This disclosure describes a system and method for generating a subsurface model representing lithological characteristics and attributes of the subsurface of a celestial body or planet. By automatically ingesting data from many sources, a machine learning system can infer information about the characteristics of regions of the subsurface and build a model representing the subsurface rock properties. In some cases, this can provide information about a region using inferred data, where no direct measurements have been taken. Remote sensing data, such as aerial or satellite imagery, gravimetric data, magnetic field data, electromagnetic data, and other information can be readily collected or is already available at scale. Lithological attributes and characteristics present in available geoscience data can be correlated with related remote sensing data using a machine learning model, which can then infer lithological attributes and characteristics for regions where remote sensing data is available, but geoscience data is not.

Processes and systems described herein are directed to imaging a subterranean formation from seismic data recorded in a marine survey with moving marine vibrators. The marine vibrators generate random sweeps with random sweep signatures. Processes and systems generate an up-going pressure wavefield from measured pressure and vertical velocity wavefield data recorded in the marine survey and obtain a downgoing vertical acceleration wavefield that records source wavefields, directivity, source ghosts, and random signatures of the random sweeps. The downgoing vertical acceleration wavefield data is deconvolved from the up-going pressure wavefield to obtain a subsurface reflectivity wavefield that is used to generate an image of the subterranean formation with reduced contamination from source wavefields, directivity, source ghosts, and random signatures of the random sweeps.

A method for waveform inversion, the method including receiving observed data d, wherein the observed data d is recorded with sensors and is indicative of a subsurface of the earth; calculating estimated data p, based on a model m of the subsurface; calculating, using a trained neural network, a misfit function J.sub.ML; and calculating an updated model m.sub.t+1 of the subsurface, based on an application of the misfit function J.sub.ML to the observed data d and the estimated data p.

A media Parameter-modified method for realizing an adaptive expression of an arbitrary discontinuous surface, comprising the following steps: importing an initial forward model, importing anisotropic parameters; and setting a space step and a time step according to the initial forward model parameters; and then starting a stepped discretization of a free surface of the initial forward model; and using a corrected constitutive relationship to correct a first level parameter of the initial forward model; and bringing the corrected constitutive relationship into a displacement stress equation, and the influence of the free surface can be introduced in the case of the anisotropic media after series of operation. The present disclosure can make an accurate numerical simulation of a wave field near the discontinuous surface, and the accurate numerical simulation will contribute to the extraction and analysis of information from the seismic data.

The present invention relates to a method for determination of real subsoil composition or structure characterized in that the method comprises: —receiving a model representing the real subsoil, said model comprising at least one parametric volume describing a geological formation in said model, said volume having a plurality of cells; —for each cell in the plurality of cells, determining a quality index (QI.sub.cell) function of a respective position of the cell in the geological formation; —receiving a set of facies, each facies in said set being associated with a proportion and a quality index ordering in said formation; —associating a facies to each cell, said association comprising: /a/ selecting a cell with a lowest quality index within cells in the plurality of cells having no facies associated to; /b/ associating, to said cell, a facies with a lowest Quality index ordering within facies of the set of facies for which the respective proportion is not reached in the formation; /c/ reiterating steps /a/ to /c/ until all cells in the plurality of cells are associated with a facies.

A method for training a backpropagation-enabled regression process is used for predicting values of an attribute of subsurface data. A multi-dimensional seismic data set with an input dimension of at least two is inputted into a backpropagation-enabled process. A predicted value of the attribute has a prediction dimension of at least 1 and is at least 1 dimension less than the input dimension.

Aspects of the technology described herein identify geologic features within seismic data using modern computer analysis. An initial step is the development of training data for the machine classifier. The training data comprises an image of seismic data paired with a label identifying points of interest that the classifier should identify within raw data. Once the training data is generated, a classifier can be trained to identify areas of interest in unlabeled seismic images. The classifier can take the form of a deep neural network, such as a U-net. Aspects of the technology described herein utilize a deep neural network architecture that is optimized to detect broad and flat features in seismic images that may go undetected by typical neural networks in use. The architecture can include a group of layers that perform aspect ratio compression and simultaneous comparison of images across multiple aspect ratio scales.

An ensemble of machine learning models is trained to evaluate seismic and risk-related data in order to evaluate, value, or otherwise rank various prospective hydrocarbon reservoir (“prospects”) of a field. A classification machine learning model is trained to classify a prospect or region of a prospect based on the exploration risk level. From the seismic data, a frequency-filtered volume (FFV) for each prospect is calculated, where the FFV is a measure of reservoir volume which takes into account seismic resolution limits. Based on the risk classification and FFV, prospects of the field are ranked based on their economic value which is a combination of the risk associated with drilling and their potential reservoir volume.

Accordingly, there are disclosed herein geologic modeling methods and systems employing reference-based inversion of seismic image volumes. An illustrative method embodiment includes: (a) obtaining a measured seismic image volume; (b) determining a reference seismic image volume based on a reference model; (c) deriving a synthesized seismic image volume from a geologic model; (d) detecting at least one geologic model region where the synthesized seismic image volume and the measured seismic image volume are mismatched; (e) finding a reference model region where the reference seismic image volume best matches the measured seismic image volume; (f) replacing content of the at least one geologic model region with content of the reference model region to obtain an improved geologic model; and (g) outputting the improved geologic model.

Systems and methods for picking seismic stacking velocity based on structures in a subterranean formation include: receiving seismic data representing a subterranean formation; generating semblance spectrums from the seismic data representing the subterranean formation; smoothing the semblance spectrums; and picking stacking velocities based on the smoothed semblance spectrums.