Numerical analysis of diaphragm wall behavior in a jet-grouted deep excavation: a random field case study Authors: Huynh Huy Nguyen, Trong Nghia Le Published: 2026 ABSTRACT AbstractDeep excavations in densely built urban areas frequently employ jet grouting to enhance base stiffness and limit diaphragm wall deformation. However, most design approaches treat the improved soil mass as homogeneous and rarely consider the spatial variability inherent in jet-grouted materials, which may lead to biased deformation predictions and incomplete risk evaluation. This study investigates the influence of spatial variability in the secant stiffness modulus (E₅₀) of jet-grouted soil on diaphragm wall displacement using a probabilistic numerical framework. A two-dimensional lognormal random field of E₅₀ was generated using the spectral representation method and incorporated into a finite element model within the Random Finite Element Method (RFEM) framework. The mean stiffness of the improved soil was taken as 1336.85 MPa based on laboratory testing, while variability was characterized by coefficients of variation (COV) ranging from 0.34 to 0.8 and spatial correlation lengths defined by scales of fluctuation of SOFₓ = 3–5 m and SOFγ = 0.5–2 m. Monte Carlo simulations were conducted with a converged sample size of 40 realizations. The results indicate that spatial variability significantly affects both the magnitude and dispersion of wall displacement, with predicted maximum values generally ranging from 11.0 to 11.8 mm and following a lognormal distribution. For an allowable displacement of 11.44 mm, the exceedance probability decreases from approximately 58% at COV = 0.34 to about 35% at COV = 0.8. Changes in spatial correlation length have comparatively smaller effects, although vertical correlation shows a slightly stronger influence than horizontal correlation. These findings demonstrate that deterministic analyses assuming uniform ground improvement may underestimate the range of possible wall movements, while incorporating spatial variability provides a more realistic basis for reliability-based assessment and risk-informed design of jet-grouting-reinforced excavations.Keywords: Jet Grouting, Random Field, FEM, Diaphragm WallSubject terms: Engineering, Environmental sciences, Solid Earth sciences Introduction Along with the rapid pace of urbanization, the construction of underground spaces and deep excavations in densely populated cities has become an essential component of modern infrastructure development. Under conditions of limited construction space and complex subsurface conditions, excavation activities often interact with nearby buildings and underground utilities, creating significant technical and safety challenges. Controlling ground deformation and ensuring the stability of retaining structures are therefore critical requirements to prevent excessive movements that may endanger adjacent facilities and urban infrastructure1–3. Among various ground improvement techniques, jet grouting has been widely adopted to enhance soil stiffness, increase basal stability, and limit diaphragm wall deformation in deep excavations. Numerous experimental investigations and numerical studies have demonstrated the effectiveness of jet grouting in improving excavation performance and mitigating ground movements4,5. Nevertheless, most current design approaches are still based on deterministic numerical analyses in which the improved soil mass is represented by uniform or averaged mechanical properties6. Field observations and laboratory investigations suggest that such simplifications may not adequately represent the true behavior of jet-grouted soil. The mechanical properties of jet-grouted material are influenced by several factors, including injection pressure, grout composition, withdrawal rate, and the heterogeneity of in-situ soil conditions. As a result, the stiffness and strength of the improved soil mass often exhibit considerable spatial variability 7,8. Previous studies have reported relatively large coefficients of variation (COV) for cement-treated soils, indicating that neglecting this variability may lead to biased deformation predictions and an underestimation of potential risks in excavation design9,10. In recent years, Random Field Theory combined with the Finite Element Method (FEM) has been increasingly applied to incorporate spatial variability of soil properties into numerical analyses. This probabilistic framework enables soil parameters to be modeled as spatially correlated random fields characterized by statistical descriptors such as the coefficient of variation and the scale of fluctuation (SOF). Applications of RFEM have been reported in various geotechnical problems, including slope stability, tunneling, and foundation performance11–14. Recent studies have further extended these probabilistic approaches to evaluate the installation torque-capacity correlation of helical piles15 and to investigate seismic landslide behaviors under stochastic ground motions16. Furthermore, advanced numerical frameworks have been utilized to compare 2D and 3D random mechanisms in complex deformation problems17. However, its application to deep excavations reinforced with jet grouting remains relatively limited. In particular, the influence of spatial