The Test Plan for Implementing a WRF Ensemble at NCEP:
Why? How? When?
Nelson Seaman
NCEP/EMC Visiter from Penn State
Abstract:
The WRF (Weather Research and Forecast) modeling system has been
developed by NCAR, NCEP, and FSL as an end-to-end system consisting
of two dynamical cores, physics parameter- izations, pre- and post-
processors and verification software, all residing in a common
infrastructure. The WRF implementation strategy was modified in
April to focus on developing a mesoscale short-range ensemble
forecast (SREF) system, rather than a single deterministic forecast
model. NCEP computational capacity in summer of 2004 will support
running a small six-member ensemble in the High-Resolution Window
domains. Both dynamical cores will run with their native physics
suites (controls), plus two additional variations of each control
model. Experiments in late summer 2003 will support evaluation of
the control configurations and a selection of alternative
configurations that are candidates to become ensemble members.
Software development is nearing completion and the NCEP NMM has
recently completed its migration into the WRF infrastructure.
Present evidence supports the conclusion that an ensemble-mean
forecast solution based on even a few well-chosen members can
significantly reduce errors compared to the deterministic approach,
plus provide useful information about forecast uncertainty. At 8-10
km resolution, NCEP is expected to implement what should be the
world's highest-resolution operational ensemble modeling system.
Hardware upgrades following October 2004 will allow expansion of the
number of ensemble members, in addition to implementation of a North
American domain WRF and a Hurricane WRF.
The WRF (Weather Research and Forecast) modeling system has been developed by NCAR, NCEP, and FSL as an end-to-end system consisting of two dynamical cores, physics parameter- izations, pre- and post-processors and verification software, all residing in a common infrastructure. The WRF implementation strategy was modified in April to focus on developing a mesoscale short-range ensemble forecast (SREF) system, rather than a single deterministic forecast model. NCEP computational capacity in summer of 2004 will support running a small six-member ensemble in the High-Resolution Window domains. Both dynamical cores will run with their native physics suites (controls), plus two additional variations of each control model. Experiments in late summer 2003 will support evaluation of the control configurations and a selection of alternative configurations that are candidates to become ensemble members. Software development is nearing completion and the NCEP NMM has recently completed its migration into the WRF infrastructure. Present evidence supports the conclusion that an ensemble-mean forecast solution based on even a few well-chosen members can significantly reduce errors compared to the deterministic approach, plus provide useful information about forecast uncertainty. At 8-10 km resolution, NCEP is expected to implement what should be the world's highest-resolution operational ensemble modeling system. Hardware upgrades following October 2004 will allow expansion of the number of ensemble members, in addition to implementation of a North American domain WRF and a Hurricane WRF.