Exploring for oil and gas is becoming more challenging as operators venture into new frontiers, especially in deepwater settings. In the recently renamed Gulf of Mexico/America, for example, where wells can be more than 30,000 feet deep, imaging subsalt reservoirs with seismic technology can be difficult – even with improved technology.
And, while machine learning and artificial intelligence are useful tools, they cannot help characterize reservoirs if they lack the necessary input data to perform.
To overcome this issue, a team of geologists at Chevron are reexamining the “first principles” that influence controlled reservoir distribution in deepwater settings: sediment supply and gradient, said Morgan Sullivan, an emeritus fellow at Chevron. Gradient in particular has been the focus as they utilize a corporate database built from thousands of high confidence measurements of dimensional properties from hundreds of wells to better understand the presence and distribution of deepwater reservoirs that would otherwise remain unknown.
After performing a number of blind tests using well data of once unknown origins, the team has been able to successfully classify reservoirs in terms of type and fairway width. This has opened up a new way to forecast and predict sand presence and distribution in the absence of high-quality seismic data. Chevron is now actively using this approach to support exploration.
“There are a lot of dry holes being drilled by the industry. We have drilled most of the easy opportunities, and exploration has moved to more data-challenged situations where prediction of reservoir presence and distribution are difficult. If seismic saw everything, we would be making a lot more discoveries,” Sullivan said. “This approach is not a silver bullet, but it is a tool to improve the understanding of sand distribution in data-challenged situations.”
The Significance of Gradient
Approximately 20 years ago, geologists at Chevron to create a database from its deepwater wells that included measurements such as thickness and reservoir type at a variety of scales – from a single channel to a complete depositional sequence. Its purpose was to assess net-to-gross ratios and provide reservoir modeling parameters for appraisal and development.
However, some geologists recently proposed that the database be utilized for exploration.
“We use well logs to support pre-drill and post-drill predictions of sand. We use them for resource density and determining how many pay sands there are,” Sullivan said. “But can we use those wells – and I am going to suggest ‘single wells’ – to actually determine reservoir type? This approach does not add significant value if it takes 10 to 20 wells to determine reservoir extent. We have already defined it by that point.”
With new eyes and a fresh vision, geologists scrutinized thousands of datapoints with the goal of improving reservoir prediction and characterization in the absence of critical data.
That process brought them back to first principles. While sediment type clearly affects reservoir type, Sullivan emphasizes the significant role played by gradient.
Gradient is critical in determining reservoir type because it controls the width of the reservoir fairway. As gradient decreases, sand distribution increases. Yet, gradient in the subsurface cannot be measured at the level of detail required, as changes of fewer than 1 to 2 degrees can have a significant impact on reservoir distribution.
From the data they had from known reservoirs, geologists determined the relative gradient and fairway width of three of the most common types of deepwater reservoirs: erosionally confined systems with moderate to steep gradients and fairway widths between 1 and 3 kilometers; weakly confined systems with moderate gradients and fairway widths from 5 to 10 kilometers; and fully unconfined systems with the lowest gradients and fairway widths greater than 10 kilometers.
The scale of these observations is important to note, Sullivan said. To predict reservoir widths, the characterization must be made at the complex scale where there is “stationarity” of reservoir characteristics. In other words, individual channels are similar in thickness, grain size, permeability and porosity.
“We’ve got to look at this at the complex scale when we are trying to predict sand distribution because we want to make sure the physical conditions aren’t changing during the deposition of the sand that we are trying to predict,” he explained.
The differences between these three deepwater reservoir types can typically be distinguished with confidence using seismic data. But how can this distinction be made in the absence of this data?
Relationship Between Gradient and Channel Element Thickness
Geologists searched their database of known reservoirs looking for relationships between fairway thickness and reservoir types in various formations. For example, in offshore Angola, reservoirs with high gradients have narrow fairways that span just 1 to 3 kilometers in width. They are considered confined systems. Yet in certain areas of the Miocene formation in parts of the Gulf, there are reservoirs with low gradients and broad fairways that are tens of miles wide. These are weakly confined or unconfined systems.
However, when looking at these reservoirs at the complex scale, geologists saw no relationship between fairway thickness and reservoir type. But as they dialed down, a relationship emerged at the individual channel scale.
Approximately 2,800 measurements were used to determine that for weakly confined or unconfined systems with a low gradient, the average thickness of an individual channel was 17 feet. Yet in a higher gradient system, individual channels had an average thickness of 24 feet – a 30-percent increase in width.
“There is a strong relationship between gradient and channel element thickness,” Sullivan said. “The thickness of the individual channels and sheets directly reflects the reservoir that they are a part of.”
Sullivan stressed that the thickness of the channel can be used as a criterion for inferring relative gradient and reservoir type at the complex scale. Thinner channels suggest a low gradient, while thicker channels suggest a higher gradient. Therefore, a low gradient indicates a weakly confined or unconfined reservoir and broad fairway, while a high gradient indicates a confined reservoir with a narrower fairway.
“Scale is key. Explorationists typically look at the large-scale, entire depositional systems, but we don’t observe a relationship between reservoir fairway width and the thickness of the total reservoir,” Sullivan said. “But let’s look at the thickness of the individual channels in that reservoir. It is at the channel scale where that relationship between thickness and relative gradient is observed. The channels are the building blocks and best reflect the relative gradient of the system.”
Chevron is currently using applications from its database for exploration and appraisal in the Miocene formation in the Gulf of Mexico/America and in the Niger Delta, yet Sullivan said it can be used for deepwater reservoir characterization around the world.
“When we drill, we want to know as soon as possible if we have a broad or narrow fairway so we can determine the economic viability of a project,” he said.
For Sullivan, in the absence of seismic imaging and other data, going back to the first principles that affect controlled reservoir distribution in deep water can be a viable solution.
“We need to trust machine learning and AI, but we have to be able to validate it. This gives us an independent validation,” he said. “You need to understand why sand is where it is in situations where there is not enough data. Sometimes you cannot depend on a computer for an answer.”