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

We evaluated six empirical and semianalytical models of the diffuse attenuation coefficient at 490 nm (K_d(490)) using an in situ dataset collected in the Pearl River estuary (PRE). A combined model with the most accurate performance (correlation coefficient, R² = 0.92) was selected and applied for long-term estimation from 2003 to 2017. Physical and biological processes in the PRE over the 14-year period were investigated by applying satellite observations (MODIS/Aqua data) and season-reliant empirical orthogonal function analysis (S-EOF). In winter, the average K_d(490) was significantly higher than in the other three seasons. A slight increasing trend was observed in spring and summer, whereas a decreasing trend was observed in winter. In summer, a tongue with a relatively high K_d(490) was found in southeastern Lingdingyang Bay. In Eastern Guangdong province (GDP), the relatively higher K_d(490) value was found in autumn and winter. Based on the second mode of S-EOF, we found that the higher values in the eastern GDP extended westward and formed a distinguishable tongue in winter. The grey relational analysis revealed that chlorophyll-a concentration (C_chla) and total suspended sediment concentration (C_tsm) were two dominant contributors determining the magnitude of K_d(490) values. The C_tsm-dominated waters were generally located in coastal and estuarine turbid waters; the C_chla-dominated waters were observed in open clear ocean. The distribution of constituents-dominated area was different in the four seasons, which was affected by physical forces, including wind field, river runoff, and sea surface temperature.

An attention U-Net was proposed to reconstruct the missing chlorophyll-a concentration (C_chla) data. The U-Net is a lightweight full convolution neural network architecture consisting of an enccoder-decoder (i.e., down-sampling and up-sampling). The attention gates (AGs) were integrated into the U-Net. Training the U-Net with AGs could implicitly teach it to suppress irrelevant areas and highlight the salient features in the missing data areas, which would increase the network sensitivity and reconstruction accuracy. The neural network uses the satellite-derived C_chla anomalies and its variance as the input, and the reconstructed fields along with their variances as outputs. The trained network was applied to long-term daily MODIS/Aqua C_chla products in the Pearl River estuary (PRE) and adjacent continental shelf area. The model performance was evaluated by using an independent test dataset from both satellite-derived and in-situ measurements. The results showed that the proposed neural network not only had good performance in the reconstruction of valid pixels, but also provided a more reasonable reconstruction compared to the standard U-Net without AGs. This study provided a feasible method for the reconstruction task in the field of ocean color, which should be helpful in producing a creditable dataset to study the ecological effects of extreme weather conditions such as typhoons on the upper ocean in the PRE waters. Based on the reconstructed C_chla products, the footprints of the typhoons were studied. An increase in surface C_chla near the typhoons’ track and a decrease in estuary were found. The composite results illustrated that the C_chla increases occurred for almost the entire area within a radius of 100 km. The time series analysis showed that the C_chla peak appeared on the fifth day after the typhoon’s passage.

Different substrate types, such as seagrass, sand, stone and silt, can be identified by detecting substrate reflectivity. In this study, high-resolution remote sensing data of Quickbird were used to map seagrass distribution in the northeast coast of the Xincun Bay. Radiance transfer model in optical shallow water was used to retrieve bottom reflectivity; relationship between hyperspectral band and leaf area index (LAI) was used for processing satellite remote sensing data to retrieve distribution and density of submerged seagrass. Hyperspectral results showed that spectral bands at 550, 650 and 675 nm were sensitive to LAI. Based on the relationship between LAI and hyperspectral bands, Bands 2 and 3 of Quickbird could be used for retrieving seagrass distribution. Satellite results showed that seagrass was mainly distributed in clusters in the northeast coast of the Xincun Bay, tens of metres away from the bank. Density gradient was very clear, with coverage between 40 and 60% occupied by a great part of seagrass bed and with coverage greater than 80% distributed in stripe pattern in the main seagrass bed. The accuracy of seagrass retrieval was more than 80%, and the density of seagrass could also be distinguished evidently.

Phytoplankton dynamics, which are highly sensitive to the ecosystem condition and change, are different in coastal waters and open ocean. Previous researches mainly focused on the open ocean dynamic in the South China Sea (SCS), but few research on the Eastern Guangdong (EGD), especially during a period of downwelling-favorable winds. In fact, the influence of topographic irregularities, downwelling, continental outflows, and cold, nutrient-rich currents from Fujian-Zhejiang coastal waters (ZFC) of the East China Sea (ECS) on the spatial distribution of Chlorophyll a (Chl a) in EGD coastal waters could be significant. This study utilized ocean color time series and meteorological and hydrographic data jointly from July 2002 to June 2020 to investigate the heterogeneous regional specific distribution pattern of Chl a in EGD and the environmental determinants in different regions subdivided by water depth. Then the temporal and spatial distribution pattern of Chl a in EGD in winter were discussed in detail by applying an Empirical Orthogonal Function (EOF) analysis, GaoFeng-1 (GF-1) satellite data and in situ measured dataset. The sea surface temperature (SST) in February was negatively related with Chl a in shallow coastal waters less than 60 m deep due to the nutrient-rich, cold waters from ZFC. The monthly mean photosynthetically active radiation (PAR) and precipitation showed significant effects on the phytoplankton growth over regions with a depth less than 10 m. An area with higher Chl a concentration in the downwelling zone were detected in winter. By an examining the temporal variability in meridional distribution of the mean Chl a at 22.41°N and 22.21°N, a symmetrical peak was observed. The coastal fronts extended southwestward from the southeastern coast of Guangdong Province to Dangan Island in the SCS. In addition, a cross-shelf filament was detected near the coast of Shanwei, Guangdong, China on 8 January 2020.

A coastal front was detected in the eastern Guangdong (EGD) coastal waters during a downwelling-favorable wind period by using the diffuse attenuation coefficient at 490 nm (K_d(490)). Long-term satellite data, meteorological data and hydrographic data collected from 2003 to 2017 were jointly utilized to analyze the environmental factors affecting coastal fronts. The intensities of the coastal fronts were found to be associated with the downwelling intensity. The monthly mean K_d(490) anomalies in shallow coastal waters less than 25 m deep along the EGD coast and the monthly mean Ekman pumping velocities retrieved by the ERA5 dataset were negatively correlated, with a Pearson correlation of − 0.71. The fronts started in October, became weaker and gradually disappeared after January, extending southwestward from the southeastern coast of Guangdong Province to the Wanshan Archipelago in the South China Sea (SCS). The cross-frontal differences in the mean K_d(490) values could reach 3.7 m⁻¹. Noticeable peaks were found in the meridional distribution of the mean K_d(490) values at 22.5°N and 22.2°N and in the zonal distribution of the mean K_d(490) values at 114.7°E and 114.4°E. The peaks tended to narrow as the latitude increased. The average coastal surface currents obtained from the global Hybrid Coordinate Ocean Model (HYCOM) showed that waters with high nutrient and sediment contents in the Fujian and Zhejiang coastal areas in the southern part of the East China Sea could flow into the SCS. The directions and lengths of the fronts were found to be associated with the flow advection.