GS-3DMG/GM-ConvCNP: GM-ConvCNP v1.0.0-alpha

Characterization of realistic subsurface structures can provide a prior reference for understanding geosystem behavior in spatial scale. Hydrogeological modeling is a well-established method of describing the structure of subsurface spaces. One of the main issues in the application of statistical-learning-based methods to the characterization of hydrogeological phenomena is the complex parameterization of the high-quality reconstruction. Insufficient amount of training data has become a hindrance to the application of deep-learning-based hydrogeological modeling methods. In this work, we propose a novel method to reconstruct the entire spatial structures of subsurface hydrogeological attributes and channels from a limited amount of conditioning data, named GM-ConvCNP. The proposed approach is able to significantly reduce training consumption and improve the performance of realizations compared to a set of different benchmark tests. Experimental results confirm that the GM-ConvCNP model can extract heterogeneous patterns by using meta-learning from limited training data and reconstruct multiple-scale hydrogeological structures. We show that it can be applied not only to modeling hydrogeological variables, but also to geophysical fields such as seismic data reconstruction.