Uncertainty Representation in the NCEP GEFS to Improve Medium Range and Subseasonal Weather Forecasts

Abstract:

The National Centers for Environmental Prediction (NCEP) Global Ensemble Forecast System (GEFS) has been in daily operations to provide probabilistic guidance for the public since December 1992. The initial uncertainty of the NCEP GEFS was introduced through the Breeding Vector method in the early stages, then replaced by the initial perturbations derived from the Ensemble Kalman Filter (EnKF) data assimilation in 2015. Representation of the model uncertainty was also changed from Stochastic Total Tendency Perturbation (STTP) to Stochastic Perturbed Physical Tendency (SPPT) and Stochastic Kinetic Energy Backscatter (SKEB) techniques. Recently, the GEFSv12 was unified to couple with WaveWatch III and aerosol components. It has extended forecasts from 16 to 35 days to support the subseasonal prediction and seamless forecast system based on the Unified Forecast System (UFS) framework.

There are challenges in further improving the GEFS' capability to forecast high impact weather and extreme events across various temporal and spatial scales, such as winter storms, heat waves, tropical storms, MJO and tropical waves. In addition to coupling with ocean and sea-ice models, representation of model uncertainty will play a key role in advancing the GEFS. The process-level stochasticity should be incorporated into the current auto-regression (AR) stochastic physics perturbations to optimize the contributions and interactions from the sub-grid scale physics due to imperfect parameterizations. Both the initial and model uncertainties from ocean, sea-ice, waves, and land systems should also be integrated into this fully coupled GEFS using advanced stochastic techniques that are physically and dynamically consistent. This talk will explore such options for advancing the GEFS into a fully coupled medium-range to sub-seasonal forecast system.