Improving Global and Hurricane Forecasts Using Time-lagged Ensembles in the GFS 4DEnVar Hybrid Data Assimilation System

Three dimensional ensemble-variational hybrid data assimilation system (3DEnVar) based on the gridpoint statistical interpolation (GSI) system was developed and implemented for the Global Forecast System (GFS) at the National Center for Environment Prediction (NCEP). As an extension of 3DEnVar, four dimensional ensemble-variational (4DEnVar) hybrid data assimilation system for the GFS was developed. Experiment results show a significant improvement of the 4DEnVar over the 3DEnVar for both the global forecasts and hurricane forecasts (Wang and Lei 2014; Kleist and Ide 2015). 4DEnVar is currently under pre-implementation tests at NCEP. Furthermore, recent studies on the GFS 3DEnVar indicated that increasing the ensemble size could further improve the forecasts (Lei and Wang 2015). It is expected that the increase of ensemble size may also improve the performance of the GFS 4DEnVar system. In this study, we use the time-lagged ensemble in GFS 4DEnVar system to increase the ensemble size while incurring the minimum extra computational costs.

In our study, a dual-resolution GFS 4DEnVar system is used with the T670 for the control and T254 for the ensemble members, mimicking the dual-resolution operational implementation of the hybrid data assimilation system. The time-lagged ensemble is created from the 20-member low-resolution ensemble forecasts initialized at various times preceding the current analysis time but valid at the same analysis time. This approach is mimicking the time-lagged operational Global Ensemble Forecast System (GEFS) forecasts. Given the fact that the time-lagged ensemble is composed of the forecasts at different leading times, methods such as equal weighting, skill weighting are explored to optimally combine the lagged ensemble forecasts with the regularly available 80 member EnKF forecasts.

As the first step of our study, the control experiment using the GFS 4DEnVar system with the regularly available 80-member EnKF ensemble was conducted. In the second experiment, in addition to the 80-member EnKF ensemble, the time-lagged GEFS ensemble are used to double the ensemble size. The two sources of ensembles are weighted equally first. Experiments are conducted for a 35-day period during July-August in 2013 and verification is made for the global forecasts and hurricane forecasts. Results will be presented at the conference. Additional results of further including the time-lagged high-resolution control forecasts are also planned to be presented in the conference.