The architecture of magma plumbing systems plays a fundamental role in volcano eruption and evolution. However, the precise configuration of crustal magma reservoirs and conduits responsible for supplying eruptions are difficult to explore across most active volcanic systems. Consequently, our understanding of their correlation with eruption dynamics is limited. Axial Seamount is an active submarine volcano located along the Juan de Fuca Ridge, with known eruptions in 1998, 2011, and 2015. Here we present high-resolution images of P-wave velocity, attenuation, and estimates of temperature and partial melt beneath the summit of Axial Seamount, derived from multi-parameter full waveform inversion of a 2D multi-channel seismic line. Multiple magma reservoirs, including a newly discovered western magma reservoir, are identified in the upper crust, with the maximum melt fraction of ~15–32% in the upper main magma reservoir (MMR) and lower fractions of 10% to 26% in other satellite reservoirs. In addition, a feeding conduit below the MMR with a melt fraction of ~4–11% and a low-velocity throat beneath the eastern caldera wall connecting the MMR roof with eruptive fissures are imaged. These findings delineate an asymmetric shallow plumbing system beneath Axial Seamount, providing insights into the magma pathways that fed recent eruptions.

Existing methods for estimating formation boundaries from well-log data only analyze the formation along the wellbore, failing to capture changes in the 3D formation structure around it. This paper presents a method for real-time 3D formation boundary interpretation using readily available well logs and seismic image data. In the proposed workflow, the mean formation boundary is estimated as a curve following the well path. 3D surfaces are then fitted through this boundary curve, aligning with the slopes and features in the seismic image data. The proposed method is tested on both synthetic and field datasets and illustrates the capabilities of accurate boundary estimation near the well path and precise representation of boundary shape changes further away from the well trajectory. With this fully automated geological interpretation workflow, human bias and interpretation uncertainty can be minimized. Subsurface conditions can be continually updated while drilling to optimize drilling decisions and further automate the geosteering process.

Full waveform inversion (FWI) can be expressed in a Bayesian framework, where the associated uncertainties are captured by the posterior probability distribution (PPD). In practice, solving Bayesian FWI with sampling-based methods such as Markov chain Monte Carlo (MCMC) is computationally demanding because of the extremely high dimensionality of the model space. To alleviate this difficulty, we develop a deep convolutional autoencoder (CAE) that serves as a learned prior for the inversion. The CAE compresses detailed subsurface velocity models into a low-dimensional latent representation, achieving more effective and geologically consistent model reduction than conventional dimension reduction approaches. The inversion procedure employs an adaptive gradient-based MCMC algorithm enhanced by automatic differentiation-based FWI to compute gradients efficiently in the latent space. In addition, we implement a transfer learning strategy through online fine-tuning during inversion, enabling the framework to adapt to velocity structures not represented in the original training set. Numerical experiments with synthetic data show that the method can reconstruct velocity models and assess uncertainty with improved efficiency compared to traditional MCMC methods.