Dec 25 2021
Atlantic hurricanes hurt people and destroy property around the Gulf of Mexico every year. Whether climate change is increasing their frequency is a serious question. Don Wheeler just had a column on this subject in the latest Quality Digest. It’s about Torturing the Data. He argues that we should be careful about not force-fitting models to arrive at pre-ordained conclusions.
His way of not torturing the history of hurricanes in the Atlantic is plotting yearly counts on, what else, an XmR chart. It’s just as he would for hole diameters in metal plates coming off a production line in 1945. Hurricanes and holes in metal plates, however, have different backstories.
The first step in showing respect for this awesome natural phenomenon is to work with raw data, not summaries like the number of years with a given number of hurricanes of category 3 or higher.
Starting from the NOAA database, we first visualize the population of hurricanes as objects that start off the coast of Senegal, ravage the Caribbean and the Eastern United States before dying in the North Atlantic in numbers rising over recent decades. Then we look at the seasonality of hurricane occurrences to characterize its pattern. It doesn’t appear to have changed over the decades.
Finally, as each hurricane season is contained with the year, you can treat counts per year and as a time series. You can then quantify the increase in their numbers. There are many more analyses that can be done, for example on times between consecutive hurricanes or on the radius of the affected areas along the path.
We generated all the charts in this post directly from the NOAA database, using a few R packages. In a few seconds, these tools provide information that is not available from 1940s SPC.
The NOAA Database of Atlantic Storms
The National Oceanic and Atmospheric Administration (NOAA) is part of the federal government of the US and was created in 1970 as the merger of preexisting agencies going back to 1807. The NOAA’s HURDAT2 database contains 52,717 observations made about storms of various intensities from 1851 to 2020.
It’s multiple observations for each storm as it moves, with geographical coordinates, sustained wind speeds, and many other parameters. A storm often starts out as a tropical depression, then graduates to a tropical storm and possibly a hurricane, and fades away.
The older records have many missing values. Many storms are likely missing altogether because they didn’t encounter people in a position to collect data. Hurricane strength categories are assigned based on wind speed sustained for 1 minute. The threshold for Category 3 is 110 mph. How anyone could measure it in Louisiana in 1866, the date of the earliest Category 3 hurricane in the database?
In fact, NOAA researchers Vecchi and Knutson have adjusted the older records for missing hurricanes using ship track density. Since1965, the NOAA has used weather satellites to detect storms, and these data need no adjustments.
Visualizing the Hurricane History
From the looks of the database, visualizing its content is challenging. You find, for example, that the storm names are not unique identifiers. There are three hurricanes named “MARIA,” and three named “KATRINA,” in 1981, 1999, and 2005, respectively with 20, 16, and 34 records:
AL211981, KATRINA, 20,
19811103, 0000, , TD, 16.9N, 81.2W, 25, 1005,…
19811103, 0600, , TD, 17.2N, 81.3W, 25, 1005,…
19811103, 1200, , TD, 17.5N, 81.4W, 25, 1004,…
AL151999, KATRINA, 16,
19991028, 1800, , TD, 11.4N, 80.9W, 30, 1001,…
19991029, 0000, , TD, 11.6N, 81.6W, 30, 1001,…
19991029, 0600, , TD, 12.0N, 82.0W, 30, 1001,…
19991029, 1200, , TD, 12.6N, 82.6W, 30, 1000,..
AL122005, KATRINA, 34,
20050823, 1800, , TD, 23.1N, 75.1W, 30, 1008,…
20050824, 0000, , TD, 23.4N, 75.7W, 30, 1007,…
20050824, 0600, , TD, 23.8N, 76.2W, 30, 1007,…
20050824, 1200, , TS, 24.5N, 76.5W, 35, 1006,…
KATRINA in 2005
In R, with the graphics package ggplot2 and the geographical mapping package sf, we can plot the path of a hurricane. The path of the KATRINA that ravaged New Orleans in 2005 was as follows:
All Category 3 Hurricanes from 1851 to 2020
For a global view, we can plot the paths of the 162 storms in the database that were, at least at one point, labeled a Category 3 hurricane:
It shows most of the storms originating off the coast of Senegal, gaining strength across the Atlantic, blanketing the Caribbean, and hitting the Eastern United States before dying in the North Atlantic. As West Side Story’s Anita belted out about Puerto Rico, it’s “Always the Hurricanes Blowing.” This picture, however, provides no hint as to whether the hurricanes are becoming more frequent.
