Application And Improvement Of Ensemble Optimal Interpolation On Hybrid Coordinate Ocean Model (HYCOM) And Regional Ocean Modeling System (ROMS)

Data assimilation is one of the most important parts of an ocean operationalsystem. Advances in computing machine over the past20years have enabled theimplement of eddy-resolving operational systems. Many countries have developedtheir operational systems for daily and seasonal forecast. The operational systemshave to provide forecasts in several hours and the ocean models are usuallytime-consuming, so the operational systems require the data assimilation to beefficient and computationally cheap. EnOI (Ensemble Optimal Interpolation), anapproximation of EnKF (Ensemble Kalman Filter), is a computationally very efficientdata assimilation method. It is widely used in coarse climate models and highresolution forecasting system. In this paper, we carried out the following works basedon EnOI:Firstly, the scheme about assimilation Argo profiles into HYCOM based on EnOIis improved. Based on characters of the Hybird coordinate, the diagnose process isupdated. Also all barotropic variables including ubavg, vbavg and pbavg are updatedwhen assimilating layer thickness to avoid potential degradation on dynamic balance.Five experiments are conducted to illustrate the improvements. The results show:(1)the improved scheme significantly reduced the RMSD (Root-Mean-Square-Deviation)for temperature and salinity. In200m, the RMSD of temperature is reduced by1.5℃,and the RMSD of salinity is reduced by0.1psu while XZ-S scheme cannot reduce theRMSE for salinity.(2) By comparing with WOA05and TAO observation, theimproved schemes act to tighten the thermocline and the temperature gradient is muchsteeper. Also the improved schemes lifted the thermocline depth.For salinity, theimproved schemes can depict the high salinity water (>35psu) exactly for bothpattern and area. But the XZ-S scheme cannot depict the water well.(3) By comparing with zonal velocity from TAO observations, the improved schemes cansignificantly enhance equatorial undercurrent and westward surface currents wasweakened.(4) The improved schemes seem to overestimate the variance in thetropical Pacific and increase the spatial correlation of the model sea level withobservation.(5) By reducing or diagnose depth, the adjustment on equatorialcurrents can be reduced but still have great disgrepency with observations.Secondly, EnOI scheme is applied to ROMS (Regional Ocean Modeling System)with the ability to assimilate TSLA (along-track Sea Level Anomaly). This system istested with an eddy-resolving system of the South China Sea (SCS). A twinexperiment without FGAT (NoFGAT) is also conducted for comparison. The dataassimilation system is tested from January2004to December2006. The resultsshow:(1) The Root Mean Square Deviation (RMSD) of the Sea Level Anomaly for7days forecast is decreased from10.57cm to6.70cm (NoFGAT) and7.01cm (FGAT),which represents a36.60%and33.68%reduction of the error and the reduction isrelatively stable for the whole assimilation time.(2) Data assimilation reduced theerror for temperature within the upper800m and for salinity within the upper200m,although the error degraded slightly at deeper depths.(3) Surface currents were inbetter agreement with trajectories of surface drifters with data assimilation thanwithout.(4) the variance of sea level is improved significantly in terms of both theamplitude and position of the strong and weak variance regions after assimilatingTSLA. Results with FGAT performed better than NoFGAT when considering theimprovements of temperature and salinity.(5) With a regional assimilation experiment,reason for the extremely strong variance of sea level in north of the SCS isinvestigated. The results show the extremely strong variance is caused by anextremely strong Kuroshio intrusion and it is necessary to assimilate TSLA in order tobetter simulate the SCS with high resolution models.Lastly, the predictability of SLA in the SCS is evaluated. A warm eddy in west ofPhilippines in10, June,2009, which may have great influence for typhoon LINFA ispicked out for case study. A hindcast experiment is setup for monthly forecasting. Theresults show the strong warm eddy compares well with observations but some small eddies around it is not forecasted.Then an eddy detecting scheme is adopted to detectthe center and trajectory of eddies from out forecast field, but it is found that detectingmethod for eddies is not appropriate for model forecast and the method seems to besensitive to the model noise. Hindcast experiments are employed to investigate thepredictability of this system. it is shown that for January and October, the forecasterror is relative small compared with model free run, and the forecast while theforecast error increase rapidly for April and July.