Transdimensional inference in the geosciences.

Area of Science:

• Geophysics
• Computational Seismology

Background:

• Seismologists image Earth's interior using surface seismic data.
• Traditional inverse problems often seek a single 'best' Earth model, which can be problematic due to inherent data non-uniqueness and regularization dependencies.

Purpose of the Study:

• To introduce transdimensional inference methods for seismological inverse problems.
• To highlight the benefits of parameter space sampling over traditional optimization approaches.

Main Methods:

• Embracing non-uniqueness through parameter space sampling.
• Utilizing transdimensional and hierarchical sampling for complex problems with multiple parameter types and uncertain data errors.
• Informing model parametrization, data noise levels, and data weighting through the data themselves.

Main Results:

• Transdimensional inference allows for an ensemble of solutions, directly addressing model non-uniqueness.
• These methods are particularly beneficial for seismological problems with complex parameterizations and uncertain error estimates.
• Sampling methods are becoming increasingly feasible and widely applicable in geosciences.

Conclusions:

• Transdimensional inference provides a more robust approach to Earth structure imaging by directly handling non-uniqueness.
• This methodology offers flexibility, allowing data to inform critical choices often fixed in optimization frameworks.
• The growing feasibility of these computationally intensive methods promises advancements in seismological research.