Accurately inferring realistic subsurface structures poses a considerable challenge due to the impact of morphology on flow and transport behaviors. Traditional subsurface characterization relies on two primary types of data: hard data, derived from direct subsurface measurements, and soft data, encompassing remotely sensed geophysical information and its interpretation. Existing deep‐learning‐based methodologies predominantly focus on the transition from multiple observations to subsurface structures. However, implicit non‐linear correlations among diverse data sources often remain underutilized, leading to potential bias and errors. In this study, we introduce a multiple‐condition fusion network (MCF‐Net) to characterize subsurface structures based on both hard and soft data. To harness the full potential of multiple‐source subsurface observations, two distinct neural networks extract implicit features from hard and soft data. The integration of these features is achieved through multiple‐condition fusion blocks, designed to capture representative characteristics. These blocks are also adept at reconstructing heterogeneous structures and facilitating hydrological parameterization. MCF‐Net exhibits accuracy in estimating subsurface structures across various types of subsurface observations. Experimental results underscore the utility and superiority of MCF‐Net in applications of hydrogeological modeling.

Multiple-point statistics (MPS) has shown promise in representing complicated subsurface structures. For a practical three-dimensional (3-D) application, however, one of the critical issues is the difficulty in obtaining a credible 3-D training image. However, bidimensional (2-D) training images are often available because established workflows exist to derive 2-D sections from scattered boreholes and/or other samples. In this work, we propose a locality-based MPS approach to reconstruct 3-D geological models on the basis of such 2-D cross sections (3DRCS), making 3-D training images unnecessary. Only several local training subsections closer to the central uninformed node are used in the MPS simulation. The main advantages of this partitioned search strategy are the high computational efficiency and a relaxation of the stationarity assumption. We embed this strategy into a standard MPS framework. Two probability aggregation formulas and their combinations are used to assemble the probability density functions (PDFs) from different subsections. Moreover, a novel strategy is adopted to capture more stable PDFs, where the distances between patterns and flexible neighborhoods are integrated on multiple grids. A series of sensitivity analyses demonstrate the stability of the proposed approach. Several hydrogeological 3-D application examples illustrate the applicability of the 3DRCS approach in reproducing complex geological features. The results, in comparison with previous MPS methods, show better performance in portraying anisotropy characteristics and in CPU cost.