research Research programme Module 1

Module 1

4D Quantitative Interpretation (4DQI) has now evolved into a recognised discipline, with distinct sub-topics such as PEM, 4D inversion, log2seis, mech2seis, sim2seis, and SHM.  Implementation of these tools and improvements gained in the context of different field settings forms the basis for improvements in 4D QI. Quantitative analysis is often challenged by the need to move with speed through the workflow, often neglecting the physics in pursuit of deadlines. Multi-disciplinary connections are seldomly in place, and links to, and consistency with the later part of the workflow may converge. In this module we address three areas which we have identified as requiring improvement in the next three years. The first arises out of observations made during Phase VIII, in which we challenged ourselves to measure and utilise pre-stack time-shifts in the gathers. We saw that this information, when interpreted through 4D tomography, could provide a strong uplift in our understanding of velocity changes. In this phase, we also re-visited our 4D Inversion tools, and determined that some upgrading was necessary due to enhancements in seismic data quality and processing. Finally, with the revision of our petro-elastic modelling tools for our toolbox we see the need to mature and consolidate the PEM across a variety of new datasets. Thus, in this module we promote three research threads: Pre-stack analysis, 4D Inversion and the Petro-elastic model.

Pre-stack 4D analysis

Pre-stack data in various formats forms the basis of many 4D interpretations. For example, restricted-offset stacks are widely used for the separation and estimation of pressure and saturation changes either as a deterministic, probabilistic or machine learning solution (Corte et al. 2022, 2023).  Such data may also be used for the interpretation of changes in stress or strain in the overburden and reservoir using pre-dominantly time-shift. Phase VIII taught us that we may move further than limited-offset stacks and include more data from the pre-stack workflow. It also gave us guidance on what processing modules to avoid (for example, trim statics) and where to extract the data (typically after residual RMO). We also saw how the impact of choices made during the pre-stack processing biased and distorted our interpretations on the post-stack data (Hatab and MacBeth 2022, 2023; see Figure 4). It was demonstrated that using 4D tomography applied to well-conditioned pre-stack time-shifts, the geometric footprint from raypath illumination could be disentangled from the in-situ velocity changes (Izadian and MacBeth 2021, 2022; see Figure 5). This technique has now been applied to two PRM datasets with success (Ekofisk and Valhall). Whilst the pre-stack domain does provide access to additional finer detail, there remain some problems to address:

• how best to measure and condition pre-stack time-shifts,
• the ideal angle ranges to utilise,
• the diversification to other datasets,
• incorporating azimuthal data,
• the need to handle large datasets,
• the need for time-shift sensitive processing to enhance the pre-stack information. 

The above challenge our use of pre-stack data, and there is a need to work in a more interactive framework where many choices can be made quickly and iterated on large datasets, such as that offered by PrestackPro (Sharp reflections). Is there value in applying the methods of data science to these large datasets?  In Phase IX we set out to develop this approach further across a broader range of datasets. This work is also linked to the Seismic Geomechanics topic in our second module.

---

4D Inversion

This sub-module links with the pre-stack research, as it may be implemented in both the post- and pre-stack domains. 4D Inversion has been defined as the inversion of time-lapse seismic traces to independent measures of impedance and velocity change. This differs from objective to estimate pressure, saturation, stress or strain changes. Instead, our desirable outcome is to provide high fidelity, broad bandwidth, quantitative estimates of impedance, velocity (and density) changes for further interpretation using a PEM. The Phase IX development follows on directly from our research initiated in the later stages of Phase VIII, where a wide various post-stack (historic and new deterministic/model-based) workflows have been investigated across a range of datasets (see Figure 6). This work identified several distinct directions for progression of this topic:

Summary of specific themes:

• Physics constraints- the need to study the relationship between changes in velocity, impedance and density using a calibrated PEM. How to best capture and introduce these non-linear correlations into the inversion is also being investigated.
• The need to test different initial models, for example: those from time-shift inversion, those from the fluid flow simulator, or fast-track quadrature results.
• The need to apply regularisation to smooth the results.
• Smooth analytic target estimates using techniques such as Gaussian reconstruction.
• The relationship between changes of amplitude and time-shift needs investigated – we know they are not correlated well, why is that?
• Joint time-shift and amplitude inversion
• The use and value of the baseline data?
• The application of probabilistic methods
• The application of machine learning

---

The Petroelastic model (PEM)

The PEM is central to many of our modelling and estimation tools in 4DQI. For example: fundamental feasibility analysis, log to seismic modelling, simulator to seismic modelling, the estimation of pressures and saturations, seismic history matching, 4D inversions (partially) and in Seismic Geomechanics the estimation of the strain and stress components. It is used to connect via modelling or inversion the physical variables of pressure, stress and strain, saturation, temperature, or fluid constituent changes with the measured seismic observables of amplitude, time-shift, impedance, velocity and density. Our ETLP toolbox now has separate rock and fluid physics models for clastics, carbonates, CO2-specific problems, geomechanical analysis, in addition to a variety of helpful proxy models. Each PEM has different challenges and degrees of complexity. For example in the case of carbonates, the objective is to capture the pore type variations whilst for CCS it is to capture the properties of different fluid mixtures which contain CO2. Finally the PEMOPT procedure is the way in which we calibrate the hyper-parameters of each PEM to the existing log data. This also includes the possibility of optimising hyper-parameters of an EOS to match fluid data.

For Phase IX we recognise that one of the main challenges is to build a database of optimised PEMs across the range of available data provided by sponsors. This broad library of diverse geological properties will prove beneficial to staff, students and staff who are exploring new datasets and wish some degree of calibration to past work. Such a database may also be beneficial when preparing data for training our machine learning solutions. Finally, the direct comparison between clastics and carbonates might help offer some insights into the use of 4D seismic in carbonate fields. An important line of development will include completing the extensions to the PEM family, which include the CCS and carbonate (hard-rock and chalk) modules.

Other items of focus are:

• The need to finalise our EOS-based fluid property code, our strain/stress tensor to seismic properties,
• Saturation heterogeneity issues for CO2 injection,
• Pressure or stress sensitivity of the rocks under depletion and injection,
• A full TOE PEM,
• Greater speed and accuracy (plus appropriateness) in the PEMOPT,
• Data-driven PEMs via machine learning,
• Develop a library of optimised PEMs for all of our UKCS and NCS datasets, and then broaden this out via data from the national repository to wider areas in the UKCS, NCS and globally.