Statistics and comparative experiments for summer background error covariance in 3DVAR of central China regional model

Based on twice-a-day forecast products of central China operational numerical model, a statistic calculation of the background error covariance (B) is conducted using the NMC method to obtain the regression coefficients, eigenvectors, eigenvalues and length-scales of summer B matrix with features of China central operational model framework, resolution and regional geography.The contrast analyses of the statistical structure characteristics of B in each triple-nested model domain denotes that B in different domain is of obvious difference, which indicates B has a close relationship with the regional geographical feature and model resolution.In order to investigate the impact of different upon the model forecasts, batch experiments and statistical verifications in the period from June to August 2013 are performed with the universal background error covariance CV3-B of the WRF model and the statistically-obtained localized background error covariance.The results indicate that, when using the localized, the threshold scores of the 24-h forecasts of light rain, moderate rain, and heavy rain and the 48-h forecasts of moderate rain, heavy rain, and torrential rain are higher than those when using the CV3-B.The root mean square error of meteorological element fields forecast such as 850-hPa wind and temperature and 2-meter temperature is decreased.The root mean square error of 500-hPa height is, however, increased.Analysis for a torrential rain case shows that different B has significant influence on model initial fields.The localized gives a more reasonable initial field and therefore improves the rainfall forecasts.