Modified Observation Error Inflation Scheme for All-Sky Infrared Radiance Assimilation Based on the Model–Observation Agreement

For all-sky infrared radiance assimilation, the heteroscedasticity and non-Gaussian behavior of observation-minus-background (OMB) departures are two major difficulties. The Geer–Bauer observation error inflation (GBOEI) scheme is a universal way to handle the issues. However, it fails to take into account the consistency between model and observation, resulting in unreasonably large observation errors where the simulations agree with the observations. Thus, this study modifies the GBOEI scheme to rationalize the observation errors in such areas. With Advanced Himawari Imager water vapor channel data, the test results show that the normalized OMB with the new observation error approach leads to more Gaussian form than the GBOEI method and constant observation errors. Hence, the assimilation experiments with the new scheme produce better brightness temperature analysis than other methods, and also improve temperature and humidity analysis. Furthermore, a real case experiment of Typhoon Lekima (2019) with the new observation error scheme exhibits more accuracy, especially in track prediction, and substantial error reductions in wind, temperature, and humidity forecasts are also obtained. Meanwhile, 5-day 6-hour cycling experiments in the real case of Typhoon Lekima (2019) with the new observation error scheme confirm that the new method does not introduce extra imbalance compared to the experiment with constant observation errors; plus, more accurate typhoon forecasts can also be obtained in both the analysis and forecast, especially in track prediction.