In order to compare any particular year to the average year, we can see how far ahead or behind that year is to the average for each day of the year. In other words, if there are 500 tornadoes at the end of May, that would be about 100 behind the long-term average of about 600. If there are 700, it would be about 100 ahead of the average year. A time series of this departure through the year gives an idea of how far ahead or behind the season is throughout the year.

Tornado anomalies through the year

On this graphic, positive departures indicate above average and negative departures are below average. If the line is horizontal, that period of the year is exactly average. The years shown include the biggest years in the tornado record (2004, 2008) and the smallest years (1987, 1988), as well as 1973 and 1992, which demonstrate different ways that a big year can happen. In 1973, there aren’t any really large periods for tornadoes, but a steady increase ahead of normal throughout the year. In 1992, the early part of the year looks a lot like the really slow years of 1987 and 1988, reaching more than 300 below normal by 14 June. After that, a short period of frequent tornado occurrence got the season back to normal within a month and then ending up more than 200 ahead by the end of the year.

The black dots represent 2009 at the end of each month and mid-June. Reports after March 2009 have been adjusted as discussed here. The gray line is for 1996, the closest analogue to 2009, where the season was ahead of normal through early May, then fell almost 100 below normal by the middle of June before finishing about 50 ahead.

In reality, despite the slow period that coincided with the VORTEX2 field project when, for a 5-week period, there were over 100 tornadoes fewer than normal, the 2009 season has stayed within 100 of the normal throughout the year.  Normally, by 22 June, more than 60% of the median 1290 tornadoes per year have occurred.  We’ll have to wait to see how 2009 turns out.

Yesterday, May 21, 2009, marked only the second time that a May 21st has had no reports of severe weather (tornado or hail or wind event) since 1955. Surprisingly, the last time no recorded tornado, hail, or wind event occurred was just six years ago, on May 21, 2003.

Every other May 21st in every year since 1955 (the first year that NWS started official records for severe thunderstorm hail and wind events) has had at least a single report in the database for this day and most May 21sts have had numerous reports given this is about the peak in severe weather activity during the annual cycle.

How about severe thunderstorm and tornado watches issued by the NWS Storm Prediction Center (SPC)? If the SPC does not issue a watch today or tomorrow, that would be a substantial “first” for this particular week in May. Since 1970, there has not been another May 17th-23rd period without at least 6 watches issued (either a severe thunderstorm watch or a tornado watch somewhere in the country). The SPC has not issued a watch since Saturday evening, May 16, 2009. Since 1970, there have been about 28 watches issued during this week in May. The greatest number of watches ever issued during this week-long period was 59 (averaging almost 9 a day!), in 1989. It now appears likely that SPC will break the previous minimum number of watches issued for the week, 6 in 1976, by Saturday, achieving a record of 0 watches for the week of May 17th-23rd, 2009. That has not happened before and can be considered quite unusual.

If we look at the first three weeks in May with respect to SPC watches, then this unusually quiescent week is masked by watches issued earlier in the month. Through the first three weeks of May 2009, SPC has issued 93 watches. This is a little above the long-term average for the period of 88 but well short of the maximum number issued (a remarkable 226 watches in three weeks), in 2003. The fewest number of watches ever issued for the first three weeks of May was 28, in 1976.

What is the cause for such tranquility in the atmosphere? The overall flow regime has quickly transitioned into one more characteristic of July or August, or even September. Fast mid-level flow is confined to the U.S./Canada border and very weak flow is left across the rest of the continental U.S. There has been a highly anomalous low pressure system drifting slowly westward from Florida over the past few days. This low formed in the wake of the last significant weather system to bring severe weather to the Central and Eastern U.S. last week. Low level moisture has persisted along the Gulf Coast, and is starting to return to parts of the Southern Plains as the Florida/Gulf low moves westward, yes westward. At the same time, and also characteristic of July/August, an early monsoon pattern has started in the Southwest U.S. This will result in an increase in thunderstorms across that region, and northward across the interior west and Rockies, in the days ahead.

It’s still possible that the overall flow pattern can revert to a more typical and active late May weather pattern, thus resulting in more widespread severe thunderstorms and a greater potential for tornadoes. However, the chances of that happening decrease as the summer months approach and the weather pattern, as established, becomes more typical for the time of year we find ourselves in.

