In this paper, subcatchments of small rivers in Hunan are determined based on DEM data and D8 algorithm. According to the network propagation model for flood forecast of main rivers in Hunan Province, a directed graph related to the subcatchments has been built, which represents the topological relationship among these areas both in temporal and spatial dimension. To build the relation between precipitation and runoff in a reasonable space, the concept of valid rainarea related to the flood graph is proposed. By using the valid rainarea data series and runoff series, the hidden Markov model is built to calculate critical rainarea threshold of flood disasters at different time scales in the subcatchments in Hunan Province. Finally, the observed flood warning data (collected in 2010-2015) are used to test the proposed method. The results show that the new method has similar performance to traditional statistical ways but is more accurate and stable. In addition, the successful application of the new method to the operation of forecasting flood disaster on 13 June 2015 further proved that the new method is suitable for flood forecasting.