Are Hurricanes Getting Worse?

To scientists who study them, there are two mysteries surrounding hurricanes that stand above the rest: Why do they exist at all, and why aren’t there many more of them? This may strike you as a paradox, but these are serious questions that arise when burrowing deep into the theory, modeling, and observations of these storms. And they bear on the question posed by the title of this essay.

What powers hurricanes

We know what powers hurricanes. Tropical oceans continually absorb sunlight, and to balance this energy source they must find a way to shed energy. Were it not for greenhouse gases, they would simply radiate heat to space in the form of infrared radiation. But the greenhouse gases in our atmosphere re-radiate the infrared radiation back down to the surface, so much so that on average the surface gets more than twice as much radiant energy from the atmosphere (and clouds within it) as directly from the sun. To balance all this energy input, the oceans shed heat by evaporating water into the atmosphere. This cools them just as water evaporating from your skin cools you after you get out of a swimming pool. To permit this evaporation, the atmosphere just above the ocean must have a relative humidity less than 100%; in fact, it is closer to 80%. This difference, or ‘evaporative potential’, is what powers hurricanes. As a tropical disturbance forms over warm oceans, increasing winds locally increase the rate of evaporation, putting more heat into the atmosphere when the water vapor condenses into rain. If conditions are right, this feedback increases until the heat energy added to the hurricane heat engine is balanced by frictional dissipation of wind energy as air blows across the sea surface.

The simple theory of heat engines allows us to calculate rather precisely how strong the winds can be in a hurricane if we know the evaporative potential and some information about how deep the convection can be. This thermodynamic wind speed limit is known as the potential intensity. It is large across much of the tropics during most of the year.

So if all that fuel is there, why aren’t hurricanes ubiquitous?

In fact, hurricanes are ubiquitous in regional models of the tropical atmosphere that have uniform sea surface temperature and no large-scale atmospheric circulation. We have a very good theoretical understanding of what controls the intensity, size, and spacing of hurricanes in such models. Then why don’t we see hurricanes throughout the tropics, all the time, in the real world?

One reason is that the tropical atmosphere is often quite dry above 1 or 2 km. As convective storms develop, rain evaporates into the dry air, causing downdrafts that bring the dry air down into the boundary layer, countering the effect of surface evaporation, like quenching a fire with buckets of cold water. To start a hurricane, it is first necessary to create a column of air with nearly 100% humidity to stop these downdrafts from forming. This is not easy, especially if the ambient winds vary with altitude, which disrupts columns of humid air. This wind shear is known to be an important factor preventing many disturbances from becoming hurricanes.

Climate change is… changing things

With this crash course in hurricane physics, we can start thinking about how climate change might affect hurricanes. First, putting more greenhouse gas into the atmosphere should increase the evaporative potential, and thus hurricane potential intensity, since the oceans receive more radiation from the atmosphere. Potential intensity should go up with global warming, and this is indeed both observed in nature and predicted by climate models. We can expect to see at least a few hurricanes that are more intense than any in the record books, and indeed we keep breaking hurricane intensity records. Typhoon Haiyan of 2013 set the all-time record for hurricane winds speed with 195 MPH winds, a record broken only two years later by eastern Pacific Hurricane Patricia, with winds of 215 MPH. Atlantic Hurricane Irma of 2017 broke all records for continuous duration at Category 5. While these are anecdotes, careful analysis of satellite records of global hurricane activity show that the proportion of high intensity storms is increasing.

But as the tropical atmosphere warms, it also becomes drier above the boundary layer, making it harder to form hurricanes in the first place. Wind shear also changes, although climate models disagree with one another on whether the shear will become stronger or weaker.

With hurricanes, as with lawn mowers, bigger engines are harder to start. So we can say that we should see more powerful, but not necessarily more frequent storms. The competing influences on hurricane initiation lead scientists to express low confidence in predictions of how the number of hurricanes might change in response to global warming.

