TY - JOUR
T1 - Non-Monotonic Transformation for Gaussianization of Regionalized Variables
T2 - Modeling Aspects
AU - Khorram, Farzaneh
AU - Emery, Xavier
AU - Maleki, Mohammad
AU - País, Gabriel
N1 - Publisher Copyright:
© International Association for Mathematical Geosciences 2024.
PY - 2024
Y1 - 2024
N2 - This paper proposes an extension of the traditional multigaussian model, where a regionalized variable measured on a continuous quantitative scale is represented as a transform of a stationary Gaussian random field. Such a model is popular in the earth and environmental sciences to address both spatial prediction and uncertainty assessment problems. The novelty of our proposal is that the transformation between the original variable and the associated Gaussian random field is not assumed to be monotonic, which offers greater versatility to the model. A step-by-step procedure is presented to infer the model parameters, based on the fitting of the marginal distribution and the indicator direct and cross-covariances of the original variable. The applicability of this procedure is illustrated with a case study related to grade control in a porphyry copper-gold deposit, where the fit of the gold grade distribution is shown to outperform the one obtained with the traditional multigaussian model based on a monotonic transformation. This translates into a better assessment of the uncertainty at unobserved locations, as proved by a split-sample validation.
AB - This paper proposes an extension of the traditional multigaussian model, where a regionalized variable measured on a continuous quantitative scale is represented as a transform of a stationary Gaussian random field. Such a model is popular in the earth and environmental sciences to address both spatial prediction and uncertainty assessment problems. The novelty of our proposal is that the transformation between the original variable and the associated Gaussian random field is not assumed to be monotonic, which offers greater versatility to the model. A step-by-step procedure is presented to infer the model parameters, based on the fitting of the marginal distribution and the indicator direct and cross-covariances of the original variable. The applicability of this procedure is illustrated with a case study related to grade control in a porphyry copper-gold deposit, where the fit of the gold grade distribution is shown to outperform the one obtained with the traditional multigaussian model based on a monotonic transformation. This translates into a better assessment of the uncertainty at unobserved locations, as proved by a split-sample validation.
KW - Gaussian anamorphosis
KW - Geostatistical modeling
KW - Indicator covariance
KW - Multigaussian model
UR - http://www.scopus.com/inward/record.url?scp=85204793033&partnerID=8YFLogxK
U2 - 10.1007/s11053-024-10400-x
DO - 10.1007/s11053-024-10400-x
M3 - Article
AN - SCOPUS:85204793033
SN - 1520-7439
VL - 33
SP - 2567
EP - 2588
JO - Natural Resources Research
JF - Natural Resources Research
IS - 6
ER -