Abstract Joint estimation of discrete (e.g., lithology or geofluid type) and continuous (e.g., porosity, pore geometry, related connectivity) subsurface parameters represents a challenging inverse problem due to strong parameter trade‐offs, effective non‐uniqueness, and multi‐modal posterior structures. In many conventional deterministic or single‐physics inversion approaches, discrete and continuous parameters are treated separately, and posterior uncertainty is often simplified, limiting the ability to evaluate competing model interpretations in a consistent probabilistic framework. Here we propose a generalized hybrid probabilistic inversion framework for mixed discrete–continuous parameter estimation using multi‐physics observations. The framework integrates three key components: (a) grid‐based marginal screening to down‐select plausible discrete models over the joint parameter space; (b) targeted Monte Carlo refinement within the retained subspaces to reconstruct posterior distributions of continuous parameters for competing discrete models; (c) Discrete inference is conditioned on high‐posterior‐density regions of the continuous parameter space. Synthetic benchmark tests under noise‐free demonstrate robust recovery of continuous parameters and lithology–geofluid combinations. Application to real seismic velocity and electrical conductivity data from the northeastern Japan Arc further shows that the proposed framework yields stable posterior estimates with transparent uncertainty quantification and well‐resolved parameter trade‐offs across different noise levels. By explicitly representing multi‐modal posterior structures while maintaining computational efficiency, the proposed framework provides a practical and scalable solution for mixed discrete–continuous geophysical inverse problems, demonstrated here for lithology–geofluid inversion using seismic velocity and electrical conductivity data.
Zhang et al. (Wed,) studied this question.