Jump to content
  • Member Statistics

    17,606
    Total Members
    7,904
    Most Online
    ArlyDude
    Newest Member
    ArlyDude
    Joined

The GFS set to update its Data Assimilation Scheme: The Introduction of Hybrid-Kalman Filtering


Recommended Posts

Yeah the H100/H85 thicknesses are all messed up too. This has created some large errors in the MOS temps across higher terrains...about a 10-15 degree F error. I heard NCEP is experimenting with a run reverting back to the old way of calculating H100 heights.

I've been out of the loop to be honest, so apologies for not replying sooner. This has nothing to do with the hybrid nor the model itself. The (unified) post-processor (which creates the files/fields we send out to the world) was also updated on Tuesday, most notably to add some new fields (specifically for wind energy and fire weather customers). I didn't even realize that the 1000 hPa Z calculation was modified in this new code (it is possible something was inherited from another non-GFS model through unification, I don't know).

From what I gather, a fix to the model post-processor has already been developed and tested, and they may get it fixed as earlier as tomorrow. As mentioned, the fix is necessary to help mitigate issues with MOS temps.

Link to comment
Share on other sites

Wednesday, May 23, 2012 5:31 PM

As you may know, there was a major GFS package upgrade at NCEP on Tuesday, May 22, beginning with the 1200 UTC cycle. The specific change to the GFS involved introduction of the hybrid EnKF-3dvar global data assimilation system. After this upgrade, some unrealistically warm MOS temperature forecasts were noted at several sites. Today, we have been attempting to diagnose the reasons for these forecast errors in collaboration with modelers at NCEP. After concluding our discussions this afternoon, we think we may have pinpointed a possible cause of the problem:

The majority of the anomalous MOS temps were reported by WFOs in elevated terrain regions. When analyzing the values of predictor fields used in the MOS equations for these sites, we noticed some dramatic changes in the 1000 -850 hPa thicknesses between the old operational and new Hybrid EnKF runs of the model. Changes to this variable alone were responsible for differences as large as 15 degrees F in the MOS temperature forecasts from the two versions of the model. NCEP confirmed that they had made a change to the way the 1000mb heights are calculated in the new version of the model. They also relayed a report from another user who independently reported that the height of the 1000mb surface at her location unexpectedly decreased after the upgrade. This would certainly make sense with what we have seen in the behavior of thicknesses that rely on the 1000 hPa height.

NCEP is currently preparing some test GFS runs which revert back to the original 1000mb height calculation to ensure this corrects the issues seen in MOS in regions of elevated terrain. MDL will also continue to examine forecasts at suspect sites to make sure that this problem is confined to the contributions from the 1000 hPa heights. MDL and NCEP are both hopeful that a fix to this problem could be implemented in the NCEP jobstream relatively quickly assuming that our initial diagnosis of the problem is correct.

Sincerely,

Kathryn Gilbert

Chief, Statistical Modeling Branch, MDL

Mark Antolik, Sr. Developer

Dave Rudack, Sr. Developer

Scott Scallion, Operations Lead

Link to comment
Share on other sites

Friday, May 25, 2012 10:47 AM

A change in the computation of the GFS 1000 mb height fields in the model's post processing was introduced with the implementation of the hybrid EnKF-3dvar global data assimilation system. This was determined to be the cause of the extreme GFS MOS temperature busts this week, particularly noticeable in Colorado and New Mexico. Beginning with the 1200 UTC GFS model runs today, the 1000 mb height field calculations will revert back to their former computational method. The 1000 mb heights from the GFS are used to compute 1000-850mb and 1000 - 500 mb thicknesses which are frequent predictors in the MOS temperature equations. This change at 12z will bring the GFS MOS temperatures back to their pre-Tuesday's implementation behavior.

Sincerely,

Kathryn Gilbert

Link to comment
Share on other sites

  • 1 month later...

Some of the forecasters I work with have been complaining that the GFS MOS has been much too cold across North America recently. I've not been paying that close of attention to the issue, but it is this an actual thing? When I look at the raw GFS anomaly correlations and biases, I don't see anything obvious.

Link to comment
Share on other sites

Some of the forecasters I work with have been complaining that the GFS MOS has been much too cold across North America recently. I've not been paying that close of attention to the issue, but it is this an actual thing? When I look at the raw GFS anomaly correlations and biases, I don't see anything obvious.

There has been a cool bias on the MOS according to our local NWS office:

And the upgrade didn't correct the raw 2m temps that were too cool in the past and have still been running too cool.

This was the 24 hr forecast for the day that Newark reached 102:

I

Link to comment
Share on other sites

NCEP sent out a notice after complaints about how the GFS underestimated the heat wave back in June. IIRC, they claimed it may have something to do with an evapotranspiration issue. Don't quote me on that...I'm just going by memory. They were working on a fix.

