Layered refinement algorithm for coastal wind speed reconstruction from sparse buoy observations

Abstract

Coastal wind fields play a critical role in marine meteorology, offshore wind energy assessment, and coastal disaster prevention. However, complex terrain features, such as mountains, islands, and irregular shorelines, introduce significant spatial heterogeneity, leading to pronounced biases, distortions, and frequent missing values in nearshore wind speed observations from reanalysis datasets or satellite products. Existing correction methods often suffer from weak structural sensitivity and heavy reliance on dense data. To address these limitations, we propose a Layered Spatial Data Reconstruction (LSDR) algorithm for accurate coastal wind speed correction under sparse data conditions. Based on the Inverse Distance Weighting–Random Forest (IDW-RF) framework, the method first develops a more flexible Layered Residual-Corrected Random Forest (LRC–RF) model. Second, global IDW interpolation is replaced with a clustered radial basis function (RBF) interpolation scheme, enabling adaptive, region specific weighting that better preserves terrain-induced wind gradients. Third, sparse buoy observations are strategically integrated to guide residual correction through the Random Forest regressor, providing physically grounded anchors that mitigate error propagation in data scarce areas. The approach is evaluated using wind speed datasets along the Guangdong–Fujian coast, under two realistic coastal masking scenarios: edge erosion (simulating shoreline data loss) and random circular masks (mimicking isolated observation gaps). The results show that the proposed LSDR algorithm, enhanced by sparse buoy truth data, achieves substantial improvements over baseline methods, reducing RMSE by 6.97%, MAE by 11.36%, and GRMSE by 9.08% on average across the experimental settings. The reconstructed fields exhibit enhanced physical consistency, such as realistic wind speed gradients near complex coastlines and reduced artificial smoothing artifacts. In conclusion, the LSDR approach offers a robust, data-efficient solution for coastal wind field reconstruction, with strong potential for applications in renewable energy planning, numerical weather prediction downscaling, and marine environmental monitoring in observation-limited regions.