UNCERTAINTY IN GEOLOGICAL MAPS AND MODELS – RECENT DEVELOPMENTS AT BGS

Geological maps and latterly 3D framework models are the outputs used by GSO’s to communicate their understanding of geology at diverse resolutions. Maps and models divide the area or volume considered into units based on an appropriate classification scheme. Lithostratigraphical schemes predominate. The definition of the unit boundaries, as 2D lines or 3D surfaces is the prime objective. The accuracy of the boundaries have traditionally been depicted on some maps using three line styles that reflect the uncertainty of the boundary, e.g. observed, inferred, conjectural, models may follow suit.

We present results from some recent BGS studies:

1. Uncertainty in geological map linework. We used formal methods to elicit the "tacit model" of uncertainty that experienced field geologists use to interpret mapped boundaries. Uncertainty models for scenarios defined on geological setting, exposure, observation density etc. were refined by structured discussion and feedback.
2. Uncertainty due to differences of interpretation between geologists. A study asked five geologists to model an area of unfaulted Quaternary and Palaeogene geology. Each modeller was provided a unique set of borehole logs from which to build their model, each dataset had boreholes withheld. The resulting models were compared with the validation observations. There was no evidence of significant between modeller variation nor of systematic bias. Future studies will treat more complex geological scenarios with more sparse control data.
3. A Confidence Index has been developed by expert elicitation of data quality and statistical analysis of the spatial complexity of model contacts. Expert elicitation is commonly employed to establish values used in statistical treatments of model uncertainty. The best solution appears to be using panels of experienced geologists to elicit these values. Examples of the Confidence Index plots will be given.

Treatments of uncertainty in maps and models yield relative rather than absolute values even though many of these are expressed numerically. This makes it extremely difficult to devise standard methodologies to determine uncertainty or propose fixed numerical scales for expressing the results. Furthermore, numbers may give a misleading impression of greater certainty than actually exists.