variability in the stiffness of jet-grouted soil on diaphragm wall deformation has not yet been systematically quantified. To address this gap, the present study investigates the effect of spatial variability in the secant stiffness modulus (E50) of jet-grouted soil on diaphragm wall displacement using a probabilistic numerical framework. A two-dimensional random field representing the variability of E50 is generated and coupled with finite element analysis through the Random Finite Element Method. Monte Carlo simulations are performed to evaluate the influence of statistical parameters, including the coefficient of variation and spatial correlation lengths, on the deformation response of the retaining system. The results provide insight into how stiffness heterogeneity within jet-grouted ground affects excavation behavior and offer a probabilistic basis for assessing the likelihood of exceeding allowable wall displacements in urban deep excavations. Case study Project background The case study presented in this paper concerns the headquarters office building of Phat Dat Group, located in Ho Chi Minh City, Vietnam, and constructed by Central Construction Joint Stock Company18. The overall project layout is illustrated in Fig. 1, while the typical excavation geometry and geological cross-section are presented in Fig. 2. The project comprises four basement levels with a maximum excavation depth of approximately − 16.9 m. Excavation was carried out using the semi top-down construction method, which allows the superstructure and basement slabs to be constructed concurrently with the excavation process. This approach enhances the overall stiffness of the support system and effectively reduces ground deformation around the excavation. Overall site plan of the Phat Dat Office Building. Geological stratigraphy and excavation cross-section. The excavation was supported by a diaphragm wall system with a thickness of 800 mm and an embedded depth of approximately 32 m, serving as the primary retaining structure surrounding the excavation. Subsurface conditions were established based on comprehensive site investigation data and consist of six distinct soil layers with varying geotechnical characteristics, as shown in Fig. 2. The groundwater table fluctuates between elevations of approximately − 0.55 m and − 3.4 m relative to the ground surface. The diaphragm wall toe is embedded in Layer 4b, which is identified as a fully saturated sand stratum; under these conditions, the embedment depth of the diaphragm wall is insufficient to function as an effective hydraulic cut-off, thereby posing a significant risk of basal hydraulic instability (bottom heave) during excavation. To enhance basal stability, a jet-grouting ground improvement layer with a thickness of approximately 5.5 m was installed at the excavation base prior to excavation. This treated zone was formed by overlapping jet-grouting columns to create a continuous improved soil mass. Owing to the inherent spatial variability of jet grouting, the mechanical properties of the improved ground were evaluated based on laboratory testing of core samples obtained from both test columns and production columns. A statistical summary of the secant stiffness modulus E50 derived from these samples is presented in Fig. 3. For the purpose of deterministic back-analysis, representative average values were adopted to characterize the overall behavior of the improved ground. Accordingly, the secant stiffness modulus E50 of the jet-grouted soil was taken as the average value of 1336.85 MPa and was used consistently in the subsequent analyses. Statistical distribution of the E50 values of jet-grouted soil. Numerical model and validation Input parameters for the calibration model The model calibration strategy was developed based on the principle of minimizing the number of adjustable parameters while ensuring an accurate reproduction of the dominant deformation mechanisms governing the diaphragm wall system. Initially, the geotechnical parameters were determined from laboratory test results, field investigations, and empirical correlations recommended in technical literature and the PLAXIS user guidelines. Subsequently, the calibration was carried out through a controlled iterative procedure, in which the secant stiffness modulus E50 of each soil layer listed in Table 1 was sequentially adjusted according to the discrepancies between numerical predictions and field monitoring data. Table 1Representative soil parameters used in the numerical model. Representative soil parameters used in the numerical model. The calibration process was conducted from shallow soil layers to deeper strata to ensure that the stiffness distribution with depth realistically reflects the increase in effective stress and the corresponding development of soil shear strength. Field monitoring during excavation was performed at four locations corresponding to the four sides of the excavation, as illustrated in Fig. 1. Lateral displacement data obtained from inclinometer measurements were used as the primary input for the inverse analysis and model calibration. Unlike basic physical properties such as unit weight, water content, and shear strength, which can be determined with reasonable confidence from borehole samples, the assessment of soil