Tomographic inversion using 4D seismic data

Abstract

Time-lapse (or 4D) seismic data has become an acknowledged technique used to monitor the development of subsurface reservoirs due to human interaction such as production of hydrocarbons or injection of CO2 for storage purposes. While qualitative interpretation approaches have been successful in the industry, a major remaining challenge is to further develop tools for quantitative analysis of the time-lapse data. To this end, the focus in this thesis is on estimating changes in P-wave velocity by using least squares inversion on 4D time-shifts and amplitude changes.

Most 4D time-shift analysis is done using stacked data, since a higher signal to noise ratio is obtained in the stacking process. However, this removes the angle information present in the time-shift data. We propose instead to analyze pre stack time-shifts, as these will contain information along many ray paths as the signal has traveled through the target layers, and thus ’see’ different slices of the earth. In our tomographic approach, we use time-shifts for several offsets an input to a least squares inversion algorithm in order to make estimates of parameters describing changes in P-wave velocity. To obtain this, we investigate different parameterizations of the velocity changes.

We find that a 2D Gaussian representation of the velocity changes has the possibility to describe many types of 4D velocity anomaly shapes, while requiring estimation of a limited amount of parameters. The weakness is that fine detail in the 4D changes are unresolved. Synthetic inversion tests are done using different types of true models, and we find that if the signal to noise ratio is sufficiently high, the estimation of parameters describing the overall shape and velocity of a 4D anomaly is possible, and our results indicate that using this parameterization we can obtain a low resolution smooth fit to a more complex true model. Inversion using the Gaussian parameterization of the velocity changes is applied to two field data cases. One from the CO2 injection site above the Sleipner field in the North Sea, where a thin layer of injected CO2 is investigated. In the second case, a shallow gas chimney resulting from a well blowout is identified in 4D data from a repeated shallow hazard site survey. We find that using this parameterization, inversion of pre-stack time shifts are able to aid interpretation by helping to identify where gas is present.

In order to investigate the properties of a thin CO2 layer in more detail, we propose to parameterize the thickness and velocity changes due to this layer by using cubic b splines. This allows more detail to be described than with the Gaussian parameterization. We find that using only pre-stack time shifts, a priori assumptions are needed in order to avoid ambiguity between velocity and thickness in the resulting estimates. This information may be obtained from rock physics. Another approach is to include the seismic amplitudes in the inversion, and we find that first inverting for time shifts, and using the result as input into amplitude inversion reduces the ambiguity problem and. This combined approach is applied to field data from the Sleipner field, and the properties of a thin CO2 layer is investigated.

Noise in the seismic data is a major source of uncertainty, and while all the proposed methods are tested for robustness on synthetic data with noise added, field data noise can have many variations. Pre stack data has a lower signal to noise ratio than stacked data, so data of a sufficient quality is required for the methods to work. Furthermore, in our inversions, we make assumptions about the background model before 4D changes has occurred, and this is a source of uncertainty. Repeatability of the seismic acquisitions is also an important factor, and is paramount for a good quality 4D signal, especially if the 4D response is weak. Permanent ocean bottom systems are maybe the most effective way to increase repeatability, and if the development of such systems becomes the trend in the industry, it has the potential to form a good basis for 4D tomography approaches in the future.