Abstract: The aim of this work is to confront the problem of the high rate of missed alarms for flash flood warnings and forecasts due to the random nature of rainfall. In order to improve the accuracy of flash flood warnings,considering the transmission and diffusion principle of probability distributions,and by taking the characteristic parameters of rainfall patterns as the control conditions,a random rainfall pattern method based on parameter control is proposed. In addition,a critical rainfall model of flash floods based on random rainfall patterns and an early warning model considering the risk appetite of decision makers are established. Taking the Peihe watershed in Henan Province,as an example,we performed a comparative analysis of the 6 h critical rainfall under different rain pattern sets,and determined the 6 h critical rainfall threshold space for the study area. The results are as follows:the influence ranges of the precipitation peak position coefficient (r),peak multiple ratio (b_max) and antecedent precipitation (P_a) on the critical rainfall are 32% to 34%,33% to 34%,and 15% to 22%,respectively;the rainfall pattern is found to have a greater impact on the critical rainfall than that of the antecedent precipitation;the 6 h critical rainfall threshold space determined by the double factors of r and b_max is 98 to 185 mm. Therefore,the critical rainfall warning model considering the different risk preferences of decision makers is effective,and has the potential for guiding the more reliable implementation of flash flood warning systems.