Wednesday, February 7, 2007

Undercounting of Mid-Atlantic Tropical Storms

It is common to see claims that the incidences of tropical storms in the Atlantic basin are accounted for in the relevant database. For example, Kerry Emanuel has stated, "Tropical cyclone detection rates have been close to 100% globally since around 1970, when global satellite coverage became nearly complete 1. In densely travelled oceans, such as the North Atlantic, detection rates were probably reasonably high well back into the 19th century."

Certainly, since the mid-1960's when satellite coverage of the basin began, every tropical storm has been identified. However, in the period from 1900 until the mid-60's it is not so clear. This note will attempt to show that the evidence leans to the conclusion that the storm databases seriously undercounts storms that never leave the Mid-Atlantic.

A Definition:

For the purposes of this study the term "Mid-Atlantic Storm" is defined as any storm that remains east of the 60 degrees west line of longitude.

The 60 degrees west line was chosen because it is close to the chain of the Windward Islands, the first land masses a Mid-Atlantic storm could likely encounter. It also seems likely storms to the east or north-east of this line of islands which remained there, would be the most difficult to identify and, thus, the most likely to remain uncounted.


For the purposes of this study I am using the Tropical Prediction Center Best Track Reanalysis (archived here). Specifically, I will be looking at the data from 1907-2006. This somewhat arbitrary selection gives me an even 100 years worth of data to examine. However, it seems no more arbitrary than selecting another date, say 1900, as a starting point because it also is a nice round number. An advantage to the dates I've selected is that it will give me 10 equal decades with which to make comparisons. (Note: The data for the 2006 season is not part of the raw Hurricane Database file, but was included from the Tropical Prediction Center Advisories.)

Mid-Atlantic Tropical Storms:

I first examined the records to see just how many Mid-Atlantic tropical storms there have been indicated. Fig. 1 shows the storms averaged by decade from 1907-2006. (Click on any Fig. # to get a larger view of the graphs.)

Fig. 1

It is obvious there is a vast difference in the numbers of storms we see in the database. The era from 1967 to 2006 sees over three times the numbers of Mid-Atlantic storms when compared to the previous six decades.

A similar story is seen if we look at the same numbers as averages of all tropical storms. (Fig. 2.)

Fig. 2

From a low of 1.4% of all storms in the decade 1907-1916 (1 out of 70 total storms), Mid-Atlantic storms made up nearly 1/4 of all storms in the decade 1987-1996 (26 of 106 storms or 24.5%). One might claim these numbers represent the variability of storm incidence in the Atlantic basin, but it seems unlikely when you consider that the range can vary 30 fold. There was one (1) recorded Mid-Atlantic storm from 1907-1916 and 31 recorded for the time period of 1997-2006.

It is my contention that a good part of the difference in Mid-Atlantic storm numbers over time is due to the undercounting of Mid-Atlantic storm.


In order to test my contention that the discrepancies in the storm numbers are due more to undercounting as opposed to natural variability, I will look at storm numbers from the same time period that are not in contention. Namely, I will look at the incidences of U.S. landfalling storms and compare them to the numbers and rates we found in the Mid Atlantic tropical storms. If the differences are primarily due to natural variation in storm rates we would expect to find those rates duplicated in the landfalling storms. Conversely, if my contention is correct and there has been undercounting of storms we should find, at least, that the ratio of U.S. landfalling storms (or any subset thereof) to all storms was higher in the past (due to inaccurate Mid-Atlantic storm counts), and that the ratio will decline once better storm detection techniques are in place.

In order to have multiple points for comparison I divided U.S. landfalls into two categories. The first consists of storm landfalls on the coast of Florida. The second consists of storm landfalls on the coasts of Louisiana, North Carolina and Texas inclusive. The two categories have roughly similar landfall rates to each other so it seemed like a good match. In reality, it shouldn't matter how I divide up U.S. landfalls (or even if I divide them) as we are looking for the rates of landfalls over time.

The next figure looks at the decadal average for land falling storms in Florida.

Fig. 3

As can be seen, the decadal average of landfalling storms in Florida is nearly equal when comparing the satellite years (1967-2006, 14.50 storms per decade) and the pre-satellite years (1907-1966, 14.67 storms per decade.)

For Louisiana, North Carolina and Texas the numbers tell a similar story:

Fig. 4

In this data the satellite years averaged 17.75 landfalling storms a decade, as opposed to 17.16 a decade for the pre-satellite years.

We do not find the large variation in numbers here that we found in the Mid Atlantic storms. This is suggestive of undercounting, but it could be argued that some closer examination of the data would find large known variations elsewhere in the Atlantic basin. This seems unlikely (to put it mildly.) In order to confirm that these results are probably due to undercounting, we need to compare the ratios of these landfalling storms to all storms in the basin. If other parts of the basin were experiencing more storms than we should find that the ratio of landfalling storms to all storms should be fairly constant over time. The data does not show that to be the case. (Figures 5 and 6)

Fig. 5

Fig. 6

As I consistently find that the ratio of land-falling storms to all storms is higher in the pre-satellite era, I am forced to conclude that the hypothesis is well founded. While I cannot say with 100% certainty that it is true, I am after all postulating large tropical storms for which there are no records, the basic logic of my approach seems inescapable.


It seems very probable the Mid-Atlantic storm counts are undercounted in some fashion. It is a trickier question to determine the degree of undercounting. However, if we take the rates of Mid-Atlantic storms found during the satellite era (see Figure 2 above) and apply them to the pre-satellite era the results are startling. Broken down by decade, the percentage of Mid-Atlantic storms to all storms in the satellite era looks like:

1967-1976: 17.20%
1977-1986: 14.44%
1987-1996: 24.53%
1997-2006: 21.23%

and for the entire period:

1967-2006: 19.77%

If we look at the minimum (14.44%) and maximum (24.53%) values as defining a range for the pre-satellite number (which today sits at 40 Mid-Atlantic storms out of 495 total storms, or 8.08%,) we are left with a range of an additional 1.28 to 2.46 storms per year. That would mean a difference for the sixty year period of plus 77 to 148 storms.

Really, these numbers strike me as too high, but I do not see how to avoid them. One might postulate there is some new mechanism at work that is increasing the incidence of Mid-Atlantic storms during the last 40 years, but there isn't such a mechanism described in the literature that I have seen. And based on my look at U.S. landfalling rates any proposed mechanism will have to apply to the Mid-Atlantic only. (It is for this reason that I have to reject the "Warm Atlantic" "Cold Atlantic" work of Landsea, Pielke, Mestas-Nuñez and Knaff, as being the answer, for such an effect would be basin wide and not limited to the Mid-Atlantic.)

Please note, this is a preliminary version of this paper which will be added to and fine tuned almost constantly. Any comments or criticisms are most welcome.

No comments: