Quantify the Coupled GEFS Forecast Uncertainty for the Weather and Subseasonal Prediction
Yuejian Zhu, Bing Fu, Bo Yang, Hong Guan, Eric Sinsky, Wei Li, Jiayi Peng, Xianwu Xue, Dingchen Hou, Xin‐Zhong Liang, Sanghoon Shin
Abstract
Abstract The Global Ensemble Forecast System version 12 (GEFSv12) has been implemented into National Centers For Environmental Prediction operations since September 2020, which was uncoupled, but increased the horizontal resolution from 34 to 25 km, increased ensemble members from 21 to 31, and extended forecasts from 16 to 35 days. It significantly improved probabilistic forecast skills in many categories, such as precipitation, tropical storms, Madden‐Julian Oscillation (MJO), etc. The improvements resulted from many aspects, including model resolution increase, dynamical core upgrade, advances in hybrid data assimilation and physical parameterizations, and more importantly, from using new stochastic schemes to improve forecast uncertainty. To further improve GEFS's sub‐seasonal forecast skill, a coupled GEFS was built up on the Unified Forecast System prototype version 5 that fully couples an atmospheric model with a land surface model, ocean model, ice model, and wave model. A set of coupled GEFS experiments were conducted to test different horizontal resolutions at approximately 50 and 25 km while adjusting the stochastic parameterization schemes for the atmosphere to better represent forecast uncertainties. The experiments were run for a 2‐year period from October 2017 to September 2019, with one initialization per week at Wednesday 00 UTC, 11 ensemble members, and were forecasted out to 35 days. The coupled GEFS significantly improves 500 hPa height anomaly correlation in week‐1, week‐2, and MJO skills compared to the current operational forecast. The forecast spread of tropical wind is greatly reduced by improved stochastic schemes and matches well with the forecast root mean square errors. The correlation of forecast error variance and ensemble variance is improved for the coupled GEFSs. Meanwhile, the spread of MJO has been greatly reduced for the coupled GEFSs to improve the MJO forecast uncertainty.