Blueback ODiSI (One Dimensional Stochastic Inversion) is a transparent and innovative approach to Seismic Inversion. Capturing all of the associated prior knowledge about a given field, prospect, or reservoir. The ODiSI user builds a robust and comprehensive model of the subsurface through the creation of pseudo-wells (PW's). These PW's contain a wealth of information including estimated well logs (Vp, Vs, Rho, Facies, Sw, porosity, etc.), rock physics models (per facies), and fluid information. The PW's are then constrained by a stratigraphic framework, each with their own facies definition and expected bed thickness distribution statistics.
The result of the inversion process is that Blueback ODiSI produces high-quality reservoir property estimates as well as captures associated uncertainties for any given reservoir property output. Generating volumes such as Net Fraction, Lithofacies distribution, and individual property volumes for Porosity, Saturation, HPV and Facies probabilities - all valuable in their own right.
Creating the ODiSI volumes is only the beginning, though! Analyzing and interrogating the volumes is where a lot of the real value comes from. It is this process that challenges the user to think about the 'what if's' associated with their particular reservoir problem and how the information from ODiSI can help to resolve some of these questions.
Figure 1: The figure above details the ODiSI process from the input seismic, the creation of pseudo-wells from prior information, the accept/reject matching to generate the best-matched pseudo-wells and the creation of outputs.
In a generic reservoir modeling workflow, the construction of a discrete ‘facies’ model usually precedes the property model. The goal of the facies model is to capture contrasts in rock type based on geometry and/or petrophysical properties. Typically, the facies model is a product of a detailed sedimentological description, which is then adapted and often simplified for the particular reservoir modeling goal. Regardless of the purpose of the model (volumetrics, infill drilling, flow dynamics, etc.), one of the key tasks is to identify how the reservoir properties are distributed between the wells. If you interpret sand in ‘Well A’, how far does it extend away from the well? Are the sands connected between wells? Are there field wide shale barriers? The list of questions continues.
This is where ODiSI can help by providing information about the facies distributions based on the calculated probability volumes. The aim is not for ODiSI to replace the current methods of facies delineation, such as building a robust conceptual model, analyzing analog fields, etc. However, it can be very powerful when combined with these methods by providing invaluable information from seismic that may otherwise be left untapped. In the following sections, we look at the ODiSI outputs in more detail surrounding facies prediction and provide practical examples of how to analyze and incorporate the ODiSI volumes into a facies model.
In order to know how to best use the ODiSI results, you need to understand what the outputs represent. ODiSI will output different properties as follows:
Figure 2: ODiSI outputs: 1) Facies Probability; 2) Probable Lithofacies; 3) Percentile Lithofacies Probability.
The volumes generated from ODiSI are a collection of seismic datasets in time domain. Below, we look at three methods for taking these outputs and transforming them to a format usable within the geomodeling workflow:
The ‘Seismic Resampling’ option within Geometrical Modeling enables a seismic volume to be sampled into the static model. If the static model is in depth, the ODiSI volumes need to be depth converted prior to re-sampling to ensure the input seismic and target model are in the same domain. Once sampled into the static model, the individual facies probability volumes (P-Shale, P-Clean Sand, etc.) are available to be used as a trend within the facies modeling process.
As these are probability volumes, each facies is represented by a continuous value range of 0-1, where the closer the sample is to 1, the more likely it is to be associated with that particular facies code.
Figure 3: P-Clean Sand volume from ODiSI.
Figure 4: P-Clean Sand volume resampled in to geomodel within reservoir zone.
Having the ability to extract ODiSI derived facies probability maps across reservoir zones is a simple and effective way to gain an insight into the likely distribution of a particular facies across the field. Use the ‘Surface Attribute’ process in Petrel E&P software platform* to extract the Mean Amplitude from the individual facies probability volume (E.g., P-Shale) within a given reservoir interval defined by a top/base surface.
The advantage of creating these maps across key reservoir intervals is that they can be extracted on the TWT volumes, but as they are a lateral trend, they do not need a vertical component and therefore can be used directly on either TWT or Depth geomodels.
Figure 5: Petrel 'Surface Attribute' process whereby maps can be generated across zones of interest and then used as trend maps later in the modeling workflow.
Statistics extracted from the ODiSI volumes can have multiple uses, but here we focus on their use as input to Facies Modeling. The easiest way to interrogate and extract trends from the seismic volumes is by using Blueback Investigator.
Load the P-Lithofacies ODiSI volume to a Blueback Investigation, set up a Spatial Selection to isolate the reservoir zone of interest, and it is then possible to visualize the relative fractions for each of the discrete facies codes across the data area. These numbers can be used directly in facies modeling when parameterizing the property, and provide a useful comparison to compliment or supersede well-derived statistics when specifying a target fraction.
Figure 6: Histogram from Blueback Investigator showing facies proportions within reservoir interval as observed in well data (red) compared to the reservoir interval within the full ODiSI volume (blue). Large discrepancy can be seen between the two relative fractions. Without the ODiSI data, there is a risk that the target fraction for this dataset could be set too optimistic for facies such as Clean Sand and therefore misrepresent what is likely to be present across the full field area. At the very least, this should act as a discussion point within the project team as to why the discrepancy between the datasets exists (targeted drilling, proximal/distal well location within depositional setting, etc.). These fractions can be used directly when parameterizing Facies Modeling as seen fit.
Having looked at what the ODiSI output volumes represent in 'Step 1', and then at different methods for extracting the data into various formats in 'Step 2', the next stage is how to use these outputs within the Facies Modeling workflow.
