Creep characteristics and poromechanics-based viscoelastic-plastic constitutive model of warm and ice-rich frozen sand Authors: Fulong Ma, Jian Li, Chunhui He, Enlong Liu, Haoran Zhou, Wei Jin, Yihui Zhong, Gang Wu Published: 2026 ABSTRACT AbstractMotivated by the pronounced rheological behavior of warm and ice-rich frozen soils, we developed a specimen preparation method tailored to ice-rich materials. We conducted a series of triaxial creep tests on saturated frozen sand with 20% ice content and on ice-rich frozen sand with 67% ice content; comparative analyses revealed distinct differences in their creep responses. Building on poromechanics and a binary-medium framework, we propose a viscoelastic–plastic constitutive model for warm frozen soils that accounts for interactions among the soil skeleton, ice, and unfrozen water, as well as the progressive breakage of ice-cement bonds. Validation against experimental data demonstrates that the model accurately captures the creep behavior of warm frozen soils across a range of ice contents and stress levels.Keywords: Triaxial creep test, Binary-medium model, Warm and ice-rich frozen soil, Creep constitutive modelSubject terms: Engineering, Materials science, Solid Earth sciences Introduction Frozen soil is a multiphase porous composite comprising mineral particles, ice, unfrozen water, and gas. The presence of ice and viscous unfrozen-water films promotes plastic flow of ice, ice-crystal reorientation, and stress-induced partial melting, producing time-dependent deformation (rheology). Higher contents of ice and unfrozen water intensify the rheological response of warm and ice-rich frozen soils. In permafrost engineering, beyond the instantaneous deformation caused by structural loads, long-term rheological settlement of foundations warrants close attention. Monitoring of the Qinghai–Tibet Highway subgrade shows that long-term deformation can be substantial, particularly in warm and ice-rich permafrost1. To mitigate creep-related engineering problems, it is critical to develop a viscoelastic–plastic constitutive model that quantitatively describes the creep behavior of warm and ice-rich frozen soils. Early frozen-soil creep models were predominantly macroscopic and empirical. Creep tests indicate that frozen soils commonly exhibit decelerating (primary) creep and non-decaying creep. Numerous concise empirical constitutive models have been proposed to simulate these behaviors2–5. While useful for early permafrost design, such models depend heavily on data quality and quantity, limiting their generality, and they offer limited insight into the underlying rheological mechanisms. Subsequent work developed element-based creep constitutive models. These assemble idealized mechanical elements—springs (elastic), dashpots (Newtonian viscous), and sliders (Saint-Venant plastic)—in various combinations to represent material behavior, yielding classic models such as Kelvin–Voigt, Maxwell, Bingham, and Nishihara6. For example, Wang et al.7 proposed a frozen-soil creep model based on a parabolic yield criterion by integrating Maxwell, Kelvin, and Bingham elements, effectively reproducing frozen-sand creep. Luo et al.8 formulated a viscoelastic–plastic model derived from the Nishihara framework, assigning a stress- and time-dependent viscosity to the viscoelastic branch and introducing a damage variable in the viscoplastic branch to capture steady and accelerated creep stages. The Abel dashpot exhibits behavior intermediate between ideal fluids and solids9,10, making it well suited to strongly rheological materials such as frozen soil. Accordingly, Wang et al.11 constructed an Abel dashpot based on fractional-order derivatives and combined it with classical elements to develop a fractional-derivative creep model capable of representing frozen-soil deformation. Li et al.12 proposed an improved Nishihara model by introducing damage and weakening factors to account for stress- and temperature-induced degradation and strengthening of frozen soil properties. Although element-based models can fit rheological test data well due to their flexibility, they provide limited insight into the intrinsic mechanisms governing the rheology of frozen soils. Consequently, researchers have developed damage-based rheological constitutive models. Using computed tomography (CT), Wu et al.13 monitored microstructural evolution during creep and observed the initiation and growth of microcracks in frozen-soil specimens. Electron microscopy studies14,15 likewise documented structural damage and analyzed its relationship with creep deformation. These observations laid the groundwork for damage-mechanics creep models. For example, Li et al.16 improved the viscoplastic element of the Nishihara model by expressing the plastic damage variable via the yield function, establishing a viscoelastic–plastic damage constitutive model for frozen soil. Li et al.17 further developed a creep model that incorporates temperature and stress effects by introducing hardening and damage factors. However, most existing creep constitutive models treat frozen soil as a homogeneous continuum and focus primarily on macroscopic influences