Dataset for Deep semantic segmentation for identifying groundwater preferential flow channels in heterogeneous porous media

This dataset supports the study “Deep semantic segmentation for identifying groundwater preferential flow channels in heterogeneous porous media.” It contains synthetic two-dimensional and three-dimensional datasets developed for deep-learning-based identification of groundwater preferential flow channels in heterogeneous porous media. The two-dimensional datasets include log-permeability fields, hydraulic head fields, Darcy velocity components, and preferential-flow-channel labels generated from groundwater flow simulation and particle tracking. The three-dimensional datasets consist of heterogeneous log-permeability volumes with dominant and competing high-permeability corridors, together with aligned hard-channel, soft-corridor, and centerline labels. The data support the training, validation, and independent testing of the semantic-segmentation framework described in the associated study.