|dc.description.abstract||The problem of quantifying pressure and saturation changes from 4D seismic data is an area of active research faced with many challenges concerning the non-uniqueness of seismic data inversion, non-repeatability noise in the data, the formulation of the inverse problem, and the use of appropriate constraints. The majority of the inversion methods rely on empirical rock-physics model calibrations linking elastic properties to expected pressure and saturation changes. Model-driven techniques indeed provide a theoretical framework for the practical interpretation of the 4D seismic response but pressure and saturation separation based on this approach are inconsistent with the observed 4D seismic response and insights from reservoir engineering. The outcome is a bias in estimated pressure and saturation changes and for some a leakage between the two. Others have addressed some of this bias using the causality between the induced-production and the observed 4D seismic response to formulate a direct, quick and less compute-intensive inversion - characterised by data-driven techniques. But challenges still remain as to the accuracy of the causality link- as defined by the reservoir’s sensitivity to production effects, and in defining appropriate constraints to tackle non-uniqueness of the seismic inversion and uncertainties in the 4D seismic data.
The main contributions of this thesis are the enhancement of data-driven inversion approach by using multiple monitor 4D seismic data to quantify the reservoir’s sensitivity to pressure and saturation changes, together with the introduction of engineering-consistent constraints provided by multiple history-matched fluid-flow simulation models. A study using observed 4D seismic data (amplitudes and times shifts) acquired at different monitor times on four producing North Sea clastic fields demonstrates the reliability of the seismic-based method to decouple the reservoir’s sensitivity specific to each field’s geological characteristics. A natural extension is to combine multiple monitor 4D seismic data in an inversion scheme that solves for the reservoir sensitivity to pressure and saturation changes, the pressure and saturation changes themselves and the uncertainties in the inversion solution. At least two monitor 4D seismic datasets are required to solve for the reservoir’s sensitivity, and offset stacks (near, mid, and far) are required to decouple pressure, water and gas saturation changes. The generation and use of geologically-constrained and production-constrained multiple simulation models provided spatial constraints to the solution space, making the inversion scheme robust. Within the inversion, the fitness to spatial historical data, i.e. 4D seismic data acquired at different monitor times is analysed. The added benefit of using multiple monitor data is that it allows for a soft “close-the-loop” between the engineering and the 4D seismic domain. One step in the inversion scheme is repeated for as many history-matched simulation models as generated. Each model provides pressure and saturation input to the inversion to obtain maps of the reservoir’s sensitivity. By computing the norm of residuals for each inversion based on each model input, the best model (having the lowest norm of residuals) can be identified, besides the use of a history-matching objective. The inversion scheme thus marks the first step for a seismic-assisted history matching procedure, suggesting that pressure and saturation inversion is best done within the history-matching process.
In addition, analysis of uncertainties in quantitative 4D seismic data interpretation is performed by developing a seismic modelling method that links the shot timings of a real field towed streamer and a permanent reservoir monitoring (PRM) acquisition to the reservoir under production. It is found that pressure and saturation fluctuations that occur during the shooting of monitor acquisitions creates a complicated spatio-temporal imprint on the pre-stack data, and errors if 4D seismic data is analysed in the post-stack domain. Pressure and saturation changes as imaged across the offset stacks (near, mid and far offset) are not the same, adding to the problems in separating pressure and saturation changes using offset stacks of 4D seismic data. The approximate modelling relay that the NRMS errors between offset stacks (up to 7.5%) caused by the intra-survey effects are likely at the limit of 4D seismic measurements using towed streamer technology, but are potentially observable, particularly for PRM technology. Intra-survey effects should thus be considered during 4D survey planning as well as during data processing and analysis. It is recommended that the shot timestamps of the acquisition is used to sort the seismic data immediately after pre-stack migration and before any stacking. The seismic data should also be shot quickly in a consistent pattern to optimise time and fold coverage. It is common to relate the simulation model output to a specific time within the acquisition (start, middle or end of survey), but this study reveals that it is best to take an average of simulation model predictions output at fine time intervals over the entire length of the acquisition, as this is a better temporal comparison to the acquired post-stack 4D seismic data.||en_US