[MIS28-01] Data assimilation of historical weather using Gaussian transformation
Keywords:data assimilation, historical weather, Gaussian transformation
Data assimilation is widely used in atmospheric prediction because it is an effective method to combine model forecasts and various types of meteorological observations. For instance, assimilation of total cloud cover converted from old descriptive weather records is significantly important in reconstructing the historical weather on a daily scale. However, cloud cover often shows a highly non-Gaussian distribution and seriously violates the basic assumption of normal error statistics in most data assimilation schemes. This study aims to apply a Gaussian transformation (GT) approach into local ensemble transform Kalman filter (LETKF), and to increase the reconstruction accuracy by assimilating cloud cover. The reference “truth” data with a 6-h time interval is produced from the Global Spectral Model (GSM). GT approach is successfully applied grid by grid to the 3-year truth data during 2015-2017, where the range of cloud cover is transformed from 0 – 100 to -3 – 3 as the upper/lower limits. A control run without assimilation of any observation and three experimental runs are performed for two months from 1 July 2017 with 30 ensemble members. Results indicate that assimilation of cloud cover with GT reduces 1% of 2-month average RMSE around Japan, when compared to the experiment without GT. The accuracy of wind direction, temperature, and humidity can also be improved by GT of cloud cover, especially when larger ensemble size is used (for example, 80). Particularly, if rainfall is also Gaussian transformed, its 2-month average RMSE can be reduced by 25% by GT of cloud cover. In consequence, the application of GT has the potential to improve cloud cover assimilation accuracy, and can be a more effective approach in reconstructing historical weather.