What global models don’t tell us

Unfortunately, the main tool we have for modeling climate change, the global, coupled, general circulation model (GCM), is not up to the task of predicting hurricanes.  The computational nodes of today’s best climate models are separated by a few tens of kilometers, whereas simulations with specialized hurricane models show that the nodes must be separated by no more than a few kilometers to adequately simulate hurricanes. GCMs do produce hurricane-like storms, but their size and intensity are distorted by the poor spatial resolution. Nevertheless, such models have been widely used to project how global warming will affect hurricanes, and most show a declining frequency of storms but with higher wind speeds and more rainfall.

One way to get around the resolution problem is to embed within the GCM a regional model (e.g. covering just the tropical North Atlantic) with much higher spatial resolution. This has been done, but generally for only a few of the hurricane-producing regions of the world, and for only a few models, because they are time-consuming to create and expensive to run.

Is there a better way to solve the resolution problem?

Back in 2005, I started to wonder if there might be a better way around the resolution problem. I had already developed a specialized, coupled ocean-atmosphere numerical model for simulating hurricane intensity along specified tracks through a known large-scale environment. Instead of using distance from the center of the storm as a model coordinate, it uses a quantity proportional to the square root of the absolute angular momentum with respect to the storm center. Among many advantages of this model coordinate is that it yields high spatial resolution in the storm core, where it is needed, and the resolution gets better as the storm intensifies. But the model assumes axisymmetry, and therefore interactions between the simulated storms and wind shear in their environment must be represented parametrically. And yet when run along the official forecast tracks of hurricanes, in real time, it produces intensity forecasts that are competitive with much fancier models[1]. The storms are realistic, even having secondary eyewall replacement cycles.

Could we somehow use this model to estimate the hurricane climatology of GCMs?  To do this, we would first have to generate hurricane tracks to feed into the intensity model. At first, we did so by randomly drawing from the observed space-time distribution of historical storm genesis points, and then creating tracks by assuming that the storms move (as real storms do) with the large-scale airflow, in this case simulated by the GCM. This seemed to work pretty well, but there was no reason to assume that this genesis distribution wouldn’t change as the climate evolved. I wanted the model itself to figure out where hurricanes ought to form.

 Attacking the genesis problem

Perhaps as an act of desperation, I hit upon the idea of just starting hurricanes as weak vortices located randomly in space and time (not constrained by any historical distribution), and let the intensity model figure out whether they would survive or not. When I tried it, not surprisingly, the intensity model threw out well over 99% of the trial storms. What was surprising was that the surviving storms formed a truly excellent hurricane climatology when driven by global climate analyses; this includes the global distribution of storms, their seasonal cycles, and even their response to El Niño events. We call this method the random seeding and natural selection approach, drawing an analogy with biological natural selection. We now have a technique we can apply to global climate models without having to use historical hurricane data except to test how well it does. And it is fast, very fast. Generation of a single hurricane takes just a few seconds on an ordinary laptop, so one can in practice generate tens of thousands of synthetic hurricane events.

So we went ahead and applied this new approach to seven models (selected on the basis of the availability of their output) run in support of the IPCC 4th Assessment Report (AR4), for the periods 1981-2000 from simulations of the historical climate, and for 2181-2200 for simulations of a future climate under the relatively moderate greenhouse gas emissions scenario A1b. The results, published here, show almost no change in the global frequency of hurricanes, but with strong variations from region to region and among the different GCMs used. Hurricane intensity mostly increases, except in the southern hemisphere. We repeated this exercise five years later using six AR5-generation climate models, this time finding a marked increase in both global hurricane frequency and intensity, including in the South Indian Ocean, which had shown decreasing activity in the AR4 downscaling. The emission scenario, called RCP 8.5 was more aggressive than what we had used for the AR4 models, but we looked at the whole period 2006-2100 rather than 2181-2200. Now we have just published a study downscaling nine climate models run in support of the most recent assessment, AR6, but this time we focused on the sensitivity of hurricane activity to CO2, using a scenario in which emissions increase at a rate of 1% yr. (The current rate of increase is about 0.6% yr. Note: Real time hurricane intensity forecasts using this model are posted at http://wind.mit.edu/~emanuel/storm.html) 

