research Research programme Module 2

Module 2

This module maintains the popular and well-recognised topic of pressure-saturation estimation (seis2PS) and adds that to the recently expanded topic of Seismic Geomechanics which we find more easily digested as separate post-stack and pre-stack flavours. We continue with Seis2PS as it remains an extremely exciting area, and solutions are still seen to be largely dependent on field type (geology and production). Despite the obvious maturity of this topic and the growing number of applications by ETLP, getting it working on stacked reservoirs, when we have structural complexity or for thin reservoirs remains a challenge. Some technical aspects remain unsolved such as saturation heterogeneity and pressure sensitivity. For the Seismic Geomechanics, post-stack time-shift analysis is typically utilised to close the loop. ETLP’s familiar mech2seis modelling helps with this analysis. However, this has not yet been fully implemented in practice, in the many workflow formats possible. Finally, pre-stack time-shift analysis which couples the benefits of the 4D tomography mentioned previously with comprehensive modelling is required for further research to provide new information on the horizontal as well as vertical stress/strains. This provides much needed input into the interpretation and de-risking of field for which geomechanics plays a prominent role. The latter capitalises as its foundation on the developments from Phase VIII in pre-stack TVO and AVO interpretation.    

Pressure and saturation estimation

As our webinars have outlined, many possible paradigms exist to estimate pressure and saturation changes due to production. Starting with early fast-track screening using AVO, one can progress to full inversion of restricted offsets stacks and multiple seismic attributes including time-shifts, constrained to some degree by predictions from the simulation model. In some specific cases, inversions to impedance products are required where 4D signals overlap vertically or laterally. For example, the case of a  producing reservoir unit overlying another, with variable thicknesses and reservoir quality. This situation arises in datasets such as GoldenEagle. ETLP has shown many examples of applications to clastic reservoirs in Phase VIII using Bayesian inversion and also machine learning (Corte et al. 2021, 2022; see Figures 8 and 9). Despite the maturity of estimation procedures much still remains to be achieved in terms of obtaining robust, high-resolution estimates directly from 4D seismic signatures. Our intention would be to continue with these applications in Phase IX to refine our existing techniques and diversify across a range of fields so that we may focus on the most robust and flexible methodology. This research will also build on our understanding from other modules, such as the saturation heterogeneity and pressure sensitivity. Fields such as Snorre offer us the opportunity to investigate the estimation of the dynamic signals for recovery mechanisms that are complex and cyclical such as water alternating gas (WAG). We find that production setting is important and hope by application of our insights to field datasets to be able to extend our methods and understanding to situations such as a stacked reservoir environment or structurally complex field setting. As we progress, strong and obvious links to the fluid flow simulator will be made, suggesting a possible seismic history matching or closing of the loop in the pressure and saturation change domain. Our analysis will link to our research on PEM, sim2seis, 4D inversion and SHM. 

Pressure and saturation estimates for a UKCS field using a DNN. In this example we use a pressure prior from the simulation model.

Post-stack Seismic Geomechanics

The use of 4D seismic as a method for interpreting stress and strain variations in the subsurface has become widespread in industry. ETLP research tools for dealing with this include post-stack time-shift estimation (see the review of MacBeth et al. 2020), mech2seis and mech2ts. Also required in this domain is the geomechanical PEM, which in our case is created by the Hatchell-Bourne-Roeste R-factor model or the Prioul third order elasticity (TOE) model. These combine with our geomechanical modelling capability to define both forward and inverse workflows. It is also possible to invert overburden time-shifts for reservoir pressure in clastics (Hodgson and MacBeth 2006; Garcia and MacBeth 2012) and chalk (Corzo et al. 2010; Wong et al. 2017) reservoirs. Our approach operates predominantly in the post-stack domain, in which modelling assumes vertical raypaths. We understand from modelling but also by processing of 4D seismic data (Hatab and MacBeth 2023; see Figure 10) that such assumptions may have limitations but are fairly accurate for most reservoirs with horizontally layered structure and wide regions of pressure evolution. It is possible to use post-stack signatures to close the loop with the geomechanical modelling and inform us about the R-factors in the reservoir and overburden. This has been achieved in the past for Shearwater (Jaramillo et al. 2019), Valhall (Tian et al. 2022; ) and Snorre (Corte et al. 2022). As we progress with our research, we see a need for:

