DT's  RULES  FOR MEDIUM / EXTENDED  FORECASTING RULES

 

 

RULE 1  CONSENSUS  RULE:

The    MODEL CONSENSUS  RULE    says when  1  model  stands  out   from what    most of the other models are showing  the  CONSENSUS     is  often  right. .  But this  rule   has limitations...  in  complicated   synopic situations     the  CONSENSUS   Breaks down.   But  as a a general run in Most cases when   9 models that say A followed by B     but  1 model says     C followed by D...     the   consensus   of the  9 models are far more likely to be right than    1 Model. . 

Often  MR ( medium range) ensemble and SR  (Short Range)   Ensemble data -- such as individual 500 MB plots  and  /or the Spaghetti plots  should be viewed to see if   whewre  the   TRUE consensus  lies. .

RULE 2 FLIP FLOP RULE:

The  FLIP FLOP RULE  is useful whenever a model shows a significant change—be it at the surface …500 MB... or 300 MB -- from what it was showing OR from what the Consensus is.    The  Model   that shows a significant change OR has a solution that is particularly  different  from the Consensus will be referred to as a   FLIP FLOP .

The BIGGER   the change from then previous model run...  the LESS likely to is to be right and the more likely it is to be a "hiccup"   sometimes known as  a  Bad Model run.  A surprisingly high amount of time the Operational Run of the  GFS   is at odds with its own ensemble and may have only 1 or 2  (if that)   of the ensemble members   supporting the Operational run.

 

RULE 3-- THE RULE OF O.M.S - Other Model Support

This is a ‘corallary" to the CONSENSUS rule.  Sometimes the  Model which is  far  away from the  CONSENSUS    (called the Model Outlier)  turns out to be  correct!   This rule --RULE  OF O.M.S.   has proven time after time to be a excellent criteria in determining  whether the Model which is Model Outlier  is a "Bad   run"  of the  model   OR    whether  the Model Outlier  is  "on to something".       Most of the    short  and long  range Models   these days  are   of  sufficent   quality  and skill   so  that   when  1 model comes  out with a  seemingly    whacky    or  "way out there"  solution    IF   that   solution is   valid then  the  other models   will  begin to  shift   towards the  Model outlier   solution  shortly.

EXAMPLE   The 5  DEC 2002  Northeast   US snowstorm.  Most of the SR data was showing  the heaviest snow occurring over VA MD  lower  PA  and lower NJ... and Just clipping   NYC Long island and coastal  CT/ southeast Mass.  The 18z run of the   DEC 3  ETA showed a significant  shift in the  northward track of the  Precipitation shield and the strongest Vertical Velocities  (VVs)... with the 850 Low tracking into WVA and bringing all rain into  DCA PHL and NYC..   However the other 18z  Models-- the MM5 and GFS-- did NOT  show   that and therefore the 18z ETA DEC3  idea was discarded.  AT 00z DEC   the ETA shifted back to the south....

On DEC 4  the 18z ETA  again shifted the precipitation shield much further northward and that trend was strongly supported by the 18Z GFS   and MM5.

RULE  4- STAY WITH THE CONSISTENT MODEL

When the overall pattern has been a mild and dry one Models that show  major Low pressure development  should be viewed with a great deal of skepticism...unless the Overall pattern changes FIRST.   For example in the winter months... during the  event of a strong southeast US Ridge...  System that  models develop into Major events  that appear to be bringing cold and snow/ice  or rain often  ended up being often much weaker and /or displaced to the north by at least a few hundred miles.

By the same token...   Models that shows a large trough coming into the Eastern Pacific or Western US without removing a blocking pattern over Canada or without the negative  NAO flipping to Positive Phase... OFTEN over warm the atmosphere... may take western US Lows (short waves)  into the Upper Plains and or Midwest/  Great Lakes. 

In the MR period (Day 3 to day 7) during  Stable patterns t is often useful to use the NEC performance scores to see which is the HOT model. Link…

 

RULE  5- RULE OF MODEL TRENDS   -  unstable trends

This is especially true at 48 hours and beyond with synoptically unstable patterns as well as Tropical cyclone forecasting.  MODEL TRENDS are more important than the what the Model is actually depicting. For example suppose Model "A" is the M.O.  (Model OUTLIER)  and the majority or consensus depict some other solution… in most cases the consensus model group is often correct. BUT…

When the model consensus or majority cluster begins to breakdown… WATCH THE SHIFT TOWARDS the Model A  or  M.O.  solution. Incremental changes towards the M.O.  solution will often lead a forecaster to make incremental changes in the forecast usually because of concerns over model uncertainty and consistentency issues. This is often a mistake. Once there is a discernible shift towards the Model A / M.O.  solution… it is often wise to make large changes in the forecast that is reflective of the Model A OUTLIER solution.

