Deep learning-based efficient drone-borne sensing of cyanobacterial blooms using a clique-based feature extraction...

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Harmful algal blooms (HABs) pose significant threats to inland water ecosystems.
• Monitoring HABs spatiotemporally is crucial for environmental management.
• Hyperspectral remote sensing offers a promising approach for HAB detection.

Purpose of the Study:

• To develop an efficient and accurate deep learning bio-optical model for HAB monitoring.
• To apply the model for retrieving algal pigment concentrations using hyperspectral data.
• To assess the model's performance in bloom-prone regions.

Main Methods:

• A hierarchical concatenated variational autoencoder (HCVAE) was developed.
• Layer-wise clique-based latent-feature extraction was employed to reduce computational load.
• Graph-based clique detection grouped related reflectance spectra.
• The model simultaneously estimated chlorophyll-a (Chl-a) and phycocyanin (PC) concentrations.

Main Results:

• HCVAE achieved high accuracy in estimating Chl-a (R²=0.76) and PC (R²=0.82) concentrations.
• Spatial distribution maps from drone-borne data successfully identified blooming spots.
• Shapley additive explanations identified key spectral regions for pigment estimation.
• A lightweight network (DNNsel) performed comparably to HCVAE.

Conclusions:

• The HCVAE's multilevel hierarchical architecture is effective for near-real-time HAB assessment using drone-borne sensing.
• HCVAE demonstrates utility for environmental big data analysis with numerous features.
• The model provides a comprehensive approach for monitoring HABs in inland waters.