of external factors on deformation or damage evolution. They seldom elucidate the failure mechanisms underlying creep, nor do they capture the gradual fracture of ice cementation from a microscopic perspective. This motivates the development of a damage-based creep model for warm and ice-rich frozen soil that explicitly considers micromechanical processes. Building on rheological damage mechanisms, researchers have introduced the binary-medium framework into frozen-soil creep modeling. Within a damage-mechanics context, Wang et al.18 employed the binary-medium concept to analyze creep damage and proposed a new binary-medium creep model. Wang et al.19 further advanced a rate-dependent model within this framework that accounts for ice-cementation failure in saturated frozen soils; leveraging micromechanical field equations and thermodynamic principles, it overcomes the difficulty prior binary-medium models had in reproducing rate-dependent responses. In parallel, poromechanics has been applied to frozen-soil constitutive modeling: Liu et al.20 developed an elasto-plastic model that captures interactions among the soil skeleton, unfrozen water, and ice, while Li et al.21 proposed a macro–mesoscopic poromechanics-based model that incorporates pressure-melting in warm frozen soils. In sum, whether empirical, element-based, or damage-oriented, most existing creep models for frozen soils are formulated within solid mechanics, treating frozen soil as a monolithic solid and neglecting the effects of both pore-water and pore-ice pressures. Given the high ice content, appreciable unfrozen water, and strong temperature sensitivity of warm and ice-rich frozen soils, a purely continuum representation cannot adequately capture their deformation mechanisms. In this study, we develop a poromechanics-based constitutive model that explicitly accounts for interactions among the soil skeleton, ice, and unfrozen water during deformation. To capture the progressive transition from bonded (cemented) to frictional contacts under loading, we embed a binary-medium framework, yielding a viscoelastic–plastic constitutive model tailored to warm and ice-rich frozen soils. The specimen preparation methods and test methods The specimen preparation method of different ice contents This study investigates saturated frozen sand and ice-rich frozen sand. Triaxial creep tests were conducted at − 1.5 °C to examine the effects of ice content and stress level on the mechanical behavior of frozen sand. Note that “ice content” here refers to mass ice content, defined as the ratio of the mass of ice to the mass of sand. Because specimen preparation depends on ice content, different methods were used for saturated and ice-rich specimens. Preparation of saturated frozen sand (conventional method) are as follows: (1) Screen the standard sand to obtain three size fractions: 0.5–1 mm, 1–2 mm, and 2–3 mm. Mix them uniformly at a mass ratio of 3:2:1, then place the blended sand into a three-lobed mold using the pluviation method. (2) Place the mold in a vacuum saturation chamber; evacuate for 3 h and saturate for 12 h. (3) Immerse the specimen in water and freeze it in a cold room at subzero temperature for more than 3 days. (4) After the water and mold have frozen into a single ice block, fracture the ice, remove the mold, and extract the frozen-sand specimen. The resulting saturated frozen sand had a dry density of 1.72 g/cm³ and an ice content of 20%. This specimen preparation method of frozen sand has a better guarantee for the uniformity of the specimen. Similar methods have been widely used in both earlier and more recent studies22,23, and is generally accepted in the field. The following is the preparation method of ice-rich frozen sand (water content exceeding saturation). Given the large particle size, lack of cohesion, and tendency to segregate, an improved preparation method was developed for ice-rich specimens: (1) Screen the standard sand into the same three size fractions (0.5–1 mm, 1–2 mm, 2–3 mm), mix at 3:2:1 by mass, and pre-freeze in a subzero cold room. (2) Freeze deionized water to form ice, crush it, pass it through a 2 mm sieve, and retain ice particles smaller than 2 mm. (3) Mix the pre-frozen sand and ice particles uniformly at a prescribed mass ratio, then place the mixture into a three-lobed mold in three equal lifts. Achieve uniformity by controlling the number of hammer blows per lift. Between lifts, lightly scarify the surface with a knife to keep it loose and prevent delamination. (4) Equip the mold base with a custom water-supply device and connect it to near-0 °C liquid water to infiltrate from bottom to top. Stop the water supply when it exudes from the specimen top and clamp the supply line. Freeze the fully saturated specimen in the cold room for 24 h before demolding. By controlling only the sand-to-ice mass ratio in step (3), ice-rich specimens with different ice contents can be produced. This method (i) ensures uniform distribution of sand particles, (ii) minimizes bubble formation via bottom-up water infiltration, and (iii) prevents layering caused by compaction in lifts. Using a sand-to-ice particle mass ratio of 1:0.4 