By representing hydraulic events via constructing multiple features of the watershed spatio-temporal scene, this study proposed a construction method of watershed scene pattern library via the spatio-temporal multiple features. The original hydrological data was firstly divided into events to remove the spatio-temporal redundancy of scene element data. Based on the analysis of element association relation, the corresponding features of scene elements were constructed via multiple ways. Afterwards, key features of watershed scene were selected by the feature selection algorithm to realize the scene initialization. Finally, the initial scene was regarded as the feature space, where the cluster extraction of scene pattern and scene pattern library construction could be carried out. Experimental results show that the proposed method can not only extract the key spatio-temporal scene data of hydrological events, but also mine scene patterns to form a scene pattern library, thus providing accurate and efficient prediction results for the hydrological event with small dataset.