Neuro-Physical Inverter for Magnetotelluric Data

This repository contains code accompanying a research study on a Neuro-Physical Inverter (NPI) for 1D magnetotelluric (MT) data. The method combines:

• 1D MT forward modeling,
• ensemble-approximated conditional Gaussian Processes (EnsCGP),
• neural residual refinement using a ResNet-based architecture.

The code is organized as a Python package and includes example scripts and notebooks demonstrating forward modeling, ensemble-based inversion, and neural refinement.

---

Installation

We recommend using a conda environment.

conda create -n npi-mt python=3.10
conda activate npi-mt
pip install -e .

---

Usage

See the examples/ directory for Jupyter notebooks demonstrating:

forward MT modeling,

EnsCGP applied to real data,

the full Neuro-Physical Inverter workflow.

---

Citing

If you use this software in academic work, please cite it as:

Kim, J. D. (2026). npi-mt: Neuro-Physical Inverter for 1D magnetotelluric data.
Zenodo. https://doi.org/10.5281/zenodo.18462490