Uncertainty in Geoscience Interpretation: Statistical Quantification of the Factors That Affect Interpretational Ability
Euan Macrae¹, Clare Bond², Zoe Shipton³, and Rebecca Lunn³
¹Senergy, Banchory, UK
²University of Aberdeen, Aberdeen, UK
³University of Strathclyde, Glasgow, UK
Understanding the subsurface through geological modelling is extremely important to modern civilisation, e.g. the extraction of resources and the geological storage of wastes. Geological data are commonly sparse, with the result that geological models are under-constrained and multiple structural interpretations are often valid. Geoscientists are also affected by cognitive biases, so individual interpretations may not be equally likely. A better understanding of how geoscientists should be trained, and what interpretational approaches are most effective, is therefore required.
In this Ph.D. research (2008-2012), an international survey was conducted at conferences, in university departments and in oil and gas companies, gaining a substantial amount of industry participation. Respondents interpreted the same 2D seismic image and completed a questionnaire about their background (i.e. education, work environment and professional experience). The techniques used in respondents' interpretations were defined and analysed. To ensure a high-quality sample, 'inexperienced' respondents were filtered out, leaving 444 respondents with a median experience of 10 years, and different technical backgrounds, for analysis. The filtered sample was considered to represent the population of those geoscientists who work in the USA or Europe.
Respondents' interpretations were compared against the interpretations completed by six hand-picked reference experts (REs) who had a median experience of 24.5 years. Although the RE interpretations could be classified into two distinct tectonic styles, there were differences in their fault geometries, the timing of faults and even in the terminology used. A 'response' variable was thus created to quantify 'how similar' respondents' interpretations were to the reference experts' interpretations. More similar interpretations were deemed to be better. Hence, a multivariate statistical analysis was used to quantify the 'effect' of individual factors (e.g. respondents' education, work environments, experience and the techniques that were used) on the response variable. The most influential factors could therefore be identified in terms of coming up with an interpretation that was similar to one of the RE interpretations.
Key findings included that the significant factors relating to respondents' backgrounds, as captured by the questionnaire, were: 'structural geology experience', 'frequency of seismic interpretation', 'experience working for a super-major or major oil company' and 'range of geographical experience'. Factors relating to respondents' education were found to be non-significant. The interpretational techniques of 'writing about geological time', 'drawing cartoons', 'writing about geological processes', 'explicitly stating the tectonic concept' and 'drawing arrows on faults' were found to be significant even when accounting for the differences in respondents' backgrounds.
Furthermore, the techniques of 'writing about geological time' and 'drawing cartoons' were more significant than all background factors, showing that training geoscientists to use these techniques is important regardless of their experience level. 'Writing about geological time' was the most effective technique; respondents who had applied it were 4.5 times more likely to attain a more similar interpretation than the respondents who had not applied it, regardless of their backgrounds.
In addition to the survey, a separate workshop experiment was conducted to determine how effective the technique of 'writing about geological time' really was. The experiment tested whether there was evidence to suggest that 'writing about geological time' caused better interpretations or whether it was just a feature of good interpretations. The 49 workshop participants had a median experience of 17 years and came from one of four companies in the oil and gas industry. Participants were split into two groups (called 'control' and 'evolution'), and unknown to them, were given different instructions on how to approach the interpretation exercise. As before, participants' backgrounds were captured by the questionnaire. The control group (24 participants) was told to 'interpret the whole seismic image', while the evolution group (25 participants) was told to 'focus on and state the geological evolution' of their interpretation, which is a form of writing about geological time.
The result was surprisingly strong. Although the evolution group were less experienced than the control group, they attained similarity scores that were 62% higher, on average, than the control group. The result was statistically significant, thus establishing a causal link between 'considering the geological evolution' and 'producing a similar interpretation to the reference experts' interpretations'.Finally, the results from this research were combined with relevant literature from geoscience and psychology to produce recommendations that mitigate the risk arising from uncertainty in geoscience interpretation. Hence, at a time when geoscience interpretation is becoming more complex, and ever more important, this research documents how geoscientists can become more effective when working with uncertain data.
AAPG Search and Discovery Article #120140© 2014 AAPG Hedberg Conference 3D Structural Geologic Interpretation: Earth, Mind and Machine, June 23-27, 2013, Reno, Nevada