Tropical Cyclones: Cat 4+ Counts and the Annoying 1990 Boundary

In Recent Trends in Tropical Cyclones last year we looked at the data for 1990-2024 from IBTrACS and compared this to a paper, Trends in Global Tropical Cyclone Activity: 1990–2021, Philip Klotzbach et al (2022).

In a few subsequent articles I tried to tease out some others trends and didn’t really succeed.

There were a two main points in Klotzbach’s paper that made me want to dig:

1. We worked from 1990 because the global satellites prior to 1990 didn’t have proper visibility of the Indian Ocean

2. Observing changes between 1980-1990 made trend data before 1990 challenging

When someone screens out data it makes me want to dig a little deeper. Is that screening really justified? If we include that data what do we find?

My question was simple:

Are there more very intense tropical cyclones now than in the 1980s?

By “very intense” I mostly mean Category 4 and 5 storms. Cat 5s are rare, so Cat 4+ is a better metric if we want enough storms to say anything useful.

Here’s the plain English version:

Weak, short-lived tropical cyclones are a major problem for trend analysis because modern instruments are much better at finding them.

Cat 4+ storms are different. They were not invisible in 1980. But their exact intensity estimates still changed with satellite capability, the Dvorak technique, agency practice, and aircraft reconnaissance.

If we exclude the Indian Ocean and start in 1980, Cat 4+ storms appear to increase.

But if we allow a step change around 1990, the continuing trend disappears.

That last point is the main finding.

The key distinction is this: weak storms have a detection problem; Cat 4+ storms have an intensity-classification problem.

A weak, short-lived storm might not have been noticed or named in 1980. A Cat 4+ storm almost certainly was noticed. But the question is whether its peak wind would have been analysed the same way: 105 kt, 110 kt, 115 kt, 125 kt? That matters because Cat 4 begins at about 115 kt in the best-track data. A storm does not need to vanish from the record to change the trend; it only needs to move across a category boundary.

Klotzbach’s 1990 boundary

Klotzbach et al. used 1990–2021. Their reason was that the global observing system before 1990 is less clean.

They write that before 1990, global TC intensity estimates are likely compromised by data issues. In particular, the North and South Indian Oceans lacked direct geostationary satellite data until Meteosat-5 in 1989. They also note a possible artificial jump in eastern North Pacific intensities after forecast responsibility shifted to NHC following the 1987 season.

That made me wonder:

What if the Indian Ocean is the main problem?
What if we simply remove the Indian Ocean and start in 1980?

This seems reasonable at first. The Indian Ocean is not the largest contributor to global TC activity. If its pre-1990 record is the problem, perhaps excluding it gives us a longer usable record.

Here’s the graph from 1980, by category of TC, excluding the Indian Ocean:

Klotzbach et al. found that short-lived named storms increased significantly from 1990–2021. They explicitly say this is likely due to improved sensors. Over the same period, global hurricane counts and ACE decreased, mainly from lower western North Pacific activity.

The key distinction is this:

For weak storms, the question is often: did we even know the storm existed?

For strong storms, the question is: did we estimate its peak intensity the same way?

Those are different problems.

A detour into ADT-HURSAT

I spent some time looking at ADT-HURSAT because it looks like the kind of dataset we might want.

IBTrACS is best track. ADT-HURSAT is more like a homogeneous satellite-only intensity reconstruction. That sounds attractive. If the issue is changing agency practice, maybe ADT-HURSAT solves the problem.

Mission failed.

That’s a little unfair. ADT-HURSAT is useful for some questions. But it is not useful for the question I had in mind: absolute counts of Cat 1, Cat 2, Cat 3, Cat 4, and Cat 5 storms comparable to IBTrACS.

HURSAT is a storm-centered satellite dataset. ADT-HURSAT applies the Advanced Dvorak Technique to HURSAT infrared imagery. Importantly, the HURSAT history files use IBTrACS to initially identify storms. So ADT-HURSAT is not independently finding missing historical storms. It is estimating intensities for storms already in the storm-centered archive. NOAA’s documentation also says that ADT-HURSAT should not be used to determine actual storm intensities, because individual peaks and valleys are not accurately represented with the reduced spatiotemporal sampling.

