Research On Reconstruction Method Of Sea Surface Temperature Off Qinhuangdao

Sea surface temperature is one of the most important elements in the marine environment.Make an intensive study of the characteristics and changing laws of sea surface temperature is of great significance to climate change,marine ecosystems and sea-vapor interactions.The coastal off Qinhuangdao is located in the western part of the Bohai Sea and is an important channel for material and energy exchange between the Liaodong Bay,the central Bohai Sea and the Bohai Bay.In recent years,due to changes in marine environmental elements such as sea surface temperature,ecological disasters such as jellyfish,red tides and green tides have frequently occurred.It has an important impact on industry,fishery and tourism in Qinhuangdao and surrounding areas.At present,the methods of obtaining sea surface temperature in Qinhuangdao waters include cross-sectional observation,buoy observation and satellite remote sensing.Among them,the cross-sectional observation has problems such as low temporal resolution,easy to be affected by weather and large consumption of human resources;although the buoy observation has a high temporal resolution and is not easily affected by other factors,it has a small number and low spatial resolution,etc.question.Therefore,in order to obtain sea surface temperature data with higher temporal and spatial resolution and accuracy,it is necessary to study the reconstruction method of sea surface temperature based on cross-sectional observations and buoy data.In this thesis,the data collected from the Qinhuangdao sea area in May,July to September 2021 and the buoy temperature data in the corresponding time period are taken as the research objects,and a sea surface temperature reconstruction method based on Kmeans++-SVR is proposed.In order to solve the influence of sea surface heat input(solar radiation)on sea surface temperature in discontinuous time cross-sectional observation,Kmeans++was used to mine its time information,so as to obtain the optimal clustering results of sea surface temperature in each month.The study found that the optimal clustering in May,July-September is 2 types;when the sea-going dates are consecutive,in August,the sea surface temperature is clustered at 10:30,and the temperature measured earlier than 10:30 The temperature is relatively low,and the temperature measured later than 10:30 is relatively high;when the sailing dates are discontinuous,in May,July and September,the sea surface temperature takes the date as the cluster dividing point,and the sea surface temperature in May and July In the early days,the temperature was lower,and the temperature was higher in the later days.In September,the result was the opposite.In order to improve the accuracy of data reconstruction,an SVR model integrating spatial information features and temporal information features(Kmean++clustering results)was constructed,Kmeans++-SVR model was trained for the features of different months,and the model was validated by grid search and k-fold cross-validation.The effect is compared with methods such as linear model,KNN and elastic network,and the root mean square error,mean absolute error and R square are used for evaluation.The research results show that in May,July and September,Kmeans++-SVR model achieved The RMSE is less than 0.5?,the mean absolute error is less than 0.4?,the R-square is greater than 0.8,and most of the absolute errors of the reconstructed temperature are within 0.5?.It can be seen that the reconstruction of sea surface temperature on coastal off Qinhuangdao based on the Kmeans++-SVR model has feasibility,superiority and practical value.