A method includes receiving, via a processor, a first seismic dataset generated using a first type of survey system. The method further includes receiving, via the processor, a second seismic dataset generated using a second type of survey system. The method additionally includes determining a frequency band in which to combine the first seismic dataset with the second seismic dataset to generate a combined dataset and generating a seismic image based upon the combined dataset, wherein the seismic image represents hydrocarbons in a subsurface region of the Earth or subsurface drilling hazards.

A device may include a processor that may recover the signals misallocated in the deblending process of seismic data acquired with simultaneous sources. The processor may update the primary signal estimate based at least in part on a separation operation that separates coherence signals from noise signals in an output associated with the residual determined to be remaining energy for separation. The processor may be incorporated into the iterative primary signal estimate of the deblending process or be applied towards preexisting deblending output. In response to satisfying an end condition, the processor may transmit a deblended output that includes the weak coherence signals recovered from the misallocation or error in the primary signal estimate. The processor may also transmit the deblended output for use in generating a seismic image. The seismic image may represent hydrocarbons in a subsurface region of Earth or subsurface drilling hazards.

A method for improving a backpropagation-enabled process for identifying subsurface features from seismic data involves a model that has been trained with an initial set of training data. A target data set is used to compute a set of initial inferences on the target data set that are combined with the initial training data to define updated training data. The model is trained with the updated training data. Updated inferences on the target data set are then computed. A set of further-updated training data is defined by combining at least a portion of the initial set of training data and at least a portion of the target data and associated updated inferences. The set of further-updated training data is used to train the model. Further-updated inferences on the target data set are then computed and used to identify the occurrence of a user-selected subsurface feature in the target data set.

A system, method and program product for implementing a machine learning platform that processes a data map having feature and operational information. A system is disclosed that includes an interpretable machine learning model that generates a function in response to an inputted data map, wherein the data map includes feature data and operational data over a region of interest, and wherein the function relates a set of predictive variables to one or more response variables; an integration/interpolation system that generates the data map from a set of disparate data sources; and an analysis system that evaluates the function to predict outcomes at unique points in the region of interest.

The present disclosure includes a method for combining controlled and uncontrolled seismic data. The method includes accessing one or more controlled signals, each controlled signal associated with a respective receiver of a plurality of receivers. The method also includes accessing one or more uncontrolled signals, each uncontrolled signal associated with a respective receiver of the plurality of receivers. The method also includes generating one or more reconstructed signals based on the one or more uncontrolled signals. The method also includes generating a composite image based at least on the one or more controlled signals and the one or more reconstructed signals. The present disclosure may also include associated systems and apparatuses.

Inversion with exponentially encoded seismic data can include exponentially encoding acquired seismic data and associated synthetic seismic data, storing the exponentially encoded acquired seismic data and the exponentially encoded associated synthetic seismic data, determining a one-dimensional (1D) Wasserstein distance between the exponentially encoded acquired seismic data and the exponentially encoded associated synthetic seismic data, and generating an adjoint source based on the 1D Wasserstein distance. The example method also includes adapting a dynamic weight implementation of a sensitivity kernel to the adjoint source to build a gradient associated with the acquired seismic data and the associated synthetic seismic data, and iteratively inverting a waveform associated with the exponentially encoded acquired seismic data and the exponentially encoded associated synthetic seismic data based on the gradient. An image of a subsurface location can be generated based on results of the iterative inversions.

A process for identifying natural fracture sweet spots in a hydrocarbon reservoir by integrating reservoir modeling components and reservoir dynamic data. A fracture density index (FDI) is determined using natural fracture predictions from geomechanics and identified fluid flow paths. Natural fracture sweet spots may be identified from the FDI and additional inputs such as a reservoir matrix model and dynamic reservoir properties. Systems and computer-readable media for identifying natural fracture sweet spots are also provided.

A computer-implemented method is described for seismic facies identification including receiving a seismic dataset representative of a subsurface volume of interest; applying a model conditioned by petrophysical classifications to the seismic dataset to identify seismic facies and generate a classified seismic image; and identifying geologic features based on the classified seismic image. The method generates seismic facies probability volumes.

A method for improving a backpropagation-enabled process for identifying subsurface features from seismic data involves a model that has been trained with an initial set of training data. A target data set is used to compute a set of initial inferences on the target data set that are combined with the initial training data to define updated training data. The model is trained with the updated training data. Updated inferences on the target data set are then computed. A set of further-updated training data is defined by combining at least a portion of the initial set of training data and at least a portion of the target data and associated updated inferences. The set of further-updated training data is used to train the model. Further-updated inferences on the target data set are then computed and used to identify the occurrence of a user-selected subsurface feature in the target data set.

A method is described for property estimation including receiving a seismic dataset representative of a subsurface volume of interest and a well log from a well location within the subsurface volume of interest; identifying seismic traces in the seismic dataset that correspond to the well location to obtain a subset of seismic traces; windowing the subset of seismic traces and the well log to generate windowed seismic traces and a windowed well log; multiplying the windowed seismic traces and the windowed well log by a random matrix to generate a plurality of training datasets; and training a neural network using the plurality of training datasets. The method may be executed by a computer system.