Enhanced Imputation of Marine Wave Observations Using a Nearest-Neighbors Algorithm With Standardized Energy-Based Wave Features

Abstract

This study addresses the critical issue of missing marine wave observation data, including significant wave height, mean wave period, and mean wave direction, which are essential for oceanographic analyses and marine operations. An imputation model based on the Weighted K-Nearest Neighbors (WKNN) algorithm is proposed, using the square of wave height as the primary input feature. This height-squared formulation, physically motivated by wave energy density being proportional to the square of wave height, has been shown to improve imputation accuracy for missing wave data, particularly when combined with standardization preprocessing. It outperforms the more common but less effective practice of using unsquared wave height values. The model is evaluated using real-world data from four buoys deployed in the northeastern waters of Taiwan. This improvement raises overall data completeness from 63.1% to 98.9%. The model yields physically plausible estimates, demonstrating strong performance in smooth to moderate WMO sea states. In rough-and-above regimes, however, the imputation results can be slightly conservative, including during typhoons. Notably, the proposed approach remains effective even when data from up to half of the buoy stations are unavailable. By generating high-quality imputed data, the model directly enhances the reliability of real-time marine monitoring and provides robust support for wave climate analysis and marine energy assessments. The results highlight the computational efficiency, robustness, and practical applicability of the WKNN algorithm in operational oceanographic systems.