Abstract Empirically derived rainfall thresholds of debris flows are used for regional-scale early warning. However, triggering rainfall intensities of post-seismic debris flows evolve with time, causing high false alarms since thresholds estimated by conventional methods ignore the uncertainty of non-triggering rainfall events. Based on 172 triggering rainfalls and 2396 non-triggering rainfalls from 2008 to 2013 after the Wenchuan earthquake, we analyzed the evolution of probabilistic rainfall thresholds for post-seismic debris flows using a Bayesian technique. We found, rainfall thresholds significantly decrease compared with pre-earthquake events initially and later tend to increase annually. Meanwhile, the triggering rainfall characteristics tend to gradually change from a short-duration high-intensity pattern to a long-duration and low-intensity pattern. We also checked the effect of antecedent precipitation on debris flows by defining an IET (inter-event time). Our results suggest the antecedent precipitation plays an important role in low-intensity long-duration rainfall-induced debris flows and has little effect on the short-duration high-intensity rainfall-induced debris flows. The characteristics of triggering rainfall for debris flows after the Wenchuan earthquake can be best reflected by IET = 7 h. By employing a Naive Bayes algorithm, uncertainties of three rainfall threshold models, I-D (mean intensity-duration), IT-D (triggering intensity-duration), and IT-E (triggering intensity-cumulative rainfall) were investigated, and the IT-D model was found to perform best. The characteristics and evolution of debris flow rainfall thresholds over the years after the Wenchuan earthquake was best presented by the Bayesian probabilistic rainfall thresholds. We believe the results can be used to improve the precision of an early warning system for post-seismic debris flows.

中文翻译：

---

汶川地震后泥石流的概率降雨阈值使用贝叶斯技术

摘要 泥石流的经验得出的降雨阈值用于区域尺度预警。然而，地震后泥石流的触发降雨强度随时间变化，导致高误报，因为传统方法估计的阈值忽略了非触发降雨事件的不确定性。基于汶川地震后2008-2013年172次触发降雨和2396次非触发降雨，我们使用贝叶斯技术分析了震后泥石流概率降雨阈值的演变。我们发现，与震前事件相比，降雨阈值最初显着降低，后来趋于逐年增加。同时，触发降雨特征往往由短时高强度型逐渐转变为长时低强度型。我们还通过定义 IET（事件间时间）来检查前期降水对泥石流的影响。我们的研究结果表明，先行降水在低强度长历时降雨诱发的泥石流中起重要作用，对短历时高强度降雨诱发的泥石流影响不大。汶川地震后泥石流触发降雨的特征最能体现为IET=7 h。采用朴素贝叶斯算法，研究了三种降雨阈值模型ID（平均强度-持续时间）、IT-D（触发强度-持续时间）和IT-E（触发强度-累积降雨）的不确定性，并且发现 IT-D 模型表现最好。贝叶斯概率降雨阈值最能体现汶川地震后多年泥石流降雨阈值的特征和演变。我们相信这些结果可用于提高震后泥石流预警系统的精度。