Study On The Spatio-temporal Changes Of Eutrophication In Zhejiang Coastal Sea

Since the middle of the 20 th century,the nutrient concentration and productivity of the global coastal waters have been changed significantly in space and time.Many coastal waters have been sufferd from severe eutrophication for a long time,which leads to frequent marine disasters such as algal blooms and seriously affects the safety of marine water quality and coastal human production and life.It is of great significance to study the spatio-temporal distribution of the eutrophication in coastal waters and understand the detailed patterns of its variation,and it is also very meaningful to provide effective control measures for the ecological protection of the coastal environments.In the era of marine big data,with the inceasing of in situ marine data,the explosive growth of large-scale and high-frequency remote sensing data,and the developing of deep learning theory,the large-scale,long-time,and high-frequency retrieval of coastal eutrophication factors have acquired data sources and theory basis.In order to obtain the spatio-temporal distributions of the key elements of coastal eutrophication,including dissolved inorganic nitrogen(DIN),dissolved inorganic phosphorus(DIP),dissolved silicate(DSi),and chemical oxygen demand(COD),and to quantitatively analyze the coastal eutrophication based on these distributions,this paper proposed a retrieval and analysis method of coastal eutrophication elements based on the fusion of deep learning and spatio-temporal features.Using the ship in situ data and MODIS remote sensing reflectance(Rrs)products in the Zhejiang Coastal Sea(ZCS)from Feb.to Nov.from 2010 to 2018,the quasi-daily spatio-temporal distributions of surface DIN,DIP,DSi,and COD in 9 years were obtained.By combining with Data INterpolating Empirical Orthogonal Function(DINEOF)reconstruction algorithm,eutrophication index(EI)method,and actual eutrophication cases,we quantitatively analyzed the spatio-temporal distribution and variation of coastal eutrophication in the ZCS.The proposed method is applied,tested and discussed,and the main results of this article are summarized as follows:(1)Proposed a retrieval and analysis method for the coastal eutrophication based on the fusion of deep learning and spatio-temporal features.By making full use of the excellent fitting and calculation capabilities of deep neural networks and the spatiotemporal features of marine big data,this method overcomes the complex nonlinearity and spatio-temporal heterogeneity in the retrieval of optically inactive parameters in coastal waters.The accuracy of the method was proved by the model comparison: The fitting accuracy of DIN and DIP achieved 90.5% and 79.0%,the generalization accuracy achieved 86.6% and 75.1%;The fitting accuracy and generalization accuracy of DSi achieved 77.3% and 77.0%,respectively;In the retrieval of COD,by adding DIN and DIP obtained by the same algorithm to the input parameters of the model,the fitting accuracy was increased from 43.4% to 59.2%,and the generalization accuracy was increased from 52.2% to 64.3%.By making use of the reconstruction ability of DINEOF that can reconstruct data with high missing rate and spatio-temporal features,and using the quantification indicator of the EI method,this method successfully completed the spatio-temporal distribution and quantitatively characterized the coastal eutrophication status.The rationality of the reconstructed data and quantification indicator was demonstrated by spatio-temporal analyses.(2)Obtained the surface,near daily,500-meter spatio-temporal distributions of DIN,DIP,DSi,and COD in the ZCS from 2010 to 2018.Through the statistical analysis of annual trend,spatio-temporal analysis of monthly variation and typical daily variation,it was found that the retrieved and reconstructed results can be supported by relevant research or statistical data in bulletins,which is interpretable and reliable.Thus,the results provide an important basis for the quantitative analysis of the trends and detailed patterns of the eutrophication in the ZCS,and have high analytical value.(3)Through the spatio-temporal analysis of the EI and typical eutrophication cases,the main conclusions are as follows: In this period,the eutrophication in the estuary of Hangzhou Bay has been very serious all along,but the improvement of the north bank is greater than that of the south.The eutrophication is also serious in the inner region of the Xiangshan Bay but shows a downward trend in the 9 years.The eutrophication in the estuary of the Wenzhou river is better than that of the Xiangshan Bay,but it is necessary to control the discharge of pollutants into rivers and inshore waters in winter;The spacetime locations that have high risk of algal blooms are determined,such as the Shengsi and Nanji waters from May to August,where the average EI is low and the average N:P is high.Two peaks of N:P can be found during the period of an algal bloom,indicating that N:P is capable of reflecting the reproduction of phytoplankton in seawater(4)A series of suggestions are provided for the prevention and control of coastal eutrophication,including: strengthening the supervision of rivers and sewage outlets to control external input;using bioremediation engineering to control the concentration of nitrogen and phosphorus so as to purify water quality;making more use of remote sensing monitoring to obtain monitoring information with large area and high frequency;pay more attention to monitroing N:P when controlling the nutrients and organic pollution.This paper provides a new method for exploring the spatio-temporal distribution and variation of the coastal eutrophication,and provides evidence-based prevention and control suggestions for the eutrophication status in the ZCS.It is expected to promote the further application of remote sensing and deep learning in the study of coastal ecological environments.