Oliver Higgins – Trinity College Dublin
Linking the diversity of magma storage conditions with the eruptions they eventually produce for improving volcano monitoring
A panoply of processes operates in the magma plumbing systems underlying active volcanoes: new melt from the mantle is injected into the crust; various crystals form, grow, and separate; melts and minerals mix and mingle to form magmatic minestrones erupted to the surface.
Igneous petrologists are tasked with picking apart the pre-eruptive history of magmas. Whilst undoubtedly complex, this can be boiled down to three simple variables which chronicle their life before emergence to the surface: pressure (P), describing the depth magma resided; temperature (T), recording how hot the magma was; composition (X), documenting the melt chemistry from which various crystals grew. These three variables may interact in complex ways to control the style, duration, and magnitude of eruptions.
Part of my PhD research, undertaken at the University of Geneva alongside Professors Luca Caricchi and Tom Sheldrake, aimed to link the PTX of magma to the eruptive dynamics at the surface. This information has been quantified for many eruptions globally. However, PTX data become extremely powerful when recovered for consecutive eruptions from the same volcano, an approach which has been performed much less frequently. By looking at packages of eruptions, patterns or recurrent behaviours may begin to emerge. This may allow us to narrow down a range of future eruption scenarios. Of course, this information needs to be combined with other monitoring methods associated with restless volcanoes to build a holistic picture. Petrology alone cannot provide all the answers, but it can be a quantitative and unique complement to other fields.
We selected Mount Liamuiga volcano on Saint Kitts as our field area, which forms part of the active Lesser Antilles island arc. Aside from the friendly people, numerous beach outcrops and relaxed atmosphere, Caribbean volcanoes have been central to a huge, multi-faceted research effort in the last few decades. This makes placing results into a regional context and combining with existing data a straightforward task. Strata (packages of sequential eruptions) on Saint Kitts were fresh and easy to access and through digging trenches we were able to access the freshest sequences. We sampled part of the infamous Mansion Series stratigraphy, which represents a series of eruptions spanning decades to millennia (Fig. 1).
After running through the playbook of classical geochemical and textural analyses, we began to think about how best to quantify magma storage. Typically, this is achieved with thermobarometry (PT) and chemometry (X). In a nutshell, the compositions of minerals are passed through equations which relate mineral compositions from experiments to the PTX at which the experiments were performed. However, we encountered a series of problems. Firstly, many of these equations also require the composition of the glass (liquid surrounding the crystals) which was difficult to reliably measure due to high microlite contents. Secondly, crystal fracturing and disaggregation during magma assembly and ascent precluded the simple identification of “touching pairs” for mineral-mineral thermobarometry equations (Fig. 2). With this in mind, we calibrated a suite of new single-phase clinopyroxene and amphibole thermobarometers and chemometers to assess our samples using machine learning.
The machine learning approach chosen was random forest which takes an input dataset of experimental mineral compositions and designs several hundred uncorrelated decision trees (Fig. 3). The tree structures are designed to best distinguish PTX by mineral composition (e.g., temperature based on amphibole composition) and are saved for prediction of natural data. Prediction is simply achieved by passing natural mineral chemistry through the decision trees to recover PTX values at the base. The reason that I enjoy this approach is that the random forest is acting a little bit like a giant encyclopaedia of experiments. The algorithm is just comparing your natural mineral composition with that of the experiments. Where the algorithm cannot decide between the degree of similarity of several experiments it recovers the average (or median) of possible experiments which are inseparably similar to the natural mineral data. Many experienced petrologists already use this approach in semi-quantitative form, with their brains replacing the algorithm. Machine learning may suffer from experimental gaps but deals with this in a quasi-human way by hedging its bets between experiments that it cannot decide between.
Applying our calibration to the minerals of the Lower Mansion Series produced some interesting results, most notably in P and T (Fig. 4). The earliest eruption of the sequence, a thick pyroclastic flow deposit, originated from the shallow crust. We envisioned this as acting a little bit like the cork of a champagne bottle which, when removed, allowed material to stream through from greater depth. Magma became progressively hotter and more representative of the primitive magma feeding the system, with crystal chemistry also homogenising through time. Together, these data suggest that the largest, and most threatening, eruptions of Saint Kitts are stored in the shallow, cooler crust. As the sequence progresses, higher temperature, and slightly deeper, eruptions are generally sub-Plinian (lower volume, thinner deposits).
We present our methodology and results in recent papers published in Contributions to Mineralogy and Petrology, JGR Solid Earth and Bulletin of volcanology. These articles formed a collaborative effort with Luca and Tom, as well as Corin Jorgenson, Maurizio Petrelli and Florence Begue. We all enjoyed exploring a novel method with an open mind and developing our coding skills.
In general, making a link between where magma is stored and how it erupts is useful. The addition of the time dimension through our use of volcanic strata will hopefully make our conclusions yet more applicable to the volcanological community. In ongoing work, we find the relationship between magma storage and eruptive behaviour is repeatable on Saint Kitts for several strata outside of the Mansion Series.
We encourage others to focus on time-constrained mineral chemistry at other volcanoes to bolster the limited dataset. Studying the PTX of these decadal-millennial eruptions has proven an interesting part of my (now complete!) PhD research project. It remains to be seen how these recurrent magma storage and mineral chemistry behaviours manifest themselves on the scale of a single, protracted eruption. The 2021 eruption of Cumbre Vieja (La Palma), lasting a disruptive 85 days, forms the latest natural laboratory for me to test this hypothesis.
Dr. Oliver Higgins (email@example.com) is a SNSF postdoctoral research fellow at Trinity College Dublin. His PhD at the University of Geneva examined the crystal cargo from a series of volcanic eruptions from Mount Liamuiga (Saint Kitts, Lesser Antilles) spanning decades to millennia. The aim was to link the diversity of crystal chemistry and pre-eruptive magma storage to the evolving eruptive dynamics of the volcano. Now Oliver applies the same temporal-geochemical approach on the scale of a single protracted eruption (Cumbre Vieja 2021 eruption, La Palma, Canary Islands).
Would you like to be featured on our blog?
We’re on the hunt for fellow VIPS enthusiasts to share their stories, whether you’re a student, an early career researcher (ECR) or more experienced researcher – we want to hear from you!
Do you have a new paper to tell the world about? Or have you been involved in some exciting fieldwork recently? Get in touch at firstname.lastname@example.org to tell us what’s going on in your field of VIPS.