Tomer Burg - Blog

All Archived Posts

Current Weather

Meteorology Analysis

Non-Weather Posts

GFS May Hurricane: More Fiction Than Reality

Tomer Burg • 15 May 2022 • Current Weather

Post Highlights

As the 2022 Atlantic Hurricane Season approaches its onset, numerous deterministic GFS runs have been consistently depicting a hurricane entering the Gulf of Mexico late this month. If this were to verify, this would be highly unusual so early in the hurricane season. But a deeper look shows many red flags that lower confidence in such a scenario.

Deterministic models such as the GFS only depict one of many possible outcomes; ensembles are better suited for determining the range of possible solutions. But like any model, ensembles are susceptible to their own set of biases, such as false alarms for tropical cyclogenesis in the region. All these factors suggest to take the GFS forecast with high caution.

Date of first Tropical Cyclone formation in the North Atlantic from HURDATv2. Years with a storm prior to April 1st use the formation of the second tropical cyclone.

Atlantic hurricane seasons have been starting earlier and earlier in recent years. During a 7-year consecutive stretch between 2015 and 2021, every season had its first tropical cyclone form before the official start of the hurricane season on June 1st. This prompted discussions last year of moving up the official start of the hurricane season. While this did not occur, the National Hurricane Center subsequently moved up the start of routine Tropical Weather Outlooks to May 15 instead of June 1.

Date of first hurricane formation in the North Atlantic from HURDATv2.

Interestingly, this earlier trend in first tropical cyclone formation doesn't appear to translate over to first hurricane formation, which remains between late July and early August. It should be noted there are important caveats to the last two plots – for example, before the start of the satellite era, it was difficult to detect tropical cyclones far from land, resulting in an unknown number of tropical cyclones missing from the HURDATv2 tropical cyclone database. Even for tropical cyclones that subsequently affected North or Central America, it is difficult to know when and where they formed, as some were already mature hurricanes by the time they were first detected. Lastly, some early-season cyclones tend to be subtropical, and the operational classification and handling of subtropical cyclones has changed quite a bit over the years. Considering this, data for first tropical cyclone/hurricane formation especially before the satellite era should be taken with high caution.

Number of tropical cyclones per 1 degree gridpoint in the month of May from 1851-2021. Plot made using Tropycal.

Climatologically speaking, tropical cyclones in the month of May are most common off the Southeast U.S. coast. A second maximum exists east of the Bahamas, potentially associated with tropical transition as a baroclinic (non-tropical) low becomes cut off from the jet stream and under the right conditions (e.g., warm sea surface temperatures and/or low tropospheric stability) may acquire subtropical or tropical characteristics.

Note that there is a third maximum in May tropical cyclone tracks over the western Caribbean. These can be associated with Central American Gyres (CAGs), which tend to occur in the spring and fall, produce heavy rainfall over Central America, and can be associated with tropical cyclogenesis. Recently, Tropical Storm Alberto (2018) formed in the western Caribbean in late May as a subtropical cyclone.

Trend over the last 8 deterministic GFS runs in 6-hour precipitation rate and mean sea level pressure (MSLP). Image from Tropical Tidbits.

Now that we've reviewed some basic May tropical cyclone history, let's take a look at the latest forecasts from the Global Forecast System (GFS) model. At a quick glance, one would notice a hurricane moving up from the Caribbean Sea into the Gulf of Mexico, a region that has been prone to hurricane landfalls over the last few years. Not just one, but eight consecutive runs have had this hurricane but with varying locations. But – there should be a couple of red flags that immediately jump out:

Forecast valid date: This forecast is still roughly 9-11 days away. Generally, deterministic forecasts beyond day 6-7 should be taken with caution as many factors can change in the meantime.
Forecast consistency doesn't translate to accuracy: Especially at such long lead times, deterministic models such as the GFS may be consistent for multiple runs in a row, giving a false perception of high confidence, but ultimately fail to verify. Part of this is a perception bias – we remember the major events that verified, but forget those that were forecast to occur but did not. For just one of many examples, the GFS consistently showed Hurricane Joaquin (2015) making landfall in the U.S., but instead it recurved well offshore.
Climatological Context: A hurricane in May should immediately raise red flags. May hurricanes are extremely rare – there are only four known May hurricanes in the HURDATv2 database (1851-2021), the most recent one in 1970. Of course, highly anomalous events can and do occur, but additional evidence beyond a few GFS deterministic runs is needed to further assess this possibility.

GEFS probability of wind gusts above 40 mph, and ensemble member low locations. Plot from PolarWx.

One tool that can help us better assess forecast uncertainty is ensemble guidance. Simply stated, ensembles are obtained by taking a forecast and adding small perturbations to its initial conditions and other parameters (e.g., physics tendencies) to account for uncertainty. This yields a set of simulations that start out with small differences that grow larger with time, which should be indicative of the range of plausible solutions.

Looking at the GEFS forecast above, we see a fairly large number of ensemble members showing a weak tropical cyclone in the western Caribbean, with a small number of members showing a more intense tropical cyclone. At later lead times (not shown), just under half of GEFS members develop a hurricane over the Gulf of Mexico, while the remaining members either have a weak tropical cyclone or no tropical cyclone at all.

To better understand the GEFS forecast, we also need to understand its systematic biases. For one, the GEFS are known to have an under-dispersive bias, meaning that the ensemble spread is often too small, resulting in the true solution falling outside of the ensemble distribution. Think of this as a forecast showing the range of possible high temperature between 68 and 74 degrees, but verification is 78 degrees. The relatively recent upgrade to the GEFSv12 with an increased number of ensemble members (from 20 to 30) improved this bias, but nonetheless some under-dispersiveness remains.

Comparison of EPS and GEFS ensemble member low positions. Plot from Pivotal Weather.

The GEFS are just one set of ensemble guidance; The ECMWF Ensembles (EPS) provide a set of 50 ensemble members, and are generally more dispersive than the GFS. In this case, notice that the EPS have a weaker and farther south/west cyclone than the GEFS, with only a handful of members showing a stronger and farther north tropical cyclone like the GFS and a subset of the GEFS members. This lack of multi-model support also aids in lowering confidence for the GFS solution of a Gulf of Mexico hurricane.

Additionally, it should be noted the currently operational GFS has a higher false alarm ratio (FAR) than the previous version of the GFS, with one hotspot of FAR over the western Caribbean – the region where the GFS and GEFS show this tropical cyclone developing.

EPS ensemble mean QPF (mm), valid through day 10.

As with every weather forecast, especially at longer lead times, it's impossible to make definitive statements this far out. There is always the possibility of a highly anomalous event occurring, such as a May hurricane in the Gulf of Mexico. Nonetheless, considering both climatological factors, systematic biases in the GFS and GEFS, and lack of multi-model ensemble support, this scenario appears unlikely to occur.

A more realistic scenario I could speculate on would be a weaker and broader tropical cyclone that forms near the western Caribbean but does not become a full-fledged hurricane in the Gulf of Mexico. Regardless, even weak tropical cyclones – or even tropical waves – can produce catastrophic rainfall in Central America, and both GEFS and EPS ensemble means show upwards of 5 inches of rain in parts of Central America. For this reason alone, this requires monitoring regardless of tropical development.

In any case, as today marks the start of NHC Tropical Weather Outlooks, and the latest outlook does not anticipate any tropical cyclone development over the next 5 days. Should one become more likely to develop, stay tuned to NHC's outlooks for additional information.

Share this post: