AMS 15TH CONFERENCE ON WEATHER ANALYSIS AND FORECASTING
AMS 11TH CONFERENCE ON NUMERICAL WEATHER PREDICTION
Spaceflight Meteorology Group, Houston
Key Theme: Ensemble Forecasting
Other Model News
Ensemble forecasting is clearly a primary future direction of NWP and forecasting. There
are several bright, energetic researchers, many outside of NCEP, who are getting involved.
Ensemble concept (much simplified): Vary model initial conditions and then run to generate
varied solutions with time. Spread of solutions may be useful in probabilistic forecasting.
Change in concept from "deterministic" (i.e., single solution) to "stochastic" (i.e., multiple
solutions related to probability distributions).
"Ensemble Mean" has shown improvement over deterministic forecasts, especially one- to
NCEP, ECMWF, UK Met Office all using ensembles operationally or semi-operationally for
medium range forecasting.
Experiments using ensembles comprising different models (e.g., ECMWF and MRF), rather
than same model with different initializations, show improvement over deterministic
forecasts, but larger "spread" between solutions than with single-model ensembles.
Experiments in short-range (6-48 hr) ensembles, including QPF, are underway.
Improvements are less dramatic in short-range than medium-range.
UK Met Office ensemble study conclusions: (1) substantial benefit to forecasting, (2)
alternate scenarios and probabilities available, (3) introduces new perspectives and new
"culture" in forecasting, (4) forecaster training/education essential.
Success with ensemble forecasting has led UK Met Office to recommend phasing out
medium range "deterministic" forecasts and replacing with ensembles only.
Caution on ensemble "mean" charts: Despite better long-term average success, the means
can "dilute" significant events.
At the NCEP Ensemble Workshop in 1994, one modeler proclaimed, "Determinism is dead."
TDL is studying ensembles for max/min MOS forecasts for medium range. Challenge will
be how to display data.
NCEP Year 2000 Plans:
Eta model 4km, >50 layers
- Rapid Update Cycle (RUC) 15km, >50 layers
- Hurricane model: nested grid 5-10 km, track and intensity predictions to three days
- Window model: Relocatable storm model
- RSM (regional spectral model) 50 km
ECMWF model slightly outperforming MRF, but gap closing yearly.
ECMWF increasing to T319 resolution in 1997-98.
ECMWF runs a three-day model off 0000 UTC data, in addition to ten-days from 1200 UTC
Eta vs Meso-Eta QPFs: Meso-Eta better in cool season, Eta better in warm season, but
Mesoscale modeling occurring at some WFOs, some collocated with universities or labs:
Penn State/NCAR MM5, RAMS, MASS model, LAPS mentioned. It takes ten hours to run
24-hour forecast on MM5 in Seattle (NWSFO in collaboration with University of
Washington), using Eta for boundaries, but results are still useful for forecasting and
understanding local weather controls.
24-hour improvement in temps past 30 years (day 2 now as good as day 1 was then).
- Two-day improvement in MSLP progs past 20 years.
NAOS: North American Observing System. NAOS Program Council set up, chaired by Ron
McPherson (NCEP). Looking at future obs systems (U/A, surface, satellite, ACARS, etc.).
There's a growing concept that predictions will "feed back" to observations, such that we can
"selectively observe" (or concentrate obs) on areas of concern. This could save money and
manpower, as well as further improve model predictions. "This (NAOS) will be a big deal"
in coming years, said Dr. McPherson.