Publications

Cui, JY; Cao, XY; Du, C; Dong, W; Liu, SW; Guo, J; Xu, MM; Yasir, M (2025). Chl-a Concentration Inversion Methods for Water Bodies With High TSM Concentrations Based on Waterbody Classification and Deep Learning. IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 18, 5673-5686.

Abstract
In water bodies with high total suspended matter (TSM) concentrations, the scattering and absorption of light by suspended solids alters the spectral characteristics of the water, leading to errors in chlorophyll-a (Chl-a) concentration inversion models. The accuracy of inversion can be improved by classifying water bodies into different types and building Chl-a concentration inversion models for each class. In this study, based on measured TSM and Chl-a concentration data from simultaneous satellite-ground sampling in the Yellow River estuary and nearshore waters, a classification method suitable for high TSM water bodies was developed using a spectral normalization method. A multiband 1D CNN model optimized with the efficient channel attention (ECA) mechanism was built for Chl-a concentration inversion. The incorporation of the channel attention mechanism allows the 1D CNN model to dynamically adjust the weight of each channel, separating the light absorption and scattering characteristics for TSM and Chl-a across different bands, thus improving inversion accuracy in high TSM waters. The study found that spectral normalization reduces atmospheric effects on remote-sensing reflectance, enhancing image classification consistency. TSM concentration determines the water body category. For different types of water bodies, the combination of the 1D CNN and the channel attention mechanism captures interchannel dependencies, further improving Chl-a inversion accuracy.

DOI:
10.1109/JSTARS.2025.3536487

ISSN:
2151-1535