stiffness remains challenging due to the discrepancy between in-situ stress conditions and laboratory testing environments. In this study, the Hardening Soil model was adopted owing to its capability to realistically simulate the nonlinear stress-strain behavior of soils as well as the loading-unloading processes associated with deep excavation construction, as demonstrated in previous studies19,20. In PLAXIS, stiffness parameters such as E50 and Eoed are typically derived from triaxial and one-dimensional consolidation tests; however, such data are often unavailable in practical engineering projects. Therefore, the modulus E50 was estimated using empirical correlations, and the relationship E50 = (1800–2500)NSPT proposed by21–23 was adopted as a calibration confidence range in this study. The numerical model of the excavation was developed using PLAXIS 2D and is illustrated in Fig. 4 at the final excavation stage. Structural components, including the diaphragm wall (DW800), basement slabs (B1, B2, B3, B4), and the ground floor slab (L1), were modeled using plate elements, assuming B30-grade concrete with a design compressive strength of 17 MPa, as summarized in Table 2. Differences in deformation characteristics among these structural components primarily arise from variations in their thickness, while the material properties were assumed to be uniform. Finite element model of the excavation in PLAXIS 2D. Table 3Investigated parameters for spatial variability analysis of the jet grouting block.CASESOFYSOFXCOV 00.530.340.337.141130.50.477.092150.50.477.093250.50.477.094250.80.706.95 Investigated parameters for spatial variability analysis of the jet grouting block. During staged excavation, the groundwater level inside the excavation decreases while the external level remains relatively constant, creating a hydraulic head difference across the diaphragm wall. To simulate this condition, a steady-state seepage analysis was adopted with appropriate external hydraulic boundary conditions. The computed pore water pressure distribution was then used to define the groundwater table behind the wall and to evaluate the resulting hydraulic loading. Consequently, groundwater effects were incorporated within an effective stress framework. The analysis was conducted under short-term excavation conditions. At the study site, the soft clay layer (Layer 1) is relatively thin (approximately 2 m), while the underlying Layers 2 and 3 consist of medium stiff to very stiff clay with relatively low compressibility. Beneath these clay layers, a sand stratum is present, where consolidation effects are negligible due to its high permeability and predominantly drained behavior. In addition, the effective basement slab area, after excluding openings, is relatively small, and the semi top-down construction sequence allowed rapid installation of slabs after each excavation stage. Consequently, the time available for consolidation between stages was limited, resulting in negligible consolidation deformation. Therefore, consolidation phases were not modeled between excavation and slab construction. This assumption is considered appropriate for the primary objective of this study, which focuses on the influence of spatial variability in jet-grouted soil stiffness on diaphragm wall displacement. Nevertheless, long-term consolidation effects are acknowledged as a limitation of the present study. Table 2Input structural parameters.ParameterUnitDW800L1B1; B2; B3B4EA1kN/m20.1E65.5E65.5E616.5E6EA2kN/m20.1E65.5E65.5E616.5E6EIkNm2/m1.1E618.4E318.4E318.4E3Thicknessm0.80.20.20.6 Input structural parameters. In addition to comparing the maximum lateral displacement of the diaphragm wall, the calibration strategy also considered the overall shape of the displacement profile with depth, ensuring that the numerical model not only reproduces the deformation magnitude but also realistically captures the soil-structure interaction mechanisms throughout the excavation process. The numerical results obtained using PLAXIS 2D, together with the comparison between monitored and simulated diaphragm wall displacement profiles at the final excavation stage, demonstrate a reasonably good agreement with field measurements in terms of both displacement distribution and maximum displacement magnitude (Fig. 5). Although the numerical model generally predicts slightly larger displacements than those measured in the field, the largest discrepancy-where the computed maximum displacement reached 11.44 mm, approximately 20% higher than the inclinometer-measured value of 9.16 mm-remains within an acceptable range for deep excavations in soft soil conditions. Given that the primary objective of this study is to reasonably reproduce excavation behavior for subsequent random field parametric analyses and predictive modeling, this level of discrepancy is considered acceptable. Furthermore, comparison with the results reported by H. D. Lin and S. C. Lin (2008)24 confirms that the numerical model adequately captures the deformation characteristics of the diaphragm wall system (Fig. 6), particularly the influence of stiffness contrast induced by jet-grouting improvement within the depth range of approximately 16–20 m. Comparison of DW displacement obtained from the numerical model and calibration results. DW