When is the best time of year to schedule a storm chasing vacation?

Most tornadoes are ephemeral. The act of finding something that tends to last seconds to minutes can be daunting. When you find one, it’s all the more rewarding. It makes you want to find more, with better results and less wasted effort.

Here, I will use the tornado clustering methods in my previous post to address the question: If my goal is to see tornadoes, when should I schedule my storm chasing vacation (or as we storm chasers like to call it, the “chasecation”)?

While hoping to find the secret to make chasecation successful, I also found a what seems to be a compelling way to rank big chase days.

Measuring tornado days

First off, we need to develop a climatology of tornado clusters. Then it becomes a matter of finding when clusters happen most often. Here, we’ll restrict the tornadoes we look at to ones in the months of April, May, and June. Tornadoes examined must also satisfy the following criteria:

• Tornadoes must be clusterable. From our previous post, the “clusterable dataset” spans from 1996-2017.
• The tornado start time must fall between 12:00 and 23:59 CST. This more or less captures the typical hours of chasing; other times are (mostly) reserved for sleeping and positioning.
• The tornado start location must lie within prime chase territory, as outlined by the following blue polygon (all tornado tracks plotted):

This polygon roughly follows the chase territory map compiled by Jeremy Perez, with minor revisions (inclusion of the Nebraska Sand Hills, truncation of areas east of the traditional Great Plains). The boundaries are up to some interpretation, and these are not necessarily exact, but they are sufficient for the purposes here.

Keep in mind that I’m analyzing and writing this from the perspective of a storm chaser. I only care about maximizing my exposure to tornado minutes, or to put it another way, the probability that I am witnessing a tornado at any given time. This in mind, I care less about all the different clusters than I care about any single cluster, because I can only be at one location at a time.

So in this examination tornado days will be judged by the biggest cluster of the day. This is regardless of whether or not that cluster did actually produce the “tornado of the day”, since we cannot know that a priori. To reiterate from my previous post, the size of a cluster is just the sum of the each tornado’s longevity for all the tornadoes in the cluster. For example, three 10-minute long tornadoes in a cluster yields a cluster size of 30 minutes.

Regarding the clustering parameters, I used eps_km=80, eps_min=60, min_samples=15 in my introduction post, and that will be our starting point. Recall that we can only call a day a “tornado day” if there is at least one tornado cluster found, given those parameters.

Applying to a select sample

To test that my parameters make sense, I applied clustering of the most well-known chase days with tornadoes that show up in the first few pages of YouTube and Google Image queries. The results of this work is summarized in the following table:

The size of the smallest cluster in this sample is 20 minutes, and none of the tornadoes were cast off as outliers. Also, the majority of these tornadoes did indeed belong to the biggest cluster of the day. These results justify the decision to use the biggest cluster only, and to maintain the parameters we started out with.

Digging deeper, I arbitrarily bucketed days into four cluster size ranges:

• 15-29 minutes: 2 events (Goshen WY, Campo).
• 30-59 minutes: 3 events (Moore, Wray, Dimmitt).
• 60-119 minutes: 8 events (El Reno, Bowdle, Dallas, Rozel, Wynnewood, Attica, Simla, Spencer).
• 120+ minutes: 9 events (Pilger, Dodge City, Shawnee, Anadarko, Marquette, Manchester, Coleridge, Bradshaw, Canton Lake).

From this, we can see that not all clusters should be weighted equally.

Many more of the well-known tornadoes are sourced from 60+ minute clusters, compared to 15-59 minute ones. So the bigger clusters should probably be weighted more heavily when we figure out when to go on our chasecations.

To summarize, we consider a tornado day as a day with at least one tornado cluster. We can divide tornado days based on cluster size ranges for more detailed analysis, and consider “big” days as days with 60+ minute clusters.

Now it’s a matter of finding out the distributions of those tornado days by month and day.

Results

Below we have the distribution for all tornado days (as defined by our clustering method).

The left-hand side of the chart shows the raw counts of clusters by calendar day. For instance, between 1996-2017, 12 of 22 June 12s were tornado days. On May 10, that number is nine.

On the right-hand side is the rolling sum for seven, 14, and 21 days respectively. The 14-day rolling sum on June 12 indicates that a little over 80 clusters have occurred between 1996-2017 on the 14 days leading up to and including that date.

To put it another way, if you took a two-week chasecation ending on June 12 every year between 1996 and 2017, and you nailed the biggest tornado cluster every time (good luck!), you would’ve experienced a little over 80 clusters. Or around four per year.

Now let’s break that down by the size ranges we defined earlier. Again, these are distributions of the biggest cluster of the day, so events counted on the 15-29 minute plot for example do not overlap with events on any other plot.

A few observations here:

• Tornado day concentrations actually max out in June.
• If we weigh the 60-119 and 120+ minute clusters more heavily, those max out in late May.
• From the rolling sum graphs, the lower-size groups increase linearly from April-June, peaking in late June. That peak shifts towards May as the sizes increase. Part of this is due to larger clusters masking out smaller clusters in May. But part of this is due to the fact that in May, the average tornado event is bigger, even if there are fewer total events compared to June.
• April sucks for chasecation across the board. The frequency of higher-end events on the Plains in April is less than that of early June. The frequency of all events is much less.

Visualizing the bigger tornado days, we get the following:

The sample size isn’t that great, but the distribution on the left looks like a bell curve with a sharp peak on May 24 (which is close to the raw count per day peak near that period as well).

So what’s our conclusion?

LATE MAY IS KING

Looking at the rolling sum to the right, we can see that a two-week chasecation that ends on May 31 has resulted in about 30 big tornado days in the past 22 years. In other words, we should expect 1-2 such events if we chase two weeks in late May, and that’s not counting the smaller events that can also yield an incredible chase.

Looking back to previous decades, late May tends to produce big as well. Well-known events in the tornado research community such as Binger, OK (May 22, 1981), Union City, OK (May 24, 1973), and the famous “Seven Days in May” from May 16-21 in 1977 have flirted around that peak.

Before everyone starts purchasing their plane tickets, let me argue for June. To start, it’s comparable to early May in terms of big days and the peak in total tornado days resides in June.

A two-week chasecation ending on June 17 has resulted in almost 20 big tornado days, and 80 total tornado days. That’s one and four per year, respectively. Outside our dataset we already know Junes have produced big events such as the June 15-16, 1992 Plains outbreak or the TX Panhandle outbreak of June 8, 1995. The biggest perk of June of course is the reduced crowds.

Early May isn’t bad either, but it’s more boom or bust. June can squeeze something out of an unfavorable pattern; early May typically doesn’t. Most chasers probably already knew that, but the data shows it plainly.

Finally, don’t plan an extended Plains chasecation in April. Just don’t.

Appendix – additional content

These are the top 10 days in the dataset.

Some big days show up in this collection: the Bridge Creek-Moore outbreak on May 3, 1999; the Dodge City, KS tornadofest on May 24, 2016; the Greensburg, KS tornado of May 4, 2007. Interestingly, six of the top seven events are centralized in either western KS or central OK.

And below is a map of the biggest clusters from all tornado days.

As you might expect, western Kansas and central Oklahoma are juiced. If you’re familiar with some of the bigger days in recent chase lore, you can see the footprints them left behind in the map.

The vertices of the chase territory polygon are enumerated in the below table. Toggle more entries to see all the rows.

This is the second of a multi-part series that introduces a technique developed by the author to classify tornado events and tornado days. We will follow up with additional posts that demonstrate its uses. Introduction post.