To reduce the serious consequences of flood disaster events, the river discharge should be predicted timely and accurately to provide decision support for the flood forecasting. In this study, an intelligent simulation method of runoff process based on the spatiotemporal feature mining was proposed. Firstly, the topological diagram of hydrological stations was established from the spatial point of view. Then, this study proposed a novel framework that incorporated the graph convolutional networks (GCN) for mining the spatial feature. The gated recurrent unit was adopted to capture temporal features for the hydrological prediction. The experimental results show that the intelligent hydrological forecasting model based on the spatiotemporal feature mining is better than other models with single feature.