Assessment of multiple point simulation quality focusing on connected geobodies

Guillaume Rongier and Pauline Collon and Philippe Renard and Julien Straubhaar and Judith Sausse. ( 2014 )
in: GeoEnv 2014 Proceedings - 10th conference on Geostatistics for Environmental Applications, 9th-11th July, 2014, Paris, France

Abstract

Multiple-point simulations (MPS) are booming stochastic simulation methods due to their ability to better take into account higher-order statistical structures than classic variogram-based approaches. They borrow the multiple-point statistics not from the data but from an external representation of the expected geology. However, there is no objective way to evaluate the quality of the reproduction of the geological structures and realizations are generally reviewed based on a visual quality control. If studying the uncertainties calls for multiple realizations, all those realizations are not equal in term of quality depending on the structures to reproduce and the chosen methods and/or parameters. In this work, we propose a general methodology to objectively assess the quality of a realization and distinguish unsatisfying ones in terms of structure reproduction, saving their useless treatment. This quality assessment is based on several indicators used to compare the realizations with a reference image, corresponding to the training image when using MPS. In addition to classic indicators such as the facies proportions, we propose to check parameters linked to the connected geobodies. Among them, some describe the spatial repartition of the connected geobodies (e.g. density, proportion of crossing geobodies,...) while others give information on their global shape (e.g. volume, box volume ratio,...) or on their topology though a skeleton extraction. To facilitate the quality analysis, we propose to compute dissimilarities between the images from those indicators. Then, the dissimilarity matrix is analyzed using heat maps and two-dimensional representations based on multidimensional scaling. The application of this methodology to a synthetic case and various associated realizations gives rise to some practical considerations. Whereas multidimensional scaling is a powerful visualization tool, it induces some mis-representations of the dissimilarities and should only be used for a first-order analysis. Details considering the relationship between the realizations and the methods should be preferably analyzed on the heat map as it represents directly the dissimilarities. If the visualization and analysis process of the dissimilarities is quite satisfying, further work should be done to improve the indicator capacity to capture the realization characteristics.

Download / Links

    BibTeX Reference

    @INPROCEEDINGS{,
        author = { Rongier, Guillaume and Collon, Pauline and Renard, Philippe and Straubhaar, Julien and Sausse, Judith },
         title = { Assessment of multiple point simulation quality focusing on connected geobodies },
         month = { "jul" },
     booktitle = { GeoEnv 2014 Proceedings - 10th conference on Geostatistics for Environmental Applications,  9th-11th July, 2014, Paris, France },
          year = { 2014 },
      abstract = { Multiple-point simulations (MPS) are booming stochastic simulation methods due to their ability to better take into account higher-order statistical structures than classic variogram-based approaches. They borrow the multiple-point statistics not from the data but from an external representation of the expected geology. However, there is no objective way to evaluate the quality of the reproduction of the geological structures and realizations are generally reviewed based on a visual quality control. If studying the uncertainties calls for multiple realizations, all those realizations are not equal in term of quality depending on the structures to reproduce and the chosen methods and/or parameters. In this work, we propose a general methodology to objectively assess the quality of a realization and distinguish unsatisfying ones in terms of structure reproduction, saving their useless treatment. This quality assessment is based on several indicators used to compare the realizations with a reference image, corresponding to the training image when using MPS. In addition to classic indicators such as the facies proportions, we propose to check parameters linked to the connected geobodies. Among them, some describe the spatial repartition of the connected geobodies (e.g. density, proportion of crossing geobodies,...) while others give information on their global shape (e.g. volume, box volume ratio,...) or on their topology though a skeleton extraction. To facilitate the quality analysis, we propose to compute dissimilarities between the images from those indicators. Then, the dissimilarity matrix is analyzed using heat maps and two-dimensional representations based on multidimensional scaling. The application of this methodology to a synthetic case and various associated realizations gives rise to some practical considerations. Whereas multidimensional scaling is a powerful visualization tool, it induces some mis-representations of the dissimilarities and should only be used for a first-order analysis. Details considering the relationship between the realizations and the methods should be preferably analyzed on the heat map as it represents directly the dissimilarities. If the visualization and analysis process of the dissimilarities is quite satisfying, further work should be done to improve the indicator capacity to capture the realization characteristics. }
    }