Systematic Review of Machine Learning Applications in Geophysics: Integrating Seismic, Potential Field, Geological Modelling, and Natural Hazard Domains

Abstract

Machine learning (ML) drives transformative advances across geophysics, revolutionising seismic interpretation, subsurface modelling, potential field analysis, and geo-hazard forecasting. This Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-compliant systematic review synthesises insights from 19 studies published between 2018 and 2025, including in-press and early-access articles, to ensure comprehensive coverage of recent developments. Using a structured screening and data extraction framework, we combine thematic and quantitative analyses to map the evolution of ML applications across four key geophysical domains. Results reveal a strong concentration in seismic domains (63 % of studies), with limited adoption in geological modelling (21 %), hazard prediction (26 %), and potential field data (11 %). Convolutional Neural Networks (CNNs) and U-Nets dominate seismic workflows, while physicsinformed models and Bayesian methods emerge in tasks requiring interpretability and uncertainty quantification. Transfer learning and synthetic data generation, present in 21 % of studies, are being leveraged to address data scarcity, while GPU-accelerated models (16 %) support large-scale seismic processing. Despite these advances, the review identifies key limitations: sparse labelled datasets, computational bottlenecks, and a lack of cross-domain frameworks. Only 16 % of studies integrate multimodal or hybrid ML-physics workflows. Thematic synthesis highlights emerging opportunities in explainable AI (XAI), energy-efficient modelling, and the embedding of geophysical constraints within ML architectures. The review calls for benchmark dataset development, opensource toolchains, and modular ML systems to promote reproducibility and domain convergence. By evaluating gaps and mapping methodological trends, this study offers a strategic roadmap for advancing scalable, interpretable, and physically consistent ML applications across the geoscientific ecosystem.