This project uses GLOBE data: GLOBE Task Team and others (Hastings, David A., Paula K. Dunbar, Gerald M. Elphingstone, Mark Bootz, Hiroshi Murakami, Hiroshi Maruyama, Hiroshi Masaharu, Peter Holland, John Payne, Nevin A. Bryant, Thomas L. Logan, J.-P. Muller, Gunter Schreier, and John S. MacDonald), eds., 1999. The Global Land One-kilometer Base Elevation (GLOBE) Digital Elevation Model, Version 1.0. National Oceanic and Atmospheric Administration, National Geophysical Data Center, 325 Broadway, Boulder, Colorado 80305-3328, U.S.A. Digital data base on the World Wide Web (URL: and CD-ROMs.

Elevation Code using GLOBE data:

Uses: NOAA's International Best Track Archive for Climate Stewardship (IBTrACS) data, accessed on 29 Nov 2020


This page outlines the noteworthy changes in Atlantic hurricanes found in the HURDAT2 dataset. All wind speed numbers below are given in knots. Code now reads IBTrACS but the results are the same as HURDAT2. Almost all of the domain logic is in storm.summarizeStorm() and year.summarize().

The data to determine whether storms are over land or not comes fron GLOBE. That code ( is relatively simple although installing the python library dependencies (GDAL) is cumbersome.

Code: | Instructions: README


The first thing to note is the poor quality data before 1900. When the majority of storms are not noticed until they achieve hurricane status, the data is not really suitable for most comparisons with modern data. There is lower quality data prior to the satellite era as shown by years with a majority of tropical storms not being recorded before becoming tropical storms.

Another reason to avoid the older data is the very significant drop in the percentage of storms near land. There's no reason for any change in that percentage in reality. But that drop could be created by a change in detection from inconsistent detection away from land to consistent detection both near and away from land.
I found that using fewer categories of storms results in more storms per category which leads to somewhat more consistent statistics in those categories. However storm tracks (and whether they move near land) may have multi-year cycles. Using 1920 to present is somewhat of a cherry pick being 100 years, but also starting with some strong hurricane years in the late 20's and early 30's. That has the effect of taking into account some natural cycles.
trends above (i.e. a drop of ten percentange points for the strongest category):
	35-68	-0.19963145415440647
	69-90	-0.17906983251976877
	91-119	-0.07756536641604327
	 >=120	-0.10121398188401555
I use those trends to adjust storm counts for past years. Using a linear trend is an oversimplification, as the data shows more of a step function before and after the satellite era. Having fewer categories increases the adjustment slightly since I do not adust zero storms in any year to be nonzero. For example, a single strong storm a century ago will be adjusted to be 1.1 storms. Zero storms in a given historical year will remain zero. With that adjustment there is still a rise in the count of the strongest storms from 0.65 such storms per year to 1.7 storms per year over the past century, shown in black:

I use the same trends to adjust ACE by adjusting a separate ACE calculated for each category for each year, then summing those for ACE for each year. ACE is rising by about 27 units per century.

All of the ACE increase appears to come from the strongest storms, which is rising at 30 units per century. One factor may be storm longevity at peak intensity. Also the very slight drop in 91 to 119 knot storms may be because more of those increase to 120+ knots.

Changes in Percentages

Percentages of storms with particular ranges of measurements should stay constant unless there are changes in environmental factors, detection, measurement, or recording. It's fairly obvious that storms near land declined due to changes in detection and recording before and after the use of satellites. Some increases in wind speeds may be due to changes in measurement. Before soundings, doppler radar, and hurricane hunters, the peak winds of storms winds were measured by happenstance and estimated from storm positions and available measurements including barometric pressure.

However there appears to be a general trends towards higher peak winds in the strongest storms, more rapid intensification above certain thresholds (e.g. >= 40 knots in 24 hours), and slower storm motion. These do not appear to be caused by measurement changes but by the global warming and possibly natural cycles.

The percentage of storms that "rapidly intensify" does not appear to be changing. But the main caveat is that the majority of hurricanes have at least one 24 period of >= 30 knot increase in winds. Thus the definition seems to be a bit loose, shown in blue:

A tighter definition using >= 40 knots shown in purple above shows a small increase. Also using an ri ratio (24h RI periods >= 35 knots divided by all 24 periods) shows an increase. For more details and a complete list of rapidly intensifying storms see Percent of rapidly intensifying storms The 36 hour RI is similar to the 24h:

Some 6 hours RI categories are increasing a bit:
There is a trend towards slower storm speeds at landfall, for both tropical storms and hurricanes. The recent trend is less clear (flat since 1990) for the stalled storms like Hermine (1980) and Mitch (1998). But Harvey (2017), Florence (2018) and Dorian (2019) also fall into the slowest category with <= 3 mph of storm motion shown by black dots below. For more details and a list of the slower storms see Percent of slow storms at landfall That page shows, at least anecdotally, that slower storms lead to higher rainfall.

Changes in Averages

Like percentages, Averages of storm measurements should also be constant except for environmental changes like global warming and natural cycles, and measurement changes. Averages are useful when there are very few storms meeting category criteria.

There is a trend to less weakening after landfall since the 1950's. There is a lot of random variation that could be created by random stronger landfalls. For example Michael landed in 2018 in Florida at 135 knots and weakened to 45 knots in 24 hours, a drop of 90 knots due to the very strong starting point. Also criteria for calculation is landfall as a hurricane and 24 hours over land as at least TD. Only one storm a year meets the criteria since 1920, more than 1 a year recently. The 1950-present slope is notable at 14 knots less weakening over the past 70 years. However the weakening trend since 1980 or 1990 is flat, and part of the 1950-2020 slope may be an artifact of outlier storms in years with few landfalls.

Average strengthening before landfall is shown for the 12 hours and 6 hours before landfall. The problem with this averaging is that more storms are zero or negative (ignored) or nonzero outliers like storm #1 in 1952. Here's the list of landfalling storms with wind increases before landfall The table basically shows 0 or 1 a year, sometimes 2 with no trend. That implies that strengthening before landfall is just chance.
Above I speculated that strong hurricanes may stay stronger longer, thus increasing ACE. But the long term ACE increase appears to be driven by more numerous strong storms (>= 120 knots). The average number of hours of strength per storm is highly variable weather, with low numbers of such storms in any given year driving the average up (e.g Allen in 1980) or down (e.g. Katia and Ophelia in 2011 with just one 6 hour period each)

Changes in Totals

Like the counts above, totals are probably a better way to capture an effect and changes, provided there adjustments for missing storms (e.g. pre-satellite). The main reason is that trends for totals include all years including the zero years. There seems to be about 1 hour more per year of >=120 knots. Same slope whether looking at 70 years or 30 years.