What question did this study set out to answer?

To develop a robust marine ecosystem assessment and prediction system employing LiDAR technology and deep learning.

March 25, 2026

Marine ecosystem assessment and prediction system based on LiDAR

YZYuanzhang ZhaoChinese Academy of Tropical Agricultural Sciences BZBen ZhangNanjing University of Science and Technology HZH. ZhaoNanjing University of Science and Technology

Key Points

To develop a robust marine ecosystem assessment and prediction system employing LiDAR technology and deep learning.
Utilized ship-borne LiDAR for data collection with dual-band configuration.
Applied CNN-LSTM for wave-height forecasting and a Raman-normalized 1D-CNN for chlorophyll-a inversion.
Employed an adaptive subsurface phytoplankton layer detection method (IASPLDM).
Implemented GPU-accelerated data preprocessing and Bayesian optimization for training.
Phytoplankton profiling achieved beyond 50 m under moderate sea states.
Biological noise suppressed by over 60% relative to conventional acoustic methods.
Wave forecast correlation exceeded 0.998.
Chlorophyll-a inversion error maintained at ≤0.5 μg/L.
Layer detection recorded ≤5% miss rate and ≤3% false-alarm rate.

Abstract

We present a ship-borne, LiDAR-enabled marine ecosystem assessment and prediction system that overcomes sparse coverage, shallow sensing, and daylight dependence of conventional platforms. A dual-band configuration (905 nm surface LiDAR; 532 nm fluorescence/Raman channels) is coupled with deep learning: a CNN-LSTM for wave-height forecasting, a Raman-normalized 1D-CNN for chlorophyll-a inversion, and an adaptive subsurface phytoplankton layer detection method (IASPLDM). The pipeline spans data acquisition, GPU-accelerated preprocessing, and Bayesian-optimized training. Under moderate sea states, the system profiles phytoplankton beyond 50 m and suppresses biological noise by >60% relative to acoustics; wave forecasts reach correlation >0.998; chlorophyll-a inversion error is ≤0.5 μg/L; and layer detection attains ≤5% miss and ≤3% false-alarm rates. These capabilities enable early warning of harmful algal blooms, fishery habitat management, and blue-carbon assessment with high spatiotemporal resolution, robustness in low-light or turbid waters, and readiness for scalable operations.

AI से पूछें

Bookmark

View Full Paper