[M-IS27] History X Earth and Planetary Science

[MIS27-03] Historical atmospheric analyses by weather information assimilation using Gaussian transformation

★Invited Papers

*Xiaoxing Wang¹, Kei Yoshimura², Kinya Toride² (1.Graduate School of Frontier Sciences, The University of Tokyo, 2.Institute of Industrial Science, The University of Tokyo)

Keywords:historical weather reconstruction, data assimilation, Gaussian transformation

Historical climate provides a reference for predicting future climate change by analyzing the past climate variability. Before the modern instrumental weather record, old descriptive sources (i.e., diaries) are important in reconstructing daily weather conditions. Data assimilation is recently widely used in climate reconstruction because it optimally combines meteorological information with a climate model. In Japan, there is a Historical Weather Database (HWDB) derived from diaries, recording daily weather information at more than 18 observation sites during the 1750s to 1870s. Cloud cover can be converted from the descriptive weather records but is difficult to assimilate because of its non-Gaussian characteristic. The objective is to reconstruct historical weather with a 6-hour temporal resolution by cloud cover data assimilation using a Gaussian transformation method. This study conducts a practical experiment in 1995 with the Global Spectral Model (GSM) and a local ensemble transform Kalman filter (LETKF) and validates the reconstruction results by reanalysis data. The observation information used in data assimilation is cloud cover categories converted from the JMA description data, keeping the consistency with the diary weather information. A control experiment without any observation assimilated (NoDA) and two experimental runs with cloud cover assimilated (NoGT and GT) are performed over Japan. Assimilation results indicate the small fluctuation amplitude in NoDA and NoGT, but GT impacts are evident in cloud cover estimation. 2-month average RMSE over the 17 observation grids are improved by 9.9% by GT, compared to NoGT. GT also contributes to RMSE reductions in three-dimensional atmospheric variables, especially in the higher troposphere. Gaussian transformation shows the great potential to improve cloud cover assimilation in actual reconstructions. This study provides a reasonable possibility to accurately reconstruct weather (for example, in the 1820s) with a higher temporal resolution.