Abstract: Based on the data from flood disaster and precipitation in Yingxia railway from 2004 to 2014,the temporal variation and spatial distribution of flood disaster were analyzed,and rainfall thresholds and their probability forecast model of the Yingxia railway for blocking alert were studied.The results show that there is a remarkable change in the flood disaster year by year along the Yingxia railway,normally occurring during the flood season,and meanwhile its spatial distribution is non-uniform with a strong localization feature.The rainfall thresholds for alerting to block the Yingxia railway include 1-h rainfall amount reaching 47 mm before the flood disaster occurs,24-h rainfall amount reaching 108 mm before the flood disaster occurs,the cumulative rainfall amount reaching 228 mm,1-h rainfall amount reaching 30 mm before the flood disaster occurs and meanwhile the cumulative rainfall amount reaching 160 mm,or the 24 h rainfall amount reaching 100 mm before the flood disaster occurs and at the same time the cumulative rainfall reaching 160 mm.The hindcast accuracy rate of the probability forecast model is 90.2%.The accuracy rate to judge whether block the railway or not according to the weather processes is higher when combining the probability forecast model with the alerting rainfall thresholds.This work provides useful guidance for operating the railway safely and efficiently.