Carolina Bays are oriented and sandy-rimmed depressions that are ubiquitous throughout the Atlantic Coastal Plain (ACP). Their origin has been a highly debated topic since the 1800s and remains unsolved. Past population estimates of Carolina Bays have varied vastly, ranging between as few as 10,000 to as many as 500,000. With such a large uncertainty around the actual population size, mapping these enigmatic features is a problem that requires an automated detection scheme. Using publicly available LiDAR-derived digital elevation models (DEMs) of the ACP as training images, various types of convolutional neural networks (CNNs) were trained to detect Carolina bays. The detection results were assessed for accuracy and scalability, as well as analyzed for various morphologic, land-use and land cover, and hydrologic characteristics. Overall, the detector found over 23,000 Carolina Bays from southern New Jersey to northern Florida, with highest densities along interfluves. Carolina Bays in Delmarva were found to be smaller and shallower than Bays in the southeastern ACP. At least a third of Carolina Bays have been converted to agricultural lands and almost half of all Carolina Bays are forested. Few Carolina Bays are classified as open water basins, yet almost all of the detected Bays were within 2 km of a water body. In addition, field investigations based upon detection results were performed to describe the sedimentology of Carolina Bays. Sedimentological investigations showed that Bays typically have 1.5 m to 2.5 m thick sand rims that show a gradient in texture, with coarser sand at the bottom and finer sand and silt towards the top. Their basins were found to be 0.5 m to 2 m thick and showed a mix of clayey, silty, and sandy deposits. Last, the results compiled during this study were compared to similar depressional features (i.e., playa-lunette systems) to pinpoint any similarities in origin processes. Altogether, this study shows that CNNs are valuable tools for automated geomorphic feature detection and can lead to new insights when coupled with various forms of remotely sensed and field-based datasets.

Yes. Equilibrium shoreline models, which simulate wave‐driven cross‐shore erosion and accretion, are mathematically equivalent to a discrete convolution (i.e., a weighted, moving average) of a time series of wave‐forcing conditions with a parameterized memory‐decay kernel function. The direct equivalence between equilibrium shoreline models and convolutions reveals key theoretical aspects of equilibrium behavior. Convolutions (representing quasi‐low‐pass filter operations) provide an intuitive theoretical description of shoreline erosion and accretion behavior in response to waves: that is, shoreline position often mirrors the weighted moving average of wave time series. Model‐convolution equivalence also provides a conceptual basis to interpret, evaluate, and construct data‐driven Machine‐Learning/Deep‐Learning (ML/DL) models that use convolutions to extract features from data and then apply them for prediction (e.g., Convolutional Neural Networks (CNNs)). Finally, our findings provide a methodological pathway (based on Fourier transforms) for future understanding of wave‐driven shoreline change, which can be used to interpret the coherence between the frequency spectrum of the processes of waves and shoreline change and construct more computationally efficient and effective shoreline‐modeling approaches.

Satellite-derived shorelines (SDS) have the potential to help researchers answer critical coastal science questions and support work to predict coastal change by filling in the spatial and temporal gaps present in current field-based and remote-sensing data collection methods. The U.S. Geological Survey conducted comparison analyses of traditionally sourced shorelines and SDS in diverse coastal landscapes to determine how SDS could be used in ongoing and future work across varied coastal environments and provided some initial findings that could be used for implementation. Using CoastSeg, a browser-based program for SDS detection and mapping, SDS for the period 1984–2023 for multiple locations across the United States were compared to shoreline positions from traditionally sourced shoreline data. In this report, the authors present these comparisons alongside lessons learned and challenges encountered when building SDS workflows in different coastal locations. Results show that individual SDS have larger uncertainty and yet produced similar linear trends to sparser, traditionally sourced shoreline data; because SDS methods provide orders of magnitude more data than traditional shoreline-detection methods, they can be used to evaluate shoreline behaviors. Refining average scalar slopes used in tidal corrections did not result in substantial decreases in uncertainty. Using lessons from this work to outline needs for regional implementation, initial setup time would be considerable, being on the order of weeks. However, once complete, shoreline detections and analyses are fast (on the order of minutes to hours) and achievable using a desktop computer.

For over a century, the enigmatic Carolina bays have captivated geologists and spurred contentious debate on their origin. These circular to ovate and shallow (median diameter of 222 m, median depth of 2.17 m, median area of 26,249 sq. m) depressions span the Atlantic Coastal Plain (ACP) from northern Florida to southern New Jersey, with total counts ranging between 10,000 and 500,000. Using 1 meter gridded, 1.7 km by 1.7 km LiDAR digital elevation models (DEMs) of Delaware as training images, a convolutional neural network (CNN) was trained to detect Carolina bays. With such a large population size and with such uncertainty around the actual population size, mapping the Carolina bays is a problem that requires an automated detection scheme. Manual detection of bays from LiDAR across the entire Atlantic Coastal Plain would be extremely time intensive and prone to human annotation errors. Using Faster R-CNN within the TensorFlow Python library, a network was trained on 978 LiDAR images for 24 hours (42,450 iterations) on an Intel Core i7-4790K CPU at 4.00 GHz. This network automatically detects bays from LiDAR images with a bounding box and a confidence level. These bounding boxes can then be used to subset and then analyze regions of the DEM for statistics on the bays’ three-dimensional shape. Extending this algorithm to DEMs from other areas of the ACP will provide a better understanding of the bays’ geographic distribution as well as any differences in morphology between different geographic regions. This method for detecting geomorphic features is a highly efficient process that will provide better means for mapping various types of abundant geomorphic features in the future.

This study describes a machine learning model cascade for automating large‐scale shoreline mapping from time‐series of coastal satellite imagery. Our model cascade consists of independent supervised deep learning models for image suitability, image segmentation, and image segmentation suitability. The segmentation model determines shoreline locations, while the other models in this cascade filter imagery not suitable for segmentation or segmentations not suitable for shoreline detection. The SegFormer‐based segmentation model is trained with almost 2,300 labeled satellite images from a large variety of globally distributed coastal environments. The image segmentation model operates on imagery that has been deemed suitable by an image classification model that classifies whole satellite images as either suitable or unsuitable for segmentation. Another supervised whole image classification model is used to determine the suitability of each segmentation output. Finally, a spatial kernel density filters removes outlier shoreline points, with further timeseries analysis techniques to remove any remaining outliers. This study demonstrates the effectiveness of this system at two sites that are challenging to map: one at a sandy spit near Moses Point Fishing Village, Alaska, USA, and one at the mouth of the Elwha River, Washington, USA. This study demonstrates that these automated techniques produce shorelines with similar or better accuracies when compared to semi‐automated techniques. Ensembling model outputs better constrains the error and uncertainty of shoreline positions. Together, the set of models presented in this paper constitute an operational workflow for the generation and quality control of large‐scale records of satellite‐derived shoreline change.