and subsequent water replenishment/freezing, a specimen with a final ice content of 67% was obtained. The specimen was cylindrical, with an inner diameter of 61.8 mm and a height of 125 mm. To verify near-complete saturation, bubble porosity was estimated from the measured masses of sand, ice particles, and unfrozen water during preparation. For three specimens with 67% ice content, bubble porosities were 1.01%, 2.46%, and 1.62%, respectively. Although a small amount of entrapped air is unavoidable, all values were below 5%, satisfying the criterion for saturated specimens. Surface and cross-sectional images of the 67% ice-content specimen (Fig. 1) show generally uniform particle distribution with no obvious bubbles. Surface and internal cross-section of a finished specimen with 67% ice content. Test device To investigate the effects of ice content and stress level on the mechanical behavior of warm frozen soil, triaxial creep tests were performed using an MTS-810 materials testing system (Fig. 2). The apparatus comprises five subsystems: a triaxial pressure cell, an axial loading actuator, a confining-pressure controller, a temperature-control unit, and a data acquisition/control system. The temperature-control unit maintains − 30 °C to 20 °C with high precision. The control system regulates confining pressure, axial load, loading rate, and loading frequency, and records time, confining pressure, axial load, axial displacement, cell (confining) displacement, and temperature in real time. MTS-810 material test device. Test procedure is as follows: (i) Set the chamber to − 1.5 °C and condition the specimen for 12 h to reach thermal equilibrium. (ii) Apply the confining pressure once temperature is stable. To minimize pre-creep effects, ramp to the target confining pressure within 5 min, then wait 5 min before axial loading. (iii) Apply axial load under load control: ramp to the target within 5 min and then hold constant. (iv)Terminate the test when axial displacement reaches 30 mm or when the constant-load duration reaches 24 h. The specimen’s volumetric strain was tracked during loading. The volume of hydraulic oil entering the cell was computed from the confining (cell) piston displacement, and the volume of the axial loading rod intruding into the cell was computed from the axial displacement. The sum of these volumes represents the specimen’s volumetric contraction. In order to ensure the accuracy of volume variation, there is an error calibration process. An iron block of the same size as the specimen is used to record the volumetric strain during confining pressure loading. Normally, there is almost no volume change under test pressure, so the volume change measured in the present test is the error that needs to be corrected. The experimental data shown in this study are the results after eliminating the error. Test plan First, the peak strengths of saturated frozen sand (20% ice content) and ice-rich frozen sand (67% ice content) were determined by triaxial compression at − 1.5 °C and a confining pressure of 1.0 MPa. The measured strengths were 7.54 MPa and 3.68 MPa, respectively. These values were then used to define the stress levels for creep testing. For saturated frozen sand, creep tests were conducted at 40%, 50%, 55%, 60%, and 80% of the triaxial strength, corresponding to deviatoric stresses of 3.02, 3.77, 4.15, 4.52, and 6.03 MPa. For the ice-rich frozen sand, creep tests were run at 20%, 30%, 40%, 50%, and 60% of the strength, corresponding to deviatoric stresses of 0.74, 1.10, 1.47, 1.84, and 2.21 MPa. Different stress levels were selected for the two materials to capture their distinct sensitivities to loading. The triaxial creep texts plan is shown in Table 1. Table 1The triaxial creep texts plan.SpecimenIce contentConfining pressureStrengthStress levelsSaturated frozen sand20%1.0 MPa7.54 MPa0.4, 0.5, 0.55, 0.6, 0.8Ice-rich frozen sand67%3.68 MPa0.2, 0.3, 0.4, 0.5, 0.6 The triaxial creep texts plan. Results and analysis of triaxial creep tests Figure 3 presents the triaxial test results for saturated frozen sand (20% ice content) and ice-rich frozen sand (67% ice content) at -1.5 °C under a confining pressure of 1 MPa. The peak strengths are 7.54 MPa for the saturated frozen sand and 3.68 MPa for the ice-rich frozen sand. Both materials exhibit strain softening; however, the saturated frozen sand achieves a higher peak strength and shows more pronounced post-peak softening. In terms of volumetric behavior, the saturated frozen sand undergoes initial contraction followed by dilation, whereas the ice-rich frozen sand exhibits continuous contraction. These results indicate that the strength of saturated frozen sand is substantially greater than that of ice-rich frozen sand. This difference is attributed to the closely packed soil particles in the saturated frozen sand, which form a soil skeleton that shares load with the ice. In contrast, in the ice-rich frozen sand the soil particles are largely suspended within the ice matrix, so the load is borne mainly by the ice. The observed dilatancy of the saturated frozen sand contributes to its higher shear strength. Triaxial tests of specimens with different ice contents under 1 MPa confining pressure (-1.5 °C). Creep tests were conducted on saturated frozen sand at stress levels of 40%, 50%, 55%, 60%, and 80% of peak strength, corresponding to deviatoric stresses of 3.02, 3.77, 4.15, 4.52, and 6.03 MPa, respectively. Additional creep tests were performed on frozen sand with 67% ice content at 20%, 30%, 40%, 50%, and 60% of peak strength, corresponding to deviatoric stresses of 0.74, 1.10, 1.47, 1.84, and 2.21 MPa, respectively. The results are shown in Figs. 4 and 5. Axial strain-time curves of creep tests on saturated frozen sand and ice-rich frozen sand under different stress levels (-1.5℃). Volumetric strain -time curves of creep tests on saturated frozen sand and ice-rich frozen sand under different stress levels (-1.5℃). As seen in Figs. 4 and 5, both materials share certain creep characteristics but also display notable differences. At lower constant loads, both sands exhibited decaying creep. However, unlike the classical three-stage non-decaying creep reported in previous studies24–26, the specimens entered a steady (secondary) creep regime at higher stress level, without showed the onset of accelerated (tertiary) creep. For saturated frozen sand, decaying creep persisted up to the 55% stress level. When the stress level exceeded 60%, axial strain surpassed 15% within a few hours, indicating specimen failure. For the ice-rich frozen sand (67% ice content), decaying creep was observed only up to the 30% stress level, and axial strain exceeded 15% within a few hours at stress levels above 40%. These results indicate that creep is much more pronounced in ice-rich frozen sand than in saturated frozen sand. Volumetric responses also differed. For saturated frozen sand, volumetric contraction dominated at lower stress levels, transitioning to volumetric expansion as stress increased; overall, volumetric strain shifted from contraction to expansion with increasing stress level. In contrast, the ice-rich frozen sand exhibited volumetric contraction at all stress levels, with the magnitude of contraction increasing with stress. Analysis suggests that the particles in saturated frozen sand are tightly packed, leading to particle displacement and volume expansion at higher stress levels. However, soil particles in ice-rich frozen soil are suspended in ice, so volumetric contraction is mainly caused by the compaction and pressure melting of ice. To ensure the test duration was sufficient to capture any accelerated creep, we conducted a 70-h creep test on saturated frozen sand at a 55% stress level. As shown in Fig. 6, the axial strain exceeded 15% during the 70-h test, yet no accelerated creep stage was observed. Given that failure is commonly defined at 15% axial strain, further testing was deemed unnecessary. Axial strain-time curves of creep tests on saturated frozen sand under 55% stress levels (-1.5℃). The multi-scale viscoelastic-plastic constitutive model based on poromechanics The elastic constitutive model based on poromechanics Poromechanics studies porous materials whose mechanical behavior is strongly influenced by pore fluids27. Its applications span rocks, soils, gels, and biological tissues. In this framework, a porous medium is modeled as the superposition of a solid (skeleton) continuum and one or more fluid continua. These continua evolve independently while interacting mechanically and exchanging mass and energy. For saturated frozen soil, the soil skeleton is treated as the solid continuum, whereas ice and unfrozen water are represented as two distinct fluid continua. The porosity of the specimen consists of the ice porosity and the unfrozen water porosity , and thus can be represented as: where the subscripts “” and “” denote the parameters associated with ice and unfrozen water, respectively. Under undrained triaxial test conditions, due to the total mass conservation of unfrozen water and ice, the following incremental equation can be obtained28: where denote the initial porosity, and are the densities of ice and unfrozen water respectively, is the saturation degree of unfrozen water during loading, represents the change of ice porosity or unfrozen water porosity caused by skeleton deformation. According to the thermodynamic equilibrium, the ice pressure and unfrozen water pressure satisfy the following relation: where and are respectively the melting entropy and the initial melting temperature of water; and T is the test temperature. In the theory of poromechanics27, the free energy of the solid skeleton in a linear elastic system can be expressed as follows, where is the free energy of the solid matrix; is the strain tensor; is the temperature, is the porosity. By introducing the Legendre-Fenchel transformation of , can be represented as: For isotropic and linear elastic porous media, has the following form29: where , G and K are the thermal volumetric dilation coefficient, the shear modulus and the bulk modulus of the solid skeleton, respectively. is the volumetric strain. and are the generalized Biot coupling moduli and the generalized Biot coefficients29,30. According to Maxwell’s symmetry, we have . and are the heat capacity of the soil