Figure 1. 13-hour WRF-NMM forecast of simulated reflectively at 1 KM AGL and valid at 1300 UTC 8 May 2009 (left), Verifying observed base reflectively, and severe thunderstorm warning polygon, valid at 1300 UTC 8 May 2009 (right). Both images are centered over middle Tennessee/south-central Kentucky.

The 13-hour WRF-NMM forecast of the Missouri Bow Echo (see earlier post with this title) is remarkable in both its accuracy and structure, particularly given the severity of the event. As model forecasts go, it appears to be a perfect piece of numerical guidance. But shifting the grid about 450 miles to the east, the exact same 13-hour WRF-NMM completely missed the MCS (albeit less severe) moving through eastern Tennessee. So while there is little doubt that the model provided an essentially perfect prediction of the intense bow echo over southern Missouri, in a purely deterministic sense, the same model provided little-to-no short-term convective guidance with respect to convective mode and QPF over much of Tennessee.

The information provided by these high-resolution NWP models is revolutionary, and will likely lead to a quantum increase in high-impact services provided by the NWS. But let’s be careful not to oversell the capabilities of a single, deterministic model forecast. In order to fully realize the potential of future NWS forecasting and warning services, an ensemble of convective-resolving models will be required to address the uncertainty that accompanies all weather forecasts. The HWT has evaluated convective-resolving models over a large portion of the CONUS for the past several years, and it is encouraging to see the improvements these models have made in high-impact convective guidance and in their ability to predict intense, realistic convective structures such as the bow echo over southern Missouri. But a single high-resolution NWP forecast, regardless of its ability to reproduce intense convective structures, is unlikely to meet the future uncertainty requirements of the entire NWS at all times and locations. That said, the development and evaluation of these convection resolving models is and will continue to be an essential part of future high-impact, life saving, decision support services provided by the NWS, likely realized through a blend of deterministic guidance, well constructed ensemble systems, and related ensemble interrogation tools.

Dr. David Bright, Storm Prediction Center, Science Support Branch, Norman, OK

Above (left) is an example of a forecast from numerical models being evaluated in this year’s SPC/NSSL Spring Experiment. The actual radar image on the right shows the intense bow echo that raked across southwest Missouri early last Friday morning. These images are centered on Joplin, Missouri (JLN). The 13-hour forecast of simulated reflectivity from the WRF-NMM (Weather Research and Forecast Model – Non-hydrostatic Mesoscale Model) is exceptionally accurate and depicts the form and intensity of the bow very well. This is a 4km resolution numerical model that was initialized at 00Z 08-May-2009 and valid at 13Z that Friday morning.

The verifying 1km base reflectivity (radar mosaic) image is on the right with the model fields for winds overlaid on the radar. The barbs in each of the images are the model’s instantaneous 10m winds in knots (the grid is skipped to lessen the clutter of the wind barbs). The isotachs (lines of constant wind speed at 10m) are plotted from the WRF “history variables” (no grid skip). The wind history variables are the maximum 10m wind speeds in the model over the past hour ending at 13Z.

Instantaneous 10m winds in the model at 13z, near the rotating bow head, are at least 50 knots. The maximum model 10m winds over the past hour range from 60-70 knots near and north of the weak echo channel and around the comma-head of the model bow. The intensifying low was coexistent with deep convection later in its life-cycle and produced significant wind damage in parts of Missouri and Illinois. Tornadoes also developed near and ahead of this feature, killing one person in Missouri in the morning, and two people in Kentucky later in the afternoon.

This was only one of several exceptional forecasts of this severe event from the models being evaluated in this year’s SPC/NSSL Spring Experiment. To see more output on this case and more, check out the Spring Experiment website.

A composite chart valid for 9 am CDT:

Composite Chart at 9 a.m. Thu. 2009-04-09

A potent mid level jet will aid deepening surface cyclone across OK/KS through this afternoon. The strong ascent and wind fields/dynamics with this system will be tapping returning low level moisture to contribute to thunderstorm development later this afternoon across KS and OK. Eventually, these storms will spread east into MO/AR. The situation bears watching and SPC has issued a  MDT risk with high probability for significant hail and perhaps a few strong tornadoes if moisture can increase sufficiently in the low levels.

One of the really neat aspects of working in Norman at the National Weather Center where various government, academic, and private entities peruse the latest in forecast technology is that we are always trying out new ideas. Below I have placed a couple of images derived from numerical model data valid for 7 p.m. CDT this evening.

