Seismic imaging with remote sensing for energy applications
Seismic imaging for exploration/monitoring purpose is a large-scale computational problem. First, it involves a large amount of seismic measurements, with 10,000’s of experiments, every experiment shooting in 10,000’s of channels, each channel containing a time-series recording. This adds up to 10’s of Tbyte of data. Next, the imaging process is a computationally heavy process that can easily take weeks to months, especially if run in an inversion mode, where the model parameters (the subsurface reflectivity image and propagation velocity model) should be updated to fit the data. In addition to the computational challenges, there is ill-posedness of the inverse problem. We foresee that AI can assist in bringing in a priori knowledge - such as satellite data and geologic constraints - in a proper way such that the inverse problem is sped up and optimally steered into physically consistent solutions.
Figure 1: Schematic of Deep Learning Neural Network approach to assist seismic inversion
Figure 2: 3D seismic image of a subsurface salt structure
Remote sensing is an effective technique that can reduce the high costs related to oil and gas exploration and monitoring missions. It can detect and observe the characteristics of the Earth's surface by measuring its reflected and emitted radiation at a distance (typically from satellite or aircraft platforms). Remote sensing images include very detailed spatial, spectral and temporal information that can be integrated in other exploration tools, e.g. seismic imaging, well, gravity, magnetic data, etc. Modern Earth observation programs provide free and dense remote sensing image time series (images regularly acquired on the same areas at different times), which enable a systematic global monitoring of e.g. hydrocarbon seepage phenomena, but also demand significant computational requirements.