Quantifying uncertainty of geological 3D layer models, constructed with a-priori geological expertise

Quantifying uncertainty of geological 3D layer models, constructed with a-priori geological expertise

Gunnink, J.J.
Maljers, D.
Hummelman, J.

2010

Uncertainty quantification of geological models that are constructed with additional geological expert-knowledge is not straightforward. To construct sound geological 3D layer models we use a lot of additional knowledge, with an uncertainty that is hard to quantify. Examples of geological expert knowledge are trend surfaces that display a geological plausible basin, additional points that guide the pinching out of geological formations along its depositional extent, etc. All the added geological knowledge, together with the stringent assumptions of normality and second-order stationarity, makes the kriging standard error in our modeling not usable as a measure of uncertainty. We developed a procedure to quantify the uncertainty of our geological 3D layer model that uses cross-validation in a moving window environment to calculate mean deviations and standard errors on a sub-regional scale. Subsequently, we rescaled the x-validation standard error to account for local data configuration and clustering. Summary statistics (Root Mean Squared Prediction Error, Root Mean Error Variance and prediction interval) indicate that there is no bias in the geological model estimation and that the absolute values are trustworthy. An additional check on the above described results was provided by a spatial bootstrapping procedure. Based on 100 bootstrap samples that were "redrilled" in the model, the variance was not comparable to the cross-validated results. To validate the results of the uncertainty quantification we used a sample of (6% randomly selected) drillings as an independent dataset. Results indicate that for datasets with lots of data, the uncertainty quantification provided satisfying results, in terms of RMSE and RMEV. In cases of sparse data, setting aside 6% of the drillings leads to unfavorable statistics, indicating that a minimum of datapoints is needed to obtain a reliable quantification of uncertainty. Organisations: Golder Magyarorszag ZRt.; Mecsekerc Zrt.; MOL

EELS - Earth, Environmental and Life Sciences
Earth / Environmental
Energy / Geological Survey Netherlands
Geological Survey Netherlands
Cross-validation
Geological modeling
Uncertainty quantification
D region
Error statistics
Geologic models
Additional knowledge
Cross validation
Geological formation
Geological modeling
Measure of uncertainty
Prediction interval
Summary statistic
Uncertainty quantifications
Uncertainty analysis
GM - Geomodelling

http://resolver.tudelft.nl/uuid:33ffe9aa-79b2-415b-b11b-b40b431026da

507177

WECO Travel Ltd

9789630698290

14th Annual Conference of the International Association for Mathematical Geosciences, IAMG 2010, 29 August 2010 through 2 September 2010, Budapest, 1-13

conference paper