Research On Sea Surface Temperature Fusion And 3D Model Based On Optimized Neural Network

Sea surface temperature(SST)affects the physical conditions of the ocean and the atmosphere.The three-dimensional spatial distribution of seawater temperature reflects the characteristics of the interior of the ocean.The SST and the three-dimensional spatial distribution of seawater temperature are of great significance in the field of marine research.The SST measurement methods include on-site measurement,satellite infrared remote sensing,satellite microwave remote sensing,and so on.Using only one of these measurement methods cannot fully meet the application requirements,and usually requires different measurement methods to combine different measurement methods to obtain higher quality of the SST data.Causing the existing SST method cannot meet the requirements of higher spatial coverage,time resolution and data accuracy at the same time,it is necessary to choose an effective method.Regarding the measurement of the three-dimensional spatial distribution of seawater temperature,only the buoys provide data at certain discrete points along the depth of the seawater temperature.Therefore,establish a three-dimensional distribution model of seawater temperature to obtain the three-dimensional distribution of seawater temperature.In order to obtain a higher quality of the SST data,an optimized neural network method is proposed in the thesis for infrared and microwave remote sensing sea temperature data fusion.Genetic Algorithm(GA)is applied to optimize the BP(Back Propagation)neural network.The optimized BP neural network is used to fuse infrared remote sensing data and microwave remote sensing data while taking wind speed as a parameter.This method utilizes the ability of neural networks to learn complex mapping relationships,and uses the global optimization advantages of genetic algorithms to avoid BP neural networks falling into local optimal solutions.At the same time,the method concerns with influence of wind speed on sea temperature to improve the accuracy of data fusion.Using this method to fuse the sea surface temperature data of the infrared radiometer AVHRR and the microwave radiometer AMSR-2 in the region(10°N?40°N,20°W?60°W).The thesis obtains fusion data with a time resolution of 12 hours.In terms of spatial coverage,the fusion data is 67% higher than AVHRR and 33% higher than AMSR-2.The information entropy of the fusion data is increased by about 164% compared to AVHRR,about 51% than AMSR-2,and contains more information.The definition of fusion data is increased by about 20% than AVHRR and about 45% than AMSR-2,and image quality is better.Compared with the buoy data,the error of the fusion data during the day is 0.10±0.49?,and the error of the fusion data during the night is-0.10±0.42?.The results show that the fusion data has higher data quality.In order to obtain the three-dimensional distribution of seawater temperature,the BP neural network optimized by genetic algorithm is used to establish a three-dimensional distribution model of seawater temperature in this thesis.The model uses the sea surface temperature data covered by a better horizontal space and the vertical temperature distribution data of the seawater with limited buoys,and uses the optimized BP neural network to simulate the complex nonlinear relationship among them to obtain the threedimensional seawater temperature distribution.This model is used to model the threedimensional distribution of seawater temperature in a three-dimensional area(0-2000m)of water depths(20°N?40°N,20°W?40°W).The results show that the mean value of the root mean square error for 2016 and 2017 obtained by the model is 0.36? and 0.31?,respectively.Expanding the application area of the model,the mean value of the root mean square error for the whole year is 1.08?.Compared with the results,the accuracy of this model is improved by about 30%.It shows that the model in this paper has stronger time generalization ability,space generalization ability,and more advantages.