ncep model and mos sfc tds

Hi and thank you in advance!

The east coast has been arid from ern nc to sne last 4 weeks including 16+ days in a row here without meas rain...yet nam and gfs dewpoint forecasts are wayyyyy tooo high during the afternoons in return flow behind the sfc high and usually too low just ahead of the cf....

I saw a soil moisture contribution graphic for the next week, and it looks far overdone.

We at kbox lower mos dewpoints during mid aftn from the top of the bl as viewed in bufkit or tsecs.

bottom line, why cant the ncep models do a better job with dewpoints which anecdotally in my mind now rank as one worst predictive elements of the model output.

This is troubling, especially since dewpoints are no longer the throwaway mos guidance we once thought... having a huge bearing on Value ADDED fire weather outlooks... as well as other applications I suspect such as air quality. I get calls from researchers on rh. Convective sounding applications.

We grid edit dewpoints but we tend not to modify the guidance often enough during the transition season...or just dont think its important enough.

any comments on additional approaches?

I presume EMC is very aware of the shortcomings of the td fcsts in mos and the models (both gfs and nam).

Just woeful the past several days (21 hr fcst 10f in error ydy from the lowest avbl 00z/30 ncep mos /fwc,met,mav/ guidance for the bulk of southern new england (sne) and ydys 45 hr fcst tds for today were in the 40s to near 50 (00z/30 fcst)!---not going to happen aznd a probable day 2 error of 8-13 degrees too wet!). we know we have a + anomaly pattern right now, yet we think the "top" of the bl is modeled very well but degrades toward the sfc on any well mixed day.

Probably most on this board agree that considerable improvement in this element would give us even better model qpf results...

in the end, we wonder if the models are mixing out the boundary layer (bl) properly to get good sfc dews and resultant bl sounding.

A byproduct of this poor element, is time devoted to dewpoint grid editing. This is an element with more importance than once thought...

HPC I suspect knows what we are taliking about with regard to the dews fcst days that they offer us in day 4-7.

Thanks for considering,

Walt Drag
Taunton-Boston (KBOX)

Hey Walt --

Could you put up graphics to illustrate your points? I'll put this message on the GFS topic site and discuss how GFS handles soil moisture and what that might imply to the near-surface moisture/specific humidity/dew point. I'll also discuss MOS and extreme events.

All --

Re: your approach to adjusting Td's using BUFKit and the top of the planetary boundary layer....that makes a lot of sense to me. That there is insufficient mixing out may also imply the BL isn't deep enough, so it may require further adjustments from even beyond that.

Re: MOS ... you say in your message that the east coast has been arid from NC to southern New England. The answer to the MOS problem is found in this. MOS works best when conditions are reasonably close to normal; that isn't the case at all here because MOS does a best linear fit given all observational and synoptic scale model data matchups. Dryness to this degree is well off that line *and* is poorly sampled in the data used to come up with MOS guidance, so it's not at all surprising that MOS is performing poorly. I suspect that MOS temps have been off also (too low during the day, too high at night, because of dry soils; this would be more likely true during the warm season). There really isn't much you can do about MOS in extreme circumstances except be aware of the problem and make appropriate adjustments

Re: the models themselves ... remember that soil moisture observations are not included in data assimilation systems used at NCEP. There is an attempt to keep soil moistures in line in the NAM (Eta, and soon WRF) assimilation by using actual precipitation in the data assimilation system, to avoid problems with precipitation bias affecting soil moisture.

Additionally, tuning models to get the best scores (i.e. so that they work well when conditions are close to normal) also can result in problems when extreme conditions exist. I'm not sure what contribution this would make here to the dew point problems you are seeing, though I'd suspect they'd go in the same direction as the *direct* soil moisture problems.

i am at home and cant recall the predictive moisture graphic i saw, tho i have seen gfs related which are better... i'll try and find the link late tonight or sat night when am on mids at work...currently at home.

on other graphics..not sure what you need...

but

i did see your response to mos and seasonal normal regime. Thats got to be a problem, especially if climate variability is going to be greater in whatever cyclic global warming cycle we are in????!!!!

Forecasters need to be encouraged to adjust the dewpoint guidance. I think we in the field have a small disconnect on the value of dewpoints for the customer.

additiionally... we rely heavily on either our prior gridded database and/or "MOS" for a starting point on the forecast. Big time problem if MOS cant flow from normal to extremes... that means the MOS database needs to be expanded in duration sample 'yearly'. Otherwise, it cant replace forecaster hands on adjustments.