displacement characteristics in excavations with jet-grouting ground improvement (H. D. Lin and S. C. Lin, 2008). The discrepancies between numerical predictions and field measurements may be attributed to several inherent limitations of numerical modeling approaches. The simplification of a three-dimensional excavation problem into a two-dimensional plane strain model may lead to conservative estimation of earth pressures acting on the diaphragm wall because load redistribution and spatial stiffening effects at structural corners are not considered. In addition, the two-dimensional model cannot fully capture the three-dimensional interaction effects of the support system, which have been demonstrated to significantly influence diaphragm wall deformation25,26. Moreover, natural geological variability and uncertainties associated with the determination of soil parameters from both field and laboratory tests may also contribute to differences between computed and observed results. Random finite element method Random field modelling The spatial variability of soil mechanical properties is an inherent feature of natural ground conditions and cannot be adequately represented by conventional deterministic models that assume uniform material parameters. To account for this uncertainty, the secant stiffness modulus E50 of the jet-grouted block used to reinforce the excavation base is modeled in this study as a two-dimensional random field. Previous studies have shown that jet grouting effectively reduces diaphragm wall displacement27, and that increased soil stiffness leads to reduced deformation in deep excavations28. As the improvement effect of jet grouting is primarily associated with an increase in the stiffness of the treated zone, the secant modulus E50 is adopted in this study as the representative parameter governing wall deformation. Spatial variability is assigned only to the jet-grouted soil (E50) to isolate its influence on wall response and maintain computational efficiency. The surrounding soils are treated as deterministic based on calibration against field data. Their interaction is captured by FEM, although combined variability is not considered. To ensure that the elastic modulus remains strictly positive and consistent with the statistical characteristics of jet-grouted material, E50 is assumed to follow a lognormal distribution, as expressed Eqs. (1), (3), (4), and (5), and is generated through a logarithmic transformation of an underlying Gaussian random field produced using the spectral representation method. In the numerical model, the jet-grouted block is discretized into square elements with dimensions of 0.5 m × 0.5 m in PLAXIS, allowing random E50 values to be assigned to each element during the Monte Carlo simulations (Figs. 7 and 8). The mean value of E50 is taken as 1336.85 MPa and is used as the central parameter for constructing the random field in the subsequent analyses. Geometry of the jet grouting block. Discretized random field mesh of the jet grouting block. The geometry of the jet-grouted block is assumed to be continuous and fixed, based on the design layout and construction records. In practice, geometric variability may arise from installation uncertainties; however, this study focuses on material variability (E50) as the primary factor governing deformation response. Therefore, geometric variability is not explicitly considered. The spatial variability of soil properties within the study domain is characterized by the SOF in the horizontal and vertical directions (SOFx, SOFy) and the COV. These SOF parameters are incorporated into the model through an exponential autocorrelation function (Eq. 2) to describe the degree of spatial dependence. Specifically, the correlation between any two points is defined as a function of their absolute separation distances (), where SOFx and SOFy govern the rate of correlation decay. While a larger SOF implies a smoother spatial variation, a smaller SOF results in stronger fluctuations; as SOF approaches zero, the values at different locations become statistically independent. Complementing this spatial structure, the COV is utilized to quantify the inherent dispersion and magnitude of uncertainty in soil properties. In this study, the interplay between the spatial correlation defined in Eq. 2 and the material uncertainty represented by the COV constitutes a vital framework for assessing the behavior of the deep excavation system. Based on the E50 dataset obtained from the project site, the stochastic parameters for this analysis were rigorously calculated. Specifically, the Coefficient of Variation (COV) was determined to be 0.34 (Eq. 4), reflecting the inherent natural dispersion of the soil stiffness as illustrated in Fig. 3. To establish the spatial correlation parameters, the vertical scale of fluctuation (SOFy) was quantified using Vanmarcke’s mean-crossing method29(Fig. 9). By analyzing the average spacing of intersections between the actual E50 profile and its mean value, while applying a correction factor of 0.8, a site-specific SOFy of 0.5 m was established. Regarding the SOFx, due to the practical limitations in obtaining continuous horizontal test data at the site, this parameter was determined by referencing established geotechnical literature. To account for the