An Assessment of Subseasonal Forecast Skill Using an Extended Global Ensemble Forecast System (GEFS)

In order to provide ensemble based subseasonal (weeks 3 & 4) forecasts to support NCEP CPC’s operational mission, experiments have been designed through the SubX project to investigate potential predictability in both tropical and extratropical regions. The control experiment is the current operational GEFS version 11 extended from 16 days to 35 days with a 52 km horizontal resolution and 21 ensemble members. In addition to the control, parallel experiments have been designed to focus on three areas: 1) Improving forecast uncertainty representation for tropics through stochastic physical perturbations; 2) Considering the impact of the ocean by using a 2-tiered SST approach; and 3) Testing a new scale aware convection scheme to improve model physics for tropical convection and MJO forecast. All experiments are initialized every 5 days at 00 UTC during the period of May 2014 - May 2016 (25 months).

In the tropics, a MJO forecast skill has been improved from an average of 12.5 days (control) to nearly 22 days by combining all three parallel GEFS configurations. For the experiment with the best overall score, a skill of 23 days could be reached for strong MJO period. In the extratropics, pattern anomaly correlation (PAC) of 500 hPa geopotential height for the ensemble mean improved over weeks 3 & 4. In addition to tropical MJO and extratropical PAC of 500 hPa height, CRPS of the Northern Hemisphere raw surface temperature (land only) improved as well. A similar result has been found for CONUS precipitation, although forecast skill is extremely low. Our results suggest that calibration may be important and necessary for surface temperature and precipitation forecast in the subseasonal time scale due to the large model systematic errors.