Forecasting maximum magnitude of fluid-induced earthquakes: Bridging statistical extrapolation and physics-based forecasting

The forecasing of maximum magnitude (Mmax) for fluid-induced seismicity is a critical challenge in risk management for geothermal energy development, unconventional hydrocarbon extraction, and wastewater disposal. This review systematically examines eight forecasting models spanning three categories: statistical models (G-R relation, extreme value theory, order statistics), semi-physical models (injection volume/time correlations), and physics-based models (rupture mechanics, rate-state friction, pressure diffusion). Through tripartite analysis of parameter sensitivity, scientific interpretability, and engineering validation, we reveal fundamental constraints: (1) Model performance is strictly governed by parameter coverage. Statistical models relying solely on seismic catalogs neglect fault mechanics parameters. Semi-physical models establish empirical ΔV/T-Mmax relationships but fail to explain dynamic slip processes and post-injection ''trailing effects''. Physics-based models integrate fault stress, friction evolution, and fluid diffusion, offering unparalleled mechanistic insights, yet their computational demands pose challenges for real-time field applications (only 7.5% of 228 global cases). (2) Scientific interpretability exhibits hierarchical deficiencies. While physics-based models quantitatively resolve stress transfer and friction evolution (e.g., rate-state models distinguishing seismic/aseismic slip), forecasting of far-field triggering and multi-fault interactions remain reliant on immature theories like poroelastic coupling. (3) Systemic validation biases emerge: 57.5% of cases concentrate in stable North American terrains with underrepresentation of Pacific Rim high-stress zones; 79.8 ​% involve geothermal/shale gas operations lacking CO2 sequestration validations; M ​≥ ​4.0 events constitute merely 12.7% of datasets, leaving M ​≥ ​5.0 predictive capacity unverified. We propose transformative pathways including hybrid “physics-core ​+ ​machine learning” architectures, standardized multi-scenario validation protocols, and dynamic traffic-light systems, aiming to advance cross-disciplinary forecasting paradigms for engineering risk governance.