A novel approach to improve numerical weather prediction skills by using anomaly integration and historical data. Issue 16 (20th August 2013)

• Record Type: Journal Article
• Authors: Peng, Xindong; Che, Yuzhang; Chang, Jun
• Is Part Of: Journal of geophysical research. Volume 118:Issue 16(2013)
• Page Start: 8814
• Page End: 8826
• Publication Date: 2013-08-20
• Subjects: anomaly integration -- model bias correction -- historical data -- numerical weather prediction; Atmospheric physics -- Periodicals; Geophysics -- Periodicals; 551.5
• DOI: 10.1002/jgrd.50682
• Languages: English
• ISSNs: 2169-897X

Abstract

Abstract : [1] Using the concept of anomaly integration and historical climate data, we have developed a novel operational framework to implement deterministic numerical weather prediction within 15 days. Real‐case validation shows pronounced improvements in the forecasts of global geopotential heights in 20 out of 30 cases with the Community Atmosphere Model version 3.0. Seven other cases are marginally improved, and only three are deteriorated, in which all are ameliorated within the first‐week period. The average of the 30 cases shows an obvious increase of anomaly correlation coefficient (ACC) and a decrease of root mean square error (RMSE) of the geopotential height over global, hemispherical, and tropical zones. Significant amelioration on tropical circulation is displayed within the first‐week prediction. The forecasting skill is extended by 0.6 day in terms of days of the ACC greater than 0.6 for 500 hPa 30 case averaged geopotential height on global scale. The 30 case mean ACC and RMSE of 500 hPa temperature show the increment of 0.2 and −1.6 K, respectively, in the first‐week prediction. In the case of January 2008, much more reasonable horizontal distribution and vertical structure are achieved in bias‐corrected model geopotential height, temperature, relative humidity, and horizontal wind components in comparison to reanalysis data. In spite of a need for additional storage of historical modeling data, the new method does not increase computational costs and therefore is suitable for routine application. Key Points: modeling improvement with historical data new method for model bias correction