Uncertainty in seismic interpretation - what factors influence interpretational ability? Clare Bond [email protected] Zoe Shipton, Alan Gibbs, Serena Jones, Rebecca Lunn, Euan Macrae, Frank Richards, Nicholas Richardson
Uncertainty in seismic interpretation - what factors influence interpretational ability?
Clare [email protected]
Zoe Shipton, Alan Gibbs, Serena Jones, Rebecca Lunn, Euan Macrae, Frank Richards, Nicholas Richardson
Politics
Fashion
Fieldwork as a training toolEnables us to make predictions – a natural lab for formulating and testing hypotheses
Twice folded quartzite beds in Donegal Bog - Geoff Tanner
Fieldwork
Geological Datasets - spatially limited data, collected in different dimensions
Often data is collected by remote sensing
Uncertainty
Hard data and soft data is combined and used to make predictions and interpretations
Data collection, processing and interpretation all have an inherent uncertainty
Data / Imagery is only as good as the model on which it is based
Uncertainty
Concepts - are applied to data during interpretation
Concepts are based on analogues formed from previous experience: direct personal experience and indirect gained from others
Concepts
Interpretation
How uncertain are we?
Data will support a number of solutions
Some solutions are invalid
The valid solutions are non-unique
Non-unique solutions
Non-unique solutions
No, it’s a hanging wall anticline with growth strata
It’s an inverted roll-over
It’s……. time for beer
1975 2D Seismic Line though South Brae
Non-unique solutions
1981
Geological cross-section through South Brae
Non-unique solutions
Harms et al.
1983Planar Faults
Non-unique solutions
Gibbs
Non-unique solutions
1984 Alternative Graben Margin Structures – Listric Faults
Gibbs
1988North Brae & Beinn Structures - Dip-Slip Extensional Model
Beinn Field
North Brae Field
NW SE
Non-unique solutions
Beach Associates
Line 541
BALDER
LOWER CRETACEOUS
EKOFISK
BCU
Base Cretaceous Time Structure
N.Brae
C.Brae
S.Brae
Non-unique solutions
1991Inversion
Russell
Changing the Brae concept modeldid not change the outcome
But, changed the thought processes and understanding
Impact on outcome?
Salt loading & doming
x2 vertical exaggeration
ENEWSW
Extension and slumping on top saltSalt domes – passive down-building
16
0
Dep
th, k
mConcept change - revolution
Salt welds
x2 vertical exaggeration
ENEWSW
36km extension on northern end of Miocene ‘trough’
Good quality deepwater depth seismic data images compressional belt at the deformation front
extension, rafts, thin/welded autochthonous saltcompression, salt canopies
16
0
Dep
th, k
mConcept change - revolution
Changing the Gulf of Mexico concept model changed the outcome
And the thought processes and understanding
Kuhn, T.S. 1962 The structure of scientific revolutions
Impact on outcome!
Silver Pit
Controversy and uncertainty – non unique solutions
Salt tectonics?
Impact tectonics?
Stewart & Allen (2002) Nature
Underhill (2004) Nature
Psychology
Tverskey and Khaneman, Science (1974)"Judgment Under Uncertainty: Heuristics and Biases”
Khaneman – nobel laureate (economics) for his work with Tverskey in 2002.
Looked at the psychologies of making decisions from uncertain datasets – including the use of ‘rules of thumb’ (heuristics) and the impact of bias, introducing the idea of anchoring.
Psychology
Heuristics and biasesHuman Bias DescriptionAvailability bias The decision, model or interpretation that is most readily to
mind, or dominant
Confirmation bias To seek out opinions and facts that support ones own beliefsand hypotheses
Anchoring bias Failure to adjust from experts’ beliefs, dominant approaches or initial ideas
Optimistic bias It won’t happen to me mentality (or there is definitely oil in this prospect!)
Positive outcome bias Wanting things to turn out for the best and putting the most positive spin on data interpretation
Hypothesis testing bias Starting with an initial hypothesis and trying to fit the data to it (similar to confirmation bias)
Bond et al., (2008) after Krueger and Funder (2004)
Human bias
Filling in the Gaps
human bias and conceptual models
Filling in the Gaps
human bias and conceptual models
human bias and conceptual models
A small experiment
Whos Biased?
Was it an impact crater?
human bias
It’s my sweet Satan. The one who’s little path would make me sad whose power is Satan.
Oh he’ll give you, give you, 666.
There was a little toolshed, where he made us suffer, sad Satan.
www.reversespeech.com
Is it an impact crater?
We are all biased
Stewart (1999)
Petroleum Geoscience
1:2 chance of an impact crater in the North Sea
Stewart open to the idea of an impact crater
We are all biased
We are all biased
What Allen saw looked like a crash site. “Iwas flabbergasted”, he says. “I’d neverseen anything like it”. It wasn’t until ameeting with Simon Stewart, a BPstructural geologist who also thought itlooked like a crater, that Allen took the ideaseriously.
Silver Pit
We are all biased
Salt tectonics?
