During active seismic measurements, reflected waves from the subsurface are recorded by sensors at different distances (called offsets) from the source. After extensive processing and imaging, the measured data at these different offsets can be used to find reservoir properties, a method termed AVO (Amplitude Variation with Offset) inversion. This method is used to distinguish between fluid content in porous sandstones, i.e., to distinguish between water, oil, or gas. However, there are many pitfalls in this AVO analysis which are partly connected to a too simplistic model of the wave propagation in the earth. Saskia Tschache has, in this industrial Ph.D. project, focused on two main topics (i) investigate and implement a more suitable modelling tool and (ii) using this tool in a Bayesian classification and inversion method to quantify the uncertainty in the estimation of reservoir properties.
Saskia’s first paper of the Ph.D. was published in Journal of Applied Geophysics in November 2022 and named "On the accuracy and spatial sampling of finite difference modelling in discontinuous models " (Saskia Tschache, Vetle Vinje, Einar Iversen). In this paper, she compares two popular methods to simulate the propagation of sound waves in the subsurface: the Reflectivity (RT) method and the Finite Difference (FD) method. She shows that the RT method is extremely accurate by comparing it with known analytical solutions. In the paper, she also derives the sampling constraints and the optimal model parametrization as a function of signal frequency required for the FD method using the RT method as a reference.
The 2nd and 3rd papers of Saskia’s thesis use the RT method as a modelling engine and combine it with Bayesian methods, where many possible realizations of reservoir models are considered. The elastic characteristics of these models are created by a combination of well log measurements and established rock physics.
Paper 2 entitled “Quantifying amplitude-variation-with-offset uncertainties related to calcite-cemented beds using a Monte Carlo simulation” was published in Interpretation in April 2023 and quantifies the uncertainty when calcite-cemented beds are present in a sandstone reservoir. These hard, thin layers create a detectable effect on the measured AVO and the AVO-based pore-fluid classification. Saskia shows in a North Sea case test that the probability of a false-positive hydrocarbon indication increases from 3-5% to 18-21% when the calcite-cemented beds are included. The results confirm that calcite-cemented beds can create a pitfall in AVO analysis.
Paper 3 was submitted to Interpretation in April 2023 and is named “Estimation of net-to-gross ratio and net pay from seismic amplitude variation with offset using Bayesian inversion“
Using a similar methodology as in paper 2, Saskia developed an inversion that can be applied to AVO maps to produce maps of the most likely estimates of important reservoir properties, such as the fraction of porous sand layers and the hydrocarbon content in a reservoir in addition to the uncertainties of the estimates.
In this work, Saskia has worked in close cooperation with her CGG and UiB supervisors, with co-students in CGG in Oslo, students and co-supervisor Jan Erik Lie from Lundin/AkerBP, the world-renowned expert in quantitative seismic interpretation Dr. Per Avseth, and AVO experts and geologists both at CGG in Norway and internationally.
Saskia har utviklet en inversjonsmetode som kan brukes på seismiske AVO-målinger for å produsere kart over de mest sannsynlige estimatene av viktige reservoaregenskaper, som andelen porøse sandlag og hydrokarboninnholdet i et reservoar i tillegg til usikkerhetene i estimatene. I tillegg har hun kvantifisert usikkerheten som sementerte, tynne kalksteinslag medfører i AVO inversjon. Begge disse anvendelsene av hennes metode kan bidra til en bedre forståelse av risiko ved boring av brønner.
The recent availability of powerful GPUs and open source software have enabled artificial neural networks (ANNs) to be applied to several practical and industrial scale problems. In seismic data processing, ANNs have the potential to be applied to many of the key processing steps (swell noise attenuation, seismic interference attenuation, deblending, deghosting, etc.) which today involve significant testing time and computational power. Once trained, ANNs are computationally very light and potentially adaptable to different datasets. Their use could, therefore, save processing times and, in the long term, impact the whole business sector.
The proposed doctoral work is about the usage of ANNs for processing of marine seismic data, esp. denosing and deblending. The goal is to achieve similar or better quality results compared to conventional processing methods.