inherent anisotropy caused by geological depositional processes and the systematic overlapping of jet-grouted columns, the horizontal scale of fluctuation (SOFx) was set at 3.0 m. This selection results in an anisotropy ratio (SOFx/SOFy) of 6.0, which is consistent with the established empirical range of 5.0 to 10.0 commonly reported in geotechnical literature30,31. These combined values ensure that the probabilistic model is directly informed by site-specific measured mechanical properties while remaining consistent with well-established empirical observations in the field. Determination of SOFy via Vanmarcke’s method. To facilitate a more rigorous investigation into the system’s performance under stochastic conditions, five simulation scenarios were established, as delineated in Table 3. While Case 0 represents the baseline model anchored in site-specific data, Cases 1 through 4 constitute a systematic parametric study designed to explore the sensitivity of the excavation behavior to varying degrees of spatial variability. The rationale for these extended scenarios is based on established geotechnical principles, which observe that the SOFx is typically significantly larger than the SOFy due to the inherent anisotropy of geological depositional processes. Extensive literature reviews suggest that SOFx commonly spans from 1 to 40 m 32,33, while SOFy varies between 0.5 and 10 m34,35, and the COV generally falls within 0.2 to 0.836,37. By adopting diverse combinations within these widely reported empirical ranges, this research ensures a comprehensive and reliable assessment of how material heterogeneity influences the overall mechanical response of the jet-grouting reinforced deep excavation. Based on the adopted random field model, together with the selected SOF and COV values, a probabilistic analysis was performed to evaluate the probability of failure, Pf. In this study, the failure state is assumed to occur when the displacement obtained from the random field-based analysis, denoted as DRF, exceeds the allowable displacement DDET derived from the deterministic analysis. The value of DDET was calibrated using field monitoring data to ensure its representativeness for the actual working conditions of the excavation system. Accordingly, the probability of failure Pf is defined as the probability that the limit state condition is not satisfied, which is equivalent to the probability of the event DRF > DDET38. Implementation procedures of random finite element method Prior to performing the RFEM analyses, it is necessary to determine the required number of Monte Carlo simulations, N, to ensure the reliability of the computed results39,40. Selecting an appropriate value of N not only achieves the desired level of accuracy but also improves computational efficiency. In this study, a convergence analysis was conducted for the reference scenario (Case 0) with SOFy = 0.5, SOFx = 3, and COV = 0.34, where the mean E50 was taken as 1336.85 MPa. A total of 100 Monte Carlo realizations were performed to evaluate convergence, as shown in Fig. 10, which illustrates the evolution of the mean and standard deviation (STD) of the diaphragm wall lateral displacement with increasing number of simulations, N. Convergence of the random field model. Figure 10b indicates that both the mean and STD gradually stabilize as the number of simulations increases and approach convergence at approximately N ≈ 40. In addition, the convergence ratio (CR), calculated to quantify the degree of convergence following the procedure described in Ref41., is also presented in Fig. 10a. The CR results confirm that the variations in the mean and STD of the wall displacement become negligible and remain below 5% when N ≥ 40. Therefore, N = 40 was adopted for the subsequent RFEM analyses for Cases 1 to 4, as summarized in Table 3. The implementation of the random finite element method in this study follows the procedure outlined below: An idealized finite element model was established in PLAXIS 2D, and the centroid coordinates of all zone elements were extracted. An idealized finite element model was established in PLAXIS 2D, and the centroid coordinates of all zone elements were extracted. Based on the element grid coordinates and the predefined random field parameters listed in Table 3, the corresponding random fields of the stiffness modulus E50 were generated using a Python-based script implemented in Google Colab, yielding N realizations. Based on the element grid coordinates and the predefined random field parameters listed in Table 3, the corresponding random fields of the stiffness modulus E50 were generated using a Python-based script implemented in Google Colab, yielding N realizations. The generated random field values were mapped onto the corresponding elements in the numerical model, after which N finite element simulations were performed. The generated random field values were mapped onto the corresponding elements in the numerical model, after which N finite element simulations were performed. The maximum lateral displacement and deformation response of the diaphragm wall were extracted to evaluate their statistical distribution and the probability of exceeding the allowable displacement threshold. The maximum lateral displacement and deformation response of the diaphragm