Impact crater?
expectation
Fashion Human Bias
Experience Knowledge Personality Psychology
Interpretation is subjective
Subjectivity
Opportunities for Uncertainty
• Objective and subjectivity uncertainty in seismic image data and interpretation
Image: VSA
Plenty of opportunity for uncertainty in industrial structural geology – both objective and subjective
Why Seismic Images?• Commonly used in geoscience industry
• Represent the uncertainty common in geological datasets
How do geoscientists deal with the uncertainty of there not being a right answer?
Human nature - the ditherer and the decisive... Are you someone who draws dashed lines on your map and dare not ink it in?Or do you... grab the pen and go for it, potentially regretting it later?
Not clear that we understand well what makes an individual effective at dealing with such uncertainty.
Never mind how to teach,or deal with it.
That’s why it’s called interpretation
Geology and Petroleum GeologySchool of Geosciences
Rankey and Mitchell, The Leading Edge (2003). That’s why it’s called interpretation: Impact of horizon uncertainty on seismic attribute analysis.
Rankey and Mitchell (2003)
“Artistic license of the interpreter will be revoked entirely by the year 2000.... There will be so much data [in 3D seismic] that you will simply follow the mechanics of what the data set reveal to you.” Marion Bone, SEG Past-President (quoted in Johnston, TLE, 1994)
Rankey and Mitchell (2003)
Geology and Petroleum GeologySchool of Geosciences
Key findings“Seismic interpretations likely are based on previous experiences, preconceived notions, types of data available, data quality, and geologic understanding.”
“Uncertainty in interpretation should be evaluated and minimized using geologic insight, analysis of end-member possibilities, and seismic modeling.”
Evidence for anchoring “I did ... not want to change any of my picks based on the additional well data—looks like I had it nailed.” - but only on a small sample of 6 interpreters.
Other papers
Geology and Petroleum GeologySchool of Geosciences
Polson and Curtis (2010)Seismic interpretation – elicitationDynamics of uncertainty in interpretation, the spread of prior knowledge and subjectivity of individuals, disagreement and confidence (4 individuals).
Bond et al. (2008) and Rowbotham et al. (2010)Multiple scenarios advocated, rather than single deterministic models
Bond et al. (2011)Seismic interpretation – experience cohortsAnalysis of the change in approach and confidence of individuals interpreting with experience (36 individuals).
Chellingsorth et al. (2011)Top reservoir map from wells and outcrop – 4 individuals to calculate volumes.
Richards et al. (2015)Geo seismic interpretation – by 8 groups to create a top reservoir map from an intersecting 2D grid.. Risks to predicted volumes highlighted.
How many different interpretations?
Experimental evidence
Seismic created from known model
Odin project used synthetic seismic based on a structural model created by forward modelling to test subjective spread in interpretation
Synthetic seismic
Experiment 1
What influences the interpretation?
– Competence
– Level of Experience
– Field of expertise
– Personality and problem solving approach
– “Cultural” background
Experiment 1
One dataset – many conceptsExperiment 1
Many structural modelsExperiment 1
1) multiple conceptual models can be applied to the samedataset (conceptual uncertainty)
Experts - model or data driven?
24 experts at the AAPG DWFTB Hedberg 2009 interpreted the seismic dataset.
Concepts vs data
Expert- experienceStructural geology experience -specialist c.60%
Seismic interpretation experience -good working knowledge c. 65%
Interpretation frequency - daily c. 40%
Experiment 2
Torvela and Bond (2011). Journal of Structural Geology
Experiment 2
Diversity and Spatial Correlation of Interpretations
Different styles and formsDifferent interpretations of faults and distributed deformationDifferent implications for the structural evolution
Experiment 2
Dominating models
Trishear model most dominant.
Experimental 2
2) even when the concept is universally recognised (e.g. fold-thrust belt) and the imagery is of good quality large uncertaintiesin fault placement exist and the resultant implied mechanisms of structural evolution cover a broad range.
Effect of Prior Knowledge?
Student –MSc sequence stratigraphy
Student – PhD salt tectonics
Anecdotal – evidence from the results….
But what about statistically?
Experimental evidence – experiment 1
Effect of Prior KnowledgeDominant Tectonic Expertise - more likely than others to produce an interpretation based on this expertise (i.e. dominant thrust tectonic experience - 29% produced a thrust interpretation, compared to 27% of participants with other expertise). BUT not statistically significant.
Length of Experience - had no obvious overall effect (i.e. students were just as likely as those with 15+ years experience to produce an incorrect interpretation, 76%).
Experiment 1
23%27% 35%
184
Self-defined experts in structural geology
445
Experiment 1
TechniquesEffective expertsuse lots of techniques.
100%
79%67%
23%18%
35%
Experiment 1
Specific Techniques
35%
94%
44%51%45%37%
10%
Effective expertsused specific techniques –notably thoughts about the geological evolution (reasoning).
