Error Analysis Of Ensemble Prediction And AMDAR Data Under THORPEX And Model Simulation

THORPEX (The Observing System Research and Predictability Experiment) is a 10-year international research and development programme sponsored by WMO (World Meteorological Organization) to accelerate improvements in the accuracy of one-day to two-week high impact weather forecasts for the benefit of society, the economy and the environment stewardship. Its sub-programmes mainly include global observing system design and demonstration,targeting and assimilation of observations,predictability research, among which TIGGE (THORPEX Interactive Grand Global Ensemble) is a key component of THORPEX. Therefore, this paper is focused on the prediction ability and improvement of CMA (China Meteorology Administration) T213 model under THORPEX background, analysis of the MCGE (Multi-Center Grand Ensemble) which includes T213 forecast and the performance of single T213 prediction is presented, as well as the characteristics of China AMDAR (Aircraft Meteorological Data Relay) reports. Finally, a model test with T213 products illustrating the improvement of numerical prediction by the use of AMDAR data is given. The main conclusions are summarized as follows:1. In order to evaluate the forecasting ability of TIGGE control data, a verification is conducted on products of China Meteorology Administration (CMA T213 model),National Centers for Environmental Prediction (NCEP T126 model) and European Center for Medium range Weather Forecasting (ECMWF T399 model) of TIGGE project for the 12 month time period of 1 Feb.2008 to 31 Dec.2008. Results of deterministic verification illustrate that the forecast skill of ECMWF is the best while that of CMA is the worst, furthermore, mean value of the control runs of three EPSs involved in the Multi-Center Grand Ensemble (MCGE) (hereafter control mean) could improve the forecast skill of single center (greastest improvement for CMA), especially for specific humidity. Analysis of Anomaly Correlation Coefficient (ACC) time series shows that most forecasts are successful to day 6, meanwhile the daily forecast exhibits a seasonal trend that the prediction accuracy in summer is poorer than other seasons, the seasonal variation is slight for specific humidity. Root Mean Square Error (RMSE) time series analysis illustrates that forecast error of specific humidity is the worst in summer, while forecast error of temperature and geo-potential height in summer is smaller than other seasons. Spatial analysis of control mean data in summer shows the forecast ability on land is better than that over ocean. Probabilistic verifications of spread indicate forecast probabilities decrease as the forecast time increases, according to the Brier Skill Score (BSS) analysis, the potential improvement over the climatological forecast is good for 5-8 days onward.2. In order to evaluate the forecasting ability of Chinese mid-range numerical weather prediction model-T213, error analysis,lag correlation and singular value decomposition (SVD) are performed on the forecasting field and objective analysis fields of T213 model from 2003 to 2007. Results show that the average error of extended-range forecast (240 hr) is nearly a magnitude larger than that of short-term forecast (24 hr), the forecast result is not so good after 120 hr (for 700-hPa temperature,500-hPa geo-potential height,850-hPa specific humidity and 300-hPa wind). The maximum error region of specific humidity is steady at middle-low latitude, while the maximum error region of other variables (resultant wind,geo-potential height and temperature) of 240 hr is further north than that of 24 hr. The average error of variables increases with the forecasting time increasing. There is a distinct positive correlation between virtual temperature error and geo-potential height (or resultant wind) error at the same prediction time, and the correlation coefficient is largest at 500-hPa.24-h SVD analysis shows, the key area for the influence between virtual temperature error and geo-potential height (or resultant wind) error locates off the coast of East China,Sea of Okhotsk and over eastern Russia. SVD analysis of 240 hr forecasting field indicates the key area for the influence between virtual temperature error and geo-potential height (or resultant wind) error locates consistently in mid-high latitude.3. A study of meteorological reports from Aircraft Meteorological Data Relay (AMDAR) System has been performed to estimate the characteristics of observation errors. Results show that the spatio-temporal distribution of data is non-uniformed: after quality control (QC) AMDAR reports mainly locate on southeast China and the littoral nearby, diurnal observations (00～12UTC) are much more than nightly observations (13～00UTC), and the height of data is from 0.5 to 8000 meters. Valid number of AMDAR reports after QC is almost 28% of original data, which indicates that the usability has to be improved. Results of comparison between AMDAR reports and T213 objective data,between AMDAR reports and NCEP reanalysis data and between AMDAR reports and sounding data illustrate that RMSE of temperature and wind are around 2℃and 3～4 m s-1 respectively, temperature observation is more accurate than wind, which is also proved by scatter plot and correlation coefficient analysis. On average, temperature error is largest in low level and slightly decreases with the increase of height, while wind error decreases from low to middle level and then increases to high level. RMSE of temperature and wind at 00UTC are larger than other observation time. Profile comparison of AMDAR reports and sounding data shows temperature and wind matches well, especially for temperature. According to error analysis, variable systematic error and periodic systematic error are likely to be included in observation data.4. A simulation of Mei-Yu front associated with heavy rainfall in 2005 are studied in order to testify the influence of aircraft observations on NWP (Numerical Weather Prediction) model forecast performance. Analysis shows that with the modification of first-guess field by AMDAR data, there is a positive impact on mesoscale simulation, and the rain intensity are more consistent with the observation. Forecast error of wind speed is around 4 ms-1, temperature forecast error is from 2 to 3℃, the accuracy rate is improved by AMDAR data.