Height Assignment Error of FY-2E Atmospheric Motion Vectors and Its Application to Data Assimilation

Atmospheric motion vectors (AMVs) can provide plenty of useful information for synoptic analysis and numerical weather prediction, because of their excellent temporal and spatial coverage. It is of great value to apply FY-2E AMVs efficiently with the purpose of improving the initial fields and numerical forecasts. The existing assimilation systems in China are in lack of systematic guidance on the quality of FY-2E AMVs, and therefore call for research on optimizing the parameters in data assimilation system, which is very important and foundational to numerical weather prediction.The error characteristics of FY-2E AMVs are investigated on the basis of quality indicator attached. Statistical results demonstrate that the quality indicator of FY-2E AMVs has relatively weak implication of quality, because speed biases and root mean square errors with high quality indicator are still very large. With the study of inversion theory of AMVs, it is found that the inaccuracy of height assignment is the main problem that causes large observation error. According to this problem, the one-dimensional variational method is employed to adjust the height of AMVs. At the same time, the improvement is compared before and after height adjustment by means of radiosonde observations. As a result, it shows that the quality of FY-2E AMVs can be improved greatly after height adjustment. The absolute error of wind speed at every level in the northern hemisphere extra-tropics is reduced from 4 m·s^-1 to 2 m·s^-1 or less, while root mean square error from 10 m·s^-1 to 6 m·s^-1 or below. And the situation improved in the Southern Hemisphere extra-tropics is even more apparent. Height bias has been controlled within 20 hPa at every level below 150 hPa after height adjustment, and the characteristic of single-level wind field is evident. These facts reflect that the height of AMVs designated is systemically high in the Northern and Southern Hemispheres extra-tropics.Furthermore, the method of innovation vector and zero-order Bessel fitting function which is based on the theory of least square method are adopted together to partition background and observation error variances in the observation space, and thus the observation error of AMVs and the quality control coefficient can be estimated according to the statistical distribution of the innovation vectors. In order to validate the assumption on uncorrelated observation errors required in 3DVAR method, observation errors are inflated to avoid the influence caused by correlated errors. Finally, the predictability and impact of FY-2E AMVs data assimilation schemes is assessed in GRAPES global numerical prediction system. The results confirm that short-range forecast ability for the global numerical weather prediction system can be improved in the Northern Hemisphere, by introducing new observation error schemes with height adjustment. The improvement above high levels appears better than those of middle and low levels.