matrix and the thermal dilation of the pore volume occupied by phase J. And: where is the solid matrix bulk modulus. Traditional constitutive models typically relate stress to strain within a single continuum, without explicitly accounting for multiphase pore-pressure effects. In contrast, poromechanics incorporates the influence of pore pressures in unfrozen water and ice on deformation of the solid skeleton. Using a free-energy-based formulation, Liu et al.29 derived the stress-strain equations for a three-phase medium comprising the solid skeleton, water, and ice as follows: where is the stress tensor. Based on the incompressible assumption of solid matrix and the small deformation assumption, we can obtain that the volume change of the specimen is equal to the change of porosity: where the volumetric strain is positive for compression. Combining Eqs. (3), (7), (8), (9), (11), (12) and (13), the control equations for water pressure and ice pressure can be solved: Substituting Eqs. (14) and (15) into (10), the elastic constitutive model for porous media is obtained: where: By transposing the stiffness matrix to the flexibility matrix, we obtain: Material viscoelasticity is represented here with the Burgers model, which consists of a Maxwell element (spring and dashpot in series) in series with a Kelvin–Voigt element (spring and dashpot in parallel). The Burgers model is widely used because it can capture the transient (primary) and steady (secondary) stages of creep observed in frozen soils. By adopting the three-dimensional Burgers framework and substituting Eq. (17) for the elastic element, the corresponding viscoelastic constitutive equation is obtained.: where is the creep shear modulus, is the deviatoric stress tensor, and are the viscosity coefficients of the two bodies respectively. The viscoplastic constitutive model based on poromechanics Different from the definition of skeleton free energy in the previous section, part of energy will be dissipated when plastic deformation occurs in frozen soil. In this case, consists of the recoverable free energy and the surface energy of the solid matrix: where is the recoverable free energy of the solid matrix, is the plastic strain tensor, and represents the change of ice porosity or unfrozen water porosity caused by plastic deformation of the skeleton, is the hardening variable. Differentiation of the skeleton free energy , we have: According to the Clasusius-Duhem inequality in the poromechanics: where is the stress tensor, is the strain tensor, is the entropy density of the skeleton. Since , substituting Eq. (20) into (21) yields: Since the first six terms of Eq. (22) occur in the elastic category, there are: Thus, Eq. (22) can be simplified as: By introducing the Legendre-Fenchel transformation of , can be represented as: By substituting Eq. (26) into Eq. (24), the state equation can be transformed into the following form: The free energy of solid matrix can be expressed by Eq. (6) in the elastic category. However, Eq. (6) requires some modification in the plastic category. In elasto-plastic mechanics, there are: Therefore, by replacing in Eq. (6) with , the expression for the free energy of the solid matrix can be returned to the elastic category, that is: Combining Eqs. (27) and (29), we have: Thus Eq. (25) can be translated into: In the case of ideal viscoplasticity, ignoring the influence of capillary force and hardening parameters, the above equation becomes: In the above equation, represents the dissipated energy caused by viscoplastic deformation, represents the dissipated energy generated by the work done by unfrozen water pressure on the viscoplastic porosity change of unfrozen water.; represents the dissipated energy generated by the work done by ice pressure on the viscoplastic porosity change of ice. The sum of the dissipation energies from the three mechanisms is always greater than or equal to zero. In conventional creep tests of frozen soil, the pressure-melting effect of warm frozen soil can be ignored due to the constant load. Meanwhile, based on small-strain assumption, the unfrozen water saturation of frozen soil remains unchanged after deformation, leading to: In addition, under the assumption that the solid matrix is incompressible, we have: Therefore, combining Eqs. (33) and (34) yields: By dividing in Eqs. (32) into the deviatoric stress tensor and the spherical stress tensor , we obtain: Substituting Eq. (35) into Eq. (36) yields: According to the definition of effective stress, introducing effective spherical stress tensor , it can be inferred that: Combining the deviatoric stress tensor and the effective spherical stress tensor yields: The viscoplastic flow rule can be defined by means of the dissipative potential in the elastoplastic constitutive model, as follows: To account for the existence of the current elastic domain and the key features of viscoplastic behavior, the function in viscoplastic models typically adopts the following expression: where is the viscosity coefficient, is the loading function, and . Thus, substituting Eq. (41) into Eq. (40) yields the expression for viscoplastic strain: Viscoplastic strain occurs only when