Accurate mapping of nearshore bathymetry is essential for coastal management, navigation, and environmental monitoring. Traditional bathymetric mapping methods such as sonar surveys and LiDAR are often time-consuming and costly. This paper introduces BathyFormer, a novel vision transformer- and encoder-based deep learning model designed to estimate nearshore bathymetry from high-resolution multispectral satellite imagery. This methodology involves training the BathyFormer model on a dataset comprising satellite images and corresponding bathymetric data obtained from the Continuously Updated Digital Elevation Model (CUDEM). The model learns to predict water depths by analyzing the spectral signatures and spatial patterns present in the multispectral imagery. Validation of the estimated bathymetry maps using independent hydrographic survey data produces a root mean squared error (RMSE) ranging from 0.55 to 0.73 m at depths of 2 to 5 m across three different locations within the Chesapeake Bay, which were independent of the training set. This approach shows significant promise for large-scale, cost-effective shallow water nearshore bathymetric mapping, providing a valuable tool for coastal scientists, marine planners, and environmental managers.

Our paper showcases the potential gain in scientific insights about blue carbon stocks (or total organic carbon) when additional rigor, in the form of a spatial autocorrelation component, is formally incorporated into the statistical model for assessing the variability in carbon stocks. Organic carbon stored in marsh soils, or blue carbon (BC), is important for sequestering carbon from the atmosphere. The potential for marshes to store carbon dioxide, mitigating anthropogenic contributions to the atmosphere, makes them a critical conservation target, but efforts have been hampered by the current lack of robust methods for assessing the variability of BC stocks at different geographic scales. Statistical model-based extrapolation of information from soil cores to surrounding tidal marshes, with rigorous uncertainty estimates, would allow robust characterization of spatial variability in many unsampled coastal habitats. In the absence of BC data, we consider a historical dataset (the best available) on soil organic matter (OM)—a close proxy of BC—on 36 tidal (fresh and salt) marshes in the Virginia portion of Chesapeake Bay (CBVA) in the USA. We employ Bayesian linear mixed(-effects) modeling to predict OM by marsh type, soil category, soil depth, and marsh site, whereby site effects are modeled as random. When the random site effects are additionally assumed to exhibit an intrinsic conditional autoregressive (ICAR) spatial dependence structure, this more complex model clearly suggests groupings of marsh sites due to their spatial proximity, even after adjusting for the remaining predictors. Although the actual membership of each group is not a focus of our proof-of-concept analysis, the clear presence of groupings suggests an underlying latent spatial effect at the localized-regional level within CBVA. In contrast, the non-spatially explicit model provides no clear indication of either spatial influence between sites or improvement in predictive power. The polar difference in conclusions between models reveals the potential inadequacy in relying on predictor variables alone to capture the spatial variability of OM across a geographic domain of this size or larger. We anticipate that spatially explicit models, such as ours, will be important quantitative tools for understanding actual carbon measurements and for assessing BC stocks in general.