Relating Depositional Environments to Logs
Purpose
- Different descriptions of reservoir rocks include:
- Geology (facies, depositional environments)
- Core Analysis (porosity, permeability, saturations,
etc.)
- Petrophysics (log responses)
- Engineering (test analysis, etc.)
- Are the interpretations consistent and correlative?
- Which models are appropriate?
- How do we integrate the approaches?
This approach is oriented to combining geology, core analysis and petrophysics
into a single and coherent reservoir model.
Goals
- Take existing core descriptions of depositional
environments and relate to raw log responses.
- Extrapolate, from neural network training, to
intervals and wells where cores do not exist.
- Evaluate how consistently neural networks can
predict depositional environments and establish limitations to this approach.
- Define correlations between depositional environments,
core measurements of porosity and permeability, and calculated petrophysical
parameters.
- Establish guidelines for combining depositional
types that are reflected in raw log informational differences.
- Show where neural network categorization fails,
due to the presence of different rock types.
Requirements
- Consistent and reliable core descriptions.
- Accurate depth calibration of cores to open-hole
logs.
- Consistent and well calibrated log suites among
all wells in the study. As necessary, “pseudo” logs may be created using
neural networks or other methods.
- Logs where bad hole effects have been correctly
edited.
- For this approach, logs that are affected minimally
by fluids. No resistivity logs were used for this application.
Examples
Hartzog Draw, Wyoming Clastic Depositional Environment
For three cored wells, identified six different depositional facies
as described from the cored interval:
- Shelf (Shale)
- Shelf (Sandstone/Siltstone)
- Inter-Ridge
- Low Energy Ridge
- High Energy Ridge
- Central Ridge
Location
Geologic Model
Training
Trained the neural network using the Bud Christensen #2 and the Federal
AS #1 to recognize the six different facies using:
- Density
- Neutron
- Sonic
- Gamma Ray
Cross Section of Results
Billings Nose North Dakota Carbonate Depositional Environment
For the cored well (Stuart 1-19) identified six different depositional
facies as described from the cored interval:
- Open Marine
- Restricted Marine
- Intertidal
- Supratidal Flat
- Supratidal - Extended for other intervals using
log response only
- Salt - Extended for other intervals using log
response only
Location
Geologic Model
Training
Trained the neural network to recognize these six different facies using:
- Density
- Neutron
- Sonic
- Gamma Ray
Did not use resistivity, since fluid content is not of interest for
this approach. The neural network resulted in a 96.5% comparison of the
training region with the core descriptions.
Cross Section of Results
Conclusions
- Neural networks can be successfully applied using
raw logs to correlate to depositional environments in both clastic and
carbonate/evaporitic sequences.
- There is a limit to the number of different environments
that can be recognized.
- The different environments should have component
and/or textural differences, such as:
Clastics
- Shale content
- Grain size
- Porosity
- Cementation
Carbonate/Evaporates
- Mineralogy
- Shale content
- Porosity
- When the neural network encounters a rock type
which is “new” (i.e., not represented in the cored interval) results become
demonstrably unreliable.
- If logs are to be used for categorization away
from cored intervals, results from the neural network classification indicate
which depositional categories might need to be combined.
- For these examples, reliability of neural networks
for depositional environment prediction ranges from 36% (very poor) to
93% (very good). If environmental types are combined into more general
categories, the predictions are in excess of 85%.
See Also