. Next, we estimated model distributions for the four indicators. The candidate models are Log-logistic, Weibull, and exponential distributions. The model parameters for each distribution were constrained by maximum likelihood estimation. As a result, Log-logistic distribution was found to be the best model for all indicators for both craters. Log-logistic distribution is one of the common models for the repose interval of Vulcanian explosions, which suggests the existence of competing processes such as increase and decrease in pressure (Connor et al., 2003; Watt et al., 2004). Therefore, the inflation and deflation process accompanied by Vulcanian explosions at Sakurajima volcano could be controlled by a balance between pressurization and depressurization in the conduit. Moreover, we estimated the temporal change in the best model and found that Log-logistic distribution was the best model for most of the analysis period. Model parameter k of Log-logistic distribution, which represents the regularity of deformation, temporally changed. In particular, k showed high values from the end of 2011 to the beginning of 2012 and from the end of 2019 to the beginning of 2020, suggesting that the ground deformations in these periods were more regular than in other periods. These periods when k is high correspond to the magma intrusion period implied by long-term deformation (Iguchi, et al., 2019; Iguchi et al., 2022). It suggests that a large magma supply would realize the recurrent explosions. In order to obtain more realistic eruption probabilities, it may be useful to change the model parameter of the probability function according to long-term deformation and/or the inflation rate available in real time before the eruption.