Ensemble Generation Approaches in the NCEP CFSv2

This study presents ongoing research on strategies to improve the probabilistic forecast of the CFSv2 in the subseasonal time range. The CFSv2 is a low-resolution (T126L64) coupled model in operations since 2011 with an extensive hindcast dataset, which allows a comprehensive forecast bias correction and an assessment of its a priori skill performance. The CFS operational design generates a 16-member ensemble forecast per day out to 45 days – four of those members are integrated further out to 9 months.  While the long term development of the CFSv3 carries on from a multi-component Unified Global Coupled System (UGCS), we are exploring ways to revamp the coupled CFSv2 in the short term to enhance its forecast skill particularly in the weeks 3 and 4.

Forecast evaluations of the operational CFSv2 show considerable skill to predict the MJO signal beyond 2 weeks. The skill is higher than the uncoupled higher-resolution global ensemble prediction system suggesting that a source of atmospheric predictability may exist in the coupling with the ocean. A study is being started to increase the ensemble size and diversity of the coupled model prediction system to improve its forecast reliability. A first design makes its ensemble initial perturbations consistent with the perturbations in the GSI’s hybrid 4DEnVar; other ensemble generation approaches are being assessed.