CHLNET: A novel hybrid 1D CNN-SVR algorithm for estimating ocean surface chlorophyll-a

Developing a unified chlorophyll-a (Chla) inversion algorithm for cross-water types is significant challenge owing to the insufficiency of input features and training samples. Although machine learning algorithms can build consistent model different trophic waters, accuracy dependent on quality extended features. Here, we designed novel hybrid framework called CHLNET, which combines one-dimensional convolutional neural network (1D CNN) support vector regression (SVR). The 1D CNN used extract from original band features, SVR perform fit Chla. CHLNET trained tested using match-up pairs SeaWiFS remote sensing reflectance [Rrs(λ)] in situ with Chla ranging 0.009 mg/m³ 138.046 mg/m³, covers mostly ocean water types. Performance metrics log space were better than those state-of-the-art testing dataset, had best overall performance largest cover area star plot. frequency distribution predicted by was more that While spatial pattern not smooth low concentration demonstrated excellent mapping ability at global local scales high waters. Through band-shift method, transfers Rrs(λ) MERIS MODIS-Aqua visible spectral range, obtained blended OCx CI matchups, validates generalization cross-sensor results indicate avoids drawbacks manually constructing need merging type-appropriate retrieval, as well provides new idea across Thus, may serve an alternative approach inversion.