Connectivity-informed drainage network generation using deep convolution generative adversarial networks

Abstract Stochastic network modeling is often limited by high computational costs to generate a large number of networks enough for meaningful statistical evaluation. In this study, Deep Convolutional Generative Adversarial Networks (DCGANs) were applied quickly reproduce drainage from the already generated samples without repetitive long stochastic model, Gibb’s model. particular, we developed novel connectivity-informed method that converts images directional information flow on each node network, and then transforms it into multiple binary layers where connectivity constraints between nodes in are stored. DCGANs trained with three different types training compared; (1) original images, (2) their corresponding only, (3) information. A comparison demonstrated outperformed other two methods more effectively better reproducing accurate due its compact representation complexity connectivity. This work highlights can be applicable contrast common earth material sciences fractures, features important.