Category 3 Hurricanes Decade by Decade
To get a closer view over time, we can plot, for example, the data decade by decade since 1960:
The activity is visibly more intense in the last two decades, and the 2020s are off to a strong start with 5 hurricanes in Category 3 or above in just one year. We can analyze the NOAA database to see whether there is an upward trend or just fluctuations around a constant intensity. If there is a trend, we cannot determine from this data alone whether it is caused by human activity.
Pointy and Skewed Seasonality
We know that storms in the Atlantic are seasonal. The NOAA data gives us information both on the shape of this seasonality and its evolution over time.
Overall Seasonal Distribution
Density plots of the day of year storms were detected shows strong seasonality, with no events from January to May, no hurricanes of category 3 and above until July and after November, and a sharp peak in September for all categories of storms. The distributions are skewed and pointier than a gaussian, with an initial spring bump:
Stability Of Storm Season Over Decades
Has the length of the hurricane season changed? Decade-by-decade Tukey-style box plots can help us find out:
For all storms, all hurricanes, or just the hurricanes of category 3 and higher, these charts do not visually suggest that the active season is shifting or expanding over 17 decades. The chart for strong hurricanes looks chaotic because it is based on only 162 events; the chart for all storms, stable because it is based on 1924 events.
According to Vecchi and Knutson, the NOAA started relying on satellite images to detect storms in 1965, and the storm counts needed no adjustments after then. To have homogeneous data, we start the charts in 1965. We use a different y-scale for each chart, based on the number of storms plotted, because using the y-scale for all storms in the three charts would visually crush any pattern we can see on hurricanes or strong hurricanes. As it is, it appears that the number of hurricanes of category 3 and above has been rising while the total number of storms has been stable.
Can we confirm it? It’s one more instruction to fit linear trends, and it does confirm it. The “99% confidence interval” in gray, however, isn’t much of one, given that 32% of the data points are outside of it. Clearly, this model doesn’t properly account for the variability of hurricane occurrences but it does show the mean number of hurricanes of category 3 and above per year tripling in 55 years while the total number of hurricanes nearly doubled and the total number of storms stayed flat.
There is more to listening to the data than plotting XmR charts. First, you need to think about the physical phenomena the data represent and how they were collected. Then you need to apply the many tools of visualization and exploratory analysis of today’s data science to raise questions and provide tentative answers.
As Wheeler says, this data tells us that strong hurricanes are getting more frequent but it can’t tell us why. According to the NOAA, there is an Atlantic Multidecadal Oscillation (AMO) in ocean surface temperatures:
“The AMO is an ongoing series of long-duration changes in the sea surface temperature of the North Atlantic Ocean, with cool and warm phases that may last for 20-40 years at a time and a difference of about 1°F between extremes. These changes are natural and have been occurring for at least the last 1,000 years.”
For the past 150 years, this assessment is based on measurements; beyond that, it’s tree rings and ice cores. According to the NOAA, the AMO explains, at least in part, the increase in the number of hurricanes with the total number of storms remaining stable.
The NOAA neither affirms nor denies that this increase is due to global warming from human activity. They do not conclude either way:
“In the 20th century, the climate swings of the AMO have alternately camouflaged and exaggerated the effects of global warming, and made attribution of global warming more difficult to ascertain.”
J.M.K.C. Donev et al., from the University of Calgary, posted the following plot of the AMO and storm occurrences on their Energy Education site in 2017:
If you want to dig further AMO monthly data since 1856 are available from the NOAA’s Physical Sciences Lab.