Research On Spatial Sampling Scheme For Marine Data

Marine is closely related with the environmental protection, resource development, national security and so on. As the basis of information technology, marine data hasdrawn more and more attention. With rapid development of marine explorationtechnology, there are more and more methods for the marine data collection andacquisition, which brings new challenges for the speed and accuracy of statisticalanalysis. Fortunately, because sampling method is able to reduce the amount of dataoperations significantly, they can improve the computational efficiency of complexmarine data. How to select the sampling points of the method is an important contentduring sampling technique.There has been an integrated theoretical system for the sampling method based ontraditional industrial products. The sampling methods include random sampling, systemsampling, stratified cluster sampling and so on. Different from the industrial data, themarine data is a multi-scale data in the data structure, which is multi-class, multi-source,multi-dimensional, spatial autocorrelation, spatial heterogeneity and so on. Because ofthe features, traditional sampling method cannot be suitable for marine data. In view ofthis, a sampling method for the marine data is discussed in this paper, contents of whichis as follows:(1)The development of sampling techniques as well as the latest spatial samplingmethod is reviewed. By analyzing the difference between the ocean data with other data,the inadequacies of the current sampling methods are indicated.(2)The characteristics of marine data are analyzed. The observation elements anddata formats through different sample acquisition methods are summarized. Besides, themain characteristics of ocean data are analyzed, including the massive, heterogeneous,multi-dimensional, multi-class resistance, dynamics, spatial autocorrelation, spatialheterogeneity and so on.(3)Based on the characteristics of ocean data, a sampling scheme suitable formarine data is designed. Firstly, the data is stratified based on the data density. Differentsampling method is applied to different layer of the data. For the data layer with lowerdata density or small area, the total inspection method or random sampling method isused to reduce the computational complexity.(4) For the marine data of large area and high density, a flexible sampling methodis proposed. According to the degree of autocorrelation, the training set is selected bydifferent sampling methods. Based on the kriging model, the spatial autocorrelation isintroduced to determine the sampling step. Data quality should be considered duringselecting samples to ensure the quality for statistics.(5) The method introduced in the paper was validated with multiple data sets of acertain sea area. Results of the interpolation and trend surface analysis experiments showed that the samples can reflect the overall statistical characteristics of each property.At last, the superiority and robustness of this sampling method is indicated bycomparing with other sampling methods.