In my comparison, ADT-HURSAT produced far fewer Cat 1–5 storms than IBTrACS.

The explanation is probably simple. IBTrACS best tracks can use satellite imagery, Dvorak analysis, aircraft reconnaissance, scatterometers, microwave data, radar, buoys, ships, and post-storm judgment. ADT-HURSAT is a homogeneous satellite-IR algorithm. It is more consistent, but it throws away a lot of information. It also smooths and undersamples short-lived peaks.

So I discarded ADT-HURSAT for this article. Not because it is bad, but because it is not the right tool for this question.

The ≤2-day storm problem

Klotzbach et al. split named storms into storms lasting ≤2 days and >2 days. This is a very useful test.

A short-lived weak storm is exactly the kind of thing that modern instruments help detect. Before scatterometers and better microwave data, an analyst might see a cloud swirl over the ocean but not have enough evidence for a named tropical storm. With modern instruments, one good pass can show a closed circulation and tropical-storm-force winds.

QuikSCAT launched in 1999 and provided daily ocean surface wind coverage over most ice-free oceans for more than a decade. ASCAT on Metop-A launched in 2006 and continued the ocean surface wind vector capability. These instruments do not measure wind directly in the way an anemometer does; they infer near-surface winds from radar backscatter off the roughened ocean surface. But they are incredibly useful for detecting 34-kt winds in weak and marginal systems.

This does not mean those storms are fake. It means similar storms in the past may not have made it into the archive.

Landsea et al. made this point for the Atlantic. They found that the increase in Atlantic TC counts was dominated by very short-lived storms, while medium- to long-lived storms increased little, if at all. They also note that recent very short-lived systems entered the record partly because of new tools such as QuikSCAT, ASCAT, AMSU, ADT, GPS dropsondes, buoys, and cyclone-phase-space analysis.

So far, so good.

But this short-lived-storm problem is mostly a weak-storm problem. Here’s the “2 day and above” graph:

And here’s the “less than 2 day TC” graph:

IBTrACS category counts for storms with hurricane-strength duration ≤2 days, excluding the Indian Ocean. Short duration matters greatly for Cat 1 and Cat 2, somewhat for Cat 3, and hardly at all for Cat 4+. Cat 5 is essentially absent.

This graph shows what we would expect. The ≤2-day category is dominated by Cat 1 storms, with some Cat 2 and a few Cat 3. Cat 4 is rare. Cat 5 is basically absent.

That means we should not use the ≤2-day exclusion to dismiss Cat 4+ counts. Cat 4+ storms were not invisible. A Category 4 or Category 5 tropical cyclone in 1982 was going to be seen.

But that doesn’t mean the Cat 4+ record is problem-free.

The strong-storm problem is different

For Cat 4+ storms, the issue is not whether the storm existed. The issue is the assigned maximum wind speed.

A storm estimated at 105 kt is Cat 3.
A storm estimated at 115 kt is Cat 4.

Same storm, different category.

Most non-Atlantic basin intensity estimates depend heavily on satellite-based Dvorak analysis. The Dvorak technique is extremely important, but it does not directly measure wind or pressure. It estimates intensity from satellite-observed cloud patterns, eye temperature, cloud-top temperature, and related features. Velden et al. note that Dvorak estimates are often good, but they are not perfect: in one Atlantic validation, half the maximum-wind estimates were within 5 kt, 75% within 12 kt, and 90% within 18 kt of reconnaissance-aided best tracks. That is easily enough uncertainty to move a storm across the Cat 3/Cat 4 boundary.

This is especially important around 1990.

