Description

One challenge that Exploration Geoscientists face when looking for oil and gas deposits is interpreting how a rock was formed. Were these sands deposited by a river? Is this mudstone from the bottom of an ancient sea? The gamma ray logging tool (GR) can be used in wellbores to answer this question. GR measures the radioactivity of rocks and can differentiate sandstones versus the more radioactive mudstones. Over the decades geoscientists have correlated some common shapes in GR with the rock’s interpreted depositional environments (i.e. rivers). For example, the “bell” geometry in Figure 1 is often associated with sandy beaches (Kendall, 2003). These GR geometries, also known as log facies, are combined with other geoscience data to build a picture of how different rock layers were formed and make predictions about where to look for resources within them.

(from Kendall, 2003)

Despite its usefulness the interpretation of log facies from GR is still a subjective and manual process which can take an Explorer weeks to do. An algorithmic approach to interpretation would allow for a more objective interpretation of larger datasets that would aid Exploration Geoscientists in understanding how environments have changed through time.

The goal of this challenge is to identify the different log facies types on dataset of GR logs from 6,000 synthetic wells (4,000 train, 2,000 test). Each well has 1,100 rows and a random number of log facies.  Along with GR the user is given a label for four common log facies: 1 - symmetrical, 2 - cylindrical, 3 - funnel, and 4 -bell. The user must develop an algorithm that searches each well individually and generates a label for intervals that match one of the type profiles. Things to consider while building the algorithm:

• Both GR measurements and geology are inherently noisy.

• A log facies can appear anywhere in a well and can have different thicknesses.

• Log facies can appear inside other log facies.

The “row_id” column represents depth in the well, the “well_id” represents one well, “GR” is the gamma log, and “label” is the log facies label. See the Data page to look at an example of the data.

Evaluation Criteria

A successful submission must include a CSV of row_ID, well_id, and label with a notebook of the solution. Submissions will be judged by cross validation with a larger dataset of synthetic GR data.

Citations

Kendall, C.G.St.C. (2003): Character of Log Response; www.sepmstrata.org/page.aspx?pageid=168.