针对海洋中尺度涡现象,提出一种基于海平面高度异常数据(SLA)的中尺度涡泛克里金提取算法。该算法运用变差函数工具计算SLA变差场,定义为广义振幅场,再利用泛克里金插值消除数据虚警和噪声,借助广义振幅场与实际振幅的关系通过少数几条特征等值线实现涡旋及其属性的提取。选取北太平洋作为实验区,采用2012年4月的4期SLA进行了中尺度涡的定量提取、分析,共提取出841个中尺度涡(含3个多核涡),包含450个气旋涡和391个反气旋涡;与其他遥感提取方法的对比显示北太平洋中尺度涡的成功检测率接近90%,过度检测率小于20%。结果表明:(1)算法具有省时高效性,通过对SLA场的重构创建广义振幅场,避免了等值线筛选过程,相对于海面高度闭合等值线等其他遥感方法具有判据简单及提取速度快的特性;(2)可靠性好,能够通过推导得出的特征等值线确保稳定的提取准确度;算法建立在发展成熟的等值线提取方法之上,并且有变差函数与泛克里金法的理论支撑;(3)自适应性强,可以对任意海区进行实时的涡旋检测和提取,并且除振幅值统计资料及必要基础数据外无需依赖其他辅助性数据。

Mesoscale eddies play an important role in the transportation of substance and energy throughout the oceans. Several extraction/identification algorithms have been proposed by scholars during the last few decades. Algorithms based on remotely sensed data are generally acknowledged, but defects are still inevitable when extracting eddies to date. To improve extraction performance, a new Universal Kriging Algorithm based on altimetric remotely sensed sea level anomaly(SLA) datasets has been proposed to identify mesoscale eddies. SLA fields are transformed to general amplitude fields to ensure rapid and effective implementation. Eddy attributes such as polarity, radius, area, and amplitude can also be acquired using this method. Variograms are generated to determine the window width and lag distance which are used to compute variance fields. Such field data are virtually planar variance grids with each pixel number indicating the variance between one central pixel and pixels at specific lag distances and directions from it on SLAs. Variance fields are defined as "general amplitude fields", assuming that all pixels are potential eddy cores consisting of both true and false ones. The Universal Kriging interpolation is utilized to eliminate false signals and data noises of variance fields, acquire pure signal fields of general amplitudes and discriminate useful signals from noises derived from the variance calculation. Variances of true cores are equivalent to real amplitudes, whereas those of false cores differ from the real amplitudes. However, the features of some of the false cores are connected with eddy boundaries, which are sufficient to determine the characteristic isolines for identifying eddies. The deduced isolines are generated on the general amplitude fields to extract eddy boundaries and attributes from background sea surfaces. The values of these isolines are determined using specific equations of true and general amplitudes of eddy boundaries on the basis of amplitude statistics. They are separated into three latitudinal zones, namely, 0°N-30°N, 30°N-45°N, and 45°N-60°N.Northern Pacific was set as the study area. Eddies were extracted followed by quantitative precision tests using four AVISO SLA datasets in April, 2012(4th, 11th, 18th, and 25th). A total of 841 eddies were identified, including 450 cyclones and 391 anticyclones. Three multi-core eddies were also captured, with one lasting for at least 15 days(from 4th to 18th). Compared with other remote-sensing oriented methods with complex criteria, the Universal Kriging Algorithm achieved a successful detection rate of approximately 90%(88.00%, 89.18%, 88.04%, and 87.92% with a maximum of 89.18%) and an excess detection rate of less than 20%(11.50%, 14.95%, 18.66%, and 16.91% with a minimum of 11.50%). Results were acceptable allowing for the spatial resolution of SLAs. Error thresholds were less than 0.25 degree.The Universal Kriging Algorithm has three noteworthy advantages. First, this algorithm is time saving. In particular, isolines are directly generated on the general amplitude fields, thereby simplifying identification procedures. The method significantly accelerates the extraction with a magnitude of 10 s in the core algorithm routines. Second, the algorithm is stable. Variance calculations along with Universal Kriging's elimination of noises are used to extract eddies at a relatively constant accuracy based on deduced characteristic isolines on general amplitude fields. Third, the algorithm is self-adaptive. That is, it is applicable to the real-time identification of mesoscale eddies throughout oceans and seas only depending on a relatively small quantity of essential data. Further plans of our research include revealing latent spatial information in marine data fields(especially in remotely sensed datasets) and exploring the application of the methodology in other ocean-element fields, such as the sea surface temperature, to improve the flexibility of the Universal Kriging Algorithm.