Experiment 1
Statistically significant factors amongst structural geologistsMaster’s or Ph.D. degree 0.0055 Odds Ratio: 7.16Horizons 0.0010 Odds Ratio: 4.13Annotation 0.0107 Odds Ratio: 2.69Geological evolution 0.0007 Odds Ratio: 40.51
Experiment 1
“those who used evolutionary thought processes in the form of sketches or textwere >40 times as likely to produce the correct interpretation than those that had no Master’s or Ph.D. degree and used none of the three significant techniques, which translated into 94% of experts who used it being successful.” Bond et al., 2012.
Bond et al. (2012), Geology.
Non-expertsEveryone can be an expert?
27%
87%
37%40%
7%
38%30%
Experiment 1
Everyone can be effective by using multiple techniques to query the data and applying specific validation techniques (reasoning).
But not many people are:
of the 184 experts only 18 (c.10%) showed evidence of thinking about the geological evolution.
Odin Experiment - ConclusionsExperiment 1
Geological Reasoning
Reasoning
Frodeman (1995) Geological reasoning: Geology as an interpretive and historical science
Regional and horizon correlation - seismic strat’ matching
A B
AB
Geological Reasoning
Reasoning
AB
3) experience is not a substitue for good technique use, or more specifically the application of geological reasoning skills.
Geological Reasoning – is it really effective?
Reasoning - the test
Euan Macrae (2013). Unpublished PhD thesis
Reference Experts 1-3Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Reference Experts 4-6Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Reference Expert’s Key Features
Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Key Feature Score’s for 444 interpreters
Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Controlled experiment
Control group – “Please interpret whole seismic image” as the main surveyTest group - “1. Interpret the whole seismic image. Please focus your interpretation on the geological evolution of the section. 2. Summarise the geological evolution below.”
Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Key Feature Score’s for controlled experiment
Experiment 3
Euan Macrae (2013). Unpublished PhD thesis
Explicit reasoning
4) use of effective techniques results in better seismicinterpretation outcomes.
Evidence from our controlled workshop experiment suggests that reasoning through the geological evolution should be an explicit process, to ensure that it is undertaken fully, requiring geoscientists to challenge themselves to verify, and hence produce better, interpretations
Creative space for reasoning
Creativity
Bond et al. (2015)
More white space – better models?
Creativity
More white space – better models?
Creativity
69%
21%
Top reservoir maps
2d-3d
8 groups – 8 top reservoir maps
Posted Data from seismic Interpretation Top Reservoir point depth posted at grid intersection points and interpreted contours. Units = meters.
860780 825765770
770 790 820 845 880
795 810 835 870 910
800 820 850 880 920
800
800
750
825 860 895 920 960
955
940
920
900700
770
780
890 890
Richards et al., (2015)
Simple seismic sections2d-3d
Top Reservoir
S N
Mapped fault variability2d-3d
A B
4-5
3
2
1
0
Number of fault cuts
3 4
0
50
100
150
200
250
65 6556 56 54 54 53 53 50 50 53 53 58 58
18 1825 25
20 32
17 28
27 27 32 32 18 18 7 7
1012
9
2
1316 16 14
7 713
57 7
9.5 202
28
28
10
18
12
40 13
3331
31
614
10
109
9
10
10
10
11
11
12
20 20
10
10
5
105
5
4.7
9
4
3
3
6
4
4
5
4
2
15
5
3
20
19
19
A B1 B2 C D E1 E2
E3 F G H I J K
L M N O P
A AA
A A AA
B B
BB
B B B
EE
EE
E
MAP 1
> 20m throw
MAP 2
> 20m throw
MAP 3
> 20m throw
MAP 4
> 20m throw
MAP 5
> 20m throw
MAP 6
> 20m throw
MAP 7
> 20m throw
MAP 8
> 20m throw
KM2
G/3E/7
C/7
F/4H/3
I/3
A/7
B/8
D/3
K/3J/4
L/3M/2
N/1
O/1
P/1
A
CC
α α α α ααα α
B
No. and volumetrics of traps
Risking – top reservoir maps2d-3d
5) reasoning techniques can be used to risk models.
The multiple interpretation challenge
Creativity
Rabbit or duck Or Duck and rabbit?
The Importance of Creativity Wiseman et al. (2011)
Generating• Using one or more geologists with a range of prior knowledge,• Exposing geologists to a range of concepts prior to interpretation,• Removing regional and tectonic context, and• Encouraging multiple interpretations.
Assessing• Using structural reasoning evolution and restoration techniques to determine
model viability,• Considering regional and tectonic context,• Use of peer review and specialist technical assurance, and• Assessment of play impact.
Generating and Assessing Interpretations
Improved outcomes
Thoughts
Interpret datasets with the integrity of the final solution in mind.Don’t need to be a specialist - but do need to be able to apply knowledge, concepts, techniques - reasoning to the problem.Ability to question the data and the model - fresh eyes and creativity.
“It is likely that this type of reasoning will become more crucial in the next century. Many of the issues we face (global warming and various types of
risk and resource assessment) are by their nature both scientific and ethical, with the scientific aspect of the problem deeply influenced by
interpretation and uncertainty.”
Frodeman (1995) on Geological Reasoning