Study On Remote Sensing Retrieval Of Phytoplankton Species And Groups In The Eastern China Seas

Marine phytoplankton is a marine primary producer and a basic component of the marine food chain,it plays a key role in both the restoration of marine ecosystems and biogeochemical cycle in the ocean and even the whole earth.Therefore,scientific research on marine phytoplankton is of great scientific significance.For traditional method of distinguishing phytoplankton species and group,we often use a microscope in the laboratory after water sampling in the field,compared with traditional method,satellite remote sensing technology has the advantages of large-scale,long-term period,and high-efficiency.In recent years,with the development of ocean color remote sensing technology,phytoplankton species and group information retrieval through optical means has gradually developed.However,most of the researches on the retrieval of phytoplankton species and groups information are focused on the open ocean with case I optically simple water bodies,Eastern China Seas(including Bohai Sea,Yellow Sea,and East China Sea)are typical case II optically complex water bodies,the performance of conventional phytoplankton species and groups retrieval algorithms in open ocean will degenerate in the Eastern China Seas,in addition,there are few algorithms specifically for the retrieval of phytoplankton species and groups information in the East China Seas.The retrieval of phytoplankton species and groups information in the Eastern China Seas is difficult and challenging,both in using conventional methods or proposing new retrieval methods.Therefore,this research mainly focuses on the important scientific problem of remote sensing retrieval of phytoplankton species and groups in the Eastern China Seas,and the following research work has been carried out:(1)Field surveys were conducted in the Eastern China Seas,phytoplankton species and cell abundance data observed from microscopic inspection,phytoplankton pigment data,and corresponding remote sensing reflectance(R_rs)data in multiple cruises were collected.Based on the indoor culture of common species in the Eastern China Seas,we measured and acquired the absorption coefficient(a_ph)and specific absorption coefficient(a^*_ph)of single species.Afterwards,we used the numerical model(Hydro Light)and analytical model(Lee model)to build dominated single phytoplankton species R_rs simulation dataset and mixed phytoplankton species R_rssimulation dataset,through comparing the simulated R_rs data with the related records in the literature,the simulation results of R_rs are reasonable.(2)Based on the phytoplankton pigment data collected in the field,the Chlorophyll a concentration([Chla])of eight phytoplankton groups were estimated through the method of diagnostic pigment analysis(DPA),they are diatoms,dinoflagellates,prymnesiophytes,pelagophytes,cryptophytes,chlorophytes,cyanobacteria,and prochlorophytes.Then,based on the abundance method,the retrieval model with 2nd-order polynomial was constructed between the total[Chla]and the[Chla]of corresponding phytoplankton group,the accuracy validation results showed that,except for cyanobacteria and pelagophytes,the Mean Absolute Percent Errors(MAPEs)of other phytoplankton groups were within 100%,within a reasonable range,accuracy validation results of daily satellite[Chla]data of Ocean-Colour Climate Change Initiative Version 4.2(OC-CCI V4.2)also indicated the reliability of our model,specifically,the MAPEs of diatoms,dinoflagellates,prymnesiophytes,cryptophytes,chlorophytes,and prochlorophytes were 80.3%,80.3%,56.8%,65.2%,60.4%,60.1%and 42.5%.Based on the R_rs data and corresponding phytoplankton pigment data collected in the field,and estimation of phytoplankton group from the pigment data based on DPA,the phytoplankton group was retrieved through empirical orthogonal decomposition(EOF)and polynomial regression.For the EOF method,the validation results indicated that the MAPEs of most phytoplankton groups were greater than 200%,which can't retrieve PGs effectively,after validation of the retrieval results of the polynomial method,the MAPEs of these 8 phytoplankton groups were 105.55%,147.38%,162.68%,114.63%,147.39%,98.29%,142.3%,111.08%,respectively,the retrieval results were better than EOF method,but inferior to the abundance method.This study carried out an error analysis for the EOF method,and the decomposed mode cannot effectively extract phytoplankton information,which should be the main reason for the large retrieval error.(3)Based on the R_rs data simulated by Hydro Light numerical model,phytoplankton species and cell abundance data collected in the field,and R_rs data collected in the field,we conducted dominated groups retrieval in the field through Similarity Index(SI)matching,the retrieval accuracy was 52.28%,the dominant phytoplankton groups in more than half of the stations can be identified.Based on the R_rs data simulated by analytical model,phytoplankton species and cell abundance data collected in the field,and R_rs data collected in the field,deep learning algorithms were used to learn the shallow information of phytoplankton in the simulated spectral dataset,and transfer learning methods were used to combine the shallow information in the simulated dataset with the deep information in the measured R_rs dataset.The relative composition of marine phytoplankton species and groups in the natural water was detected.Through the calibration and validation of the model from the R_rs data collected in 5 field cruises from 2015 to 2018,the detection accuracy of phytoplankton species composition was:goodness of fit(R²)=0.88,MAPE=26.08%,Mean Absolute Error(MAE)=3.38%,Root Mean Square Error(RMSE)=4.4%,and the detection accuracy of phytoplankton group composition was:R²=0.99,MAPE=1.74%,MAE=0.33%,RMSE=1.28%.The algorithm was then applied to Hyperspectral Imager for the Coastal Ocean(HICO),showing that the dominant species of phytoplankton in the adjacent waters of the Changjiang Estuary was Prorocentrum dentatum,and the dominant phytoplankton group was dinoflagellates,compared with the literature records,the detected results are reasonable.(4)Based on the monthly OC-CCI V4.2[Chla]data from 1998 to 2019,and the empirical retrieval method(based on abundance method)proposed in this study,we carried out the analysis of the temporal and spatial variations and influencing factors of phytoplankton groups in the Eastern China Seas,the obtained spatial distribution results showed that diatoms and dinoflagellates were mainly distributed near shore,and other phytoplankton groups were mainly distributed in the far shore,consistent with the results of literature research.The long-term variations of the Changjiang River discharge were consistent with the variations of the[Chla]of the phytoplankton groups.The contribution of key environmental factors in the sea area affected by different water masses to the phytoplankton group were found not the same.The influence of the eastern type El Ni(?)o event and the central type La Ni(?)a event on the phytoplankton groups in the Eastern China Seas was roughly opposite.In summary,this study integrated indoor simulated and field collected R_rs data,field collected phytoplankton information data,and satellite remote sensing data.Constructing an abundance-based phytoplankton group retrieval empirical method,which was suitable for the Eastern China Seas.Constructing the hyperspectral retrieval method of phytoplankton species and groups composition based on deep learning-transfer learning.Analyzing the temporal and spatial distribution of phytoplankton groups in the Eastern China Seas on 20-year time scale and their influencing factors.Providing ideas and references for remote sensing retrieval of phytoplankton species and groups in the Eastern China Seas,providing a new data source for the study of phytoplankton species and groups information in the Eastern China Seas.