Sensitivity of Tropical Tropopause Layer Cirrus Prediction in GRAPES Global Forecast System

A warm bias with a maximum value of over 4K in the tropical tropopause layer (TTL) is detected in day-5 operational forecasts of the Global/Regional Assimilation and Prediction System (GRAPES) for global medium-range numerical weather prediction (GRAPES_GFS). Predicted temperature changes due to different processes, cirrus cloud distribution and the threshold relative humidity over ice (RHi) in the TTL are analyzed in detail. Compared with the European Centre for Medium-range Weather Forecast ERA5 reanalysis data, the TTL cirrus overprediction is found to be the exact cause of the warm bias. The induced cloud fraction and ice crystal content bias of the cirrus overprediction distributes uniformly almost in the whole TTL. Sensitivity tests indicate that the ice nucleation plays an important role in the ice crystal prediction in the TTL, and the TTL cirrus prediction is quite sensitive to the threshold RHi for ice nucleation. Further analysis shows that the ice crystal content prediction is sensitive to the saturation estimation in the TTL. Different ice saturation pressure formulae result in different RHi values, which leads to great differences in cirrus prediction. Detailed analyses reveal that the inaccuracy of the Magnus-Tetens formula for the saturation vapor pressure over ice at very low temperatures gives rise to the serious overestimation in the TTL cirrus prediction. As the overpredicted cirrus clouds arise from an inaccurate estimation of saturation, they are supposed to be formed in situ. By using the accurate Goff-Gratch formula, the Goff-Gratch experiment can well predict the TTL cirrus, and the overall forecasts in the TTL are largely improved.

The large bias of the CTRL experiment forecasts indicates that several aspects are closely associated with the sensitivity of the TTL cirrus prediction. The uncertainty of formulae for the saturation vapor pressure over ice contributes to the differences in the RHi estimation and remarkably affects the TTL cirrus prediction. Furthermore, the complicated interactions between various processes of the dynamics, cloud and radiation also have a great impact on the TTL cirrus prediction.

This study provides us some critical methods to improve TTL prediction in the GRAPES_GFS. It is found that the TTL temperature shows strong sensitivity to the microphysics scheme. The cloud scheme in the GRAPES_GFS describes microphysical processes in detail and is proved to be of good presentation for cloud forecasts in most regions around the globe. However, the forecasts are sensitive in the TTL because this is where such extremely low temperatures are found, so an alternative estimation of saturation is needed. The interplay of the processes then implies other effects than just an overprediction of the cirrus, such as the TTL warming and the geopotential height anomaly. As a result, keeping coordination of processes such as the cloud, radiation, dynamics and convection in the NWP model will lead to more reliable TTL forecasts.