Advances in Frozen Soil Creep Research: A Comprehensive Review of Experimental Methods, Micro-Mechanisms, Mechanical Characteristics, and Constitutive Models

Frozen soil creep is a primary factor affecting the long-term stability of cold-region engineering. Based on a structured, PRISMA-informed search of Web of Science, Scopus, CNKI, and ScienceDirect (last searched 31 August 2026; coverage 1950–2026), 126 peer-reviewed papers were included after screening 243 de-duplicated records (296 records initially identified; 53 duplicates removed). This paper reviews recent advances in frozen soil creep from four perspectives: experimental methods, micro-mechanisms, mechanical characteristics, and constitutive models. In experiments, laboratory temperature control has expanded from uniform fields to temperature gradient fields, and field monitoring achieves creep separation and multi-parameter sensing via plate load tests, layered settlement, and distributed optical fibers. At the micro-scale, techniques such as Computed Tomography (CT), Scanning Electron Microscopy (SEM), Nuclear Magnetic Resonance (NMR), and Electrical Resistivity Tomography (ERT) infer the synergistic mechanisms of unfrozen water lubrication, mass-transport medium for solutes and particles, and breakdown of ice cementation across the three creep stages. Concerning mechanical characteristics, the influences of temperature, water content, confining pressure, loading conditions, and salinity on creep rate and failure are summarized; the determination methods and influencing factors of long-term strength are analyzed; and the physical bases and applicability of various strength criteria are compared. In terms of models, the theoretical frameworks, key advances, and application boundaries of empirical models, element models, and classical theoretical models are reviewed. Finally, it is indicated that future breakthroughs lie in constructing a cross-scale bridge of “micro-mechanisms–meso-representation–macro-response” and developing a unified constitutive model that adopts key mechanism parameters as internal variables while balancing physical interpretability and parameter accessibility.