Remote Sensing Model For Inversion Of Sea Surface Salinity In The Northern Of South China Sea-Hong Kong

Sea surface salinity is described as an important parameter of ocean. Its properties of distribution and change play a significant role on the understanding of complex system for marine–atmosphere exchange. Remote sensing is one of the effective means to obtain information of sea surface salinity. This paper is focus on the thesis of sea surface salinity remote sensing mechanism and BP neural network retrieval model, considering the complex environment of coastal waters, use the diversity parameters to build retrieval model for sea surface salinity and explore the influence of different seasons to the sea surface salinity retrieval model.Results showed that: (1) Coastal waters are taken as the study area because of the complex and variety environmental factors. Through the analysis of potential impact factors to salinity, the factors are classified by the role and contribution played in the process of remote sensing in salinity, then, temperature and total nitrogen are selected as the higher correlation coefficient with the sea surface salinity reached 0.69 and 0.79, especially the nitrogen, the coefficient was above 0.86 to salinity in the spring, summer and autumn; Then, based on the sensitive of blue band (b1) from ETM data to nitrogen, singer factor retrieval model for salinity was built between b1 and sea surface salinity. This shows that the total nitrogen and salinity is closely related in the study area, and in the optical remote sensing, total nitrogen content as the mediator can use to screening the salinity of the offshore marine areas. (2) The factor influenced salinity is complex and diversity, and singer factor inversion model can not meet the requirements of the salinity retrieval accuracy for applications. So, based on BP neural network as weak information extraction method of salinity stress, chlorophyll a, total nitrogen, suspended particulate matter and sea surface temperature are taken as input parameters, corresponding to the salinity values, to build BP model. Results came that: "transig, transig, logsig" are function for each hidden layer, and"4-12-2-1, 4-12-3-1"are tested as the best internal structure for constructing the BP model. Then, the resulting BP parameters and salinity fitting a higher degree of correlation coefficient of 0.8 or more; Based on the sensitive to the input parameters b1, b3, b4, b3/b1 and b6 of ETM data are as optical input parameters, applied in that BP model, and then came the result, the accuracy of sea surface salinity retrieved from the BP model can be reached 0.9 psu above.With many unknown factors, the environmental factors influenced salinity is more complex than expected, and the ocean physical parameters reacted with each other between the constraints. Those seriously affected the sea surface salinity retrieval accuracy, but also the optical factors of salinity are not enough perfect studied, so the reasons that the remote sensing studies of sea surface salinity also need to be further explored.