Kriging Algorithms Applied To The Identification Of Mesoscale Eddies In The Pacific Ocean On Multivariate Remotely Sensed Data

The marine mesoscale eddies play an important role in the transportation of energy of theoceans by transmitting heat, salt, carbon, nutrient and other tracers which is different fromlarge-scale oceanic circulation and small-scale motion of waves. Efficient extraction andtracing of eddies make it possible to equip scholars with a better understanding of localoceanic gyres, thus contributing to finding the connections of mesoscale eddies and ElNino-Southern Oscillation (ENSO) with a series of parameters including Eddy Kinetic Energy(EKE), Eddy Activity Index (EAI) and Eddy Intensity (EI), making for analyzing the origin ofthe generation of El Nino and La Nina anomalies. Remotely sensed data has become widelyused in extracting eddies during the past few years, mainly consisting of altimetry, SeaSurface Temperature (SST), chlorophyll-a (CHLA) and Sea Surface Salinity (SSS) products.Methodologies of extraction based on these datasets are enclosed Sea Surface Height (SSH)contours algorithm with (without) thresholds and the Hybrid Algorithm; image processingmethods based on SST and CHLA data, yet there’re common defects in these methods, forexample, the complicated criteria, the sensitiveness to the thresholds, the limited adaption andhigh requirements for computation. We have devoted efforts to invent a new algorithm basedon remotely sensed data which has simplified criteria and is efficient, reliable and lesstime-consuming. Kriging methodology has been adopted to do experiments because of its bestlinear unbiased estimator and measurable errors.1. ObjectiveTo improve the performance of extracting, a new set of kriging methodologies includingUniversal Kriging Algorithm and Co-kriging Algorithm had been proposed for theidentification of mesoscale eddies based on multivariate remotely sensed datasets amongwhich altimetric remotely sensed sea level anomaly (SLA) datasets are the most effective,with transforming the remotely sensed data fields into general amplitude, temperature or othervaiance fields which guarantees a more rapid and effective implementation. Meanwhile,attributes of eddies: polarity, radius, area and amplitude can also be acquired at the same timeemploying the algorithm.2. MethodologyVariograms are generated to determine the window width and lag distance used to compute variance fields. Those field data are virtually two-dimensional variance grids with each pixelnumber indicating the variance between one central pixel and pixels at specific lag distancesand directions from it on SLAs or SSTs. Variance fields are then defined as “generalamplitude fields”(Specially for altimetry data) or “general temperature variance fields”assuming all the pixels are potential eddy cores consisted of both true and false ones. Toseparate valid signals from noises derived from the variance calculation, the Kriginginterpolation is utilized to eliminate false signals and data noises of variance fields to acquirepure signal fields of general amplitudes. Futhermore, kriging interpolation makes it possiblefor a skewed distributed data field to be transformed into a stationary dataset. Variances oftrue cores are equivalent to real amplitudes or other observations while false ones have valuesmiles away from real amplitudes or measurements, but features of the some of the latter onescorrelates with those of eddy boundaries, valuable enough for determining characteristicisolines for identifying vortices. Those deduced isolines are implemented on the generalamplitude, temperature or salinity fields to extract eddy boundaries out of background seasurfaces along with their attributes. Values of those isolines are determined by specificequations of true amplitudes (or other parameters) and general amplitudes of eddy boundariesusing records of amplitude statistics, and are separated into a small group of zones, as thethree latitudinal zones in the Northern Pacific located between0-30°N,30°N-45°N and45°N-60°N and the three cross-separated zones in the Southern Pacific.3. Result(1) Defining northern Pacific as the study area, we had extracted eddies followed byquantitative precision tests, based on four AVISO SLA datasets in April,2012(4th,11th,18thand25th).Totally841eddies were identified, including450cyclones and391anticyclones.Three multi-core eddies were also captured, with one lasting at least15days (from4th to18th). Compared to other remote-sensing oriented methods with complex criteria, the successof detection rate almost reaches90%(88.00%,89.18%,88.04%and87.92%respectively,with a maximum value of89.18%) and the excess of detection rate is less than20%(11.50%,14.95%,18.66%and16.91%with a minimum of11.50%). Results are acceptable allowingfor the spatial definition of SLAs. Computational errors are confined below1/4degrees.(2) There’re more eddies extracted in the South Pacific than that in its northern counterpart,where the sum is932comprising432cyclonic eddies and500anticylonic eddies. The successof detection rate ranges between80%and90%(87.76%,84.90%,89.33%and82.25%respectively, with a maximum value of89.33%). (3) Four datasets were chosen to compute mesoscale eddies in the Northwestern Pacific inApril2014, with433cyclones and310anticyclones successfully detected. Accuracy test alsocorroborate the stability of the success of detection rates.(4) As to the SST data, we used a Co-kriging Algorithm to extract eddies in the Kuroshioextension and subtropical convergence zones. Altogether111mesoscale eddies wereidentified effectively, including77cold-core eddies and34warm-core eddies. Resultssupported the practicability of remotely sensed datasets. However, the success of detectionrate is relatively low due to the limitations of SST datasets.4. ConclusionThe Kriging Algorithm has three noteworthy advantages which are (1) Time-saving andefficient: by generating isolines directly on the general amplitude (variance) fields, itsimplifies identification procedures. The method significantly accelerates the extraction into amagnitude of10s in the core algorithm routines;(2)Stable: using variance calculations alongwith Kriging’s elimination of noises and optimal interpolation, the algorithm can extracteddies at a relatively constant accuracy based on deduced characteristic isolines on generalamplitude fields.(3) Self-adaptive: the algorithm is applicable to the real-time identificationof mesoscale eddies throughout oceans and seas only depending on a relatively small quantityof essential data. Further plans of our research include revealing more latent spatialinformation in marine data fields (especially in remotely sensed datasets) and exploring theapplication of the methodology in other ocean element fields to improve the flexibility ofKrigingAlgorithm.