Grids are work, if we want accurate forecasts.
That means these dewpoints are a grid issue.

You're welcome to check a tsec in the lowest 5000 ft from 00z/30 and 00z/31 NAM and GFS for ORH, and then check the comparable 00z/30 MAV and MET forecast and then the verification... not good.

You can do that for BDL, MHT, BED, TAN, womever you choose... it will be illustrative of how to go about mitigating bad guidance...

a problem though, time consuming effort on adjusting MOS tds when mutlitasking on deadline. It can be done, but can be hectic. Thats just one elemnt that has to be modified.

The 18z TD error (42 hr fcst for today at ORH is at least 11degs on the dewpoint.

Whats this mean... it means our collaboration with SPC the past two days on significant downward adjustments on Td's was worthwhile, even tho many may not have believed. Additionally... MOS and model guidance would never have caught the Red Flag situation that occurred in sne today FQT Gusts >25 MPH, rh <25 pct and tinder dry surface debris fuels.

this is not proofed/spell checked but should get the basic flavour back to you.

Let me know what else you need... and thank you Bill!
Walt

I looked at Albany (NY) observed and forecast soundings valid at 00z from various days and found that the models were, as Walt said, way too moist, but they did a good job of mixing through the boundary layer - the whole boundary layer was too moist and often not deep enough, but not missing the PBL height by very much. Similarly, looking at raob minus first guess differences of 925 mb mixing ratio at 00z, the first guesses were all too moist, often by a lot, over the entire eastern US, not just the drought-stricken northeast. A little piece of good news is that while the parallel WRF runs were also too moist, the errors at least were smaller.

It seems a likely culprit is too much evaporation and not enough sensible heating. The GFS has for many years had too much evaporation over land, and we noted that the NAM-Eta change last May also boosted its evaporation, though before then it usually had too little, was too dry on average. See http://meted.ucar.edu/nwp/pcu2/NAMMay2005.htm for that discussion.

Right now, the mesoscale branch at EMC is focusing on the WRF implementation and tuning in WRF. Its moisture analysis method is quite different than in the Eta, so there will not be much interest in trying to fix the Eta since it will be going away in a couple of months anyway. The Eta has a known bug affecting how much water is evaporated by green vegetation. That bug is fixed in WRF. Since parallel WRF runs are also systematically too moist though not by as much, this bug might be contributing to the problem but isn't the entire cause of the problem.

For what to do in the field, perhaps someone who isn't working a forecast shift could write a smart tool that would calculate a boundary layer height from model soundings and your human forecast max temperature, then have it mix the moisture in the model grids through that depth to come up with daytime minimum dewpoints, and apply some sort of diurnal curve based on recent daily data. You will still need to add some arbitrary further drying to account for the evaporation bias in the model, just mixing the boundary layer higher won't provide the full correction needed.

In the longer run, NCEP is working on developing bias corrections for the models. These will probably show up first in ensemble statistics as bias-corrected ensemble means. This still won't solve all your problems, because if the bias statistics are long term and you have a relatively rare situation, the correction won't be good, or if the bias correction is based on recent weather and you have a pattern change, it will take a little while for the bias correction to catch up to the new pattern. But it will be a notable improvement.

Stephen

Hello Stephen,

I appreciate your helpful take on this. Thank you!

Yesterday Sunday 4/2/06 was another day of model excesses in surface dewpoints...18z and 21z errors averaged about 10 degrees too high again in Southern New England both from GFS-MAV, NAM-MET's and NGM-FWC. Another Red Flag Warning day with a PVD TF Green airport brush fire on our radar.

The NGM (frozen model 1991) had the least average error at ORH (used as a spot check) with its errors generally 6-9f too high, while the 18 and 24 hr fcsts of dewpoints by the GFS had generally 18z, 21z and 00z/3 errors of 9-15degs!!

Not to beat a dead horse (so to speak) .. basically this is an alert to forecasters (maybe most already knew this?) of value added forecast interaction with the ported grid DEWPOINT guidance in the short term (at least first 48 hrs)... via use of model BUFKIT (or AWIPS) tsecs/soundings and looking for the lowest dewpoint in the PBL at that AFTERNOON hour and modifying your forecast close to those values (non-marine influenced).