wall were extracted to evaluate their statistical distribution and the probability of exceeding the allowable displacement threshold. The above procedure was repeated for all cases summarized in Table 3. The above procedure was repeated for all cases summarized in Table 3. Results and discussions Effect of spatial variability The influence of soil spatial variability on diaphragm wall displacement was evaluated using random finite element analyses. Figure 11 compares the displacement profiles of the diaphragm wall under homogeneous soil conditions and two representative random field realizations (Realizations 6 and 28) with COV = 0.5, SOFy = 2, and SOFx = 5 (Case 4 in Table 3). For Realization 28, the maximum diaphragm wall displacement, Smax, reaches 10.69 mm, which is 6.6% lower than the value predicted by the deterministic analysis (11.44 mm). In contrast, Realization 6 yields a larger Smax of 11.89 mm, exceeding the deterministic result by approximately 3.6%. The discrepancy between the two realizations exceeds 10%. These results highlight the critical role of soil spatial variability in accurately predicting the displacement response of diaphragm walls in jet-grouting-reinforced ground. Representative random field case. The significant differences in the shapes of the displacement profiles among the random realizations arise from the spatial variability of the jet-grouted material. This variability is primarily governed by fluctuations in E50 within the jet-grouted block, as illustrated in Fig. 12, which presents the displacement fields and their correlation with the spatial distribution of E50. The two-dimensional random field exhibits pronounced heterogeneity, where the red and blue regions represent zones with relatively high and low stiffness, respectively. Two-dimensional random fields of E50 for two representative realizations. It should be noted that the two realizations discussed above (Realizations 6 and 28) are selected as representative examples to illustrate the local deformation patterns that may occur across different random realizations. The use of the complete Monte Carlo dataset in the subsequent analyses ensures that the probabilistic nature of the diaphragm wall displacement response is fully captured, rather than relying on a single realization. Effect of COV and SOF To investigate the influence of the COV of E50 on the deformation characteristics of the diaphragm wall, 40 random finite element analyses were performed for each case listed in Table 3. Figures 13a-e present the distributions of lateral wall displacement obtained from the random field analyses for Cases 0 to 4, respectively. In particular, Fig. 13a and d, and 13e correspond to Cases 0, 3, and 4 and are used to examine the effect of COV while the spatial correlation parameters remain constant. In contrast, Fig. 13b and c (Cases 1 and 2) are used to evaluate the influence of variations in SOFx. The effect of SOFy is assessed by comparing Cases 2 and 3, as shown in Fig. 13c and d. Effect of COV and SOF of E50 on DW displacement. The results indicate that the maximum diaphragm wall displacement consistently occurs near the wall crest for all investigated cases, which is consistent with the typical deformation mechanism of braced excavations reinforced with jet grouting at the base. Although the shapes of the displacement profiles along depth vary among individual random realizations, the overall distribution trend remains relatively consistent across cases, suggesting mechanical stability of the numerical model. Comparison with the deterministic FEM results shows that the maximum displacement predicted by the deterministic model generally falls within the range of the random field outcomes, confirming its capability to capture the overall trend of wall deformation. However, under several unfavorable realizations, the random field simulations produce larger displacements than the deterministic prediction, indicating that the deterministic approach may not fully account for the effects of spatial variability in E50. In the deterministic analysis, a single reference stiffness parameter () is assigned to each soil layer, whereas in reality the diaphragm wall response develops under varying strain levels with depth, and soil stiffness is inherently strain-dependent. This simplification may affect the model’s ability to accurately reproduce the detailed displacement profile along depth. Nevertheless, the convergence analysis in Sect. 3.2 confirms that the selected number of Monte Carlo simulations is sufficient to ensure statistical reliability. Within the adopted framework, spatial variability is introduced only for the jet-grouted soil (E50), while the surrounding soils are treated as deterministic after calibration against field data. As a result, the simulated responses are centered around the baseline solution corresponding to the mean E50 of the improved zone, and the predicted range does not necessarily encompass the full spread of monitoring observations. Despite these limitations, the observed trends are derived from statistical results over multiple Monte Carlo realizations and are not dependent on a specific random seed, as demonstrated by the convergence analysis. While individual realizations may exhibit local deviations due to different random