Research On Optimizing The Layout Of Moored Stations Based On EOF And Clustering Analysis

Oceanography depends on observations.Accurate and sufficient observations are the essential prerequisites for revealing ocean phenomena,understanding ocean variability,and improving ocean prediction skills.The continuous measurement of ocean variables at fixed stations,such as a moored buoy array or subsurface mooring array(collectively referred to as moored or mooring array hereafter),is one of the most important approaches to collect long-term time series of the three-dimensional data on ocean temperature,salinity,and velocity.Although the ideal number of moored array sites is “the more the better”,available sites,however,could never be adequate to cover the investigated domain due to costly instruments required for each site in reality.Therefore,it is important to determine the optimum array design using a limited number of station sites.In this study,a moored array optimization tool(MAOT)was developed and applied to the South China Sea(SCS)and the tropical Pacific with a focus on three-dimensional temperature and salinity observations.Application of the MAOT involves two steps:(1)deriving a set of optimal arrays that are independent of each other for different variables at different depths based on an empirical orthogonal function method,and(2)consolidating these arrays using a K-center clustering algorithm.Compared with the assumed initial array(the running observation array or assumed observation array),the consolidated array improved the observing ability for three-dimensional temperature and salinity.For the South China Sea,the optimization efficiencies is 19.03%and 21.38% respectively,and for the tropical Pacific,the optimization efficiencies is 8.24% and 18.04% respectively.Experiments with an increased number of moored sites in the SCS showed that the most costeffective option is a total of 20 moorings,improving the observing ability with optimization efficiencies up to 26.54% for temperature and 27.25%for salinity.In addition,compared with the previous consolidated methods,it is proved that K-center clustering algorithm is superior in consolidating different arrays.The design of an objective array relies on the ocean phenomenon of interest and its spatial and temporal scales.In this study,we focus on basinscale variations in temperature and salinity in the SCS,and thus our consolidated array may not well resolve mesoscale processes.The MAOT can be extended to include other variables and multi-scale variability and can be applied to other regions.