Klotzbach and Landsea reviewed the old claim that global Cat 4–5 storms had roughly doubled from 1970–2004. They argued that much of that increase was due to observational improvements in the 1970s and 1980s. Satellite imagery improved dramatically. The Dvorak technique became more widely used. The eastern Pacific had an agency transition after 1987. Western North Pacific aircraft reconnaissance ended in 1987. And there was no standardized global intensity database.

Their recommendation was very direct: global studies of extreme hurricanes should begin around 1990 because before that the record is incomplete and can give a distorted view of activity.

This is the context for the next graph.

Cat 4+ storms excluding the Indian Ocean

First, I removed the North and South Indian Ocean basins. This was the test:

If the Indian Ocean is the big pre-1990 problem, what happens if we take it out and start in 1980?

IBTrACS Cat4+ annual counts excluding the Indian Ocean. Starting in 1980 gives a weakly significant positive trend. Starting in 1990 gives no trend.

The result is interesting.

Excluding the Indian Ocean, 1980–2025 gives:

+1.12 Cat4+ storms per decade
95% CI: +0.03 to +2.20
p = 0.043

That looks like a positive trend.

But using the same basin exclusion and starting in 1990 gives:

−0.08 Cat4+ storms per decade
95% CI: −1.72 to +1.55
p = 0.920

That is nothing. No trend at all.

This is where a straight-line trend can mislead. If the 1980s are low and the post-1990 period is higher, a straight line from 1980 to 2025 will slope upward. But that doesn’t mean there has been a steady rise.

Maybe the data are telling us this:

Low before 1990.
Higher after 1990.
Flat after that.

That is a step, not a trend.

Step change or continuing trend?

To test that idea, I used two simple models.

The first model is just a straight line: do counts go up or down over time?

The second model allows a jump around 1990. Then it asks: after allowing the jump, is there still a continuing trend?

I also used a Poisson model, because these are counts. Counts are not quite like temperature or pressure measurements. They are whole numbers: 7 storms, 12 storms, 21 storms. The Poisson model is a standard way to check this kind of data.

Step-versus-trend models for Cat4+ counts excluding the Indian Ocean. A simple trend line slopes upward, but the step-plus-trend model shows a pre/post-1990 level change with essentially no continuing trend.]

The result is clear.

The straight-line trend says there is an increase. But once the model is allowed to include a pre/post-1990 step, the continuing trend disappears.

The step is large:

OLS step: +4.75 Cat4+ storms per year after 1990
Poisson step ratio: 1.51
In plain English: the post-1990 level is about 51% higher than the pre-1990 level.

But the continuing trend is basically zero:

OLS continuing trend: +0.063 Cat4+ storms per decade, p = 0.933
Poisson continuing trend ratio: 1.004 per decade, p = 0.916

That is not a trend. That is noise.

A simple analogy: imagine your salary is flat for ten years, then jumps once, then stays flat. If you draw one straight line through the whole period, it slopes upward. But the real story is not “steady salary growth.” The real story is “one raise, then flat.”

That appears to be what is happening here.

What caused the 1990 step?

The graph and the statistics do not prove the cause.

But the timing is suspicious.

The late 1980s and early 1990s are exactly when several known changes matter for TC intensity analysis:

• the eastern North Pacific responsibility shift after 1987;

• the end of western North Pacific aircraft reconnaissance in 1987;

• wider and more mature use of Dvorak analysis;

• better satellite imagery and interpretation tools.

Even after excluding the Indian Ocean, the eastern North Pacific and western North Pacific remain in the dataset. So removing the Indian Ocean solves one problem, but not all of them.

The western North Pacific is especially important because it contributes a large share of global intense storms. Kamahori et al. compared JMA and JTWC data and found that after the end of US Air Force reconnaissance in 1987, the two datasets diverged in their treatment of extremely intense tropical cyclones. JTWC showed a large increase in extremely intense TC days, while JMA showed a decrease. They connected this divergence to the post-reconnaissance reliance on Dvorak-based estimates.