The results are more similar to our AR5-based study than to the first study we did, based on AR4: We found a global increase in hurricane frequency of about 25% per CO2 doubling, and a roughly 30% increase in hurricane power dissipation, a measure of the rate at which hurricanes generate and dissipate wind energy. As with the earlier studies, most of the increase is in the northern hemisphere, but the locus of the increase shifts from the far western North Pacific to the central North Pacific region. It is notable that none of the three studies we undertook shows any discernible change in any measure of hurricane activity in the South Pacific. 

Figure 1 shows maps of the percentage change in genesis frequency, track density, and power dissipation in response to doubling CO2. Most of the increase in hurricane genesis occurs on the poleward margins of the current genesis zones in the northern hemisphere; there is not much change in the centers of the current genesis zones, e.g., the tropical North Atlantic. The largest increases in the occurrence of hurricane tracks are in the central North Pacific and in northern region of the North Atlantic. The increase in genesis off the southeastern U.S. coast seems to lead to more cases of extratropical transition of hurricanes, whereby they transform themselves into extratropical cyclones (like Hurricane Sandy did in 2012).

Figure 1: Maps of the percentage change of genesis density (left), track density (center) and power dissipation (right) in response to doubling CO2. Changes are only displayed where at least 7 of the 9 GCMs used agree on the sign of the change.

All of the main hurricane zones of the northern hemisphere see increasing activity, but as before, there is little projected change in the southern hemisphere, except an indication of more extratropical transition of storms in the South Indian Ocean.

Insofar as this approach yields increasing hurricane intensity and (although I have not discussed it here) greatly increasing hurricane rainfall, it is consistent with the broad consensus of studies mostly based on explicit simulations of hurricanes in GCMs. But in predicting increased overall frequency of events (using the AR5 and AR6 models, though not the AR4 models) it contradicts the majority (22 out of 27) of modeling studies that project decreasing frequency. Why is that? Some scientists have suggested that our random seeding approach does not account for a possible decrease in disturbances that serve as seeds for hurricanes. This may be the case, but that limitation does not seem to compromise our ability to simulate the observed seasonal cycles or response to natural variations (e.g. El Niño) in the current climate. Another possibility has to do with the way hurricanes are detected and counted in GCMs. For example, to be counted as a hurricane in most cases, the vorticity of a disturbance must exceed some preset threshold. But as the climate warms, disturbances tend to grow larger in size, so that for a given wind speed, their vorticity decreases, fooling the detection algorithm into thinking that hurricane frequency is declining.

So… are hurricanes getting worse?

While more work needs to be done to nail down the issue of how overall hurricane frequency should respond to global warming, there is a fairly strong, emerging consensus that the frequency of strong hurricanes (Cat 4-5) should increase, as will hurricane rainfall. Since the great majority of damage is done by flooding from heavy rain and from storm surges accompanying strong hurricanes, this consensus has strong societal implications, while what happens to weaker storms, which dominate the overall frequency, is less consequential.

All things considered, the best current answer to the question posed by the title of this essay is “yes”.

Dr. Kerry Emanuel is the Cecil and Ida Green professor of atmospheric science at the Massachusetts Institute of Technology (MIT), where he has been on the faculty since 1981, after spending three years on the faculty of University of California, Los Angeles (UCLA). His specialty is hurricane physics and he was the first to investigate how long-term climate change might affect hurricane activity, an issue that continues to occupy him today. His interests also include cumulus convection, and advanced methods of sampling the atmosphere in aid of numerical weather prediction.

Dr. Emanuel is the author or co-author of over 200 peer-reviewed scientific papers, and three books, including Divine Wind: The History and Science of Hurricanes, What We Know about Climate Change, and Climate Science and Climate Risk: A Primer.