Table2_Identifying analogues for data-limited volcanoes using hierarchical clustering and expert knowledge: a case study of Melimoyu (Chile).XLSX

Determining the eruption frequency-Magnitude (f-M) relationship for data-limited volcanoes is challenging since it requires a comprehensive eruption record of the past eruptive activity. This is the case for Melimoyu, a long-dormant and data-limited volcano in the Southern Volcanic Zone (SVZ) in Chile with only two confirmed Holocene eruptions (VEI 5). To supplement the eruption records, we identified analogue volcanoes for Melimoyu (i.e., volcanoes that behave similarly and are identified through shared characteristics) using a quantitative and objective approach. Firstly, we compiled a global database containing 181 variables describing the eruptive history, tectonic setting, rock composition, and morphology of 1,428 volcanoes. This database was filtered primarily based on data availability into an input dataset comprising 37 numerical variables for 438 subduction zone volcanoes. Then, we applied Agglomerative Nesting, a bottom-up hierarchical clustering algorithm on three datasets derived from the input dataset: 1) raw data, 2) output from a Principal Component Analysis, and 3) weighted data tuned to minimise the dispersion in the absolute probability per VEI. Lastly, we identified the best set of analogues by analysing the dispersion in the absolute probability per VEI and applying a set of criteria deemed important by the local geological service, SERNAGEOMIN, and VB. Our analysis shows that the raw data generate a low dispersion and the highest number of analogues (n = 20). More than half of these analogues are in the SVZ, suggesting that the tectonic setting plays a key role in the clustering analysis. The eruption f-M relationship modelled from the analogue’s eruption data shows that if Melimoyu has an eruption, there is a 49% probability (50th percentile) of it being VEI≥4. Meanwhile, the annual absolute probability of a VEI≤1, VEI 2, VEI 3, VEI 4, and VEI≥5 eruption at Melimoyu is 4.82 × 10−4, 1.2 × 10−3, 1.45 × 10−4, 9.77 × 10−4, and 8.3 × 10−4 (50th percentile), respectively. Our work shows the importance of using numerical variables to capture the variability across volcanoes and combining quantitative approaches with expert knowledge to assess the suitability of potential analogues. Additionally, this approach allows identifying groups of analogues and can be easily applied to other cases using numerical variables from the global database. Future work will use the analogues to populate an event tree and define eruption source parameters for modelling volcanic hazards at Melimoyu.

确定数据受限火山的喷发频次-震级（f-M）关系颇具挑战，因为这类研究需要完整的过往喷发活动记录。智利南火山带（Southern Volcanic Zone, SVZ）的梅利莫尤火山（Melimoyu）便是如此：它长期处于休眠状态，属于数据受限火山，目前仅确认存在两起全新世喷发，火山爆发指数（Volcanic Explosivity Index, VEI）均为5级。为补充该火山的喷发记录，本研究采用定量且客观的方法，筛选出梅利莫尤火山的类比火山（即喷发行为相似、通过共同特征识别的火山）。首先，我们构建了一套涵盖1428座火山的全球数据库，包含喷发历史、构造背景、岩石成分与形态特征共181项变量。随后，该数据库主要依据数据可用性进行筛选，最终得到输入数据集，包含438座俯冲带火山的37项数值型变量。接着，我们针对从输入数据集衍生出的三套数据集应用凝聚嵌套（Agglomerative Nesting）自下而上层级聚类算法：1）原始数据；2）主成分分析（Principal Component Analysis, PCA）输出结果；3）经调优以最小化各VEI等级绝对概率离散度的加权数据。最后，我们通过分析各VEI等级的绝对概率离散度，并结合当地地质服务机构SERNAGEOMIN与VB认定的关键标准，筛选出最优类比火山集合。分析结果表明，原始数据聚类得到的离散度最低，且类比火山数量最多（共20座）。其中超过半数的类比火山位于南火山带（SVZ），这表明构造背景在聚类分析中发挥关键作用。基于类比火山喷发数据构建的f-M关系模型显示，若梅利莫尤火山发生喷发，其喷发指数达到VEI≥4的概率为49%（50分位数）。与此同时，梅利莫尤火山发生VEI≤1、VEI 2、VEI 3、VEI 4与VEI≥5级喷发的年绝对概率（50分位数）分别为4.82×10⁻⁴、1.2×10⁻³、1.45×10⁻⁴、9.77×10⁻⁴与8.3×10⁻⁴。本研究证实了利用数值型变量捕捉火山间差异的重要性，以及将定量方法与专家知识相结合以评估潜在类比火山适用性的价值。此外，该方法可有效识别类比火山群，且可借助全球数据库中的数值型变量轻松推广至其他研究案例。后续研究将利用这些类比火山构建事件树，并确定梅利莫尤火山火山灾害建模所需的喷发源参数。