 This paper proposes a new approach to generate realistic atmospheric forcing structures for storm surge modelling. It uses a deep learning method called Generative Adversarial Networks, GAN, which is trained on historical typhoons and then used to generate forcings for future typhoons. This method is faster and more efficient than traditional parametric models, and it produces results that are comparable to those produced by numerical weather prediction models. This article was authored by Ian Emulia, Neonori Raider, Taekmasa Miyoshi, and others.