 

RULE  6  RULE OF  THE   SEASONAL BIAS:

SEASONAL BIAS is the most often over looked aspect to the MR forecast. The rule ‘drought begets drought´ is a form of this rule. In a drought go with the NWP that shows the least amount of rain  UNTIL   there is some sort of major pattern change. In a dry winter that has not featured a lot of storms… lean toward the weaker smaller development with less QPF scenarios.

EXAMPLE 1 The winter of 2000 – 2001 along the East coast was a classic case for Winter with respect to the issue of Seasonal Trends. In the MR (MEDIUM RANGE) period system after system appeared to be taking tracks that favored the Middle Atlantic region over the Northeast and New England. Throughout the winter season most of the MR models consistently showed systems that APPEARED to be mainly Middle Atlantic events in the day 4 to day 8 range. YET as the period in question got closer, the tracks would shift with each subsequent model run or cycle.  For example on 6 FEB 2001   there was a moderate / significant snowstorm in  New England.  Yet the MR models  showed the  500 MB  as well as 850 and surface Lows originally forecasted to propagate across the VA/NC border... then with each new model run... showed a constant shift  north. In fact  the 500 Low  would pass across DC/ lower MD.

The seasonal bias may have also played a part in the 1-3 MARCH 2001 super hype snowstorm in the Middle Atlantic and Northeast. Early on… the MR models depicted heavy snowfall threat in the Middle Atlantic and less of snow threat for the Northeast…. When in fact the New England region was the area that one area that saw the Heavy snowfall.

EXAMPLE 2 Hurricane season 2002; Of the 8 named Tropical cyclones only 1 formed in the MDR -Main development Region of the tropics. 7 of the 8 formed north of 20 degrees N latitude. Yet there were numerous T.W that various Models would develop in TC in the tropical Atlantic or Caribbean.

RULE #7   GFS  BIAS on the East Coast;

On the East Coast; in the MR period when the 00Z  GFS/ 12Z GFS is showing  pattern  or storm development that is   "opposite" of the ECMWF… 9 times out of 10 go with or lean towards the ECMWF. Note that is especially true when the MRFX has Upper air patterns (at 500 or 250 MB) that do NOT match the 00z or 06z run of the OP-AVN.... and/or match  the  ECMWF.

The AVN has 2 major problems on the East coast and both of these make this model the model LEAST likely to verify beyond the 72 hr time frame.

6.1 Because of the severe cold bias of the GFS  it loves to overdevelop troughs and as a result often takes features in the subtropical Jet and merge / phase that feature with the trough in the PJ …hence making the  Big storm or Big event.. when no other model is doing so. . Thus there are many well known cases where the OP GFS   /  OP MRF have developed large Lows well Inland…sometimes even as far west as the Ohio valley/ Great lakes…only to see these features verify on the East Coast.

6.2 In the event the GFS/ AVN does NOT forecast the Superstorm of the century…it OFTEN has the feature in the STJ passing harmlessly out to sea off the GA or SC coast as a 1010 Low.

6.3  As a result of 6.1 and 6.2  Most of the Major events that have affected the East coast in the 20  years or so the AVN/ GFS  has been way off and seriously outperformed by the most of the other MR models

6.4.   The  GFS  is the only MR model  that TRUNCATES....  HPC has recently  upgraded the GFS... to a T254  or 55Km  model resolution but this is only to Day 7 / 168 hrs  on the 00z  run. From  180 hrs to 240 hrs   the 00z GFS  runs at a T170/ 77 KM  model resolution...   and from 240 hrs to 384 hrs / Day 16  the 00z GFS runs at T126  / 108 KM resolution.

At the 06z 12z and 18z   the GFS  runs at the Respectable  T254 / 55 KM   resolution ONLY TO  84 hrs...  AFTER WHICH  it truncates at the  T170/ 77 KM  model resolution...   to Day 10 /  240 hrs...  and  from 240 hrs to 384 hrs / Day 16  the 06z / 12z / 18z  GFS runs at T126  / 108 KM resolution. 

The Ecmwf uses a 25 KM  resolution through day 7...   the Navy uses a 60 KM through day 6 at both runs... the  UKMET  runs at 60KM at 00z and 12z runs... Canadian approx.  85 KM  through day 10.

The AVN has 2 major problems on the east coast and both of these make this model the model LEAST likely to verify beyond the 72 hr time frame.

