ZLiu93/Generated_prior

3D EM Inversion - Pre-generated Prior Datasets

Overview

This dataset contains the pre-generated prior data utilized for 3D Electromagnetic (EM) Prior Falsification and Stochastic Inversion with MCMC. Due to the substantial file size (~1.29 GB), these large NumPy binary files (.npy) are hosted here on Hugging Face to supplement the main GitHub repository.

These files serve as the input/output training data or pre-computed structures for building a surrogate model to accelerate 3D EM geophysical inversions.

File Structure & Descriptions

Inside this dataset, you will find the following core files:

• EMsigma_padded.npy (962.3 MB): The padded 3D electrical conductivity ($\sigma$) models used for forward modeling or training the surrogate network.
• EMsigma_core.npy (324 MB): The core region of the 3D conductivity models, excluding padding boundaries, optimized for inversion/falsification analysis.
• dpred.npy (1.82 MB): Predicted data responses corresponding to the prior models.
• Hyper_Param.npy (64.1 KB): Hyperparameters and configuration settings used during the prior generation process.

---

Usage & Integration with GitHub

To use these files in your local environment alongside the source code, please clone the GitHub repository and place these files in the designated directory.

Recommended Directory Structure:

EM_Surrogate_Inv3D/
└── Prior falsification/
    ├── 3D EM prior falsification.ipynb
    └── Generated prior/             <-- Create this folder if it doesn't exist
        ├── EMsigma_padded.npy
        ├── EMsigma_core.npy
        ├── dpred.npy
        └── Hyper_Param.npy