Deep Learning for Chlorophyll-a Concentration Retrieval: A Case Study for the Pearl River Estuary

• DOI: 10.3390/rs13183717
• Publication Date: 2021-09-22T07:47:35Z
• Authors: Haibin Ye; Shilin Tang; Chaoyu Yang
• Keywords: Sun photometer

The abundance of phytoplankton is generally estimated by measuring the chlorophyll-a concentration (Cchla), which an important factor in photosynthesis and can be used to analyze density biomass ecosystem. band-ratio-based empirical or semi-analytical algorithms are operationally applied retrieve Cchla global oceans, experience difficulties from diversity optical properties complexity radiative transfer equations analytical analyses, respectively. With attempt develop accurate retrieval model for optically complex coastal estuarine waters, this study aimed explore deep learning (DL) methods satellite Cchla. A two-stage convolutional neural network (CNN), named Cchla-Net, was proposed, utilized spectral information remote sensing reflectances at MODIS/Aqua’s visible bands. In first-stage phase, Cchla-Net pretrained a set patches, generated existing (OC3M). results were than as initial values refine with synthetic oversampled in-situ dataset second-stage training phase. Using samples new has higher probability reach optimum. quantitative analyses showed that more likely achieve optimum optimization one-stage training. Matchups measurements evaluate models. Results proposed produced obvious better performance algorithms, implying DL method effective waters extremely high This provided applicable Cchla, should helpful studying spatial distribution temporal variability productive Pearl River estuary (PRE) waters.