Machine Learning for Geophysical Modeling and Analysis: A Comprehensive Overview

Abstract

Geophysical modeling underpins our understanding of Earth- system processes, ranging from seismic wave propagation and subsurface imaging to climate dynamics and planetary exploration. In the past decade, machine learning (ML) has emerged as a transformative resource that complements, accelerates, and sometimes replaces conventional physics-based approaches. This review synthesises the state- of- the- art ML techniques applied to geophysical problems, highlighting methodological advances, benchmark datasets, performance metrics, and open challenges. We categorise the literature into four thematic parts, namely data-driven forward modeling, inverse problems and parameter estimation, surrogate and emulation frameworks, and unsupervised discovery of geophysical patterns. For each part, we discuss model architectures, involving convolutional neural networks, graph neural networks, physics- informed neural networks, transformer- based sequence models, training strategies, interpretability tools, and integration with physical constraints. The review concludes with a roadmap outlining promising research directions, such as hybrid physics- ML solvers, uncertainty quantification, scalable high- performance computing, and the ethical deployment of ML in geoscience.