MEX long range max temps

This message was edited 1 time. Last update was at 11/04/2006 20:55:28

I noticed with a stretch of several runs late last week and last weekend that the 6 and 7 day maxes were often quite a bit higher on the 12z MEX guidance than the 00z. It seemed like the 00z would back off, then the 12z would pump them up, only to have the next 00z MEX back them off again (swings of 4-8 degrees). Given that the 00z MEX has a tendency to exhibit a bit of a climo bias, I was wondering if this bias doesn't exist, or has been accounted for somehow, in the 12z MEX guidance...it was certainly surprising to see temperatures so much higher than climo in days 6 and 7. That is definitely a rarity.

Perhaps the 12z is better at capturing significant warming trends in the longer ranges than the 00z...it has certainly anecdotally seemed like the "older" 00z MEX never quite captured significant temperature swings to the full degree.

Joe --

This is a MOS question that I'll refer to the Techniques Development Lab (TDL). They're the ones that develop MOS.

However, at day 6 or 7, it probably is synoptic variable fluctuations in the model that are resulting in what you are seeing. I *think* the same equations are used for ALL GFS MOS, no matter when the forecast cycle starts. What did the large scale flow look like for each of those forecasts?

Bill

I got the answer from Paul Dallavalle who works on MOS with the Meteorological Development Laboratory (MDL). His answer is below.
Stephen
-------------------------
The 00Z and the 12Z extended MOS guidance packages are based on completely different sets of equations. In the MOS temperature system, we use different equations for each cycle, projection, season, and station. For any MOS element, different equations are always developed for each forecast cycle, projection, and season.

What people are noticing in the differences between the 00Z and 12Z MEX guidance at days 6 and 7 is mostly due to the different developmental samples. While it is true that the MOS guidance tends toward "climo" at days 6 and 7, MOS is tending toward the "climo" of the developmental sample, not some long-term climatology. This tendency of a multiple regression system towards the mean of the developmental sample reflects the mathematics of the problem, i.e., as the information content of the independent variables (in this case, the model forecasts) declines, the regression equation is able to explain less and less variance in the predicted data and reverts to predicting the mean of the sample.

The differences in the developmental samples between the current 00Z and 12Z extended-range MOS forecast equations are quite substantial. For the 00Z equations, we used GFS data from April 1997 through September 2003. For the warm season equations, then, we had 7 seasons of developmental data. For the 12Z equations, we used GFS data from April 2002 through September 2004, and so had only 3 years of data for the warm season. As you might imagine, the mean values between the 7-year sample and the 3-year sample have to be significantly different at a large number of sites, and that's what you see at days 6 and 7. In case you're wondering, the differences in the developmental sample arose because there was never a 12Z extended GFS run until relatively recently.

We've redeveloped the GFS MOS temperature equations recently and plan on implementing them in May. To avoid a complete repeat of this variation in guidance between cycles, we tried to even out the developmental sample as best we could, though it's a balancing act between having adequate data to produce good guidance and having the same sample sizes. We used data from October 2000 through September 2005. Thus, for the warm season, we now have 5 seasons for 00Z and 4 seasons for 12Z. For the cool season, we have 5 seasons for 00Z, but only 3 seasons for 12Z. Eventually, we'll eliminate enough of the 00Z sample so that the two cycles, at least, are using the same seasons.