Model forecasts valid at 7 p.m. CDT 2009-04-09

The image above is a combination of the North American Mesoscale (NAM-WRF) Model and the higher resolution WRF-ARW run at the National Severe Storms Laboratory here in Norman. We are looking at a simulated infrared satellite (IR) image for this evening derived from the NAM-WRF. The forecast surface pressure, temperatures (F), and dewpoint values (brown and green values in (F)) are also from the 12h forecast valid for this evening.

Model forecasts valid at 7 p.m. CDT 2009-04-09

In this image, valid at the same time, I’ve removed the pressure, temps, and dewpoints to better reveal the fields from the higher-resolution WRF-ARW run at NSSL. The unique thing about the contours shown on the image is that these are based on maximum values occurring over the past hour valid at this forecast time. We are looking for areas in the model, generally near the surface, that meet certain thresholds. We can see where higher simulated reflectivity in the model indicates thunderstorms. And, where these simulated thunderstorms are particularly intense as far as organized updrafts. While not quite at the scale small enough to depict an individual supercell thunderstorm, this data is getting close to showing us “surrogate severe storms” based on the parameter thresholds that we have chosen.

It is interesting to juxtapose the coarser resolution simulated IR with the higher resolution of the WRF-ARW model reflectivity in this case. This is essentially what we do operationally with these observed data sources where IR satellite imagery is about 4km in resolution and radar data is about 1km or less.

Later today we will see how the models compare with reality.

The first dataset is the number of deaths per year from tornadoes in the United States. The National Weather Service archive goes back to 1950. Brooks and Doswell (2002) discussed the long-term history of tornado deaths, drawing on the work of Grazulis (1993) (Grazulis, T. P., 1993: Significant Tornadoes, 1680–1991. Environmental Films, 1326 pp.). Reasonably reliable estimates of deaths per year can be made back to about 1875 by using the Grazulis data.

The Brooks and Doswell paper had a graph of the annual death toll normalized by population of the US through 2000.  Here is an updated version (through 2008) of that figure, showing that the death toll per million population appears to have leveled off in the last decade or so.

The purple points are the annual death rates, the red line is a simple smoother, the solid black line is a long-term trend in two sections (1875-1925, 1925-2000) and the cyan lines are estimates of the 10th percentile and 90th percentile from 1925-2000. Brooks and Doswell (2002) have an extensive discussion of the record and its possible implications.

Normalized US Tornado Deaths (1875-2008)

At the end of this post, we have a table of the annual death tolls going back to 1875 (1875-1949 from Grazulis, 1950-2008 from the National Weather Service.)  Although the data represent our best understanding at this time, it is possible that the numbers could change, if additional information was found.  Occasionally, it’s discovered that a fatality associated with a tornado was missed or double-counted.  We’ll correct such entries if we find about about them, but that will likely be a rare event.  The death tolls are for direct deaths, i.e., someone killed directly by the impact of the tornado.  It does not include indirect deaths, which might not have occurred if the tornado had not happened, but the tornado was not an immediate cause.  Examples of indirect deaths that have occurred include a heart attack upon seeing damage to a neighbor’s house, falls when going to shelter, and a fire caused by a candle lit when the power went out after a tornado.

First, it’s important to understand how tornado information is collected.  Reports come into local National Weather Service Forecast Offices from a wide variety of sources (trained spotters, emergency personnel, the media, the general public.)  Those reports get put out relatively quickly and form the basis of the preliminary, so so-called “rough”, log of tornadoes.  As time goes on, more reports may come in, increasing the count, but the local Warning Coordination Meteorologist (WCM) may also find that some of the reports are of the same tornado, lowering the count.  Within a couple of months of the end of the month, the WCM submits a list of tornadoes that the National Climatic Data Center and the National Weather Service’s Storm Prediction Center’s WCM go through and make sure that tornadoes that cross from one office’s area of responsibility into another office’s area only get counted once and do some quality control.  Eventually, this produces the final, or “smooth”, log of tornadoes.  Asking the question “How many tornadoes have there been this year?” right after an event gives us the problem of only having the preliminary data avaiable.  Answering someone’s question with “Ask us again in a few months” is not usually seen as satisfactory by the questioner.

The approach we’d like to take is to look at the historical relationship between preliminary reports and final reports.  In general, our naive expectations would be that there might be increases from the preliminary log to the final log on days with very few, if any preliminary reports, and there might be decreases on days with long-track tornadoes that might be reported many times.  Which effect wins out, and by how much?