The key to having a reliable Facies Model is having reliable and realistic input parameters. The standard input parameters needed for the majority of modeling algorithms (Kriging, Gaussian, Object) are; Variogram, Facies Fraction, and Trends (both lateral and vertical). In simplistic terms, the Variogram will control the direction and lateral extent of the facies, the Global Fraction will control the quantity of each facies across a given zone, and Trends will control where each facies type is more likely to be present. Here we look at how ODiSI can help with these.
Depending on the dataset and geology, the ODiSI volumes may be able to identify and isolate specific depositional features such as fan lobes, fluvial channels, etc. In these cases, the dimensions and geometries of the bodies can be measured and used directly as input to the modeling algorithm as object dimensions, or variogram ranges.
In general, the starting assumption for specifying a target fraction for a particular facies within Facies Modelling is to use well data statistics. As highlighted in the histogram in Figure 5, this may not always be an appropriate method. Drilling will always target reservoir sweet spots, so well results are unlikely to represent the true fractions that exist across the full field area. To compensate for this, analog data and/or conceptual models are often used to help steer the global facies fractions. This provides more control, but has a large uncertainty and can be very subjective.
Therefore, an additional way to control the target facies fractions is to use the ODiSI Lithofacies volumes to estimate the inputs per facies, per zone.
As shown in Figure 6, there are four ways to determine the target fraction within Petrel. Option ‘A’ (Upscaled well logs) and ‘B’ (Well logs) are the standard log derived methods. Option ‘C’ allows a user-defined ‘manual’ fraction to be specified. This is where you can utilize the data analysis shown in Figure 4 to input the fraction extracted from the ODiSI P-Lithofacies volume. Option ‘D’ is to use a ‘Trend’, and this will take the fractions from a specified trend, in this case using the resampled 3D P-Facies properties that were detailed in ‘Step 2’ of this document.
As with all aspects of modeling, the appropriate method for a particular field will be governed by the data itself and the goal of the project. Using well data vs. ODiSI vs. any other method to determine the target fractions will be a decision to be made within the project team, however having the data from ODiSI available is an extremely valuable additional dataset to support and justify any decisions that are eventually taken.
Figure 7: Images demonstrating what input data can be used for each of the 'global fraction' options in Petrel facies modeling.
The lateral and vertical placement of facies within the field can be guided by the use of trends. ‘Step 2’ within this document detailed how to generate zonal facies proportion maps from the facies probability volumes. It also showed how to resample the data into the static model. Both of these can be used as trends for guiding the facies model. The advantage of using the zonal maps instead of the 3D property is that the map can be generated directly on the TWT ODiSI volume and used as input to a Depth geomodel, whereas the ODiSI volumes would need to be depth converted if they are to be used on the same model.
Figure 7 shows the input trend map used from one of the ODiSI Clean-Sand probability volumes, and the effect it has on the modeled facies property when implemented. When the trend is applied, the facies model generates a stochastic facies property, but rather than being completely random (as shown with the No Trend image), it will try to honor the input trend map when selecting where to place a particular facies type. This is extremely useful and a powerful and transparent way to gain control over your modeled outputs.
Figure 8: Input trend is generated using the ‘Surface Attributes’ process in Petrel on each of the Facies Probability volume outputs from ODiSI. These are scaled from 0-1 and can then be used as input to the Facies Modeling process to guide the lateral facies distributions.
An additional control that can be combined with the lateral trend from ODiSI is to apply a Vertical Proportion Curve (VPC) alongside the 2D map, and this will allow control over both the lateral and vertical components of the modeled facies property. The VPC in Petrel is based on layer number rather than absolute depth, so it is easiest to generate this in the standard way from Petrel within Data Analysis.
An example of the comparative effects of implementing these trends on the final facies model is depicted in Figure 8. The ODiSI P-Lithofacies volume is shown for comparison, and then three different Z-Slice extractions are made across the facies model to demonstrate that when both a lateral and a vertical trend are applied, the results start to look more like the ODiSI output, but still with a stochastic element.
Figure 9: Comparing the same Z-Slice from the ODiSI P-Lithofacies Volume to three variations of Facies Model. 1) using no trends; 2) using lateral trend; 3) using lateral and vertical trend.
ODiSI can discover trends within datasets that may otherwise be left uncovered from more traditional data analysis approaches. This in itself makes it a potentially powerful tool when used on a suitable dataset. It is then demonstrated within this article that the outputs from ODiSI can also be used and incorporated into a geomodel to help with one of the most important aspects of model building and reservoir characterization: facies distribution.
Predicting where facies are likely to occur across the reservoir is critical for understanding the likely flow patterns within the field and therefore help with various aspects of well planning and field development. ODiSI facilitates this process by enabling the user to make informed decisions when it comes to the facies modeling process and allowing the user to frame all key aspects of the facies modeling input from geometry, quantity, and locality of each modeled facies.
There are inevitable limitations with regards to seismic resolution and the scale of facies that this process will isolate, however, this is assessed in a screening process prior to running the inversion. Therefore, assuming ODiSI is deemed appropriate for the field, it has significant scope for adding value to the facies model and general quality and confidence in the final geomodel product. Finally, by extracting geological information from a geophysical dataset, it has the added bonus of bringing the asset team together by working across disciplines towards a common goal.
Contributions: Lisa Casteleyn - Lead ODiSI Consultant, Cegal., Vince Hilton - Reservoir Geologist, Cegal., John Sayer - Regional Sales Manager UK& Africa, Cegal., Arne Stoffe Norborg - Technical Sale/ Global Business Development, Cegal. & Patrick Connolly - PCA/ Cegal.
*Petrel is a mark of SLB