Bandwidth extension via deep neural networks trained on synthetic seismic datasets

A method, comprising:

• collecting a seismic dataset from a seismic survey conducted over a subterranean region of interest;
• generating a plurality of synthetic datasets, wherein each synthetic dataset comprises an input seismic dataset with a first bandwidth and an associated target seismic dataset with a second bandwidth, wherein the second bandwidth is broader, by extending to higher frequencies, than the first bandwidth, wherein generating a given synthetic dataset of the plurality of synthetic datasets comprises:
• obtaining, for the given synthetic dataset, synthetic data generation parameters,
• initializing a time domain as a collection of traces organized according to a simulated location and aligned along a time axis, each trace initialized with a value of zero or a random value at each time comprised by the time axis,
• determining, using the synthetic data generation parameters, a number of events to be superimposed into the time domain,
• for each event:
• determining, using the data generation parameters, a geometric shape, shape parameters, and an event location;
• and
• superimposing the event into the time domain by altering the value of each trace at each time that intersects the event based on the geometric shape, shape parameters, and event location,
• selecting, using the data generation parameters, a wavelet and wavelet parameters,
• duplicating the time domain to form a first time domain and a second time domain,
• setting the selected wavelet to have a first width and convolving the selected wavelet with the traces of the first time domain to form an input seismic dataset of the given synthetic dataset, and
• setting the selected wavelet to have a second width and convolving the selected wavelet with the traces of the second time domain to form a target seismic dataset of the given synthetic dataset,
• wherein the second width is smaller than the first width;
• splitting the plurality of synthetic datasets into a training set;
• selecting a first machine-learned model with a first architecture;
• training the first machine-learned model using the training set, wherein the first machine-learned model is trained to receive the seismic dataset and output another seismic dataset with extended bandwidth relative to the seismic dataset;
• using the first machine-learned model to produce an extended bandwidth seismic dataset from the seismic dataset;
• determining a location of a hydrocarbon reservoir in the subterranean region of interest using the extended bandwidth seismic dataset;
• planning a wellbore to penetrate the hydrocarbon reservoir based on the location, wherein the planned wellbore comprises a planned wellbore path;
• and
• drilling the wellbore guided by the planned wellbore path.