Ages of Reservoir Modelling – what are already there and what are coming

Geomodelling has been in service for decades in the industry. The technical and business mission has evolved remarkably in different ages.

Past: Age of Geostatistics (approx. 1970s-1990s)

representing technology: Kriging, Co-kriging, SIS, SGS

We are in the post classic-geostatistics age but when it was invented, Geostatistics is motivated by demands such as:

(1) data conditioning: putting together data of variable scale and type,  sparse and dense;

(2) Inferring geological settings away from wells, a.k.a. nonstationarity models;

(3) reproducing statistics and heterogeneity models derived from local observations and geologic conceptual models;

(4) representing uncertainty, and

(5) ease for implementation;

These approaches are still in service! Yet, there are fundamental limitations that must be considered in designing reservoir modeling workflows nowadays.

1) stationarity is the assumption that the input statistics are invariant under translation. This is imposed and however, the best practice is to maximize geologic mapping to relax this assumption it, in other words, adding more realism to the statistics. But this often degrades the uncertainty model.

2) spatial features reproduced by geostatistical are limited to a reduced set of input statistics represented by variogram, multiple point statistics or geometric parameters for objects.

3) spatial statistics the geostatistical realizations tend toward maximum entropy, which is associated with maximum discontinuity of extreme values. Given the importance of connectivity of permeability extremes to the formation of flow barriers, baffles and conduits, this attribute often has significant consequences on the simulated flow response.

4) Finally, all geostatistical simulation methods include an inherent conditioning priority. Contradictions in model inputs are settled by compromise and simulation models may not honor all input data;

5) the method works only with limited data types that are available at its birth, e.g. logs, cores, limited seismic understanding, which are only part of the current field data type that is being generated today.

Figure. Most of the efforts are paid to core samples and well logs data in the age of statistics.


Age of Seismic (approx. 1990s-2010s)

representing technology: deterministic inversion, stochastic inversion, lithology inversion, facies classification

With the advances of seismic data quality, its application to reservoir modelling has been widely used throughout the field exploration, development, and monitoring. The use of seismic interpretation and inversion technology is superior because of:

  • Realism. Unlike geostatistics, the seismic data directly sees the true subsurface, e.g. big structures, reservoir distribution, faults, and reservoir heterogeneities when the data quality is under control. Inversion creates volumetric attributes to represents the exactness in 3D.
  • Quantitative information for reservoir characterization. Tens of attributes can be extracted/inverted from seismic data to analyze the most of the reservoir properties from statics (PORO, PERM, NTG, brittleness, fluid contents, fractures, etc.) to dynamics (residual oil, pressure change, stress change etc.).
  • Joint interpretation. Crossing more disciplines becomes possible with the joint interpretation of petrophysics, geophysics, geology and reservoir engineering, which leads to lower and lower subsurface uncertainty.

The seismic inversion family is no doubt one of the most powerful tools for the industry whereas numerous challenges have been overcome. However, fundamental limitations exist:

  • Amplitude plays only. Seismic data literally can contribute little if the reservoir is not “imaged”. Due to the extreme subsurface conditions, e.g. complex overburdens, weak lithology contrasts, gas attenuation and so on, some of the reservoirs cannot be illuminated and imaged properly. In other words, some of the fields are not seismically applicable.
  • Domain conversion. Born in the TWT domain, data domain conversion is the forever pain for practitioners who are to perform true integration between reservoir engineering and seismic. Despite the complex imaging process and velocity models used (lets call this V1 in this discussion, some also call this the z-domain velocity which includes lots of stretching by velocity manipulation to match the depth domain log data and beautiful principles of seismic imaging), seismic data is usually landed in depth through a second time regional velocity estimation (V2). Reservoir models are framed in TWT or Depth from the seismic horizons based on V2, and the inverted reservoir properties such as P-velocity (V3) is then mapped on the V2 model grids for quantitative study. This is the currently most popular Gridding + Inversion + Mapping workflow. Why this bother? Because V3 ≠ V2 ≠ V1. And a lesson learned from the client project is that the V2 and V3 discrepancy caused over 20% reserve error from the problematic reservoir model that can never be reconciled by engineering material balance during history matching. In short,
    1. V1 is the data origin but it is raw and cannot be used for quantitative reservoir study.
    2. V2 is smooth and never accurate but used to build the model frame, and
    3. V3 is highly detailed and used to compare with logs to QC, but misplaced using V2;
  • Inherent illness and assumptions. Inversion is non-unique and uncertain. The inversion uncertainty is hard to be embedded into the reservoir modelling workflow with a limited number of realizations.

Figure. In the age of seismic, reservoir modelling is predominantly driven by incorporating the spatial details illuminated by the seismic data, to build more realistic models compared to the geostatistics.


Age of Integration (approx. 2010s ~ ????)

representing technology: Depth-domain Geostatistical inversion, Direct Petrophysical Inversion, Closed-Loop Reservoir Modelling

Fields are maturing up quickly with more and more field data being generated. Pinpoint extraction of the remaining oil from sweet-spots is the key for most of the operators and it is only possible with a high certainty of the subsurface. At this stage, everything has to be as precise and accurate as possible in geomodelling. Joining up data from all resources is the only way to effectively reduce subsurface and reservoir engineering uncertainty and this is the era of integration! Efforts have been put in developing an ALL-MATCHED subsurface model to honor data from all perspectives and the Closed-Loop Reservoir Modelling concepts are raised and being implemented.

Figure. In the age of integration, all data needs to be balanced and embedded to jointly reduce the uncertainty of subsurface model.

 How to perform a Closed-Loop Reservoir Modelling?

  1. An innovative inversion scheme is needed to DIRECTLY convert the seismic into CORRECT and simulation ready properties on the reservoir model grid. As discussed before, the V1, V2 and V3 problem is critical. The FWI based the methods try to match V1, V2, and V3 but far more expensive, inefficient and researchy for everyday reservoir modelling purposes. A direct reservoir model domain inversion is needed to close the loop between at least the V2 and V3 and convert true reservoir properties to the true 3D location in depth. This involves:
  • innovative seismic forward modelling from the model grid domain
  • The consistent inverse of the model domain inversion
  • Simultaneously solving V3 and the true depth location of the results on the fly

  1. An interactive workflow is needed for smart reservoir model overhaul at the real-time when new data is acquired. Two important technology components of such workflow are:
    1. Sim2seis modelling for screening purpose – identify the discrepancy between model prediction and field observation
    2. Smart model updating scheme – root analysis of the causes of discrepancy and direct fix of the imperfection of the existing models


EVP of iRes-Geo, Formerly Reservoir Geophysicist, Chevron Global Technology Centre, Aberdeen

Ph.D. in Reservoir Geophysics from Heriot-Watt University.

Is your reservoir model up to date? Please contact us to validate it for accurate remaining oil prediction!