Ensemble Methods And Experiments To Ocean Wave Data Assimilation

Ensemble methods haven’t got comprehensive and deep application inocean wave data assimilation recent years. Based on the full-spectralthird-generation wind-wave model called WAVEWATCH III, we attempt tobuild an ocean wave data assimilation system, which is attached to the globalwave forecast system, and test its short-term and long-term assimilation effectby using Optimal Interpolation (OI) and Ensemble Optimal Interpolation (EnOI).By comparing with observation data, we find that EnOI can greatly improve theaccuracy of significant wave height forecast. As the historical ensemblemembers of EnOI may exaggerate background error and cause pseudocorrelations, we initiatively manage to generate a set of dynamic ensemblemembers through overlaying pseudo random fields on wind field, and evaluatethe merits of stationary ensemble and perturbed ensemble, which wouldcontribute to the future applications of Ensemble Kalman Filter (EnKF).The main work is as follows:（1）Background error information is important to data assimilation, wemust have a full understanding of it before any assimilation experiment. A10-year global experiment is designed and compared with buoy and satellitealtimetry ocean wave data. The results indicate that WAVEWATCH III workswell in the prediction of global ocean wave. It also provides effective modelerror which is a reliable basis in building background error covariance matrixand choosing ensemble members in later work.（2）On account of massive global scale satellite altimetry observation dataand low resolution of the model, we compare four types of observation pointselection schemes in order to save computing time. It shows that small size ofobservation does not significantly weaken the assimilation effect, therefore we can filter the high frequency disturbance by using the thinning scheme of takingaverage value of every five points, which can shorten the calculation timewithout harming the assimilation effect.（3）According to short-term ocean wave forecast assimilation experiment,we evaluate the improvement effect of OI and EnOI on three days’ forecast. Theresults show that data assimilation is able to correct the deviation of the initialfield, and greatly improve the initialization process and three days’ forecast. Theimprovement effect of EnOI is more stable on the time series experiments.During the36hours forecast, EnOI performs better than OI.（4）In order to further investigate the assimilation effect of EnOI onlong-term wave forecast, a one-year assimilation experiment is designed andcompared with more observation data. The probability of absolute error of EnOIexperiment that less than0.5m is83.79%, and the probability of absolute errorthat less than1m is96.03%. The forecast accuracy of EnOI is more considerablethan OI.（5）In EnOI, background error is estimated by historical statistic ensemblemembers, which always remain time-dependent, and this may cause exaggeratebackground error and a wide range of space correlations. To solve this problem,we design a series of sensitivity experiments to discuss a better way to generatethe initial ensemble of EnKF, and compare with the statistic ensemble of EnOI.After analysis, the perturbed ensemble shows us better persistent spread andspecial correlation than the other one.