Internet IM Update, September 11, 1998

Proposed Revisions to NCEP Global Forecast System

The NCEP global forecast system upgraded operationally on June 15, 1998, after more than 5 months of parallel testing from January to May 1998. The T170/L42 implementation was a result of significantly improved forecast performance for tropical winds, surface temperature and tropospheric temperature bias, precipitation skill, and other standard scores.

Recently however, some significant problems have arisen in the global forecasts resulting in poor performance on statistical scores as well as subjective evaluations, and affecting forecast systems that use the global output as initial or boundary conditions. For example, Eta model forecasts over Alaska have performed poorly due to poor global model boundary conditions over Russia and the Arctic. Many of these problems have been traced to inaccuracies in the global analysis resulting from an insufficient number of iterations in the analysis algorithm. Increasing the number of iterations as a solution will require significantly more computer time for all global model analyses.

Currently, the C90 NCEP computer is running a saturated workload. Thus the computer time required to test and remedy the analysis problem is not available. Equally important, there are insufficient computer resources to run exact parallel tests for other existing problems within the global suite of the Eta model system. Virtually all time slots for operational jobs are filled to capacity.

Given these circumstances, NCEP is planning a a reduction of the resolution of the of the global forecast system from T170/L42 to T126/L28 until the Class VIII computer is operational. The solution will keep the improved model physics while:

1. allowing more iterations in the analysis model;
2. helping to reduce the observed over-development of small scale systems, especially in the tropics;
3. reducing the time crunch for operational products which currently results in delays of important products such as the ensembles;
4. creating time for additional investigations into model performance with parallel runs that currently are impossible due to C90 workload saturation; and
5. allowing development of improved forecast systems that can be implemented on the Class VIII computer.

Currently NCEP is in the middle of parallel testing and evaluation to ensure that model results are as expected and that the performance is suitable for operations. Target implementation of the scaled down T126/L28 forecast system is October 6, 1998.

The imroved physics package from the T170/L42 will be retained despite the scaled down resolution. Other improvements will include:

1. improved background covariance formulation that improves tropical wind forecasts, and improved use of radiance data & GOES radiances;
2. improved use of moisture data, Y2K compliance and 3-D ozone analysis; and
3. improved radiation parameterization, gravity wave drag, and land surface climatologies.

The Industrial Meteorology Office will continue to provide updates on the planned operational implementation.