The following are the main points

  • An inversion scheme to disentangle 4D signatures from different regions of pressure change in the reservoir. The scheme needs to be fast, iterative, and capture the uncertainties. It should include a separation of fault reactivations versus stress arching effects on time-shift.
  • An inversion scheme to ‘history match’ the geomechanical model
  • Capturing the realistic uncertainties between geomechanical model, reservoir pressure distribution and R-factors.
  • Speeding up the geomechanical modelling via the Geertsma-based approach and incorporating reservoir contrast and layering as part of our inversion solution. Can machine learning be use here?
  • Improving vertical resolution of pressure inversion when stacked reservoirs contribute to the geomechanics at more than one producing interval.
  • Understanding and rationalisation of the strain-based R-factor model and stress sensitivity in the reservoir.
  • Consideration of injection (inflation) and production (depletion) effects in the overburden and reservoir to estimate R-factors relevant for these processes, and how these may be linked to prior knowledge of the stress sensitivity. This is important both in the producing hydrocarbon context where pressure may experience cyclic variations due to fluctuating injector and producer rates, but also in CCS/hydrogen storage where inflation due to injection followed by final relaxation may induce mechanisms that may contribute to the risk of failure.

This sub-module will also continue to explore case study examples in this area, refining our modelling and inversion solutions, and attempting to understand the items above.

Mech2seis results versus observed post-stack time-shift. Convolution based mech2seis produces more comparable synthetic time-shifts than straight integration.
The time shift gradient computed on pre-stack time-shift gathers, showing regions of poor confidence in post-stack time-shift analysis in the reservoir and underburden.

Pre-stack seismic Geomechanics

In Phase VIII our pre-stack analysis progressed on several fronts. Firstly, with the help of a high quality Ekofisk PRM dataset ETLP was able to estimate time-shifts in the pre-stack domain, immediately after RMO correction (which we found to be the most desirable point to extract the data), with the help of some careful conditioning. These time-shifts could then be utilised in our 4D time-shift tomography scheme which unravelled the overlapping effects of geometric raypath and time-shift variation with offset to provide a reliable image of velocity change at each subsurface location (Izadian and MacBeth 2021, 2022, 2023; ). This scheme was then further extended to output vertical and horizontal velocity estimates that are linked to an anisotropic model of the reservoir and overburden (in this case we used the Prioul third order elasticity (Prioul-TOE) model which allows us to simplify the anisotropy to two R-factors). These velocity components are then related directly to the vertical and horizontal strains. The inversion scheme required some careful stacking and conditioning to ensure reliable input. We also demonstrated that we could work with these large volumes. A further application to Valhall PRM data was undertaken, which was challenging due to the noise levels, but this encouraged innovative post-processing techniques to prise out the subtle pre-stack time-shift information. To check the sensitivity of time-shifts to the pre-stack, post-migration, processing Hatab and MacBeth (2021, 2022) carried out a series of studies on the Volve and the Ekofisk data. They extended this work to include the impact of the migration velocity model. The results suggested that post-stack time-shifts are robust to choices made during processing, although several key regions exhibited large variability. For future assessment a metric was designed that highlighted how large changes in time-shift variation with offset impacted the post-stack domain. In this phase, the use of amplitude variation with offset has also been considered as a supplement to time-shift analysis. This proves beneficial to enhance the resolution, particularly in the overburden.

From our research it is clear that pre-stack analysis on the processed gathers has a future, particularly when using time-shifts. To store and manipulate the large volumes we can utilise software such as PrestackPro or work with limited sub-volumes. We have shown that by using the Prioul-TOE anisotropic model that we can extract both vertical and horizontal strain/stress information directly from the 4D seismic data. To take this promising line of research further and to extend the utility of 4D seismic for geomechnical interrogation in Phase IX we wish to investigate the following items:

Key features of this sub-module are therefore:

  • Applications to further datasets,
  • Speed up and make the 4D tomographic method more robust,
  • Calibration and inversion of the Prioul-TOE dual R-factor model,
  • Imaging of strain/stress when the principal axes are rotated (as on the edges of the stress arch or tilted structures),
  • Comparison between the fully anisotropic TOE (instead of the simplified Prioul-TOE model) – we will use the published SINTEF model to explore this possibility. How can we validate these models?
  • The effect of physical displacement on the results, and the possibility of separation physical strain and velocity change,
  • Incorporating the impact of fault re-activation, can this be resolved more easily in the pre-stack domain?
  • Cost versus value exercise: post-stack versus pre-stack time-shifts.
Inverted velocity components: a) (Δ𝑣⁄v)1 – vertical and b) (Δ𝑣/𝑣)2 – horizontal. Having these two components at hand, it is possible to determine the geomechanical regimes in place (from the studies of Izadian and MacBeth, 2023).