Geophysical Journal International

SummaryInterpolating scatter data obtained from discrete observations is essential for the continuous representation of subsurface media. Traditional interpolation algorithms typically rely on weighting the relationship between interpolation points and nearby known points, which makes it more difficult to incorporate multi-source data and prior constraints as the amount of information increases. This study explores the use of deep neural networks to replace traditional interpolants, constructing a deep learning-based scatter interpolation workflow that integrates prior information through isotropic or anisotropic smoothness loss functions, addressing traditional methods’ limitations. To enhance the capability of the deep neural network for sparse scatter interpolation, we synthesized a large number of scatter-velocity model pairs to pre-train it using supervised learning. The pre-trained network is further adapted to specific interpolation tasks by physics-guided unsupervised fine-tuning to achieve stable interpolation results. Due to the flexibility of incorporating multi-source information through input or supervised loss and imposing constraints of geophysical laws through unsupervised loss, our DL-based interpolation can be easily extended to solve geophysical inversion problems that jointly fits both data and geophysical laws. Our experiments validate the effectiveness of this workflow and demonstrate its potential in multi-information-constrained geophysical scatter interpolation, which forms the basis for multi-information inversion. This work not only advances machine learning algorithms for geophysical scatter interpolation but also provides valuable insights for deep learning geophysical inversion involving multiple data sources, and physical laws.

SummaryWe revisit the budget of 20th century true polar wander (∼1°/Myr in the direction of 70°W) using a state-of-the-art adjoint-based reconstruction of mantle convective flow and predictions of ongoing glacial isostatic adjustment that adopt two independent models of Pleistocene ice history. Both calculations are based on a mantle viscosity profile that simultaneously fits a suite of data sets related to glacial isostatic adjustment (Fennoscandian Relaxation Spectrum, post-glacial decay times) and a set of present-day observations associated with mantle convection (long-wavelength gravity-anomalies, plate motions, excess ellipticity of the core-mantle boundary). Our predictions reconcile both the magnitude and direction of the observed true polar wander rate, with convection and glacial isostatic adjustment contributing signals that are 25-30% and ∼75% of the observed rate, respectively. The former assumes that large-scale seismic velocity heterogeneities are purely thermal in origin, and we argue that our estimate of the convection signal likely represents an upper bound due to the neglect of hypothesized compositional variations within the large low shear velocity provinces in the deep mantle.

AbstractThe earthquake size distribution is well described by the Gutenberg Richter Law, controlled by the b-value parameter. In recent decades, a great variety of methods for estimating the b-value have been proposed by the scientific community, despite the simplicity of this relationship. All these methods underlie the different views of individual modelers and, therefore, often generate inconsistent results. In this study, we perform a seismological experiment in which we compare different, commonly adopted, methodologies, to estimate the completeness magnitude and the b-value, for seismicity in Central Italy. The inter-method differences are on average equal to 0.4 and 0.3, for Mc and b, respectively, but reach much larger values, especially during more intense seismic activity. This shows that epistemic uncertainty in the b-value plays a more crucial role than intra-method uncertainties, opening new perspectives in the interpretation of discrepant, single studies.