  1. Because of the severe cold bias of the AVN it loves to overdevelop troughs and as a result often takes features in the subtropical Jet and merge / phase that feature with the trough in the PJ …hence making a Big storm or Big event. Thus there are many well known cases where the MRF/ OP AVN have developed large Lows well Inland…sometimes even as far west as the Ohio valley/ Great lakes…only to see these features on the east Coast
  2. In the event the AVN does NOT forecast the Superstorm of the century…it OFTEN has the feature in the STJ passing harmlessly out to sea off the GA or SC coast as a 1010 Low. .
  3. In 1999 there was Floyd.  According to the AVN/ MRF ...   Floyd would hit FL and the AVN didn't back off of this idea until 36 hrs before Floyd made Landfall in NC. Meanwhile at day 7 the ECMWF turned   Floyd sharply to the North into NC/ VA at Day 7 and never  backed off of this solution which proved to be correct.  144 hrs this idea was supported by the  Nogaps  UKMET   and Global Canadian GGEM

    In  August 1998 Hurricane  Bonnie  the AVN   strongly and consistently showed that Bonnie would pass EAST of Hatteras when in fact Bonnie went  over eastern NC and through Norfolk area as a Borderline Cat 1 hurricane.

    In the 24 -25 January 2000  SECS ( significant east coast snowstorm) that pounded the Middle Atlantic with 10-20 inches of snow..  the AVN was furthest EAST of any of the MR or SR Models . It is my assertion  that the AVN east bias was the PRIMARY reason why so many Middle Atlantic  Forecasters both Govt and private  Blindly assumed the 24-25 January 2000 Low was headed out to sea  EVEN right up until  3PM forecast discussions 24 January

    6-8 January 1996 Blizzard: Once more the AVN  was by far and away   the weakest with the Jan 1996 Noreaster. The ECMWF from Day 7 consistently forecasted  a major Noreaster / Blizzard for the East coast and was quickly followed within  the next model run by the UKMET...Nogaps... and Canadian.  The AVN / MRF utterly missed this system and never at any time showed the a SECL  (significant east coast low)  from day 10 through 24 hrs before  the snow began at Washington / Baltimore and Philly)   I will present the actual Model data  from Jan 1996 shortly.

    The 11-13 March 1993  Superstorm:   Day 7    the ECMWF showed a 975 MB Low along the Immediate East Coast... The UKMET 968 over AOO...  with the Nogaps and other Models showing  a similar deep SECL.    The  old MRF at least this time did  show a  strong system  (986 MB over  RDU) but it was  consistently  too far inland and weakest of all the MR  models.   I will present the actual Model data  from March 1993  shortly.  When the event moved within 72 h and 60n hrs   the AVN suddenly developed  a perfect  968 MB Bomb over SBY  MD...

    The 8 -12 February 1994:   This  was actually TWO events  that came during one of the coldest winters in the Midwest / Great lakes and Northeast in the last 100 years.  The AVN consistently mishandled the SPLIT FLOW pattern and displayed surface lows/ short waves taking a  much further north track... into the Ohio valley  over an over developed  500 MB ridge in the southeast which the AVN/ MRF Models showed bringing Mild air / rain to the Northeast & Middle Atlantic.  The 2nd low 10-11 FEB 1994  which brought Crippling and Historic ice storm from RDU to RIC to DC  to PHL  and Heavy snow in New England... was depicted by the  72  and 60 hr AVN  as Rain with temps in  the 40s and 50s  (850 MB temps +5) as the main Low would go through Ohio  and into western PA/ NY.  However the ECMWF UKMET and Canadian model consistently and correctly showed   the systems sliding east... downplaying the Southeast Ridge  and bring Much colder snowier / ice forecasts.  I will present some of the Maps from February 1994   shortly.  

    29-30 December NJ / NYC/ CT snowstorm.  Correctly forecasted in the MR by  the ECMWF  UKMET Nogaps  Canadian (GGEM)  the MRF AVN   played a game of amazing flip flop. The 00z morning runs of the MRF  from Day 8 would show a SECL off the coast with significant  snow amounts over the Northeast    only to see EVERY run of the 12z AVN   either completely remove the   Low or Place it out by Bermuda.  HPC  obsessed and focused solely on the AVN NEVER forecasted a SECL  from  Day 7 right through  48hrs.  In fact   one of the less skilled forecasters at HPC   Gleefully taunted how the snow crows would be wrong  and this perspective was based solely on the 12z AVN... as the forecaster unwisely  ignored  the  ECMWF UKMET NOGAPS  ETAXX    GGEM SEF  and strong ensemble support.

    It is MY contention and feel free to disagree  that because of model Truncation problems...  the cold bias and over phasing bias  the GFS/ AVN/ MRF has  missed most of  the  major   events over the eastern US in the last 20 years or so. ONCE  the event moves within the 84 hour time frame the superior model resolution of the AVN / GFS and lack of   model resolution truncates  gives the model a lot more creditability.

 

RULE #7 DOUBLE EE RULE

When the 60 - 84 Hour ETA matches the ECMWF day 3-4 at surface and 500 MB that is a pretty good winning hand. Take it to the bank.