The widest part of the Andean orogen is between 15° and 27°S (Figure 1), where the subduction angle is 20°-30°, flanked southwards and northwards by the flat subduction segments, where the subducted Nazca plate flattens out to become nearly horizontal. The Altiplano and Puna plateaus together constitute the second largest high plateau in the world, the Central Andean Plateau (Figure 1), which is also the only one that formed under a subduction regime. The Altiplano plateau (AP), in the northern part of the Central Andean Plateau, is characterized by a single internally drained basin with an average rather uniform elevation around 3,800 m, whereas the southern part of the Central Andean Plateau is the Puna plateau (PN), which exhibits a higher altitude around 4,500 m with more rugged relief, enclosing a series of internal drained basins. The Central Andean Plateau is flanked to the west by the Western Cordillera (WC) and to

SUMMARY We present an accelerated full-waveform inversion based on dynamic mini-batch optimization, which naturally exploits redundancies in observed data from different sources. The method rests on the selection of quasi-random subsets (mini-batches) of sources, used to approximate the misfit and the gradient of the complete data set. The size of the mini-batch is dynamically controlled by the desired quality of the gradient approximation. Within each mini-batch, redundancy is minimized by selecting sources with the largest angular differences between their respective gradients, and spatial coverage is maximized by selecting candidate events with Mitchell’s best-candidate algorithm. Information from sources not included in a specific mini-batch is incorporated into each gradient calculation through a quasi-Newton approximation of the Hessian, and a consistent misfit measure is achieved through the inclusion of a control group of sources. By design, the dynamic mini-batch approach has several main advantages: (1) The use of mini-batches with adaptive size ensures that an optimally small number of sources is used in each iteration, thus potentially leading to significant computational savings; (2) curvature information is accumulated and exploited during the inversion, using a randomized quasi-Newton method; (3) new data can be incorporated without the need to re-invert the complete data set, thereby enabling an evolutionary mode of full-waveform inversion. We illustrate our method using synthetic and real-data inversions for upper-mantle structure beneath the African Plate. In these specific examples, the dynamic mini-batch approach requires around 20 per cent of the computational resources in order to achieve data and model misfits that are comparable to those achieved by a standard full-waveform inversion where all sources are used in each iteration.

SUMMARY We present a novel full-waveform inversion (FWI) approach which can reduce the computational cost by up to an order of magnitude compared to conventional approaches, provided that variations in medium properties are sufficiently smooth. Our method is based on the usage of wavefield adapted meshes which accelerate the forward and adjoint wavefield simulations. By adapting the mesh to the expected complexity and smoothness of the wavefield, the number of elements needed to discretize the wave equation can be greatly reduced. This leads to spectral-element meshes which are optimally tailored to source locations and medium complexity. We demonstrate a workflow which opens up the possibility to use these meshes in FWI and show the computational advantages of the approach. We provide examples in 2-D and 3-D to illustrate the concept, describe how the new workflow deviates from the standard FWI workflow, and explain the additional steps in detail.

Recent methodological advances and increases in computational power have made it feasible to perform full-waveform inversions (FWI) of large domains while using more sources. This trend, along with the increasing availability of seismic data has led to an explosion of the data volumes that can, and should, be used within an inversion. Similar to machine learning problems, the incorporation of more data can result in more robust and higher quality models. In this contribution, we present the new version of LASIF, an open-source LArge-scale Seismic Inversion Framework, which helps to automate many of the historically labor-intensive tasks that were bottlenecks in earlier FWI workflows and prevented the use of the larger datasets. Among other things, the framework automates data selection, data acquisition from public web services, and data processing. It also defines an inversion project structure that organizes the data and documents the progress of the inversion. The code is open-source and available on Github. Features are available through a graphical user interface (GUI), a command-line interface (CLI), and an application programming interface (API). While we will show examples for use of LASIF with the Salvus wave equation solver, the API makes it possible to use the features of LASIF for any type of wave equation solver as long as the LASIF file formats are adhered to.

We present the first‐generation full‐waveform tomographic model (SinoScope 1.0) for the crust‐mantle structure beneath China and adjacent regions. The three‐component seismograms from 410 earthquakes recorded at 2,427 stations are employed in iterative gradient‐based inversions for three successively broadened period bands of 70–120 s, 50–120 s, and 30–120 s. Synthetic seismograms were computed using GPU‐accelerated spectral‐element simulations of seismic wave propagation in 3‐D anelastic models, and Fréchet derivatives were calculated based on an adjoint‐state method facilitated by a checkpointing algorithm. The inversion involved 352 iterations, which required 18,600 wavefield simulations. SinoScope 1.0 is described in terms of isotropic P‐wave (VP), horizontally and vertically polarized S‐wave velocities (VSH and VSV), and mass density (ρ), which are independently constrained with the same data set coupled with a stochastic L‐BFGS quasi‐Newton optimization scheme. It systematically reduced differences between observed and synthetic full‐length seismograms. We performed a detailed resolution analysis by repairing input random parametric perturbations, indicating that resolution lengths can approach the half propagated wavelength within the well‐covered areas. SinoScope 1.0 reveals strong lateral heterogeneities in the lithosphere, and features correlate well with geological observations, such as sedimentary basins, Holocene volcanoes, Tibetan Plateau, Philippine Sea Plate, and various tectonic units. The asthenosphere lies below the lithosphere beneath East and Southeast Asia, bounded by subduction trenches and cratonic blocks. Furthermore, we observe an enhanced image of well‐known slabs along strongly curved subduction zones, which do not exist in the initial model.