Enhancement of Coastal Ocean Reanalysis Using Spatial Transfer of Sparse Observational Constraints

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Coastal ocean reanalyses provide spatially continuous water-level information that supports coastal science and hazard assessment beyond tide-gauge locations. Physics-based models provide the foundation for these reanalyses, particularly when data assimilation constrains the low-frequency response, but the tidal band still offers room for improvement. Here, we examine the extent to which errors in the NOAA Coastal Ocean Reanalysis (CORA) can be diagnosed and mitigated using sparse tide-gauge observations across contrasting coastal regimes. CORA mean-water-level differences are generally centered near zero, consistent with its assimilation of low-frequency water-level residuals. However, analysis of the tidal component shows that CORA predominantly underpredicts mean higher high water (MHHW), while errors in great diurnal range (GT) vary spatially in sign and magnitude. Coherent regional constituent-amplitude and phase errors are identified, and bias in the locally dominant constituent is strongly associated with spatial variation in MHHW bias.We develop a regime-aware hybrid machine-learning framework that combines geographic, tidal, and topobathymetric characteristics with nearby tide-gauge information and residual spatial covariance. Independent and spatially separated validation shows that the framework reduces MHHW and GT discrepancies at stations excluded from model development. Information-ablation experiments indicate that tidal characteristics provide key transferable information, with nearby observations and residual spatial dependence providing additional, datum-specific refinement. These results demonstrate that sparse observational constraints and data-science methods can enhance physics-based coastal reanalysis by diagnosing and mitigating spatially structured tidal-band errors.

This publication has 23 references indexed in Scilit: