Optimal water depth bias correction in LiDAR data using deviation-level analysis and advanced deep hybrid models

Water depth bias correction in Light Detection and Ranging (LiDAR) point cloud data is essential to provide accurate depth estimation, which has direct applications in hydrological modeling, infrastructure planning and environmental monitoring. The variability in water depth bias during depth correction modeling has not been a significant focus in the majority of prior research. Hence, this paper aims to propose an optimal model for water depth bias correction from bathymetric LiDAR point cloud data. To begin with, the LiDAR water depth dataset from US Geological Survey is processed with Hadoop map framework for parallel efficiency. Afterwards, clustering via Calinski–Harabaz Density-Based Spatial Clustering of Applications with Noise (CH-DBSCAN) to categorize the LiDAR points into dense and sparse areas. This is followed by Linear Interpolation (LI), Exponential Smoothing, Min-Max normalization, Independent Component Analysis (ICA), and sub-sampling. Dimensionality reduction is performed by T-distributed Radial Basis Covariance Embedding (TRBCE), with a compression ratio of 10.12 outperforming that of T-SNE by 18.1% and PCA by 398.4%. A sliding window technique is applied to segment the time-series data, from which geometric features are obtained. Bias correction was effectively achieved by Adaptive Weighted Bayesian – Linearly Scaled Hyperbolic – Long Short Term Memory (AWB-LSH-LSTM) model, and it significantly reduces the correction time by 6.67% to nearly 23.35%. Also, it demonstrates substantial error reduction across all metrics such as MAPE, RAE, MAE, MSE, and RMSE and improved the NSE and R 2 scores ranging from 3%-38% and 3%-47% respectively. The optimization strategy had recorded lowest fitness across all iterations, thereby shortening runtime compared with state-of-art methods. Overall, the proposed method demonstrates superior performance in clustering, data compression, bias correction accuracy, and computational efficiency for large-scale LiDAR depth processing.