3D stochastic inversion of potential field data using structural geologic constraints

Shamsipour, Pejman; Schetselaar, Ernst; Bellefleur, Gilles; Marcotte, Denis

doi:10.1016/j.jappgeo.2014.09.013

Cited by 6 publications

(3 citation statements)

References 28 publications

(22 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many authors incorporate prior geological information in different forms to constrain the inversion algorithms in order to derive a physical model that satisfies the expected geology and the geophysical data (Boulanger and Chouteau ; Bosch and McGaughey ; Fullagar and Pears ; Lelievre ; Shamsipour et al . ). Among the inversion methods, the algorithm developed by Li and Oldenburg (, ) can obtain reliable results and has been successfully applied to several mineral exploration problems (Phillips ; Williams and Dipple ; Farquharson, Ash and Miller ; Williams ; Spicer, Morris and Ugalde ).…”

Section: Introductionmentioning

confidence: 97%

Using constrained inversion of gravity and magnetic field to produce a 3D litho‐prediction model

Mahmoodi

Smith

Spicer³

2017

Geophysical Prospecting

View full text Add to dashboard Cite

Geologically constrained inversion of gravity and magnetic field data of the Victoria property (located in Sudbury, Canada) was undertaken in order to update the present three‐dimensional geological model. The initial and reference model was constructed based on geological information from over 950 drillholes to constrain the inversion. In addition, downhole density and magnetic susceptibility measured in six holes were statistically analysed to derive lower and upper bounds on the physical properties attributed to the lithological units in the reference model. Constrained inversion of the ground gravity and the airborne magnetic data collected at the Victoria property were performed using GRAV3D and MAG3D, respectively. A neural network was trained to predict lithological units from the physical properties measured in six holes. Then, the trained network was applied on the three‐dimensional distribution of physical properties derived from the inversion models to produce a three‐dimensional litho‐prediction model. Some of the features evident in the lithological model are remnants of the constraints, where the data did not demand a significant change in the model from the initial constraining model (e.g., the thin pair of diabase dykes). However, some important changes away from the initial model are evident; for example, a larger body was predicted for quartz diorite, which may be related to the prospective offset dykes; a new zone was predicted as sulfide, which may represent potential mineralisation; and a geophysical subcategory of metabasalt was identified with high magnetic susceptibility and high density. The litho‐prediction model agrees with the geological expectation for the three‐dimensional structure at Victoria and is consistent with the geophysical data, which results in a more holistic understanding of the subsurface lithology.

show abstract

Section: Introductionmentioning

confidence: 97%

Using constrained inversion of gravity and magnetic field to produce a 3D litho‐prediction model

Mahmoodi

Smith

Spicer³

2017

Geophysical Prospecting

View full text Add to dashboard Cite

show abstract

“…How best to extract information from auxiliary variable grids for geostatistical modelling tasks has remained an open question, but has often involved trial-anderror experimentation using manually designed filters to extract features with as much explanatory power as possible (e.g., Ruiz-Arias et al 2011;Poggio et al 2013;Parmentier et al 2014;Shamsipour et al 2014;Kirkwood et al 2016;Kirkwood 2016;Young et al 2018;Lamichhane et al 2019). For example, Youssef et al (2016) use slope angle derived from a digital terrain model as a feature to explain landslide susceptibility, but many more complex features may be useful, and these are not necessarily known in advance.…”

Section: Introductionmentioning

confidence: 99%

Bayesian Deep Learning for Spatial Interpolation in the Presence of Auxiliary Information

