4:15 PM - 4:30 PM
[SVC34-10] Attempt to understand the state of volcanic activity and extract transition patterns using multivariate analysis at Sakurajima volcano, Japan
Volcanic phenomena are diverse, and the disasters associated with them are also diverse. If we assume that the volcanic phenomena depend on the state of volcanic activity, understanding the state of volcanic activity and extracting its transition patterns is important for mitigating volcanic disasters. In order to understand the volcanic phenomena, observational and analytical data have been accumulated using a variety of approaches. In this study, we apply multivariate analysis to multi-item data to attempt to understand the state of volcanic activity and extract transition patterns.
In order to perform multivariate analysis using multi-item data as input values, we aligned each data set in time, processed missing values, and then whitened the data. We performed dimensionality reduction to remove duplicates of similar data. We performed cluster analysis to separate the states of volcanic activity. We used data on Sakurajima volcano, Japan reported by the Japan Meteorological Agency from 2010 to 2023 (i.e., eruptions, seismicity, air shocks, crustal deformation, and sulfur dioxide flux).
Through dimensionality reduction, the input data was consolidated into low-dimensional parameters. By giving the number of clusters and performing cluster analysis, temporal changes in volcanic activity were obtained. In addition, by outputting the temporal relationship between volcanic activity states as a state transition diagram, the transition patterns of volcanic activity states were expressed probabilistically. By separating volcanic activity states through cluster analysis, we were able to obtain feature values for the input data for each volcanic activity state. By selecting noteworthy volcanic activity states based on the feature values for each volcanic activity state, it is possible to extract prior states that are likely to transition to the volcanic activity state of interest, which means that it may be possible to take disaster prevention measures in advance.
In order to perform multivariate analysis using multi-item data as input values, we aligned each data set in time, processed missing values, and then whitened the data. We performed dimensionality reduction to remove duplicates of similar data. We performed cluster analysis to separate the states of volcanic activity. We used data on Sakurajima volcano, Japan reported by the Japan Meteorological Agency from 2010 to 2023 (i.e., eruptions, seismicity, air shocks, crustal deformation, and sulfur dioxide flux).
Through dimensionality reduction, the input data was consolidated into low-dimensional parameters. By giving the number of clusters and performing cluster analysis, temporal changes in volcanic activity were obtained. In addition, by outputting the temporal relationship between volcanic activity states as a state transition diagram, the transition patterns of volcanic activity states were expressed probabilistically. By separating volcanic activity states through cluster analysis, we were able to obtain feature values for the input data for each volcanic activity state. By selecting noteworthy volcanic activity states based on the feature values for each volcanic activity state, it is possible to extract prior states that are likely to transition to the volcanic activity state of interest, which means that it may be possible to take disaster prevention measures in advance.