Geophysical Inversion
Also known as: inverse problem solving, parameter estimation, model-data fitting
Geophysical inversion is the process of using observed geophysical data to estimate subsurface properties and structures. Formalized by Tikhonov (1963) and expanded by Tarantola (1987), this mathematical framework solves the inverse problem: given measurements (gravity, magnetics, seismic, electrical), what subsurface model produced them? Inversion is central to all quantitative geophysics and enables extraction of detailed subsurface information from surface or borehole measurements.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Geophysical inversion is applicable to any quantitative geophysical interpretation: gravity and magnetic surveys for density anomalies, seismic inversion for velocity structure, electromagnetic surveys for resistivity mapping, and well-log inversion for reservoir properties. It is most effective when data are abundant (dense sampling), measurement uncertainty is well-characterized, and the physical relationship between subsurface properties and data is well-understood. Inversion becomes unreliable when data are sparse or noisy, or when the assumed forward model is incorrect.
Strengths & limitations
- Quantitative framework—inversion produces numerical estimates of subsurface properties with uncertainty bounds, not just qualitative interpretations
- Integration of multiple data types—different geophysical methods can be inverted jointly to constrain properties better than any single method alone
- Efficiency—inversion extracts information from data automatically; manually trying different models is slow and subjective
- Flexibility—regularization constraints and prior information can be incorporated to guide solutions toward geologically plausible models
- Non-uniqueness—many different models can fit the same data; inversion selects the simplest or smoothest but not necessarily the true model
- Model dependence—inversion results depend strongly on the assumed forward model; if the model is wrong, results are biased
- Data inadequacy—limited spatial coverage or high noise reduces resolution; trade-off between data fit and model smoothness must be carefully balanced
- Computational cost—3D inversion of large datasets (seismic, electromagnetics) is computationally expensive; approximations or coarse grids may be necessary
Frequently asked
What is the forward problem versus the inverse problem?
The forward problem asks: 'Given a subsurface model, what data would I measure?' It involves solving physics equations to predict data. The inverse problem asks the reverse: 'Given measured data, what subsurface model produced it?' Solving the inverse problem is more difficult because the relationship between model and data may be nonlinear and non-unique.
What is regularization and why is it necessary?
Regularization adds a penalty term to the inversion objective function, discouraging extreme model features (sharp boundaries, large oscillations). Common regularization includes damping (penalizing large parameter values) and smoothing (penalizing gradients). Regularization is necessary because geophysical data typically contain less information than there are model parameters to estimate; without regularization, the solution is non-unique or unstable.
How is the trade-off between data fit and model simplicity balanced?
The balance is controlled by a regularization parameter (damping factor, trade-off parameter). Too much damping produces smooth but inaccurate models; too little produces oscillatory models that fit noise. The Pareto curve (plotting data misfit versus model roughness) reveals the optimal trade-off. Methods like the L-curve and generalized cross-validation automatically estimate the best parameter.
What is resolution and how is it estimated?
Resolution is the ability to distinguish separate features; poor resolution means closely-spaced features blend together in the inverted model. Resolution depends on data quality, spacing, and sensitivity to depth. Resolution is estimated using the resolution matrix or by computing the point spread function: smearing of a delta-like feature in the true model into the inverted model. Coarser grids improve computational speed but reduce resolution.
Can machine learning improve geophysical inversion?
Yes. Neural networks trained on synthetic datasets can learn the relationship between data and model parameters, providing fast approximations to deterministic inversion. This speeds up exploratory interpretation. However, neural network inversions are data-driven and may not generalize well to conditions outside the training set. Hybrid approaches combining physics-based inversion with machine learning for parameter estimation or uncertainty quantification show promise.
Sources
- Tarantola, A. (1987). Inverse Problem Theory: Methods for Data Fitting and Model Parameter Estimation. Elsevier. link ↗
- Constable, S. C., Parker, R. L., & Constable, C. G. (1990). Occam's inversion: A practical algorithm for generating smooth models from electromagnetic sounding data. Geophysics, 55(3), 289–300. link ↗
- Menke, W. (2012). Geophysical Data Analysis: Discrete Inverse Theory (3rd ed.). Academic Press. link ↗
How to cite this page
ScholarGate. (2026, June 3). Geophysical Inversion. ScholarGate. https://scholargate.app/en/geoscience/geophysical-inversion
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Basin Subsidence AnalysisGeoscience↔ compare
- Seismic Reflection InterpretationGeoscience↔ compare
- Well Log AnalysisGeoscience↔ compare