Volcanic ash provides information that can help understanding the evolution of volcanic activity during the early stages of a crisis, and possible transitions towards different eruptive styles. Ash consists of particles from a range of origins in the volcanic system and its analysis can be indicative of the processes driving activity. However, classifying ash particles into different types is not straightforward. Diagnostic observations for particle classification are not standardized and vary across samples. Here we explore the use of machine learning (ML) to improve the classification accuracy and reproducibility. We use a curated database of ash particles (VolcAshDB) to optimize and train two ML-based models: an Extreme Gradient Boosting (XGBoost) that uses the measured physical attributes of the particles, from which predictions are interpreted by the SHAP method, and a Vision Transformer (ViT) that classifies binocular, multi-focused, particle images. We find that the XGBoost has an overall classification accuracy of 0.77 (macro F1-score), and specific features of color (hue_mean) and texture (correlation) are the most discriminant between particle types. Classification using the particle images and the ViT is more accurate (macro F1-score of 0.93), with performances across eruptive styles from 0.85 in dome explosion, to 0.95 for phreatic and subplinian events. Notwithstanding the success of the classification algorithms, the used training dataset is limited in number of particles, ranges of eruptive styles, and volcanoes. Thus, the algorithms should be tested further with additional samples, and it is likely that classification for a given volcano is more accurate than between volcanoes.
Abstract Volcanic ash provides unique pieces of information that can help to understand the progress of volcanic activity at the early stages of unrest, and possible transitions towards different eruptive styles. Ash contains different types of particles that are indicative of eruptive styles and magma ascent processes. However, classifying ash particles into its main components is not straightforward. Diagnostic observations vary depending on the magma composition and the style of eruption, which leads to ambiguities in assigning a given particle to a given class. Moreover, there is no standardized methodology for particle classification, and thus different observers may infer different interpretations. To improve this situation, we created the web-based platform Volcanic Ash DataBase (VolcAshDB). The database contains > 6,300 multi-focused high-resolution images of ash particles as seen under the binocular microscope from a wide range of magma compositions and types of volcanic activity. For each particle image, we quantitatively extracted 33 features of shape, texture, and color, and petrographically classified each particle into one of the four main categories: free crystal, altered material, lithic, and juvenile. VolcAshDB ( https://volcash.wovodat.org ) is publicly available and enables users to browse, obtain visual summaries, and download the images with their corresponding labels. The classified images could be used for comparative studies and to train Machine Learning models to automatically classify particles and minimize observer biases.
Volcanoes are often monitored by geophysical and geochemical instruments that aim to track and anticipate their eruptive activity. However, assessment of the state of the volcano at any given time, and its evolution towards eruption or change in eruptive activity is notoriously difficult. Once explosive activity has begun, the study of volcanic ash can provide crucial insights on whether the activity is mainly driven by the hydrothermal system, shallow gas accumulation, or/and a new stalled intrusion close to the surface. We present the results of a study of ash componentry from part of the current volcanic crisis that started in December 2015 at Nevados de Chillán Volcanic complex (Chile) which we integrate with seismic and visual data. We identified three main stages: (i) an early one that lasted for about a year and includes two months of increased seismicity and significant amount of juvenile ash fragments, and thus suggesting some explosions were fed by a shallow magma intrusion. (ii) A second one which lasted for about six months with cycles of quiescence and explosions, and a predominance of lithic particles in the ash, suggesting that the explosions were probably driven by the dynamics in the upper part of the system, including shallow gas accumulation or/and the hydrothermal system, rather than by fresh magma intrusion. (iii) Finally, after about two years of unrest and intermittent explosions, seismicity increased again and the ash became dominated by juvenile particles, and led to the extrusion of a dome. The timing and sequence of events that we report is broadly similar to other volcanoes that have produced dome eruptions such as Soufriere Hills (Montserrat), Unzen (Japan) and Sinabung (Indonesia). Our study highlights the usefulness of integration of volcanic ash studies with other monitoring data and importance of integration of many case studies to gain a more comprehensive understanding of the processes and evolution of dome-forming eruptions.
Abstract Volcanic ash provides information that can help understanding the evolution of volcanic activity during the early stages of a crisis and possible transitions toward different eruptive styles. Ash consists of particles from a range of origins within the volcanic system and its analysis can be indicative of the processes driving the eruptive activity. However, classifying ash particles into different types is not straightforward. Diagnostic observations for particle classification are not standardized and vary across samples. Here we explore the use of machine learning (ML) to improve the classification accuracy and reproducibility. We use a curated database of ash particles (VolcAshDB) to optimize and train two ML‐based models: Extreme Gradient Boosting (XGBoost) that uses the measured physical attributes of the particles, from which predictions are interpreted by the SHapley Additive exPlanations (SHAP) method, and a Vision Transformer (ViT) that classifies binocular, multi‐focused, particle images. We find that the XGBoost has an overall classification accuracy of 0.77 ( macro F1‐score ), and specific features of color ( hue_mean ) and texture ( correlation ) are the most discriminant between particle types. Classification using the particle images and the ViT is more accurate ( macro F1‐score of 0.93), with performances varying from 0.85 for samples of dome explosions, to 0.95 for phreatic and subplinian events. Notwithstanding the success of the classification algorithms, the training dataset is limited in number of particles, ranges of eruptive styles, and volcanoes. Thus, the algorithms should be tested further with additional samples, and it is likely that classification for a given volcano is more accurate than between volcanoes.
Volcanic ash provides unique pieces of information that can help understand the progress of volcanic activity at the early stages of unrest and possible transitions towards different eruptive styles. Ash contains different types of particles that are indicative of eruptive styles and magma ascent-related processes. However, classifying ash particles into its main components is not straightforward. Diagnostic observations vary depending on the magma composition and the style of eruption, which leads to ambiguities in assigning a given particle to a given class. Moreover, there is no standardized methodology for particle classification, and thus different observers may infer different interpretations. In order to help improving this situation, we created the web-based platform Volcanic ash DataBase (VolcashDB). The database contains > 6,300 multi-focused high-resolution images of ash particles as seen under the binocular microscope from a wide range of magma compositions and eruptive styles. We quantitatively extracted multiple features of shape, texture, and color in each particle image, and petrologically classified each particle into one of the four main categories: free crystal, altered material, lithic, and juvenile. VolcashDB is publicly available and enables users to browse, obtain visual summaries, and download the images with their corresponding labels, and thus could be used for comparative studies. The classified images could also be used to train Machine Learning models to automatically classify particles and minimize observer biases.
'/%3e%3cuse xlink:href='%23a' transform='scale(-1 1)'/%3e%3c/g%3e%3cuse xlink:href='%23b' transform='rotate(72)'/%3e%3c/g%3e%3cuse xlink:href='%23b' transform='rotate(-72)'/%3e%3cuse xlink:href='%23c' transform='rotate(144)'/%3e%3c/svg%3e)
'/%3e%3cuse xlink:href='%23a' transform='translate(288.889 -214.211)'/%3e%3cuse xlink:href='%23a' transform='translate(108 342.749)'/%3e%3cuse xlink:href='%23a' transform='translate(469.189 342.749)'/%3e%3c/svg%3e)
