Soil salinization and moisture redistribution in the deep vadose zone are critical hydrological processes governing the regional water cycle in arid and semi-arid loess regions. However, the complex feedback mechanisms among moisture flow, heat transfer and solute transport, particularly involving phase changes, remain poorly understood due to numerical challenges. In this study, a fully coupled water-heat-salt transport model for unsaturated loess was developed based on the representative elementary volume (REV) scale. The model rigorously integrates moisture transport (Van Genuchten), convective-diffusive solute transport, and heat conduction, while explicitly accounting for source terms arising from evaporation and salt precipitation. THSVFoam, a finite volume solver developed on OpenFOAM, integrates a hybrid Picard-Newton iterative algorithm to enhance numerical robustness under multi-physics coupling. The model was validated against laboratory column experiments and subsequently applied to simulate field-scale spatiotemporal dynamics of a loess slope under rainfall and post-rainfall redistribution conditions. Results indicate that during rainfall, the moisture infiltration rate attenuates due to shifting driving forces, whereas heat conduction accelerates with depth driven by vaporphase latent heat. During the redistribution stage, moisture migration exhibits an increase-then-decrease trend, while deep soil temperature shows a fluctuating rebound pattern. Furthermore, distinct ionic fractionation is observed, with SO42 consistently lagging behind Cl due to differential retardation effects. These findings demonstrate that neglecting coupled salt-blockage and thermal-vapor feedback leads to substantial errors in estimating groundwater recharge and soil water storage.