It’s time to reflect upon what proved a peculiar yet not entirely unpredictable hurricane season in the Atlantic basin. Despite the challenges, MetSwift’s predictions offered a 31% improvement in accuracy over using an average of the past four decades (1981-2020).
This is not unexpected based on analysis of predictions made for the seasons of 2010-2020. On average, MetSwift’s predictions released in Dec-Mar are 9-19% more accurate than climatology.
In fact, MetSwift’s February issued predictions are typically 19% closer to the observed tropical storm count than the leading April or May prediction issued by other organisations.
Raised Likelihood of Major Hurricanes Like Grace & Ida
One standout success has been the prediction that two major hurricane landfalls could occur in the Gulf of Mexico during the month of August, including one striking the northern coast in the final week of the month. That week saw the highest insurance claims of the season.
Activity Summary: Roaring Peak Whimpering Retreat
The season kicked off with a flurry of mainly weak pre-season tropical cyclones, most of them in June. For Apr-Jul, the number of tropical storms was well above normal, while the number of hurricanes was typical recent decades.
Peak season was then extremely active, up there with the busiest years on record. Yet all but one of the 16 named storms and all the hurricanes occurred in Aug-Sep; Oct was unusually quiet. Extraordinarily, this early drop-off has occurred during a weak La Niña event, which usually supports a busy Oct.
Post-season has been silent, without a single named storm, let alone hurricane. The 2021 Atlantic hurricane season led with a growl, peaked with a roar, but left with a whimper.
Assessment of MetSwift’s Outlook Released 14th April 2021
MetSwift’s outlook for this hurricane season presented predictions from our in-house analogues-driven model that were produced in the middle of each month Dec 2020 through Mar 2021.
Below, these are presented again, but with predictions in mid-Apr and mid-May added, followed by the all-important observed numbers.
Bold numbers indicate above-average activity.
Based on the Dec-Mar model output, MetSwift predicted a near or below average pre-peak, a very busy peak, then slightly more post-peak activity than usual.
The pre-peak and post-peak predictions have been a little wide of the mark. However, the peak one (arguably most important) has proved a good call – it was indeed very active.
An active season as a whole was predicted, in line with most other organisations currently issuing publicly visible forecasts. The one exception was UKMO, which predicted a near-average season. Them aside, there’s very little to set the organisations apart from each other or MetSwift. TWC’s May forecast was closest for named storms, but this also had the shortest lead time.
A Fitful Season
A monthly breakdown illustrates the stuttering nature of the season; very focused in Jun and Aug-Sep:
This manner of waxing and waxing is typical of the Atlantic hurricane season, as part of sub-seasonal ‘tropical cycles’. They serve to make conditions more or less favourable for tropical cyclone development.
That’s why I’ve chosen to lead with the ‘stage of the season’ approach. By aggregating across 3-4 months at time, the overall level of activity can be better assessed, while retaining some sub-seasonal detail.
Spatial Distribution of the Hurricanes
Predicting named storm or hurricane counts is good, but ideally, we need to anticipate where they will tend to go. A very busy open Atlantic is far less of a concern than a highly active Caribbean, for example.
Generally, Caribbean and Gulf of Mexico activity is of greatest concern, as storms here are surrounded by land areas. It’s hard for them not to make landfall somewhere! So, we’ll start with those two regions.
The Caribbean Sea
For the Caribbean, MetSwift’s prediction was the right side of climatology for the active peak season, but the wrong side for the quiet post-peak; again, the early shutdown was unusual for a La Niña season.
The Gulf of Mexico
For the Gulf of Mexico, the model did well with pre-season but was no closer than climatology for the peak season.
The Main Development Region (Between Caribbean & Africa)
Tropical cyclones in the Main Development Region pose less immediate threat to land but can become some of the strongest recorded. Most years, the majority stay out over the North Atlantic, while the rest head either west or northwest and can landfall anywhere from the Lesser Antilles to eastern Canada.
MetSwift’s predictions were closer than climatology for pre-peak and much closer for the hyperactive peak season. They were on par for post-peak, but the season is nearly always over by then for this region.
The Subtropical North Atlantic
The subtropical Atlantic is of least interest from a risk perspective, as most storms here stay out at sea, but some have been known to head west and strike the East US Coast. May-Jun and Aug-Sep, this region was exceptionally active, to an extent not picked up on by the model.
This sort of behaviour fits the climate trend, as the subtropics become warmer and more supportive of tropical cyclones. This will be considered for future Atlantic Hurricane Season forecasts.
Subsequent Improvement to MetSwift’s Model
Shortly after the prediction was published, it was discovered that the advanced statistical model’s predictions were being biased downward by an artificial trend in the historical observation data.
This is down to improvements in the detection of weak tropical cyclones that only just reach the threshold required to be named. These days, hardly any are overlooked, but go back even a few decades and it’s likely that 1-3 were being missed each season. This year, there were 7 ‘only just’ storms that might once have gone without a name.
When accounting for this ‘artificial quietening’ of past seasons, MetSwift’s model has demonstrated further-increased skill when predicting the seasons of 2010-2020.
The table below shows how the upgrade affects the predictions made Dec-Mar 2021 (the Apr-May ones shown previously were already using the upgrade).
Interestingly, most of the impact on predictions is for the pre-season, with near to above average activity suggested. MetSwift’s predictions for peak and post-peak would not have been altered.
James Peacock MSc
Head Meteorologist at MetSwift