seeds or excavation configurations, the overall statistical response remains stable. It is acknowledged that the magnitude of displacement may vary with changes in the mean stiffness of the jet-grouted soil or excavation geometry; however, the general statistical trend remains consistent. This indicates that, although simplified, the adopted approach is capable of capturing the governing influence of material heterogeneity on the soil-structure interaction behavior. The influence of COV is clearly reflected by the increase in both the magnitude and dispersion of the displacement profiles as COV increases. Specifically, when COV increases from 0.34 (Fig. 13a) to 0.5 (Fig. 13d) and 0.8 (Fig. 13e), the range of diaphragm wall displacement becomes progressively wider, reflecting a higher level of material uncertainty. This observation suggests that COV plays a dominant role in governing the magnitude of wall displacement compared with the spatial correlation parameters. Regarding spatial correlation, the effect of the vertical SOF, inferred from the comparison between Fig. 13c and d, is generally more pronounced than that of the horizontal SOF, illustrated by the comparison between Fig. 13b and d. An increase in SOFy promotes the formation of soil zones with similar mechanical properties along depth, thereby modifying the displacement distribution along the wall. Nevertheless, the overall influence of spatial correlation parameters remains smaller than that of COV, highlighting the dominant role of E50 uncertainty in this problem. Statistical analyses Figure 14 illustrates the statistical distributions of the maximum diaphragm wall displacement for the five analyzed cases. The majority of the values lie within approximately 11.0–11.8 mm, with a mean concentration around 11.4–11.5 mm. The histograms reveal a slightly right-skewed distribution that is well approximated by a lognormal model, as evidenced by the consistency between the observed frequencies and the fitted curve. This behavior indicates that the displacement response is strongly governed by the spatial uncertainty of E50 and supports the suitability of a lognormal probabilistic representation for describing the variability of wall deformation. Statistical distribution of the maximum diaphragm wall displacement for the analyzed cases. Probabilistic modeling enables engineers to transcend the limitations of deterministic forecasting by explicitly accounting for the inherent uncertainties in soil mechanical properties and the construction conditions of diaphragm walls in jet-grouting reinforced ground. While deterministic methods yield only a single predicted value (e.g., a wall deflection of 11.44 mm), they fail to reflect the associated risk levels or the spectrum of potential real-world scenarios. In contrast, a probabilistic approach quantifies the likelihood of displacements exceeding allowable thresholds, thereby establishing a probabilistic risk profile that facilitates more scientific and comprehensive design decision-making. This methodology is particularly critical in high-density urban areas, where exceeding displacement limits can lead to significant deformation, damage to adjacent structures, serviceability issues, or potential risks to public safety. As the dataset of wall displacements-derived from 40 random realizations-follows a lognormal distribution, Fig. 14 presents the probability of exceedance of wall displacement under different combinations of COV and SOF of the E50 of the jet-grouting reinforced soil. Given an allowable displacement threshold of 11.44 mm, Fig. 15a indicates that the probability of exceedance reaches approximately 58% when COV = 0.34, decreasing to roughly 45% at COV = 0.5 (Fig. 15d) and further declining to about 35% for the highest variability case (Case 4) (Fig. 15e). These results indicate that the exceedance probability decreases as the COV increases. This seemingly counterintuitive trend may be attributed to the heterogeneous stiffness distribution within the jet-grouted block. A larger COV increases the contrast between relatively weak and relatively stiff zones. In the present study, the presence of locally stiffer clusters in several realizations appears to increase the overall stiffness of the improved base, thereby slightly reducing the probability of excessive diaphragm wall displacement. Failure probability (Pf) for the analyzed cases. Similarly, Fig. 15b and c exhibit a comparable trend regarding the SOF, reflecting the influence of spatial correlation length on the deformation behavior of the soil-wall system. Although the horizontal (SOFₓ) and vertical (SOFγ) components possess distinct characteristics, their impact on the exceedance probability in this study is not significantly different, resulting in a failure probability of approximately 55% across the investigated cases. It should be noted that these probabilities do not directly imply structural collapse; rather, they serve to quantify the risk level of displacements surpassing recommended limits. Furthermore, as the allowable displacement threshold decreases, the probability of exceedance-representing the risk of damage or loss due to excessive deformation-increases rapidly. This underscores the necessity of conservatively defining allowable limits while considering