Abstract: Accurate sea surface positioning and extraction are critical factors influencing the measurement precision and bathymetric accuracy of airborne oceanographic lidar. When 532 nm lidar is used for sea surveying and mapping, the scattering and absorption of sea surface waves and near-surface water after laser penetrating through the air-sea interface will degrade the accuracy of sea surface positioning at this wavelength. This study employed a Bidirectional Long Short-Term Memory (Bi-LSTM) network for sea surface positioning. According to Bi-LSTM method, the complete sea surface return waveforms from dualwavelength detection (1064 nm and 532 nm) were used for training, and then the 1064 nm detection results were used as ground truth to determine the sea surface positions of 532 nm channel. Application of the trained model to 532 nm detection data showed that the average positioning bias using Bi-LSTM algorithm was 0.03 m, while that of conventional peak detection algorithm was −0.21 m. Bi-LSTM algorithm was further used to quantitatively evaluate the influence of seawater diffuse attenuation coefficient on the bias, and the results showed that the minimum bias of 0.03 m occurred when the seawater diffuse attenuation coefficient was in the range of 0.05–0.15 m-1.

Key words: airborne laser bathymetry, marine lidar, sea surface position, deep learning, optical properties of seawater