Building PSDM velocity models in complex structure land environments is difficult. A machine-learning method using a convolutional neural network (CNN) incorporates both human and artificial intelligence to overcome these difficulties. The supervised learning process used a large representative dataset to train the CNN, learning the convolutional weights that best map the input seismic shot records to the target velocity model. With careful consideration of both training data and network architecture, the CNN can accurately predict velocity models on both synthetic and field data.

Ioannina Block is a fold-and-thrust belt located onshore NW Greece that contains fault systems and complex deformation making seismic imaging extremely difficult. Time and depth processing of ten 2D seismic lines totaling 400 km was undertaken by a team of Geoscientists from Energean and Thrust Belt Imaging. The strategy for successful seismic imaging relied on a collaborative approach integrating a seismic processing workflow that tackled each of the key processing challenges with the structural knowledge and expertise of Energean Geoscientists.

This presentation focusses on three key steps, Time Processing, Anisotropic Depth Imaging and Anisotropic Depth Imaging with Model Moveout.

Time Processing corrects for near-surface weathering effects through statics corrections and is the starting point for depth velocity model building.

Depth Imaging allows for the seismic signal to be observed in true depth from surface. To achieve this, an accurate rock velocity model is required along with the parameterization to correct for anisotropy. Building the depth velocity model and collaboration with the Geoscience team is discussed.

Depth Processing with Model Moveout was used to further improve the final image.
In place of the time statics the weathering layers are dealt with by first superimposing the Near Surface Velocity Model onto the depth velocity model and recalculating the residual statics using data that has been flattened with Model Moveout, a more accurate method than the standard approach of NMO.

A collaborative approach between TBI and Energean Geoscientists combined a careful analytical data driven approach with a structural, geologic perspective resulting in improved seismic imaging as compared to previous efforts.

In complex-structure environments like the Andes mountains and foothills, outcropping rocks generate a significant velocity variation in the near surface. In time processing, we intend to correct for these near–surface fluctuations with weathering statics corrections, however, this solution assumes vertical raypaths through the weathering layer, and this assumption is violated when we have high-velocity rocks outcropping at the surface. With depth imaging, we have an opportunity to include the weathering velocities in the PSDM velocity model and raytrace through the weathering layers to get a more accurate correction for near–surface velocity variation.

But then, what do we do with the weathering correction we calculated in the time processing? The weathering velocities from the refraction tomography are now in the PSDM velocity model, but the reflection statics calculated in the time processing are coupled to the refraction statics and time-processing velocities. These statics are therefore decoupled from the PSDM velocity model. Calculating new reflection statics using the PSDM velocity model resolves this issue. We propose a method where we discard all weathering statics from the time processing and perform all near–surface velocity corrections in the depth domain—including reflection statics.

We applied this workflow to a seismic dataset from the foothills of the Colombian Andes, which shows significant imaging improvements below the mountainous areas of the seismic survey.