site-specific spatial variability to mitigate risks during construction. In geotechnical design, these thresholds are typically determined based on technical standards, codes, and project-specific requirements. However, if the COV and SOF are not fully accounted for, the probability of exceeding wall displacement limits can remain exceptionally high, leading to an underestimation of risk and unreliable design predictions. For instance, with a threshold of 11 mm, the exceedance probability is nearly 1.0 when COV = 0.34 (Fig. 15a), whereas it is approximately 90% for COV = 0.8 (Fig. 15e). Both values represent unacceptable risk levels for safety-critical design. To limit the exceedance probability to below 1%, the maximum allowable wall displacement should be set at approximately 11.75 mm and 11.9 mm for COV = 0.34 and COV ≥ 0.5, respectively. These values provide a more robust foundation for diaphragm wall design in complex urban geological conditions where jet-grouting is utilized and the deformation modulus E50 exhibits high heterogeneity. A reliability-based design chart is developed (Fig. 16) to illustrate the relationship between allowable displacement and the coefficient of variation (COV) for different target probabilities of failure (Pf = 1%, 5%, and 10%). The results indicate that the allowable displacement increases slightly with increasing COV for a given reliability level. This trend should not be interpreted as an improvement in system safety at higher variability; rather, it reflects the necessity of adopting a less stringent displacement threshold to maintain the same probability of failure under elevated uncertainty. Conversely, more stringent reliability requirements (i.e., lower Pf) result in increased allowable displacement values. Reliability-based design chart of allowable displacement versus COV. Notably, for a commonly adopted serviceability threshold of 11 mm, the corresponding exceedance probabilities exceed 10% across the entire range of COV considered, suggesting that this limit may be unconservative when spatial variability is explicitly accounted for. Accordingly, the proposed chart provides a practical and rational framework for selecting allowable displacement values consistent with a specified reliability level, thereby supporting reliability-based design under spatial uncertainty. Conclusion This study applied the Finite Element Method combined with Random Field Theory to investigate the influence of spatial variability in the secant stiffness modulus (E50) of jet-grouted soil on diaphragm wall displacement in deep excavations. Spatial variability was represented by two-dimensional lognormal random fields characterized by different COV and SOF, and Monte Carlo simulations were performed to evaluate the probabilistic response of the excavation system. The results indicate that the maximum diaphragm wall displacement consistently occurs near the wall head, which agrees with the typical deformation pattern of braced excavations with base improvement. Although displacement profiles vary among individual realizations, the deterministic solution generally falls within the distribution range obtained from the random field analyses. Parametric investigations show that the coefficient of variation of E50 has the most significant influence on both the magnitude and dispersion of wall displacement. In contrast, the influence of spatial correlation parameters is comparatively smaller, although vertical correlation (SOFy) exhibits a slightly stronger effect than horizontal correlation (SOFx). Statistical analysis demonstrates that the maximum wall displacement follows an approximately lognormal distribution, enabling the estimation of exceedance probabilities for different allowable displacement thresholds. The results highlight that ignoring spatial variability of jet-grouted stiffness may lead to biased predictions of wall deformation and underestimate the probability of exceeding serviceability limits. While the present study provides useful insights, several limitations should be acknowledged. The geometry of the jet-grouted zone is assumed to be fixed, and geometric variability due to construction uncertainties is not considered. The analysis is based on a two-dimensional plane strain model and does not include three-dimensional effects such as corner stiffening and sequential excavation. Consequently, the influence of spatial correlation (SOF) identified in this study may differ under fully three-dimensional conditions. Future work will extend the present framework to three-dimensional random fields with more realistic representations of jet-grouting columns, including overlap and geometric variability. Acknowledgements We acknowledge Ho Chi Minh City University of Technology (HCMUT), VNU-HCM for supporting this study. Author contributions Huynh Huy Nguyen: Methodology, Investigation, Data curation, Formal analysis, Visualization, Software, Writing – original draft.Trong Nghia Le: Conceptualisation, Methodology, Supervision, Investigation, Writing – review & editing, Project administration. Funding No funding was received for this study. Data availability The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. 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 datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.