Up this way, the NAM has been way too warm. It's like MOS has gone nuts. We are resorting to look at 850 temps and do forecasting the old fashioned way. NAM is pushing 98F temps with like +16 at 850.

Link to comment
Share on other sites

Yeah, I totally agree the forecasts have been out to lunch in the out periods. But the concern for our forecasters are the D1 and D2 MOS forecasts.

This was the most recent comment that I saw from our regional forum:

Link to comment
Share on other sites

Thanks, bluewave. Hopefully one of the NCEP/HPC guys can chime in with some quantitative stats.

Adam,

The complaints we have been getting (and what we are trying to address ASAP) has to do with the GFS being too cool (and wet) in the afternoon hours, over specific regions of the country (most notably the plains and upper midwest)....particularly during these hot spells.

It doesn't show up in the plot you showed as that is depicting a hemispheric score (so for 00z, as shown, the only region affected by this issue is over central CONUS.....whereas other regions at the same time have a nocturnal warm bias ... thereby canceling out the daytime cool bias over CONUS).

This actually has nothing to do with the hybrid implementation this past May, as this issue was also prevelant last summer as well (when we were still running 3DVAR). We've recently discovered that some changes were made to a table that is used by the NOAH land surface model coupled to the GFS (I've been in and out of the office, and I'm not sure if this happened with the original T574/physics updrade in 2010, or the "bug fixes" that were put in sometime in May 2011). We're already testing the fix (going back to the old table for the LSM), but I have no idea what the time table is for getting the patch into production. We have to make absolutely sure that it doesn't mess up other regions/seasons/etc.

Link to comment
Share on other sites

NCEP sent out a notice after complaints about how the GFS underestimated the heat wave back in June. IIRC, they claimed it may have something to do with an evapotranspiration issue. Don't quote me on that...I'm just going by memory. They were working on a fix.

Up this way, the NAM has been way too warm. It's like MOS has gone nuts. We are resorting to look at 850 temps and do forecasting the old fashioned way. NAM is pushing 98F temps with like +16 at 850.

Yup even here raw nam data surface temp progs have been averaging around 3F too high and its bled into NAM MOS. Also noticed that the heat ramps up way too quickly; the run yesterday had PHL at 99F at 11 am Wed morning and 97F at 11 am Thursday morning.

Link to comment
Share on other sites

Adam,

The complaints we have been getting (and what we are trying to address ASAP) has to do with the GFS being too cool (and wet) in the afternoon hours, over specific regions of the country (most notably the plains and upper midwest)....particularly during these hot spells.

It doesn't show up in the plot you showed as that is depicting a hemispheric score (so for 00z, as shown, the only region affected by this issue is over central CONUS.....whereas other regions at the same time have a nocturnal warm bias ... thereby canceling out the daytime cool bias over CONUS).

This actually has nothing to do with the hybrid implementation this past May, as this issue was also prevelant last summer as well (when we were still running 3DVAR). We've recently discovered that some changes were made to a table that is used by the NOAH land surface model coupled to the GFS (I've been in and out of the office, and I'm not sure if this happened with the original T574/physics updrade in 2010, or the "bug fixes" that were put in sometime in May 2011). We're already testing the fix (going back to the old table for the LSM), but I have no idea what the time table is for getting the patch into production. We have to make absolutely sure that it doesn't mess up other regions/seasons/etc.

Daryl,

Does this include the expansiveness of the .01" isohyet?

Link to comment
Share on other sites

Daryl,

Does this include the expansiveness of the .01" isohyet?

I was referring to the specific humidity, but yes, this also impacts the lightest of threshold precipitation amounts (rain/no-rain, 0.01", etc.). The "fix" seems to reduce this bias substantially (though we've only run a fairly small sample).

Link to comment
Share on other sites

I just started keeping track of MOS numbers (again), statistically through excel, and have noticed a cool bias in both the GFS and NAM for the southeast. In some instances the NAM is 5-7° too cool as it anticipates more cloud cover than what actually happens.

DTK, is there any timetable for the "fix"? It would be nice to know so I can make note of it on the spreadsheet I have, which, unfortunately, only goes back a week :lol:.

Link to comment
Share on other sites

I just started keeping track of MOS numbers (again), statistically through excel, and have noticed a cool bias in both the GFS and NAM for the southeast. In some instances the NAM is 5-7° too cool as it anticipates more cloud cover than what actually happens.

DTK, is there any timetable for the "fix"? It would be nice to know so I can make note of it on the spreadsheet I have, which, unfortunately, only goes back a week :lol:.

Not yet that I'm aware of (hopefully soon, but we still have more testing to do). There will be proper notice sent out, however, once it is decided if/when the change would go into the production GFS.

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...