| Numerical prediction has rapidly developed recently along with modern technology development makes domestic and abroad weather forecast performing well. It becomes one of the main methods of modern weather prediction. Numeric models also become better and better. However, it is found that there are always errors between forecast values and practical values. The errors increase gradually with computational integration. This becomes a problem for further weather prediction improvement.Usually the problem could be solved by two methods:one is traditional method. It decreases the error by improving the numerical prediction model. However, this method is too difficult to the meteorological departments at all levels due to the lack of technicians and equipments. The other method is the way that improves model accuracy by using the historical materials. Gu and Chou has discussed importance and practicability of historic material in numeric prediction. Chou JiFan proposed an alternative method to establish numerical weather prediction system for 6-15 days. It is still in the process of operational experiments. The results show that correction effect by this method is remarkable. But it is difficult to popularize in some meteorological departments because of large amount of calculation. Therefore, based on Chou’s idea, Shang KeZheng proposed similarity correcting method which utilizes historical meteorological data to correct posterior numerical prediction product. The method is also effective but the correction effect within 72 hours is not very remarkable.How to increase the accuracy of numerical prediction products within 72 hours by correcting the errors through historical meteorological data mining and applying the newest observations? It is a main motivation of this paper.Numerical prediction errors of specific model are continuous. Especially in the same class of weather system or weather progress, the errors of different lengths exist a certain correlation. Therefore, we got a new idea called immediate error correcting method which is applied for the numerical weather prediction products to improve forecast effect. First, the errors are calculated by comparing model prediction results and its corresponding observations. Then, the future errors whose corresponding observations do not appear are calculated based on the above errors. So we can immediately correct errors for numerical weather prediction products and improve accuracy of numerical prediction products.To apply and test the proposed idea, the paper selects T213 numerical prediction products and NCEP data from 2003 to 2010 including height field, temperature field, U-wind field, V-wind field, vertical velocity field and specific humidity field. There are 8 standard pressure levels of the specific humidity field and 12 standard pressure levels of the other fields. The research area is east Asia:65(?)-152.5(?),10(?)-70(?). The resolution is set as 2.5°2.5°.To accurately calculate correction errors of future prediction, the correction equation is established through stepwise regression method and Kalman filtering method respectively.1) After selecting factors, the correction equation is established through stepwise regression method by using T213 numerical prediction products from 2003 to 2008. T213 numerical prediction products from 2003 to 2008 were return corrected and T213 numerical prediction products from 2009 to 2010 were try corrected. Return correction and try correction of all factor fields have been evaluated by correlation coefficient and root-mean-square error(RMSE). The result as follows:the correction effects of all factor fields are effective at all lengths and remarkable within 72 hours.Try correction effect is consistent with return correction.2) After selecting factors, the correction equation is established through kalman filtering method by using T213 numerical prediction products from 2003 to 2010. And T213 numerical prediction products from 2003 to 2010 were corrected. Then, the corrections have been evaluated by using correlation coefficient and root-mean-square error(RMSE). The result as follows:The correction effects of the temperature field are the most effective. They are effective within 60 hours. They are only effective on high levels and medium levels after 72 hours. The correction effects of the height field are second effective. They are effective within 60 ho(?) The correction effects of the wind field and the specific humidity field are third effective They are effective only on low pressure levels. The correction effects of vertical velocity fieL are the worst.By comparing the above methods, it is found that both have respective advantages and disadvantages. The equations established by stepwise regression method are historical rules summary, so they are stable. The equations established by kalman filtering method can adjust coefficient according to the newest observations. In order to comprehensive advantages of them, a setting correction scheme are designed by integrating these two methods. Setting correction results are smoothed by 9 points smoothing. The smoothing are effective of the height field, the temperature field, the v-wind field and the specific humidity field. The smoothing effects of the vertical velocity field are not very good.Under the setting correction scheme, T213 numerical prediction products from 2003 to 2008 were return corrected and T213 numerical prediction products from 2009 to 2010 are try corrected. Return correction and try correction of all factor fields have been evaluated by correlation coefficient, climate anomaly correlation coefficient and root-mean-square error(RMSE). The results address as: The correction effects of the height field, the temperature field, the wind field, the vertical velocity field and the specific humidity field. Try correction effect is consistent with return correction. It shows that the setting correction scheme is reliable and stable.Except the specific humidity field, the correction effects of the setting correction scheme within 72 hours are better than the stepwise regression. The correction effects of the setting correction scheme within 72 hours are all better than kalman filtering. They are also better than similarity correcting. In brief, the setting correction scheme is the most effective correction scheme.In summary, the correction results are more close to the current weather conditions because the proposed scheme immediately corrects the errors within 72 hours with historical meteorologlo(?) data and the newest observations. The method also can be conveniently applied to meteorological depa(?)ments because of small calculation quantity.The setting correction scheme can be combined with similarity correcting to futher imrove accuracy. The results can be interpreted and applied by the numerical prediction product interpretation subsystem and gradually filtering similar forecast subsystem. So the forecast accuracy of XX long-term weather forecast system can be improved.Meanwhile, because of the security of military project, the softwares of XX long-term weather forecast system are all developed independly. In order to test effectiveness of the proposed correction scheme and apply the scheme to XX long-term weather forecast system, the author developed kinds of software such as situation display subsystem, the forecast results display and pre inspection subsystem and the forecast results inspection subsystem. |
PDF Full Text Request