To address the high false-alarm rate in landslide monitoring and early warning systems, this study aimed to explore more effective methods for rainfall-induced landslide prediction and early warning. By integrating theoretical analysis, field investigations, laboratory testing, and numerical simulation, a dynamic computational algorithm for rock-soil mass softening and stability under rainfall infiltration was developed, focusing on water—the critical factor triggering landslides. This algorithm was established based on a comprehensive analysis of the entire process of rainfall-induced landslides, incorporating considerations of rainfall infiltration boundaries, unsaturated seepage, saturated seepage, rock-soil mass softening, and stability coefficient calculation. The corresponding program was further implemented using FLAC3D's built-in FISH programming language. Subsequently, through a case study, the developed algorithm and program were validated through analysing pore water pressure, saturation, moisture content, cohesion, internal friction angle, maximum shear strain increment, stability coefficient, landslide location, and extent during rainfall-induced slope failure. Finally, the relationships between monitored data (moisture content, displacement, stress, pore water pressure, and mechanical properties) at three representative locations (ground surface, slip surface, and slip bed) and the stability coefficients, along with their applicability in landslide prediction, were analysed and discussed. The research found that when the moisture content in the middle of the potential sliding surface reaches its peak, the cohesion and internal friction angle at this location drop to the trough value. At this point, the slope stability coefficient reached 1.000, indicating the timing of the landslide. Based on this finding, a new landslide monitoring and early warning method was proposed. This method focuses on monitoring the moisture content in the middle of the potential sliding surface, supplemented by cross-verification through rainfall and surface displacement data. Its feasibility was demonstrated through theoretical analysis and numerical simulations. This research provided a new numerical analysis platform for the study of rain-induced landslides and made beneficial explorations for the precise prediction and early warning of landslides.