August 11, 2020

Dr. Jeff Masters: Bogus GFS Model forecasts of Atlantic tropical cyclone genesis


Pin It

cristina-jun12-modisBy Dr. Jeff Masters From Weather Underground

There are no tropical cyclone threat areas in the Atlantic to discuss today, and none of the reliable models for forecasting tropical cyclone genesis is predicting development over the coming five days. However, the 00Z Friday run of the model predicts that a low pressure area will develop over the by Wednesday, and push northwards into the Gulf of and become a tropical storm late in the week. The GFS has been fixated on variations of this idea in all of its runs for the past five days–though the timing of when the predicted storm will form has bounced around from 5 – 11 days into the future. Should we be concerned? A 2013 study by a group of scientists led by Florida State’s Daniel Halperin found that we have three models that can make decent forecasts of the genesis of new tropical cyclones in the Atlantic: the GFS, European (ECMWF), and UKMET models. The study only evaluated the model skill for forecasts out to four days in the future, and the forecast skill declined markedly for three- and four-day forecasts. In the current scenario, we are talking about forecasts made much further into the future, which are bound to be low-skill. In addition, the study found that the GFS model had a high incidence of for tropical cyclone genesis forecasts in the Caribbean (50%). The other two reliable models for predicting tropical cyclone genesis (European and UKMET) had no hint of a low pressure area developing in the on Wednesday in their 00Z Friday runs. One additional model to consider: the 00Z Friday run of the NAVGEM model is supporting the GFS’s idea of a low pressure area forming in the by Wednesday. The predecessor to this model, the NOGAPS model, was evaluated in the Florida State study, but performed poorly in making tropical cyclone genesis forecasts. FIG 1 gfs-jun13However, when two or more models make the same genesis forecast, the odds of the event actually occurring are increased considerably, the study found.

Figure 1. Friday the 13th, GFS style: The 00Z UTC Friday, June 13, 2014 forecast from the GFS model for nine days into the future shows a powerful tropical storm in the Gulf of Mexico. The purple colors indicate winds of 50 – 60 knots (57 – 69 mph.) But is it a bogus forecast? Very likely.

Figure 2. The 2004–11 GFS forecasts for tropical cyclone genesis, showing Hits (green triangle), False Alarms (red square), and Incorrect Timing (blue circle) event locations. Numbers in parentheses are the numbers of model-predicted FIG 2 gfs-genesisevents. The model made 46 forecasts that a tropical depression or tropical storm would form in the Caribbean (purple box) during this 8-year period. Fully 50% of these forecasts were False Alarms; 11% of the forecasts verified, but the timing was off by at least a day (IT events); and 39% of the genesis forecasts verified with the right timing. Noteworthy is the model’s few False Alarms over the Gulf of Mexico: only 15% of the total. Image credit: Halperin et al., 2013, Weather and Forecasting, “An evaluation of tropical cyclone genesis forecasts from global numerical models.”

We know that the GFS model gets in trouble when making predictions of heavy thunderstorm activity via a problem called “convective feedback.” Basically, the model sometimes simulates that an unrealistically large area of thunderstorms will develop, destabilize the atmosphere, and cause an area of low pressure to form that will draw in more moisture and create more heavy thunderstorms. This vicious cycle can snowball out of control and generate a bogus low pressure area that can then modify the upper level winds, reduce the wind shear, and allow a tropical depression to form. This problem may be less of an issue in a new version of the GFS model scheduled to be released late this summer; NHC hurricane specialist Eric Blake tweeted on Tuesday a comparison of the old and new 1-week GFS model forecasts for the Western Caribbean made last Tuesday, showing that the upgraded GFS model was not creating nearly as strong of a low pressure system as the old GFS model. Arguing against any development in the Atlantic the remainder of June is the anticipated strengthening in the West-Central Pacific Ocean of the Madden Julian Oscillation (MJO), a pattern of increased thunderstorm activity near the Equator that moves around the globe in 30 – 60 days. An active MJO in that part of the tropics tends to bring large-scale sinking motion to the tropical Atlantic and increased wind shear, which puts a damper on the chances of tropical storm formation in the Atlantic. The MJO is predicted to drift slowly eastwards into the Eastern Pacific by late June, which will tend to keep odds of tropical storm formation lower than average in the Atlantic into late June. All factors considered, I am inclined to give a 10% chance that the GFS model is correct in spinning up a tropical depression late in the week in the Western Caribbean.

Top Five Strongest Early Season (May – June) Eastern Pacific Hurricanes

1973, June 6: Hurricane Ava, 160 mph, 915 mb.

2010, June 25: Hurricane Celia, 160 mph, 921 mb

2014, May 25: Hurricane Amanda, 155 mph, 932 mb

2000, June 21: Hurricane Carlotta, 155 mph, 932 mb

2014, June 12: Hurricane Cristina, 150 mph, 935 mb

See attachments for Fig 1 and Fig 2

IMAGE: True-color MODIS image from the Aqua satellite of Hurricane Cristina at 18 UTC Thursday, June 12, 2014. At the time, Cristina was a Category 4 storm with 145 mph winds. Image credit: NASA.

For more on this story go to:


Print Friendly, PDF & Email
About ieyenews

Speak Your Mind