In the spring of 2008, we started looking at this question in detail and started by looking at monthly and annual totals.  The ratio of the final total to the preliminary total for each month from January 1998 to November 2008 with at least 30 preliminary reports is shown in red in the figure below.  A running annual total for the next twelve months is shown in black.

Monthly/12-monthly Final to Preliminary Tornado Count Ratio

The result was surprising.  A rather sudden change took place in early 2006.  Prior to that, the final log tended to have about 10% more tornadoes than the preliminary log.  The ratio had, perhaps, trended slightly downward, but starting in March 2006, it dropped precipitously, such that for 2007 and 2008, there were 20% fewer tornadoes in the final log than in the preliminary log.  This, obviously, was a huge change and meant that our intuition of what the preliminary log meant in terms of the number of tornadoes was wrong.

We then looked at the differences on a day by day basis.  The figure below shows the change from preliminary to final tornado count as a function of the preliminary count.  As we might expect, there’s not much difference on the “small” days and some of those have more in the final than in the preliminary and some have fewer.  For “big” days, the final log tends to have fewer.  We broke things down in the periods 1998-2005 and 2006-2008.  Linear regression lines are shown as dashes for each of the two periods.  For the early period, for 100 preliminary reports, the final total, on average, was about 10 lower.  For the later period, it was about 30 lower.  Only once in that three-year period did a preliminary count of at least 40 end up with more in the final count.  In the earlier period, that was a common event. The effect on the 2008 totals was that, instead of adding about 200 tornadoes from the preliminary to the final totals, you had to subtract more than 400!

Changes in final report count as function of preliminary reports

We admit we’re not entirely sure what happened in March 2006.  A number of small changes in software for collecting reports were implemented about then, but it’s not clear if that’s enough to explain what happened.  We plan to monitor the situation and use the regression from the recent data to estimate the final count, based on the preliminary count.

This is another example of how challenging it can be to deal with what seems to be a simple dataset.  Caveat lector!

To illustrate, I’ve put together a series of “postage-stamp” maps showing tornado touchdown locations by week for each of the first nine weeks of the year (e.g., 1-7 January, 8-14 January, etc.) for 1950-2007, with the Oklahoma tornadoes of this week overlaid on the 6th week’s map (5-11 February).

Postage stamp maps of tornado locations

In general, there’s a fairly sharp boundary on the northwest side of the locations of tornadoes. For the 6th week (middle panel on the right side), the recent tornadoes are right at the edge of that boundary. Over the next three weeks, however, tornado occurrence becomes much more frequent west of where that boundary is earlier in the year. Thus, a qualitative answer to the question of rarity would be that the tornadoes would have been much more likely if they had been either a couple of weeks later or a couple of hundred miles southeast. Quantitative answers are much harder to develop.  I’ve developed techniques in the past to look at that question, but for events like this that are right at the edges of what we’ve observed in the past, the answer you get is very sensitive to the assumptions you make.

Another point that can be seen is that the 7 January 1992 tornadoes in central Nebraska (see the upper left panel) are a very long distance in space and time from anything else. They are likely to be among the rarest of tornadoes.

While yesterday’s data remain preliminary, here are the official top 10 wind report days occurring in either the month of January or February since 1955:

01/29/08 333
02/21/97 217
01/03/00 203
02/10/90 189
02/05/08 186
02/06/08 184
02/22/03 178
01/18/96 173
02/11/99 172
02/24/01 160

It’s no real surprise that all these days have occurred since the 90s. The NWS has undergone dramatic restructuring during this time and the emphasis on documenting severe storm events has likewise increased. That, in combination with an increase in population and a technological revolution in communications technology (cell phones, the web, etc.) have led to a rise in storm event reporting.

However, severe weather events during the winter months are not that common. Significant thunderstorm wind events during January and February are even less common (refer to the list). The first and only time until yesterday that more than 300 severe thunderstorm wind reports occurred on a single day in January or February was just last year on January 29. Wednesday’s event will be the second time we’ve equal or exceeded that number on a January or February day since 1955. Even more remarkable is that 4 out of the top 10 January or February thunderstorm wind days have all occurred in a little over the past year.

Some of this is obviously due to the factors discussed above. How much of it is due to a real trend in severe weather during the cold season remains an open question. Or, do these episodes occur every so often in the long-term record and we are only beginning to see them due to the increased density of the reporting network?

(Data source-National Weather Service Storm Prediction Center)