The mid shift SPC fire weather forecaster was also noting these recent wet dewpoint bias's.

A final note: Thanks for consolidating the forum and making yourselves available to clarify-improve the science.

Walt

All --

Stephen's suggestions for how to cope with the problem we are seeing is a great example of a use of the science we know to make a tool that allows for the intelligent use of the models. Hope someone takes him up on it that has time to write the code for such a smart tool for the IFPS grids.

Incidentally, model performance for these near-surface variables, for those of you who may NOT know, can be found at:

http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/mmbverif.namplls/

The 10-m wind forecasts, 2-m temperature and RH forecast bias and RMS error for the last 15 days of forecasts appear by region for the GFS, NAM (Eta), NAMX (WRF), NAMY (experiment with WRF). In taking a glance at the RH graphics for the Northeast Coast, there's been a high bias in RH over the last 15 days, lowest in NAMX/NAMY, highest in GFS, but drifting toward no bias in the most recent few days. It would appear then that the models themselves are performing better than MOS, which cannot handle extremes very well. In this case, it appears that though the NWP models are tuned to perform best when the weather is not too abnormal, the NAM and NAMX (i.e. Eta and WRF) are reflecting, at least to some degree, the drought being experienced in the east.

Bill

The ETA is having big-time problems once again with its 3-d var analysis. Its 2-meter dewpoints were initialized up to 17F too high this evening.

It's pretty obvious the ETA is having problems correcting back towards the actual obs yet again.

WRF-NMM is somewhat better with its GSI analysis scheme, but its dewpoints were generally initialized too high as well.
http://www.wxcaster.com/CONUS_WRF_SFC_SLPTMPDPTSBLI_00HR.gif

RUC was the closest to reality (as expected).

Below is a list of ETA/WRF/RUC initialized dewpoints (approximation using my eyeballs) for selected locations versus the actual observed dewpoints for 00z/SAT, April 8:

SITE ETA WRF RUC OBSERVED

DNV 70F 64F 65F 64F
LAF 70F 64F 65F 64F
IND 69F 64F 65F 61F
MCI 66F 60F 56F 49F
CMH 65F 64F 59F 54F
STL 65F 60F 52F 52F
SHV 65F 61F 62F 60F
DVN 65F 58F 60F 60F
ORD 64F 57F 60F 60F
PAH 64F 57F 53F 54F
MEM 64F 57F 45F 49F
BNA 63F 55F 51F 50F
OKC 62F 58F 52F 52F
LIT 62F 58F 51F 49F
ATL 62F 54F 49F 48F
SGF 61F 59F 59F 60F
TUL 61F 59F 53F 54F
ICT 60F 57F 49F 46F
DFW 59F 59F 55F 56F
HSI 49F 41F 39F 35F

Is there any way to correct the ETA 3-d var analysis dewpoints back to reality or will we have to wait for expected improvements when the WRF-NMM arrives on June 13th? How much improvement should we expect from the WRF-NMM (i.e., are the above dewpoints representative)?

Thank you for your contribution... glad to see others are noticing. Highlights the importance of the forecaster being in the mix with officially issued dewpoint assessment and therefore associated forecast product interrelationships, at least until the model surface fields improve. Walt

I'm posting this message in the "ncep model and mos sfc tds' forum because I do not appear to have the ability to start a new thread.

***************************************************************

I just finished viewing Stephen Jascourt's webcast on the NAM-NMM:
http://www.meted.ucar.edu/nwp/NAMWRF/splash.htm

He states that surface mesonet obs will not be ingested into the WRF-NMM "because quality control problems created adverse impacts."

I really have to wonder about the wisdom of this. Not all meso-networks are created equally! Some are much better than others. For example, in my experience, the Oklahoma and Texas Mesonets provide better quality observations than the ASOS/AWOS obs because they are properly sited (i.e., not sitting on/near concrete runways at the airport) and maintained by meteorologists that care about providing accurate, research-quality data.

I routinely see that the FAA-maintained Oklahoma AWOSs report a high dewpoint bias (2-6F) on afternoons with a well-mixed boundary layer compared to the Oklahoma mesonet obs and NWS-maintained ASOS obs. This problem has been noted by many other meteorologists besides myself. I have compared these obs to properly-calibrated mobile RM Young equipment and I am convinced the Oklahoma mesonet obs are more correct. I've had similar experiences with the West Texas mesonet.