the stress level exceeds the elastic domain. Following the associated flow rule, the loading function can be defined by an effective stress-modified Drucker-Prager (D-P) criterion as: where the parameters and are related to the cohesion and internal friction angle through the following conversions: When , according to the operation rule of , the viscoplastic strain does not occur; when , elastoplastic strain develops, we have: Viscoelastic-plastic constitutive model based on binary-medium framework Saturated frozen soil is a multiphase composite consisting of soil particles, ice, and unfrozen water. Owing to ice cementation, frozen soils exhibit pronounced structural characteristics: ice both binds particles and shares a portion of the load with the soil skeleton. However, when the applied load is sufficiently high, portions of the ice fracture and lose their cementing capacity, resulting in increased deformation and reduced strength. Ice-bond failure does not occur abruptly; rather, it progresses gradually during loading. The binary-medium framework is well suited to describing this internal structural evolution. Within this framework, domains with intact ice cementation are treated as bonded elements, whereas regions where ice bonds have fractured are treated as frictional elements. A specimen with fully intact ice bonding prior to testing can be viewed as composed entirely of bonded elements. As loading proceeds, breakage accumulates and some bonded elements transform into frictional elements; the specimen thus evolves into a mixture of bonded and frictional domains that collectively carry the external load. Within a representative volume element (RVE), can be used to characterize the deformation at any arbitrary position, where (x, y, z) denotes a Cartesian coordinate system. Therefore, according to the principle of volume homogenization, the macroscopic average strain can be expressed as: where, is the total volume of the RVE, which is equal to the sum of the volume of bonded element and frictional element inside the RVE. Therefore, Eq. (47) can be split into: where, and are the strain tensors of bonded elements and frictional elements, respectively. The breakage ratio is characterized by the volume ratio of frictional elements to the entire RVE: Substituting Eq. (49) into Eq. (48) yields: When applying the above equation to the creep category, and should be time-dependent functions, as follows: where the strain of the bonded element can be expressed by the viscoelastic constitutive model of Eq. (18), one gets: where superscript denotes the relevant parameters of the bonded element. The strain of the frictional element can be expressed by the viscoplastic constitutive model of Eq. (46): where superscript denotes the relevant parameters of the frictional element. The breakage ratio quantifies the progressive transformation of the bonded elements into frictional elements and thus its mathematical form critically affects the constitutive response. Because is a microscale descriptor, it cannot be measured directly; its functional form is therefore inferred by calibrating to macroscopic test data18. In conventional creep tests, the applied stress is held constant, making stress an unsuitable state variable to govern the evolution of . Instead, we adopt the shear strain as the independent variable. Accordingly, we postulate the following expression for the breakage ratio : where and are damage parameters that depend on macroscopic factors such as stress level and ice content and are identified via trial-and-error calibration. This form satisfies and as , consistent with bounded, monotonic damage evolution. Finally, substituting Eqs. (52), (53), and (54) into Eq. (51) yields the viscoelastic-plastic constitutive model for warm and ice-rich frozen soil based on poromechanics: Simplifying the above equation under triaxial stress states yields the following expression: When , we have When , we have According to the loading criterion in Eq. (43), the following can be solved: Thus, Eq. (57) can be simplified as: In general, the following assumptions were adopted in the derivation: (a) neglecting the capillary force effect; (b) isotropic assumption; (c) the incompressibility of sand particles; (d) the small deformation assumption; (e) the associated flow law. These assumptions capture the main contradiction of the problem and do not affect the main conclusions. The creep constitutive model developed here treats frozen soil, at the microscale, as a mixture of bonded and frictional elements. By modeling the damage-driven transition from bonded to frictional elements, we derive a macroscopic creep constitutive relation from micromechanical principles. Whereas traditional frozen-soil creep models are typically divided into decaying and non-decaying types, the present approach yields a unified and more concise formulation that captures both behaviors while more fundamentally reflecting the failure mechanisms governing creep. Verification of the constitutive model Determination of model parameters The parameters involved in this model mainly include the test temperature , initial melting temperature of water , initial saturation