First off, Invest 97L is in the Carib. Invest 98L is still in the middle of the Atlantic. Both have a high chance of development within 5 days according to the NHC.
Atlantic 5-Day Graphical Tropical Weather Outlook:
https://www.nhc.noaa.gov/gtwo.php?basin=atlc&fdays=5
I've avoided posting runs lately until there is some consistency. I heard one meteorologist here in Tampa talking on TV a few days ago about how social media was discussing some of them.
I'll talk about the model you posted at the end. It is an experimental version of the GFS.
I usually use these, the operational versions of these models:
GFS:
https://www.tropicaltidbits.com/analysis/models/?model=gfs®ion=atl&pkg=mslp_uv850
Euro:
https://www.tropicaltidbits.com/analysis/models/?model=ecmwf®ion=atl&pkg=mslp_uv850
To see things better that don't show up well because they aren't shown as developing much, then I use vorticity:
GFS:
https://www.tropicaltidbits.com/analysis/models/?model=gfs®ion=atl&pkg=uv850_vort
Euro:
https://www.tropicaltidbits.com/analysis/models/?model=ecmwf®ion=atl&pkg=uv850_vort
Sometimes that makes something seem worse than it might be though, so check the other to see if it has a low icon at the surface with the pressure.
Euro hasn't shown too much lately. A few times a tropical storm in the Gulf from Invest 98L in the Atlantic. The one notable run was the 12Z August 17th run where it had a powerful hurricane just offshore from the Carolinas. That was also Invest 98L.
Link to that specific run:
https://www.tropicaltidbits.com/analysis/models/?model=ecmwf®ion=atl&pkg=mslp_uv850&runtime=2020081712&fh=228
The next paragraph is just about how much inconsistency there is from run to run.
18Z August 18th GFS has a tropical storm off the east coast of Florida and has it developing into a possible hurricane while it skirts the SE U.S. coastline through the Carolinas. That was also 98L. Previous run, 12Z, has Invest 98L going over Hispaniola so nothing else happened with it after. Invest 97L on that run when into SE Texas as a possible tropical storm. 6Z GFS had 97L clip Yucatan as possible strong tropical storm and then maybe Houston area as a strong tropical storm. 98L into Hispaniola and nothing much happened after. 0Z GFS on August 18th, 97L into Central America as possible strong tropical storm, near hurricane, maybe also emerging in Bay of Campeche also being a hurricane perhaps before going back into Mexico. 98L into Hispaniola it seems and not much after. 18Z GFS on August 17th, 97L into Yucatan as possible tropical storm and emerging to head toward SE Texas as perhaps a tropical storm. Again not much from 98L. 12Z on August 17th had 97L clipping Yucatan as tropical storm and then tropical storm into Louisiana. 98L goes south of some of the islands and ends up making landfall near Brownsville a category 1 hurricane probably.
I don't want to go through all the runs, but as you can see, a lot of variation.
I'm wasn't exactly sure what version of the GFS the "GFS-Para" was that you posted. At first I thought it was the old GFS, but the older GFS was only supposed to be run in parallel through September 2019 as discussed here:
https://www.noaa.gov/media-release/noaa-upgrades-us-global-weather-forecast-model
The article is from June 2019 and talks about GFS being upgraded at the time.
Then I looked at where I think the older version had raw data and it wasn't there any longer. ("para" folder empty, current in "prod": https://ftp.ncep.noaa.gov/data/nccf/com/gfs/)
Then I did some searching on Twitter and found something that indicated it was experimental. Which I didn't know if what I found was reliable but eventually I got to where I needed to be.
"The Environmental Modeling Center (EMC) is proposing to upgrade the NWS operational Global Forecast System (GFS) and Global Data Assimilation System (GDAS) from v15.3 to v16 in February 2021. The upgrade will improve model forecast performance and provide enhanced products. See PNS for information on specific upgrades."
https://nws.weather.gov/products/viewItem.php?selrow=868
And finally a PDF file it links to:
https://www.weather.gov/media/notification/pns20-44gfs_v16.pdf
There's a lot of information in there about it.
"In preparation for the operational implementation of GFS v16, EMC is currently running a real-time parallel with fully cycled data assimilation. Real-time output from this parallel are made available through both NOAAPORT and NCEP Web Services and can be found at:
https://para.nomads.ncep.noaa.gov
The real-time data will be available at best effort and with variable timing of output availability. EMC is also conducting retrospective experiments covering the past one and half years for a comprehensive evaluation of the GFS v16 implementation.
Evaluation of both the real-time and retrospective parallels can be found at
https://www.emc.ncep.noaa.gov/users/meg/gfsv16/
This page includes relevant links to various evaluation and verification web sites."
This page does have a link to real time data normally:
https://www.emc.ncep.noaa.gov/users/meg/gfsv16/
But while raw data is current on the NOAA server it is available from, that particular site they link to only had data through August 15th:
https://www.emc.ncep.noaa.gov/users/meg/gfsv16/realtime/gfsximages.html
I wanted to verify if the "GFS-Para" at Tropical Tidbits was the experimental GFS so I compared the data I could and compared the August 14, 2020 0Z run of the GFS on both sites.
I compared the East Pacific hurricane on that date and it matches. Compare 986mb for Genevieve in East Pacific for experimental GFS and 950mb for operational GFS.
Saved imagery:
Tropical Tidbits:
https://i.imgur.com/T7vjMDF.png
Experimental GFS (top right corner is experimental version, top left is operational version):
https://i.imgur.com/5VauH5Q.png
And to verify top left corner is the operational version, here it matches Tropical Tidbits operational version for that date and frame:
https://i.imgur.com/GJ4gV2C.png
I can't speak too much about how well that model is doing. This gets beyond what I know, but it *appears* it's doing worse right now for the Atlantic. First off Genevieve in the East Pacific is not near the mainland coast of Mexico like the experimental version had it.
They have verification statistics that compare the operational and experimental GFS with each other here:
https://emc.ncep.noaa.gov/users/verification/global/gfs/para/
And specifically for tropical cyclone track error they have a chart here:
https://emc.ncep.noaa.gov/users/verification/global/gfs/para/tropcyc/trackerr/
In the later parts of the chart they have less cases. I suppose that works like my model error does. At 120 hour interval they have 14 cases as of posting this that they used to determine the data for the chart at that particular interval. AVNO, the current GFS, has error of around 200 nautical miles. Experimental version maybe around 280 nautical miles.
You can compare specific storms too. I checked the East Pacific and over there the experimental version seems to be doing better, but we aren't talking about anything other than this year so who knows about long term. The data I assume is not current for Genevieve as it is really off in reality so I assume that since the modeling isn't current the error statistics are not either. Also, I have 9 cases available for my average model error for that storm for the operational GFS at 48 hours and they have only 1. I have 1 case of model error for 96 hours for the operational GFS and they don't have data that far out yet.
Anyway, the point is the model is experimental. You can do some research on how well it's been doing, but until I see some consistency in models, from multiple operational models and also over a few runs, I wouldn't panic about things. If reservations are easily cancellable, that's up to you, I never want to encourage someone to not do something. But right now it remains pretty unclear in my opinion where these invests are going. I can't say that a model of any type can't be right in the end, but I like to see consistency from reliable operational models. Sometimes even a model that isn't good gets something right, it doesn't mean we should trust it more than others for that time it does. (I'm not saying the new version isn't good, just about models in general, I didn't know about the new version until posting this) I have no idea how good the experimental GFS is doing on various things, but I need to see some track record of how it does over a longer period. Using the statistics as I see them, it hasn't done as well for the Atlantic. We'll all be watching them. |