<p>Post-earthquake debris flows pose significant risks to property and lives in earthquake-affected areas, making the development of robust prediction models essential. While most existing models rely primarily on rainfall data, this approach often results in frequent false positives due to the neglect of hydrological conditions. Furthermore, many prediction models use power-law thresholds, which struggle to account for the complex nonlinear relationships between debris flow occurrence and triggering factors. To fill the current research gaps, based on the monitoring data from the first year following the 2022 Luding Earthquake (Ms 6.8), this study focuses on enhancing predictive accuracy for post-earthquake debris flows by leveraging hydrological monitoring data and machine learning. Unlike susceptibility or risk assessments, we target the prediction of debris flows occurrence. We employed two predictive models: logistic regression (LR) and random forest (RF). LR was selected as a benchmark to represent widely used traditional statistical methods, while RF was chosen as a representative ensemble method. The results show that the RF model outperformed the LR model. Moreover, models that included hydrological variables (such as observed soil water content and calculated peak discharge) performed better than those relying solely on rainfall features. Among the proposed models, those incorporating mean rainfall intensity, rainfall duration, and maximum soil water content exhibited the best performance. Importance analysis of the input parameters also revealed that hydrological variables, particularly calculated peak discharge, were more critical than conventional rainfall features. These findings highlight the importance of hydrological variables in triggering post-earthquake debris flows and demonstrate that including these factors may improve the accuracy of post-earthquake debris flow predictions.</p>