The MADIS data-feed provides the capability of filtering out the suspect networks (e.g., AWS Weatherbug) and permitting the use of the good networks. Surely, the obviously-good networks could be added now. I suspect the model developers already have a good idea of which networks are good, suspect, and bad. The FSL RUC developers probably have a good idea as well since they allow mesonet data ingest of T/Td obs.

hi Robert-

Thanks for posting information about this case involving NAM dew points. I have examined it a bit and have a few comments/explanations. First, as best I can tell, the analysis time is actually 00z 15 Apil instead of 8 April, as these obs match the ones you list: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/2006041500_obsdew.gif

The analyzed 2-meter NAM dew points are indeed too high over much of this region: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/2namdwpf_2006041500f00.gif

The problem is that not all of this high bias is simply due to the analysis - part of it is clearly due to the reduction of the values to the 2-m level. Here is the dew point analysis at the same time for the first model level above ground: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/1namhybldwpf_2006041500f00.gif
It's still too moist, but it is certainly an improvement. The differences between the 2-m and 1st level values are generally between 2 and 6 degrees F in the moist sector with much higher differences back in the drier air: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/namdewdiff_2006041500f00.gif

As you note, the NAMX (WRF) does offer some hope, as its 2-m values are better than those from the NAM: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/2namxdwpf_2006041500f00.gif

And the NAMX values on the 1st model level look very good compared to obs: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/1namxhybldwpf_2006041500f00.gif

Where does this leave us? I have been advocating using the 1st model level to represent "surface" T/Td for years now ( http://wwwt.emc.ncep.noaa.gov/mmb/tips/tip.oct00.html), but I haven't gotten far with it. (One big shortcoming is that it doesn't work very well over higher terrain.) 2 meters seems to be the most common level at which these values are examined in the model, but the NAM reduction is far from perfect. The reduction uses the 1st level value, the skin layer values, various fluxes, and the surface exchange coefficient. We may not be handling skin level values very well, although I think it is improved in the WRF.

Note: here is 1st layer q from the NAM: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/1q_2006041500f00.gif
now, skin q: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/skinq_2006041500f00.gif
finally, 2-m q: http://www.emc.ncep.noaa.gov/mmb/gmanikin/ndas/2q_2006041500f00.gif

I think that 2-m and 1st level dew point values should be examined together, although I acknowledge that AWIPS and other displays rarely make this available. The main point, though, is that 00-hr values at 2 meters are not just dependent upon the analysis code.

-Geoff

Robert.A.Prentice wrote:The ETA is having big-time problems once again with its 3-d var analysis. Its 2-meter dewpoints were initialized up to 17F too high this evening. It's pretty obvious the ETA is having problems correcting back towards the actual obs yet again.

WRF-NMM is somewhat better with its GSI analysis scheme, but its dewpoints were generally initialized too high as well.
http://www.wxcaster.com/CONUS_WRF_SFC_SLPTMPDPTSBLI_00HR.gif

RUC was the closest to reality (as expected).

Below is a list of ETA/WRF/RUC initialized dewpoints (approximation using my eyeballs) for selected locations versus the actual observed dewpoints for 00z/SAT, April 8:

SITE ETA WRF RUC OBSERVED

DNV 70F 64F 65F 64F
LAF 70F 64F 65F 64F
IND 69F 64F 65F 61F
MCI 66F 60F 56F 49F
CMH 65F 64F 59F 54F
STL 65F 60F 52F 52F
SHV 65F 61F 62F 60F
DVN 65F 58F 60F 60F
ORD 64F 57F 60F 60F
PAH 64F 57F 53F 54F
MEM 64F 57F 45F 49F
BNA 63F 55F 51F 50F
OKC 62F 58F 52F 52F
LIT 62F 58F 51F 49F
ATL 62F 54F 49F 48F
SGF 61F 59F 59F 60F
TUL 61F 59F 53F 54F
ICT 60F 57F 49F 46F
DFW 59F 59F 55F 56F
HSI 49F 41F 39F 35F

Is there any way to correct the ETA 3-d var analysis dewpoints back to reality or will we have to wait for expected improvements when the WRF-NMM arrives on June 13th? How much improvement should we expect from the WRF-NMM (i.e., are the above dewpoints representative)?