This study proposes a deep learning model based on a convolutional neural network, which can effectively fuse atmospheric information (wind field) and station water level information and can effectively forecast the station water level and also have a good response to the anomalous water level increase brought by storm surge.