Sentinel-2 MSI offers improved spatial resolution over sensors dedicated to water observations (e.g. OLCI). This makes it an attractive sensor to attempt water quality observations in coastal regions and inland water bodies. However, the band configuration of Sentinel-2 MSI is not optimized to resolve the optical features that are diagnostic of waterbody health. It is therefore worth investigating whether the SNR statistics of the MSI are good enough for water quality retrieval.

The predominant source of error in remote sensing of water quality is atmospheric correction, followed by selection of suitable water constituent retrieval algorithms. Here we present a comparative analysis of atmospheric correction solutions for MSI including Polymer, C2RCC, Acolite, iCOR, l2gen and Sen2Cor. In situ data from optically diverse sources (Baltic Sea, Western English Channel, and lakes in Estonia, Italy and the Netherlands) are used to generate > 1000 match-ups to compare the suitability and limitations of the different tools for different water environments. We further show how the dynamic algorithm selection of the Calimnos processing chain is implemented for Sentinel-2, leading to retrieval of a set of distinct Optical Water Types which describes how MSI 'sees' waterbodies.

The GaoFen-4 (GF-4) remote sensing satellite is China's first civilian high-resolution geostationary optical satellite, which has been launched at the end of December 2015. The GF-4 has the unique advantages of high temporal resolution (20s) and high spatial resolution (50m). In order to explore GF-4’s potential in ocean bloom monitoring, the GF-4 images were used in red tide detection in the Bohai Sea and drifting velocity of the green tide in the Yellow Sea. Results showed that the GF-4 images had great potential in small patches of red tide detection and could provide data support for accurate monitoring of green tide short-term movement.

Daily coverage of single ocean color sensor can only reach 10%~15% because of cloudy and rainy weather, solar flare, track gap, etc. Merging datasets from different missions into unified data products is a valid way to increase the spatial and temporal coverage of ocean color satellites.

ESA started the GlobColour project in 2005 with the aim of providing a continuous dataset of merged Level 3 ocean color products (1997–present). Based on long time series of merged ocean color products (2003–2016) from the ESA GlobColour dataset, spatial and temporal distributions on effective observation days of ocean color satellites of the East Yellow and East China Seas are analyzed. The results show that the average number of effective observation days for the East China Seasregion is 51±6.8 days per year. Large numbers of such days appeared for the West Korea Bay and western Liaodong Bay, at 100±8.3 days per year. Small numbers of days appeared for the southwestern southern Yellow Sea, northwestern East China Seas, and the branch of Kuroshio Current region, at 40±10.1 days per year. Effective observation days of the Yellow and East China Seas had strong seasonal characteristics. The Bohai Sea, northern Yellow Sea, and southern Yellow Sea had two peaks in March and October. The East China Seas had a single peak in July. A lack of effective observation reduces the quality of monthly ocean color products, with greater than 15% bias if the monthly data are averaged from 3 days and 30% bias if the monthly data are from 1 day only.

SeaWiFS and MERIS stopped successively in 2010~2012 and the number of ocean color sensors decreased. We tried to add Medium Resolution Spectral Imager (MERSI)/FY-3 of China in merged Chl-a products. The results show that the average daily coverage of merged products increases by ~9% when MERSI data are added in the merging process. Sampling frequency (temporal coverage) is greatly improved by combining MERSI data, with the median sampling frequency increasing from 15.6% (~57 days/year) to 29.9% (~109 days/year). The new merged products agree within ~10% of the merged Chl-a product from GlobColour. Time series of the Chl anomalies are similar to GlobColour products.

Environmental monitoring and early warning of water quality from space is now feasible at unprecedented spatial and temporal resolution following the latest generation of satellite sensors. The transformation of this data through classification into labelled, tracked event information is of critical importance to offer a searchable dataset.

Advances in image recognition techniques through Deep Learning research have been successfully applied to satellite remote sensing data. Deep Learning approaches that leverage optical satellite data are now being developed for remotely sensed multi- and hyperspectral reflectance. The combination of spectral with spatial feature extracting Deep Learning networks promises a significant improvement in the accuracy of classifiers using remotely sensed data.

This project aims to re-tool and optimise spectral-spatial Convolutional Neural Networks originally developed for land classification as a novel approach to identifying and labelling dynamic features in waterbodies, such as algal blooms and sediment plumes in high-resolution satellite sensors.

