Matthew M. Haney

Constraining magma storage and transport through the entire crust at a variety of arc volcanoes is essential to understand how subduction parameters may influence these processes. However, constraints throughout the crust are often limited to volcanoes that either fortuitously have properties that illuminate the entire crustal system, or volcanoes where high‐resolution geophysical imaging has been performed. Seismic receiver functions (RFs) offer sensitivity to the whole crust independent of local seismicity distribution using a few instruments. However, near volcanoes, they are often complex, requiring detailed analysis and potentially non‐unique velocity model solutions. Here, we introduce a novel, data‐driven method to identify and map RF variability using spectral clustering and kernel logistic regression. We apply it to 16 volcanoes in the Aleutian arc, enabling systematic comparisons of RF patterns without need for detailed velocity modeling. We observe that a subset of these volcanoes has RF variability that is spatially well‐separated in the crust, while others appear to have minimal spatial organization. Based on comparisons with previous results, we find that the former tends to occur at volcanoes with relatively simple magma storage systems, and the latter at volcanoes with complex, multi‐layer storage. This suggests that our technique reveals first‐order variability in crustal magmatic architecture. Furthermore, the ability to recover spatially coherent RF patterns shows no clear correlation with along‐arc subduction parameters but instead appears associated with edifice morphology. This work establishes a systematic framework for comparing arc volcanic structure and a foundation for future RF velocity modeling, extendable to other heterogeneous settings.

Imaging magma storage beneath active calderas requires resolving crustal structure at scales comparable to those of individual reservoirs. To identify potential reservoirs beneath Kı̄lauea, Hawai'i, we conducted a hybrid seismic experiment at its summit for 37 days in 2023 that combined more than 1,800 three‐component nodal seismometers with 396 vibroseis source positions, yielding about 15 million ‐ and ‐wave arrivals. Joint inversion of earthquake and active‐source travel times yields three‐dimensional models of , , and within the upper ∼4 km beneath the summit relative to mean sea level, parameterized on a 500 m grid, within which the resolution tests demonstrate recovery of sub‐kilometer to kilometer‐scale structure. The models delineate two principal magma storage regions beneath Halema'uma'u and the south caldera, each on the order of 2 km across. Both reservoirs are expressed as localized zones of reduced and moderately reduced with locally elevated at depths of 1–3 km below sea level. The anomalies are laterally distinct at shallow levels but appear to merge within a broader zone of reduced velocities at depth. Surrounding higher‐velocity domains likely represent mechanically stronger intrusive complexes that confine magma storage. Spatial correspondence between seismic velocity anomalies and independently modeled geodetic deformation sources supports the interpretation that the imaged structures represent magma storage regions. These results demonstrate that dense nodal arrays combined with hybrid passive‐active tomography can resolve reservoir‐scale structure at active volcanoes and provide an improved framework for interpreting magma storage and transport beneath Kı̄lauea.

Large, rapid landslides are a global hazard that can occur in remote, mountainous areas. Eyewitness reports of landslides and satellite imagery can often be limited or delayed, particularly during inclement weather. However, landslide‐generated seismic and infrasound (low‐frequency atmospheric sound) waves can be remotely detected in near real‐time. This information can significantly expedite characterization and possible landslide response activities. Here, we highlight these capabilities using a 6.1 million m ³ ice–rock avalanche in Denali National Park and Preserve (Alaska). This event was detected via a landslide‐specific seismic location and volume estimation algorithm deployed in Alaska, and—notably—by standard earthquake monitoring systems. Following rapid detection of this event, we combined its seismic and infrasound data set with optical, synthetic aperture radar, and oblique aerial imagery, multitemporal digital elevation models, and a numerical flow model to reconstruct its failure timeline and dynamics. We apply array processing to infrasound signals traveling >250 km and find that two precursory events occurred minutes prior to the main failure. We use long‐period seismic signals to infer the force exerted by the landslide on the Earth and constrain the rheological parameters of our numerical flow simulation with this result and deposit morphology. The main failure produced a steeply dipping impulsive initial downward force and reached speeds exceeding 60 m/s. This impulsive force generated relatively strong seismic body waves, which contributed to the earthquake system detection. This large, remote Alaska landslide underscores the key value of seismic and infrasound analysis for rapid landslide assessment and motivates efforts to further operationalize these approaches.