et al. 2022

View full text Add to dashboard Cite

Earth scientists increasingly deal with ‘big data’. For spatial interpolation tasks, variants of kriging have long been regarded as the established geostatistical methods. However, kriging and its variants (such as regression kriging, in which auxiliary variables or derivatives of these are included as covariates) are relatively restrictive models and lack capabilities provided by deep neural networks. Principal among these is feature learning: the ability to learn filters to recognise task-relevant patterns in gridded data such as images. Here, we demonstrate the power of feature learning in a geostatistical context by showing how deep neural networks can automatically learn the complex high-order patterns by which point-sampled target variables relate to gridded auxiliary variables (such as those provided by remote sensing) and in doing so produce detailed maps. In order to cater for the needs of decision makers who require well-calibrated probabilities, we also demonstrate how both aleatoric and epistemic uncertainty can be quantified in our deep learning approach via a Bayesian approximation known as Monte Carlo dropout. In our example, we produce a national-scale probabilistic geochemical map from point-sampled observations with auxiliary data provided by a terrain elevation grid. By combining location information with automatically learned terrain derivatives, our deep learning approach achieves an excellent coefficient of determination ($$R^{2} = 0.74$$ R 2 = 0.74 ) and near-perfect probabilistic calibration on held-out test data. Our results indicate the suitability of Bayesian deep learning and its feature-learning capabilities for large-scale geostatistical applications where uncertainty matters. Graphic Abstract

show abstract

“…Many methods can be used to address the non-uniqueness and instability problem in potential field data inversion (e.g., Oldenburg 1996, 1998;Oldenburg and Pratt 2007;Ganguli and Dimri 2013;Ganguli et al 2015). , Dong 1990;Haas and Dubrule 1994;Torres-Verdin et al 1999;Bosch and McGaughey 2001;Chasseriau and Chouteau 2003;Gloaguen et al 2005;Hansen et al 2006;Giroux, Gloaguen and Chouteau 2007;Calcagno et al 2008;Guillen et al 2008;Shamsipour et al 2010Shamsipour et al , 2011Shamsipour, Marcotte and Chouteau 2012;Shamsipour et al 2014). There are two mainstream inversion formulations that are always applied in many potential field data inversions.…”

mentioning

confidence: 99%

Three‐dimensional potential field data inversion with L0 quasinorm sparse constraints

Meng¹

2017

Geophysical Prospecting

View full text Add to dashboard Cite

The quantitative explanation of the potential field data of three‐dimensional geological structures remains one of the most challenging issues in modern geophysical inversion. Obtaining a stable solution that can simultaneously resolve complicated geological structures is a critical inverse problem in the geophysics field. I have developed a new method for determining a three‐dimensional petrophysical property distribution, which produces a corresponding potential field anomaly. In contrast with the tradition inverse algorithm, my inversion method proposes a new model norm, which incorporates two important weighting functions. One is the L0 quasi norm (enforcing sparse constraints), and the other is depth‐weighting that counteracts the influence of source depth on the resulting potential field data of the solution. Sparseness constraints are imposed by using the L0 quasinorm on model parameters. To solve the representation problem, an L0 quasinorm minimisation model with different smooth approximations is proposed. Hence, the data space (N) method, which is much smaller than model space (M), combined with the gradient‐projected method, and the model space, combined with the modified Newton method for L0 quasinorm sparse constraints, leads to a computationally efficient method by using an N × N system versus an M × M one because N ≪ M. Tests on synthetic data and real datasets demonstrate the stability and validity of the L0 quasinorm spare norms inversion method. With the aim of obtaining the blocky results, the inversion method with the L0 quasinorm sparse constraints method performs better than the traditional L2 norm (standard Tikhonov regularisation). It can obtain the focus and sparse results easily. Then, the Bouguer anomaly survey data of the salt dome, offshore Louisiana, is considered as a real case study. The real inversion result shows that the inclusion the L0 quasinorm sparse constraints leads to a simpler and better resolved solution, and the density distribution is obtained in this area to reveal its geological structure. These results confirm the validity of the L0 quasinorm sparse constraints method and indicate its application for other potential field data inversions and the exploration of geological structures.

show abstract

3D stochastic inversion of potential field data using structural geologic constraints

Cited by 6 publications

References 28 publications

Using constrained inversion of gravity and magnetic field to produce a 3D litho‐prediction model

Using constrained inversion of gravity and magnetic field to produce a 3D litho‐prediction model

Bayesian Deep Learning for Spatial Interpolation in the Presence of Auxiliary Information

Three‐dimensional potential field data inversion with L0 quasinorm sparse constraints

Contact Info

Product

Resources

About