Ensemble Kalman Filter with Linear Model Bias Correction and Application in Regional Surface Observations Data Assimilation
|School||Ocean University of China|
|Course||Atmospheric Physics and Atmospheric Environment|
|Keywords||Ensemble Kalman Filter Pattern deviation Correct Surface observation data|
Data assimilation is an important aspect to improve the quality of numerical weather prediction. Business application problems as a new generation of data assimilation methods, in order to rely on the circulation of the background error covariance as the main advantage of the ensemble Kalman filter and similar methods in recent years, data assimilation hot area of ??research. Limited area surface observations data assimilation as a starting point, the mode deviation of the ensemble Kalman filter data assimilation brought adversely affected, the linear mode deviation model and pattern deviation linear homogeneous revised method. Use Lorenz96 system to carry out the experiment, the effect of the the mode deviation of mechanisms and deviations revised. Self-established the WRF-the forecast models EnSRF data assimilation system, use the system, data assimilation on April 28, 2006 squall line, the forecast error covariance distribution characteristics, and further analysis of the different ground the mode of the observed variables deviation of the differential impact of data assimilation, further verified the validity of the linear model deviation and linear homogeneous the revised methodology, sea level pressure data assimilation and surface temperature data assimilation, this paper analyzes the different variables of ground-based observations assimilation problems that may exist, the surface observation data assimilation techniques. In this paper, the mode deviation affect the mechanism of the ensemble Kalman filter data assimilation, pattern deviation revised methods and effects were studied and analyzed, the following conclusions: (1) in the ensemble Kalman filter pattern deviation filtering observation Add item deviation is equivalent to the non-Gaussian distribution and the error correction of the background field, increasing the analysis error may cause the system to become unstable and crash. (2) for a certain mode, the deviation, the greater the dispersion of observations corresponding to the background field value (Hx), the smaller the error of the cause analysis, on the contrary, the pattern deviation analysis will result in a large error. Dispersion larger area in the collection of the background, pattern deviation is likely to cause greater analytical error. (3) the mode deviation linear homogeneous revised method proposed in this study, when there is no systematic errors, not assimilation noticeable impact, that is, the side effects are minor. When there is a system error, the revised steps taken, can significantly reduce the analytical error, increase the stability of the system. Different modes error patterns, the effect of the revised linear homogeneous deviation revised basic elimination of the adverse effects of the pattern deviation, and for other types of mode deviation revised only part of the improvements, which can only be eliminated model error can be used in some deviation from the linear homogeneous relations described the impact. (4) in the assimilation of the observed data, the forecast error co-larger scale of the observed variables (such as sea level pressure), the limited area observations often there is an imbalance in the distribution of positive and negative co-regional, at this time, mode deviation will significantly reduce the assimilation effect, pattern deviation errors caused by revised spatial scales on the background field of a meteorological elements related to the Association of the elements error scale, and observation more information, the greater will be the amplitude error fluctuations and must therefore be mode deviation revised. Conversely, for error covariance smaller scale observed variables, the actual observations tend to be more balanced distribution In this case, the adverse effects of the pattern deviation smaller. This paper analyzes the limited area ensemble Kalman data assimilation in the background field distribution of the forecast error covariance characteristics, surface data assimilation techniques, the following conclusions: (1) If the members of the ensemble prediction with ensemble mean The difference can be seen as a displacement of error field fluctuations at different times, then the scale of the background error Society can be seen as error scale fluctuations, different weather elements, and the same meteorological elements at different levels, the error fluctuations scale is different. The present study, sea level pressure fluctuations in the error field scale is much larger than the average distance of the actual ground-based observations, so there may be \Mode spatial resolution used in this study, the scale of fluctuations in surface air temperature error of showing the sub-synoptic and mesoscale surface wind field error fluctuations smaller scale assimilation of surface air temperature and wind field data should be used in the encryption observation network . Altitude weather elements forecast error of the scale of fluctuations in descending order of perturbation potential temperature and wind fields, sounding station density should be able to meet the needs of the geopotential height and high air temperature data assimilation for high-altitude wind field data assimilation, somewhat less than the density of the radiosonde data. (2) In the beginning of the start of the ensemble prediction, influenced by the atmospheric circulation, the error covariance structure superimposed on the initial field increase along with will be made to adjust the basic adaptation of the atmospheric circulation, the forecast error covariance structure following the atmospheric circulation The pace of change is much slower. (3) the ground than the the wet forecast error co distribution shows the day or night, the correlation coefficient at night, during the day correlation coefficient should be related to the vertical exchange of water vapor with the boundary layer. Strong correlation of the prediction error of sigma surface air pressure sea level pressure and altitude, sea level pressure will affect the entire troposphere of each level is equally important. Wind field, temperature and specific humidity, as the level increases, surface and elevated the same elements of the forecast error decreased rapidly, but in the daytime boundary layer, turbulent mixing, the temperature and specific humidity forecast error ground the same elements of prediction errors decrease with height more slowly. Experimental results show a strong correlation of the two meters above the ground temperature and upper air disturbance potential forecast error, the correlation coefficient is even higher than the correlation coefficient of the surface air temperature and high air temperature prediction error, showing that the surface temperature assimilation play an important role for the geopotential height. Other ground and aerial different types of meteorological elements prediction error between the poor correlation. (4) affected by the boundary layer physics program, contrary to the size of the vertical distribution of specific humidity in the boundary layer from the top of the boundary layer to near the ground than the collection of wet dispersion decreases rapidly, the solutions of the boundary layer near the ground temperature discrete also have a certain degree of inhibition. In surface air temperature and specific humidity assimilation, the smaller the dispersion, may cause false revised background. (5) The mode of sea level pressure and surface temperature deviation distribution is in line with the linear mode deviation model proposed in this study, the use of linear homogeneous mode bias correction method to achieve better results. The poor correlation of surface wind observations and model forecast values ??described the linear homogeneous revised method proposed in this study may not apply to wind field model bias correction.