Ensemble Kalman Filter With Linear Model Bias Correction And Application In Regional Surface Observations Data Assimilation

Data assimilation is important to the improvement of numerical weather prediction.Characterized by flow-dependent background error covarianee,and as new-generationdata assimilation techniques.Ensemble Kalman Filter (EnKF) has gained particularpopularity for environmental state estimation.However,there are still some obstaclesfor its fully application in operational works.In this work,oriented around surfaceobservations data assimilation in regional numerical forecast,some key scientificproblems of EnKF application are studied.The impact of model bias on EnKF isstudied,and a linear model bias model and homogeneous linear bias correctionmethodology are put forward.The impact mechanism of model bias on EnKF and theefficiency of the homogeneous linear bias correction method are studied inexperiments using Lorenz96 system.In this paper,an EnSRF data assimilation systemusing WRF as forecast model is set up independently.With the help of this system,adata assimilation case study is carried out on a disastrous weather caused by squallline occurred on 28 April,2006.The structure of background error covariance isstudied,and the impacts of model bias on different surface data assimilation areanalyzed.In sea-level pressure and surface air teraperature data assimilation,thelinear bias model and the Homogeneous Linear bias correction method are testified tobe efficient.In addition,the potential problems of different surface observation dataassimilation are discussed.On the impact mechanism of model bias on EnKF,the Homogeneous Linear biascorrection method and its efficiency,the following conclusions are drawn in thiswork:(1) In EnKF,model bias plays a simila(?) role as observation bias.The non-Gaussian distribution and erroneous background updating increase the analysis error and lead tosystem instability and cracking.(2) With certain model bias,the smaller the backgroand ensemble spread ofobservation,the larger analysis error generated in model state analysis.The modelstates in area of larger ensemble spread or with lower predictability are likely to getlarger analysis error.(3) With the help of homogeneous linear bias correction method proposed in this work,substantial improvement in analysis quality and system stability are got.The biascorrection method has little negative impact when there is no model bias in dataassimilation system.The efficiency of the homogeneous linear bias correction methodis different for different bias pattern.While the homogeneous linear bias can be fullycorrected,only part of the bias,which can be represented by homogeneous linear biasmodel in other bias patterns,can be corrected.(4) In EnKF data assimilation of real observations,the model bias on observationalvariables with spatially large scale error covariance (for example sea level pressure)should be corrected to avoid large erroneous updating to background state.Whenmodel bias exist in observations of large scale background err covariance,the morethe observations and the more imbalance in its distribution,the larger the erroneousupdating is generated in regional data assimilation.On the background error covariance structure and surface observation dataassimilation,the following conclusions are drawn in this work:(1) The spatial scales of covariance are different with different variable and ondifferent levels.The sea level pressure has larger spatial scale,which may lead toimprovement"saturation"in data assimilation with relative dense observations.Oncontrary,with the spatial resolution in this work,the er(?)or covariance scale of surface air temperature,wind and humidity is much smaller.The density of wind observationin troposphere should be increased from current density of the soundings.(2) The error covariance originally added on initial conditions adjusts fast till it fit forflow.The sea level pressure has important impact on all model levels,while theeovariance of surface wind,tenperature and humidity with the same variables withintroposphere decrease quickly with height.Different variables have small covarianceexcept for surface air temperature and perturbed geopotential.(3) Affected by physics par(?)eterization scheme,within planate boundary level,thespread of humidity decreases fast with height,and the spread of temperature are alsosuppressed.It should be a potential problem in surface air temperature and specifichumidity data assimilation,since it could bring large spurious increment analysisfield.(4) The model bias of sea level pressure and surface air tempperature are in agree withthe linear model put forward in the work.EnKF with homogeneous linear biascorrection method can get substantial improvement compared with assimilationwithout bias correction.