Understanding how pre-eruptive processes manifest in geophysical observables remains a central challenge in volcanology and volcanic hazard assessment. Among these observables, seismic tremor, a persistent ground vibration commonly recorded at active volcanoes, holds strong potential for eruption forecasting, yet its temporal evolution is notoriously difficult to interpret. Bridging tremor observations with eruption forecasting therefore requires computational frameworks that explicitly link tremor characteristics to the degree of volcanic unrest and the likelihood of eruption. Here, we present two complementary computational frameworks for eruption forecasting from continuous seismic tremor data that integrate physics-based forward modeling, inverse methods, and machine learning. Both approaches are tested using the 13 paroxysms of Shishaldin Volcano (Alaska) that occurred between July and November 2023. The first framework is physics-informed and relies on data assimilation to invert tremor observations and retrieve subsurface pressure evolution. It couples a physical model of tremor generation, rooted in multiphase gas accumulation and porous-media flow within the upper conduit, with genetic algorithm optimization and Monte Carlo simulations. This approach captures the effects of magma ascent, volatile exsolution, partial conduit sealing, and gas transport on transient tremor signals, revealing pressure increases of several MPa and a systematic rise in eruption probability hours before each paroxysm. The second framework is data-driven and applies pattern-recognition techniques to extract physically motivated seismic features (e.g., dominant frequency, amplitude, kurtosis, entropy), which are combined with a supervised machine-learning classifier (random forest) to estimate eruption probabilities. Despite their differing philosophies, both frameworks consistently relate pre-eruptive tremor evolution to probabilistic eruption forecasts. Together, these results demonstrate how computational approaches can enhance the interpretation of seismic tremor, provide quantitative insight into magma–volatile interactions, and advance eruption forecasting and volcanic hazard assessment strategies.

Because of similarities between locating an earthquake with seismic stations and locating a Global Positioning System (GPS) receiver from satellites, the Bancroft algorithm developed for GPS processing can be used to locate earthquakes. Such an approach to earthquake location differs from the conventional method of choosing an initial or trial solution and then iteratively improving the solution until convergence. The Bancroft algorithm has the advantage of being a direct, noniterative solution but with the disadvantage of only being able to accommodate a homogeneous velocity model. An additional limitation of the standard Bancroft algorithm is that it considers arrival times in a medium with a single propagation velocity. This poses no problem for GPS processing because electromagnetic waves travel at the speed of light; however, for seismic waves it means the algorithm can be applied to collections of either P- or S-wave arrival times. Here, I show how the Bancroft algorithm can be generalized to handle both P- and S-wave arrival-time measurements simultaneously. I also show how to accommodate depth-varying P- and S-wave velocity models. I apply the generalized Bancroft algorithm to microearthquakes beneath Tanaga Volcano in Alaska and compare standard locations from the widely used HYPOINVERSE location code to Bancroft locations and to the output of HYPOINVERSE when setting the trial location to the Bancroft location. I find the Bancroft locations outperform the results from the other methods for shallow earthquakes near sea level, where a quantity known as the geometric dilution of precision is large and linearized approaches such as HYPOINVERSE are expected to struggle.

From July to October 2022, a noneruptive volcanic earthquake swarm occurred within ∼15 km of Ta‘ū Island, located in eastern American Samoa. Felt reports from residents were the only available information about the swarm when it started, as American Samoa lacked a seismic monitoring network. We developed a consistent single-station catalog for the entire swarm, using seismic data from the nearest station IU.AFI, ∼250 km away. We applied the EQTransformer deep-learning model (Mousavi et al., 2020), automatically picking Pn and Sn arrivals on IU.AFI continuous data. We retained only events with Sn–Pn times of 22.5–25 s, consistent with the expected locations based on felt reports, then detected smaller swarm events with subsequent template-matching. This single-station catalog characterized the swarm’s onset and escalation to peak activity before a multiagency field response team installed a local seismic network in mid-August 2022. This permanent seismic network captured the swarm’s decline. EQTransformer identified short S–P times on the first two locally deployed seismometers, both Raspberry Shake sensors, to constrain the swarm’s distance from Ta‘ū Island. Modern seismological processing methods, combined with basic observations such as felt reports, can quickly contribute useful information during an earthquake response in a poorly monitored region.

In 2022, the Alaska Volcano Observatory responded to eruptions, volcanic unrest or suspected unrest, increased seismicity, and other significant activity at 11 volcanic centers in Alaska and in the Northern Mariana Islands. Eruptive activity in Alaska consisted of repeated small, ash-producing, phreatomagmatic explosions from Mount Young on Semisopochnoi Island; the eruption of a thick lava flow within the summit crater at Great Sitkin Volcano; and weak explosive activity and the eruption of small, channelized flows at Pavlof Volcano. Uplift and an increase in seismicity were detected at Mount Edgecumbe, a long-dormant volcano in southeastern Alaska. Anomalous seismicity was also detected at three other volcanoes, including Trident Volcano, Takawangha volcano, and Davidof volcano. Other activity documented in 2022 includes ash resuspension events at Mount Katmai and Aniakchak Crater, and Mount Cleveland had a period of unrest, but no eruptive activity took place. In the Commonwealth of the Northern Marianas Islands, hydroacoustic detections and a submarine plume observed in satellite data at Ahyi seamount indicated underwater eruptive activity there.

From the Andes to the Aleutian Islands, the Americas are rich with volcanism that spans a diverse range of tectonic settings, eruptive styles, levels of