Again, this does not prove that every post-1990 Cat 4+ storm is overestimated. It says something more modest:

The 1980s and post-1990 records are probably not homogeneous enough to treat the 1980–2025 slope as a clean climate trend.

Where I ended up

I began with the thought that Klotzbach et al. might have been cherry picking by starting in 1990. Maybe the Indian Ocean was the main reason, and removing it would let us go back to 1980.

But the data argue against that.

The apparent 1980–2025 increase in Cat 4+ storms, excluding the Indian Ocean, is mostly a low-1980s versus higher-post-1990 step. Once we allow that step, there is no continuing trend.

This is the cleanest summary I can make:

For weak named storms, modern observing systems create a serious detection problem, especially for storms lasting ≤2 days.

For Cat 4+ storms, the detection problem is much smaller. The storms were seen. But the intensity classification problem remains.

In IBTrACS USA_WIND, excluding the Indian Ocean, Cat 4+ counts are higher after 1990 than before 1990. But since 1990, there is no trend in Cat 4+ counts in this construction.

That does not mean intense TCs cannot be affected by climate change. It does not address future projections. It does not address rainfall, storm surge, rapid intensification, or landfall risk.

It only says that this particular observed-count question gives a more cautious answer than the simple 1980–2025 trend line suggests.

Conclusion

The 1990 boundary is annoying. But it seems justified.

Short-lived weak storms have a clear observing-system problem. Modern scatterometers and microwave instruments make weak, short-lived storms easier to detect and classify. That explains why raw named-storm counts are dangerous.

Cat 4+ storms are different. They are not being newly discovered by QuikSCAT. But their peak intensities are still sensitive to satellite resolution, Dvorak interpretation, agency practice, and the loss or presence of aircraft reconnaissance. A storm does not need to disappear from the archive to change the trend; it only needs to move from 105 kt to 115 kt.

So the answer to “are there more Cat 4+ storms than in the 1980s?” is:

In this dataset, yes, the post-1990 level is higher than the 1980s.

But the better interpretation is a pre/post-1990 step change, not a continuing increase.

Since 1990, the Cat 4+ count trend is essentially zero.

That’s not as exciting as a headline. But it is probably closer to the data.

References

Klotzbach, P. J., Wood, K. M., Schreck III, C. J., Bowen, S. G., Patricola, C. M., & Bell, M. M. (2022), Trends in Global Tropical Cyclone Activity: 1990–2021, Geophysical Research Letters.

Klotzbach, P. J. & Landsea, C. W. (2015), Extremely Intense Hurricanes: Revisiting Webster et al. (2005) after 10 Years, Journal of Climate.

Landsea, C. W. et al. (2010), Impact of Duration Thresholds on Atlantic Tropical Cyclone Counts, Journal of Climate.

Velden, C. et al. (2006), The Dvorak Tropical Cyclone Intensity Estimation Technique: A Satellite-Based Method that Has Endured for over 30 Years, Bulletin of the American Meteorological Society.

Kamahori, H., Yamazaki, N., Mannoji, N., & Takahashi, K. (2006), Variability in Intense Tropical Cyclone Days in the Western North Pacific, SOLA.

NOAA/NCEI, International Best Track Archive for Climate Stewardship and ADT-HURSAT dataset documentation.

Notes

The data used here are from IBTrACS — the International Best Track Archive for Climate Stewardship. IBTrACS combines tropical cyclone best-track data from many agencies into one archive. It is probably the best place to start for global historical TC work, but it is not one uniform observing system. It is a compilation of agency best tracks. Different agencies, different basins, different observing systems, and changing methods all matter. NOAA describes IBTrACS as a merged global best-track archive, while UCAR’s Climate Data Guide notes that changing operational procedures and observing systems have produced significant heterogeneities in the record.

For these graphs I used the US-agency winds, USA_WIND. In the Atlantic and eastern Pacific this is mostly the NHC/CPHC side of the record; elsewhere it is generally the JTWC side. Each storm is counted once, by its peak category. Cat 4+ means Cat 4 plus Cat 5.