The development of early warning systems for landslide hazards has long been a challenge because the accuracy of such systemsis limited by both the complicated underlying mechanisms of landslides and the lack of in situ data. In this study, we implemented a multivariate threshold criterion that integrates in situ monitoring data and data from unsaturated hydro-mechanical analyses as an early warning system for rainfall-induced landslides in the Wenchuan earthquake region of China. The results indicate that rainfall intensity is closely correlated with the probability of landslide occurrence. Variations in matric suction and suction stress were obtained from in situ measurements and used to quantify the soil water retention curve, which presented clear hysteresis characteristics. The impacts of rainfall infiltration on slope failure in post-earthquake landslide areas under transient rainfall conditions were quantified by hydro-mechanical modelling theories. Variations in the suction stress of unsaturated soil were used to calculate the safety factor. The influence of hydrological hysteresis processes on the slope failure mechanism was analysed. Multivariate threshold criteria that include the intensity–probability (I-P) threshold, soil moisture and matric suction based on in situ big data and unsaturated slope stability analysis benchmarks are proposed for use in an early warning system for rainfallinduced landslides.