of unfrozen water , melting entropy of water , solid matrix bulk modulus , initial porosity , bulk modulus and shear modulus of bonded elements, creep shear modulus , cohesion , internal friction angle , breakage parameters and , as well as viscosity coefficients , , and ,. Some of the physical parameters involved in the model, such as , , , , , and , were obtained from experimental conditions and references27,28,31. The initial porosity can be obtained by saturation and drying tests. Since micro-cracks within the specimen have not yet developed during the initial loading stage of the triaxial test, the specimen primarily consists of the bonded elements. Therefore, this stage can be utilized to determine the relevant parameters of the bonded elements. For the parameters related to the bonded elements, The bulk modulus and shear modulus of the bonded elements can be determined by the stress-strain curve during the initial loading stage of triaxial testing. The creep shear modulus ,viscosity coefficients and are mainly obtained by fitting the strain-time creep curve of frozen soil under lower stress level. For the viscosity coefficient of the frictional element, it is mainly obtained by fitting the non- decaying creep curve of frozen soil under high stress level. For the breakage parameters and , which mainly describe the conversion process of the bonded elements to frictional elements during loading, it is difficult to obtain directly through the test. Through trial-and-error analysis, it can be observed that the breakage parameters and exhibits a strong correlation with the stress state in creep tests. Consequently, a linear relationship has been established between the breakage parameters and the deviatoric stress. Model verification The creep test of saturated frozen sand under-1.5 °C and 1 MPa confining pressure is verified. Based on the parameter determination method described in the previous section, take , , , , , , , , , , , and . In addition, , and are related to the deviatoric stress , with , , . The coefficient of determination for the three fitting formulas exceeds 0.99, indicating a high accuracy in the fitting. The comparison of the experimental and predicted creep curves for saturated frozen sand is shown in Fig. 7: Comparison of test curves and predicted curves for creep experiments of saturated frozen sand (-1.5℃). Then, the creep test of ice-rich frozen soil with 67% ice content under-1.5 °C and 1 MPa confining pressure is verified. Based on the parameter determination method described in the previous section, take , , , , , , , , , , , , and . In addition, , and are related to the deviatoric stress , with , . The coefficient of determination for the three fitting formulas exceeds 0.99, indicating a high accuracy in the fitting. The comparison of the experimental and predicted creep curves for ice-rich frozen sand is shown in Fig. 8: Comparison of test curves and predicted curves for creep experiments of ice-rich frozen sand (-1.5℃). As shown in Figs. 7 and 8, the proposed model fits the creep curves across stress levels well and achieves high simulation accuracy. The results indicate that the poromechanics-based viscoelastic–plastic constitutive model provides excellent agreement with the creep behavior of both saturated and ice-rich frozen soils, and is suitable for predicting foundation deformation in warm and ice-rich permafrost regions. Experimental results show that saturated sand transitions from volumetric contraction to expansion as the stress level increases. While volumetric contraction is commonly predicted, the proposed model also captures volumetric expansion. For saturated frozen sand at a 60% stress level, Fig. 9 compares the measured and predicted curves. The fit indicates that the model slightly underpredicts the early-time volumetric expansion, but the predictions converge toward the experimental values as time progresses. The comparison of test curves and predicted curves of saturated frozen at 60% stress level. Furthermore, to illustrate the model’s ability to capture the contraction–expansion transition, we adjusted the parameters to generate predictions with volumetric strain approaching zero, as shown in Fig. 10. Taken together, Figs. 9 and 10 demonstrate the constitutive model’s capability to predict volumetric expansion. The predict curves with volumetric strain approaching 0 of saturated frozen at 60% stress level. Sensitivity analysis of model parameters The breakage ratio is characterized by the volumetric fraction of frictional elements within the representative volume element (RVE). Its evolution law is central to the binary-medium framework. The time evolution of during creep in ice-rich frozen sand is shown in Fig. 11. Evolution of breakage rate with time under different constant loads (-1.5℃). As illustrated, the growth rate of increases with increasing stress level. However, at higher stress levels, failure may occur even when remains relatively low. This is attributed to more severe stress concentrations under high loads, where frictional domains coalesce and percolate along concentrated zones, ultimately triggering rapid shear failure. We further conducted