TY - JOUR
T1 - A locally adaptive kernel regression method for facies delineation
AU - Fernàndez-Garcia, D.
AU - Barahona-Palomo, M.
AU - Henri, C.V.
AU - Sanchez-Vila, X.
N1 - Funding Information:
This work has been supported by the Spanish Ministry of Science and Innovation through project FEAR (CGL2012-38120) and the European Research Council through project MARSOL (Grant 619120 ). XS also acknowledges support of Program ICREA Acadèmia.
Publisher Copyright:
© 2015 Elsevier B.V.
PY - 2015/12/1
Y1 - 2015/12/1
N2 - Facies delineation is defined as the separation of geological units with distinct intrinsic characteristics (grain size, hydraulic conductivity, mineralogical composition). A major challenge in this area stems from the fact that only a few scattered pieces of hydrogeological information are available to delineate geological facies. Several methods to delineate facies are available in the literature, ranging from those based only on existing hard data, to those including secondary data or external knowledge about sedimentological patterns. This paper describes a methodology to use kernel regression methods as an effective tool for facies delineation. The method uses both the spatial and the actual sampled values to produce, for each individual hard data point, a locally adaptive steering kernel function, self-adjusting the principal directions of the local anisotropic kernels to the direction of highest local spatial correlation. The method is shown to outperform the nearest neighbor classification method in a number of synthetic aquifers whenever the available number of hard data is small and randomly distributed in space. In the case of exhaustive sampling, the steering kernel regression method converges to the true solution. Simulations ran in a suite of synthetic examples are used to explore the selection of kernel parameters in typical field settings. It is shown that, in practice, a rule of thumb can be used to obtain suboptimal results. The performance of the method is demonstrated to significantly improve when external information regarding facies proportions is incorporated. Remarkably, the method allows for a reasonable reconstruction of the facies connectivity patterns, shown in terms of breakthrough curves performance.
AB - Facies delineation is defined as the separation of geological units with distinct intrinsic characteristics (grain size, hydraulic conductivity, mineralogical composition). A major challenge in this area stems from the fact that only a few scattered pieces of hydrogeological information are available to delineate geological facies. Several methods to delineate facies are available in the literature, ranging from those based only on existing hard data, to those including secondary data or external knowledge about sedimentological patterns. This paper describes a methodology to use kernel regression methods as an effective tool for facies delineation. The method uses both the spatial and the actual sampled values to produce, for each individual hard data point, a locally adaptive steering kernel function, self-adjusting the principal directions of the local anisotropic kernels to the direction of highest local spatial correlation. The method is shown to outperform the nearest neighbor classification method in a number of synthetic aquifers whenever the available number of hard data is small and randomly distributed in space. In the case of exhaustive sampling, the steering kernel regression method converges to the true solution. Simulations ran in a suite of synthetic examples are used to explore the selection of kernel parameters in typical field settings. It is shown that, in practice, a rule of thumb can be used to obtain suboptimal results. The performance of the method is demonstrated to significantly improve when external information regarding facies proportions is incorporated. Remarkably, the method allows for a reasonable reconstruction of the facies connectivity patterns, shown in terms of breakthrough curves performance.
KW - Geological facies reconstruction
KW - Geostatistics
KW - Solute transport
KW - Stochastic groundwater hydrology
UR - http://www.scopus.com/inward/record.url?scp=84946559604&partnerID=8YFLogxK
U2 - 10.1016/j.jhydrol.2015.09.066
DO - 10.1016/j.jhydrol.2015.09.066
M3 - Article
AN - SCOPUS:84946559604
SN - 0022-1694
VL - 531
SP - 62
EP - 72
JO - Journal of Hydrology
JF - Journal of Hydrology
IS - Part 1
ER -