With accelerating climate change and receding summer ice cover, the problems of marine environments in the Arctic Ocean appear to be increasing. Ocean color remote sensing is one of the most effective methods with relatively low cost to monitor marine ecosystems. The quality assurance (QA) system of remote sensing reflectance (Rrs) developed by Wei et al. (2016) is used to assess the MERIS radiometric products in the Arctic Ocean over the 2002 ~2012 time period, which is scored according to the spectral shapes and amplitudes of Rrs spectra. Results show that monthly QA average scores are about 0.75 with little fluctuation (<0.1) and the life-cycle quality of the MERIS radiometric products keeps relatively steady. The majority of the Arctic Ocean has a QA score close to 0.7 while relatively lower QA scores (<0.4) are mainly found in the Kara Sea, Laptev Sea and Hudson Bay. There is less valid MERIS data in the spring and winter, which mainly distributes in the Norwegian Sea and the Denmark Straits with QA scores of about 0.7. More valid MERIS radiometric data are obtained in the summer and autumn, and the radiometric products in the Norwegian and Bering Sea are observed with relatively higher QA scores (>0.8).

Optical satellites of HY-1C and HY-1D will be launched by China in 2018 and 2019, onboard which the Coastal Zone Imager (CZI) is one of the shared payloads. CZI is designed to monitor the coastal zone by providing the optical images in the four bands (blue, green, red and near infrared) with wide swath (950km) and moderate spatial resolution (50m). Quantitative retrieval of the environmental parameters including water quality will be achieved for the coastal zone management.

In this paper, the atmospheric correction algorithm for the operational CZI data processing is proposed. Based on the atmosphere radiation transfer (6S) model and CZI band response functions, lookup tables have been established for the calculation of Rayleigh scattering, aerosol scattering and scattering transmittance for different aerosol models. With these lookup tables, two schemes of atmospheric correction algorithm have been developed. For the clear water with nearly null water-leaving radiance in the CZI red and near infrared bands, the aerosol scattering contributions are firstly estimated at these bands to determine the aerosol models close to actual situation. Then, the aerosol scattering at the blue and green bands is estimated with the lookup tables for the selected aerosol models. For the turbid coastal waters with significant water-leaving contributions in the red and near infrared bands, the aerosol optical depth observed by the Chinese Ocean Color and Temperature Scanner (COCTS), which is concurrent with CZI, is used as auxiliary data to determine aerosol models.

China is rich in neritic and tideland resources. Floating raft aquaculture is an important part of the coastal marine environment monitoring. With rapid development of the aquaculture industry and driven by interests, high-density culturing and occupation of key ecological function areas including core and buffer zones of natural reserves, and illegal use of public facilities protective zones for culturing including ports and waterways have been exacerbated year by year. Therefore, systematic and deep studies on distribution and area of Floating raft aquaculture can provide additional decision-making information for fisheries authorities to rationally plan the Floating raft aquaculture, and offer a reliable scientific basis for controlling culturing density, curbing the deterioration of culturing environment, and preventing and controlling mariculture diseases.

Taking the seawater and the floating raft aquaculture in the Jiangsu Lianyungang offshore area as the classified objects, and the domestic GF1 high-resolution remote sensing data as the data source, this paper explores the method of extracting the raft cultivation information with the domestic high- resolution satellite imagery. The GF1 satellite data are preprocessed, and spectral and texture features are combined and applied to the support vector machine(SVM) algorithm. Then, the classification results of extracted seawater and floating raft aquaculture are compared and analyzed. In this paper, blue and green bands sensitive to culturing information are firstly screened for calculation of texture features. Then, 8 characteristic variables of gray level co-occurrence matrix, namely mean value, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment and correlation, are adopted. Texture feature variables are combined with the red, green, and blue spectral data to form 19-layer feature variables. In order to capture key samples and removes redundancy, those feature variables are processed by principal component conversion. The components of the first 8 principal components with large quantities of information and well-retained culturing information are screened and used for classification experiment of support vector machine. To verify the classification accuracy, this method is compared with the classic maximum likelihood estimation and minimum distance methods. The experimental results show that the method of applying texture features in extraction of culturing information method can improve the classification accuracy of floating raft aquaculture. Compared with the maximum likelihood estimation and minimum distance methods, this method has its classification accuracy improved by 3%-5% as a whole.