Last Friday marked the end of the official hurricane season for both the North Atlantic (NA) and Eastern Pacific (EP) basins, providing a great opportunity to review how our first tropical cyclone activity forecast performed.
Before we dive into the details of each basin/intensity category here is a quick summary of what our projections looked like back in June, along with the activity to date. Note both the Western Pacific (WP) and Northern Indian (NI) seasons are still active.
*Season still active.
- The activity recorded in the North Atlantic (NA) is within our projected range for all categories, and the number of observed Cat 0 storms (15) is an almost perfect match for our mean projection (14.75)
- For both the Eastern Pacific (EP) and Western Pacific (WP) the Cat 0 and Cat 1 activity did fall within our predicted ranges and our forecasts underestimated the number of Cat 3 storms
- Our forecast for the North Indian (NI) ocean appears to have underestimated the level of activity in the basin for all categories
- Our ranges are designed to represent a modelled probability of occurrence around 60-70%. In other words we would expect around two thirds of our range forecasts to verify. Bar a very active end to both the WP and NI seasons this level of confidence from our model should be confirmed with 7-9 out of our 12 ranges likely to verify
Named storms (Cat 0)
North Atlantic
Let’s start with the basin that receives the most attention. In terms of named storm activity, and despite a very quiet month of August, 2018 will go down on record as an above average season in the NA with 15 storms reaching the Cat 0 intensity threshold.
The figure below shows the activity recorded during the season (in red), along with the average seasonal activity (purple), our June predictions (black line / coloured bins) and a summary of forecasts issued by other agencies as they were reported to the BSC/AXA-XL model inter-comparison project (in white: the average forecast is shown by a solid line and the dashed lines indicate the average prediction ranges for the May-June release).
Our projections in June were towards an above average season with our modelled distribution (colour bins above) giving a 70% chance of a season with 13 or more named storms and 67% chance of Cat 0 activity falling within the [11-18] range. Both these outcomes verified with the final tally of 15 very close to our model mean estimate of 14.75. Interestingly, and as discussed at the time in this post, this view was not shared by the majority of our peers with most agencies calling for an average season.
Eastern Pacific
With 23 named storms recorded the activity in the Eastern Pacific has been well above average, and while this reaches the higher end of our expectations (see colour bins below) this level of activity is consistent with the trend our model projected back in June.
Western Pacific
Next on the list is the Western North Pacific, and although the season has not yet reached its conclusion, the following figure provides good evidence that our forecast is here again in the right zone (above average activity):
North Indian
Finally, our projections for the North Indian basin were for a season producing only an average number of named storms. Given current counts this is shaping up to be an underestimation of 2018 activity.
Hurricanes (Cat 1)
Our June forecast for Cat 1 storms in all 4 basins are summarized below (box plots and violins representing the quartiles and our full distribution, respectively), along with the season counts to date (dots) and the climatology (crosses). All 4 basins are registering Cat 1 activity levels above the long term climatology. For all but the NI ocean this is in line with our model projections albeit in the upper range of our expectations for the North Atlantic (and Eastern Pacific).
Major hurricanes (Cat 3)
Our projections for the most intense storms (Cat 3 and above) did not foresee the high level of activity recorded in both the EP and WP basins. The EP in particular has been a near record-breaking season in terms of intense storms and registers in the tail of our modelled distribution (but well within the range of possible modelled outcomes).
New Functionality
For next season we have decided to upgrade our methodology using two new modules recently developed:
- a genesis location probability forecast (see map below)
- a full track/intensity model conditioned on our set of climate predictors
This new method, to be plugged on the back of our Cat 0 genesis module, will replace the current Cat 1 and Cat 3 models – and we are eager to see how that will perform in 2019.
Conclusion
Our model’s performance in its first outing is very encouraging to say the least. In particular projections for the level of Cat 0 activity have been very reliable and we take pride in having registered one of the most accurate forecast among the 24 agencies participating in the BSC/AXA-XL project.
We are seeking industry partners to be involved in our 2019 season forecast – please get in touch if interested.