a sensitivity analysis of the breakage-ratio parameters in the constitutive model. Using the creep test of ice-rich frozen sand at a deviatoric stress of 3.68 MPa as an example, we examined the influence of parameters and on the creep response (Figs. 12 and 13). Effect of the breakage rate parameter on time- axial strain curve(-1.5℃). Effect of the breakage rate parameter on time- axial strain curve(-1.5℃). Sensitivity analysis shows that the mid-to-late-stage slope of the creep curve decreases as increases, whereas it increases as increases. Although structural parameters such as and ϑ cannot be measured directly, they represent internal structural evolution during loading and thereby influence predicted creep responses. Calibrating their evolution from limited test data enables broader prediction of creep behavior, which underscores the utility of these parameters. Conclusions Building on poromechanics and a binary-medium framework, we developed a multiscale viscoelastic–plastic constitutive model that captures the creep behavior of warm and ice-rich frozen soils. Exploiting the binary-medium approach, the model reproduces creep curves featuring only stable and unstable stages (i.e., no rapid-flow stage), consistent with observed behavior. The model was validated against creep tests spanning a range of ice contents and stress levels. The main conclusions are: Creep tests on saturated frozen sand (20% ice) and ice-rich frozen sand (67% ice) at − 1.5 °C and 1 MPa show decaying creep at lower stress levels. At higher stress levels, both materials exhibit stable and unstable creep without entering a rapid-flow stage. Creep tests on saturated frozen sand (20% ice) and ice-rich frozen sand (67% ice) at − 1.5 °C and 1 MPa show decaying creep at lower stress levels. At higher stress levels, both materials exhibit stable and unstable creep without entering a rapid-flow stage. Within a poromechanics framework, we formulated viscoelastic and viscoplastic relations that account for mechanical interactions among the soil skeleton, unfrozen water, and ice, providing a more mechanistic representation of material behavior. Within a poromechanics framework, we formulated viscoelastic and viscoplastic relations that account for mechanical interactions among the soil skeleton, unfrozen water, and ice, providing a more mechanistic representation of material behavior. In the binary-medium framework, viscoelastic and viscoplastic components are combined by distinguishing bonded elements (intact ice cementation) from frictional elements (fractured ice bonds). The breakage ratio governs the transition from bonded to frictional elements, yielding a multiscale viscoelastic–plastic model suited to warm and ice-rich frozen soils. In the binary-medium framework, viscoelastic and viscoplastic components are combined by distinguishing bonded elements (intact ice cementation) from frictional elements (fractured ice bonds). The breakage ratio governs the transition from bonded to frictional elements, yielding a multiscale viscoelastic–plastic model suited to warm and ice-rich frozen soils. Validation against creep tests on both saturated and ice-rich frozen soils shows that the model accurately reproduces creep behavior across varying ice contents and stress levels. Validation against creep tests on both saturated and ice-rich frozen soils shows that the model accurately reproduces creep behavior across varying ice contents and stress levels. Analysis of breakage-ratio evolution under different stress levels indicates that its growth rate increases with stress. At higher stresses, frictional elements tend to cluster and percolate in regions of stress concentration, leading to structural failure even when the overall breakage ratio remains relatively low. Analysis of breakage-ratio evolution under different stress levels indicates that its growth rate increases with stress. At higher stresses, frictional elements tend to cluster and percolate in regions of stress concentration, leading to structural failure even when the overall breakage ratio remains relatively low. Acknowledgements The authors thank the participants for their contribution to this study. Author contributions Conceptualization, Fulong Ma and Enlong Liu; Data Curation, Jian Li, Chunhui He and Enlong Liu; Formal Analysis, Fulong Ma and Haoran Zhou; Funding Acquisition, Enlong Liu and Gang Wu; Methodology, Fulong Ma, Chunhui He and Enlong Liu; Project Administration, Jian Li and Yihui Zhong; Resources, Jian Li and Enlong Liu; Supervision, Jian Li, Chunhui He and Enlong Liu; Val-idation, Fulong Ma and Haoran Zhou; Visualization, Chunhui He; Writing – Original Draft, Fulong Ma; Writing – Review & Editing, Chunhui He, Enlong Liu and Wei Jin. Funding This work was supported by the funding of the Science and Technology Plan Project of Linzhi (Grant No. SYQ2024-13). Data availability The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author. Declarations Competing interests The authors declare no competing interests. Declaration of Competing Interest The authors declared that there is no conflict of interest to the work submitted. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.