The Delaware River Basin (DRB) covers portions of five states (Delaware, Maryland, New Jersey, New York, and Pennsylvania) and several geologic provinces, encompassing much of the complex geology of the Mid-Atlantic region. This data release focuses on the recently glaciated northern DRB, which includes portions of New Jersey, New York, and Pennsylvania. Groundwater storage is conceptualized to be greatest in the glacial surficial aquifers in the upper part of the basin, thus characterization of this critical zone is of primary importance for USGS Next Generation Water Observing System (NGWOS) modeling of baseflow to the upper Delaware River. In support of this effort, we trained four deep learning models to classify surficial materials in unique physiographic areas of the northern DRB, using previously published surficial geologic maps as training data. First, we compiled existing digital surficial geologic map data at various scales (1:100,000 to 1:24,000), with high-resolution data taking precedent where available. Next, we generalized the compiled map data to the following categories: alluvium, anthropogenic, bedrock, colluvium, drumlins, glaciofluvial, glaciolacustrine (coarse), ice-contact stratified, marsh, moraine, pre-Illinoian glacial, till, and water bodies. We then compiled lidar data for the entire northern DRB and generated a derivative RGB composite raster to facilitate the training and running of convolutional neural networks (Odom and Doctor, 2023; Maxwell et al., 2023). The resultant geologic data and imagery were then clipped to four distinct physiographic areas: "Eastern Valley and Ridge", "Northern Glaciated Plateau", "Terminal Moraine Plateau", & "Western Valley and Ridge". We then trained a DeepLabV3+ pixel classification model for each region, reserving 10% of geologic training polygons for validation during the training process. The resultant models, which had training times ranging from 21-41 hours, featured F1 values ranging from 0.85-0.91 at the final epoch of training. External validation using geologic maps likewise demonstrated satisfactory performance. The trained models were then run on their respective physiographic areas, and the results joined in a mosaic classification of the northern DRB. References: Maxwell, A.M., Odom, W.E., Shobe, C.M., Doctor, D.H., Bester, M.S., Ore, T.M, 2023. Exploring the Influence of Input Feature Space on CNN-Based Geomorphic Feature Extraction From Digital Terrain Data. Earth and Space Science, 10. https://doi.org/10.1029/2023EA002845 Odom, W.E., Doctor, D.H., 2023. Rapid estimation of minimum depth-to-bedrock from lidar leveraging deep-learning-derived surficial material maps. Applied Computing and Geosciences, 18. https://doi.org/10.1016/j.acags.2023.100116