Characteristics and Application of Background Errors in GRAPES_Meso

The statistic structure of background covariance is studied by NMC method of USA based on GRAPES regional model forecast data spanning one year from June 2015 to May 2016. The horizontal correlation length scale is estimated with Gauss function linear fitting method. Characteristics of background error and horizontal correlation length scale with the latitude, height and season are investigated. Results show that background error and horizontal correlation characteristic scale obviously change with height and latitude, and the unbalanced non-dimensional pressure and humidity are closely related to the season. Background errors of four control variables are nonhomogeneous, among which background errors of stream function and unbalanced velocity potential mainly change with latitude and height, background errors of unbalanced non-dimensional pressure and humidity show local and seasonal characteristics. The biggest background errors of the unbalanced non-dimensional pressure occur in the Tibetan Plateau, and they are larger in winter while smaller in summer. The biggest background errors of humidity happen in low latitude of tropical monsoon region, and they are larger in summer while smaller in winter. The horizontal correlation length scales of four control variables with Gauss function fitting are reasonable except that correlation coefficients of the unbalanced non-dimensional pressure are overestimated in close distance and underestimated in far distance. Horizontal correlation length scales of steam, unbalanced velocity potential and humidity obviously change with height are largest in tropopause. The length scale of unbalanced non-dimensional pressure obviously changes with latitude and is larger in low latitude of middle tropospheric. The horizontal correlation length scale of the unbalanced non-dimensional pressure and humidity both are larger in winter and smaller in summer. The horizontal correlation length scales changing with height are used in GRAPES-3DVar system instead of single parameter, and then the analysis and forecast experiment results of one month indicate that, qualities of 6-hour geopotential height forecast in troposphere are improved. Analysis and 12-hour forecast of wind in stratosphere are greatly improved; all levels of 24-hour accumulated precipitation forecast are obviously improved; the false prediction of 24-hour accumulated precipitation of light rain, moderate rain and heavy rain are improved; 12-24-hour accumulated precipitation of extra torrential rain in control test fails to be reported, but the experiment with changing horizontal correlation length scales improves forecasts of positions and values for extra torrential rain.