UAV-based hyperspectral imaging and deep learning for mapping fouling-related water quality indicators in seawater...

Harmful algal blooms (HABs) are increasing in frequency and persistence under climate change and anthropogenic nutrient enrichment. This proliferation poses growing challenges to seawater reverse osmosis (SWRO) desalination through accelerated membrane fouling. During HAB conditions, conventional point-based water quality monitoring often fails to capture the spatial heterogeneity of fouling-relevant water quality, limiting proactive intake and pretreatment decision-making. This study demonstrates an unmanned aerial vehicle (UAV)-based hyperspectral imaging and deep learning framework for spatially resolved assessment of fouling-related water quality indicators under HAB conditions. Hyperspectral imagery (400-1000 nm, 5 nm spectral resolution) was acquired across pre-bloom, bloom peak, and post-treatment stages. Concurrent water sampling and laboratory analyses of silt density index (SDI), total organic carbon (TOC), and transparent exopolymer particles (TEP) were used as fouling-related target variables. Partial least squares regression (PLSR) identified linearly important wavelengths, and variable importance in projection (VIP) scores quantified spectral contributions. VIP-weighted spectral features were integrated into a one-dimensional convolutional neural network (1D-CNN) to capture nonlinear relationships between spectral features and target variables. The VIP-weighted 1D-CNN achieved R² values of 0.87, 0.78, and 0.65 for SDI, TOC, and TEP, respectively (RMSE: 0.87, 2.70, 1.29). Spatially continuous prediction maps revealed pronounced gradients and localized hotspots not discernible from conventional RGB imagery, highlighting residual organic- and biopolymer-related water quality deterioration even post-treatment. UAV-based hyperspectral deep learning provides a practical framework for quantifying HAB-induced fouling-related water quality indicators and generating spatially resolved information that can support more informed intake management and pretreatment responses in coastal desalination systems.