Occam's Inversion

Key Takeaways

• The non-uniqueness problem is the central challenge that Occam's inversion addresses — given a finite number of measurements with finite accuracy, infinitely many subsurface models will reproduce those measurements within the measurement error; without additional constraints, any inversion algorithm will find one of these many equivalent models but has no way of knowing which one is geologically realistic; Occam's inversion resolves this by choosing the smoothest model (fewest sharp boundaries, gentlest property gradients) from among all the equally-valid models; this is not always geologically correct (real geology has sharp contacts and layers), but it provides a reproducible and objective minimum-structure interpretation that is preferable to more complex models whose extra detail is unconstrained by the data.
• The regularization parameter controls the trade-off between smoothness and data fit — at high regularization (strong smoothness constraint), the algorithm produces a very smooth model that may not fit the data well; at low regularization (weak smoothness constraint), the algorithm fits the data closely but may produce a rough, geologically implausible model with artifacts; Occam's inversion systematically searches for the regularization value that achieves a target data misfit level (typically equal to the estimated measurement noise level), ensuring the model fits the data as well as the data quality warrants without overfitting noise into spurious model features.
• Magnetotelluric (MT) surveys have particularly benefited from Occam's inversion — MT surveys measure natural electromagnetic field variations at the surface and use them to determine subsurface electrical resistivity structure at depths ranging from hundreds of meters to tens of kilometers; the data is inherently band-limited and has complex sensitivity to subsurface structure, making non-uniqueness a severe practical problem; Occam's 1D inversion for MT data (the original Constable, Parker, and Constable 1987 formulation) provides smooth resistivity-depth profiles that can be compared between survey stations to map regional structures like sedimentary basins, volcanic units, and conductive zones associated with fluid-bearing or mineralized rocks.
• Modern Occam-style inversion has extended to 2D and 3D problems with additional complexity — the original Occam formulation was for 1D problems (layered earth models at individual stations), but the smoothness regularization concept has been adapted for 2D and 3D inversions of entire survey datasets simultaneously; 2D Occam inversion produces smooth cross-sections of subsurface resistivity beneath a profile of stations; 3D implementations produce volumetric resistivity models beneath a grid of stations; the computational cost increases dramatically with dimensionality, requiring specialized algorithms and parallel computing resources for large-scale 3D inversions used in regional exploration surveys.
• Occam's inversion results must be interpreted with awareness of the smoothness bias — the minimum-structure philosophy means that sharp boundaries and thin layers are systematically underrepresented in Occam inverted models; a thin, highly resistive limestone sandwiched between conductive shales will appear in the Occam model as a broader, less resistive feature than it actually is; interpreters use Occam models as a starting point for understanding regional structure but apply additional analysis (sensitivity testing, parametric forward modeling, integration with well data) to assess whether sharper structures are consistent with the data; the Occam model is a reliable minimum-information interpretation, not necessarily the geologically accurate model.