Sensitivity analysis of VG model parameters for infiltration in undisturbed loess under ponding conditions using the HYDRUS-1D model

It is difficult for remolded loess to reproduce the infiltration law of water in natural loess, resulting in inaccurate parameters of the Van-Genuchten model (VG model) obtained by inversion, which subsequently affects the changes in the moisture field during slope stability numerical simulations. To address this gap, a large-scale undisturbed loess soil column with a diameter of 60 cm and a height of 100 cm was artificially excavated for a water infiltration test under ponding conditions. Based on the test results, the VG model parameters were inverted by the Hydrus-1D software. The simulation accuracy was evaluated using the root mean square error (RMSE) and the Nash-Sutcliffe efficiency coefficient (NSE). Based on the inverted VG model parameters, the single-factor perturbation method was employed to calculate different VG model parameters under a disturbance amplitude of ± 20%. The numerical simulations were carried out to investigate the wetting front arrival times at different depths under varying disturbance parameters. Finally, the sensitivity of soil water characteristic parameters was analyzed. The results indicated that the measured water contents in the column closely matched the simulated values, with minimal errors, as evidenced by the RMSE ranging from 0.010 to 0.019. Except for the NSE values at 10 cm and 20 cm depths, which were below 0.900, all other depths exhibited NSE values greater than 0.900, indicating satisfactory simulation performance. The influence of soil water characteristic parameters on simulation results was θs > ks > n > α > θr. The disturbances in θs and α were negatively correlated, while disturbances in ks, n, and θr were positively correlated. Parameter disturbances less influenced the infiltration laws of water in shallow soils. However, as soil column depth increased, sensitivity to parameter changes gradually increased, leading to a greater impact on simulation results. The research findings provide technical support for evaluating and preventing water-induced landslide hazards.