NOTES from Global Climate and Weather Modeling Branch Map Briefing
----------------
  2/17/05:

        Glenn White - New model analysis (parallel Y) diagnosis:
	  see: http://wwwt.emc.ncep.noaa.gov/gmb/noor/oct98op/textwea.html

		1. Set of 24hr forecasts: 1-14 December 2004.
		2. Pryme - parallel Y with T382 opnl SSI analysis (no AIRS data)
		3. Pryme2 - new SSI without updates to snow/ice emissivities.
		4. The differences in analysis produce small effects.  Basic 
			problems remain - longitudinal means show deep 
			difference emanating from near Wedell Sea.  Looks 
			barotropic (surface to 70mb).
		5. Difference between Opnl and Para Y seem related to contrasts 
			between land and sea (radiosonde vs satellite data??)?
		6. Suggested running a test without the new sea ice model!

	Sarah Lu - Parallel GFS Land Surface scheme, some validation!

		1. New vegatation fraction climatology tests:
			a. Neutral impacts on Anomaly Correlation at 500mb,
			b. Fits to radiosondes, mixed results (worse/better 
				for model layers below/above 950mb).
		2. Using Eta model's FVS (Forecast Verification System) on 
			Opnl, T254 and Para Y, T382:
			a. Looked at regions in Eastern/Western N.Amer and Alaska.
			b. Looked 2-meter RH and Temperature - notes that 
				temperature cools too much in Para Y.
			c. In Alaska, the Para Y 'cool-off' gets worse with 
				forecast hour.  Downward trend not seen in Eta!
   
	Yuejian Zhu - Thoughts re: next Ensembles implementation:

		1. Increased horizontal resolution (T170) for 1st part of forecast.
		2. Double members for each cycle (currently 10 each).
		3. Change to breeding cycle - current=24hrs, new=6hrs. 
		4. Some tests have been run for 24Aug-30Sep 2004 (new .vs.operational):
			a. New Ensemble mean has better 500mb anomaly correlation.
			b. NH AC 'die-off' in the New is better at all forecast hours, 
				but for SH its similar to Operational.
			c. New RMS hgt error: better in NH, worse in SH.
			d. Though sample size is small, hurricane track 
				errors show improvement with the new Ensembles..
		5. Plan to test T126 all the way to 16 days (still 28 layers),
				with 20 members for each cycle.

	Bo Cui - Statistical post-processing of ensembles (and CDC re-forecast):

		1. Preliminary tests, Spring/Summer 2004 - compare with verifying 
			analysis at 500 mb.
		2. T62L28 model, compare 'raw' Ensembles with bias-corrected 
			(remove 25-yr climatological error), and optimized 
			(remove previous 31-day error).  Done for current model 
			and the older global model used in the CDC re-forecast!
		3. Preliminary results:
			a. Ensembles with the operational model, bias-corrected 
				are better than the older model in the re-forecast 
				system through the 1st 10 days!
			b. Climate corrections add value to the probability 
				forecasts.  May not need a 25-year climatology, 
				rather using a 5-year climatology may retain 
				much of the benefit.