Font Size: a A A

Principles And Experiments Of Hybrid Data Assimilation For Multiscale Atmospheric Observations From Several Sources

Posted on:2021-03-06Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y D GaoFull Text:PDF
GTID:1360330620977915Subject:Atmospheric Science
Abstract/Summary:PDF Full Text Request
Multiscale flows constrained by different physical laws exist in atmosphere.Technology of atmospheric measurement brings multiscale information for weather forecast through updates in communication,remoting sensing and electronic computers during last decades.Data assimilation is a bridge between the atmospheric observations and the numerical weather model.Thus,a hybrid data assimilation scheme is an effective way to improve numerical weather prediction for multiscale atmospheric observations from several sources.This research explains response scales and deficiencies by investigating algorithms of three-dimensional variation?GRAPES3DVAR?and ensemble Kalman Filter?GRAPESEnKF?of the Global and Regional Analysis and Predication System?GRAPES?,proposes a novel principle of hybrid data assimilation in terms of multiscale features in atmospheric observations from servral sources,called Iterative Increments Hybrid Data Assimilation?IIHDA?,and gives three applications of IIHDA.The first one is Categorized Scale Data Assimilation?CSDA?by using several lengths of recursive filer in GRAPES3DVAR.The second one is Categorized Scale Hybrid Data Assimilation?CSHDA?,which iteratively hybrid a static increment and a flow dependent increment.The third one is Multilayers Hybrid Data Assimilation?MLHDA?,which takes GRAPESEnKF as a hidden computing layer,resembling to deep learning.This research also discusses characteristics of GRAPES3DVAR and GRAPESEnKF when airborne observations penetrating Typhoon Nida?2016?from Hongkong Observatory are assimilated.A false vortex is generated in the blind area of observations because of static background error covariance in GRAPES3DVR when the distribution of observations is not uniform.Thus,a bogus typhoon is used to improve both track and intensity forecasts.In contrast to GRAPES3DVAR,a gain matrix of GRAPESEnKF is computed explicitly by ensemble perturbations,in which a filter element of Schur product decides the influence radius of observations.The Root Mean Square Error?RMSE?of the posterior mean of GRAPESEnKF is the minimum and the subsequent track and intensity forecasts are optimal when the resolution of observations is approximately equal to the grid distance of GRAPES model.Experiments of Typhoon Hato?2017?after assimilating atmospheric movement vectors,radiosonde winds and radial winds of four Doppler radars demonstrate that CSDA can correct false function fitting of multiscale observations from several sources and improve subsequent forecasts by iterating increments at different scales.CSHDA has a significant improvement in intensity forecasts due to the flow dependent gain matrix.MLHDA shows excellent track and intensity forecasts in cycling data assimilation experiments.It eliminates the false vortex in horizontal of CSDA and noises in vertical of CSHDA and decreases filter divergence of CSHDA by using the flow dependent gain matrix as a pre-transforming operator to supply information at small scale in the blind area of observations.In addition,CSHDA and MLHDA are feasible in operation because they do not need to hindcast the optimal response scale in CSDA.
Keywords/Search Tags:hybrid data assimilation, multiscale movements, forecasts of tropical cyclones
PDF Full Text Request
Related items