Tornadoes, “threat levels,” and over-warning

There are bloggers, and then there are actual-card-carrying-journalists-who-happen-to-blog. Andy Revkin is one of the finer examples of the latter, more-select club. His post, Tough Issues with Warnings in Tornado Zones, which appeared on his blog DotEarth, provides a great illustration.

In this post Revkin addresses a problem central to tornado warnings. They warn of a danger that is highly localized in both space and time. The path of damage in the Joplin tornado? Maybe half a mile wide and six miles long. A total of 3 square miles, when you add it up. And this is for a multi-vortex, EF5 tornado, which is one of the big ones – trophy-size. Compare with 3 million square miles of US land. Most tornadoes do their damage over an even more-limited area.

In a case like this, given technologies and approaches we use for weather warnings today, it’s almost impossible not to overwarn. What does that mean? There are 3000 counties in the United States. So on average, each of them covers an area of about 1000 square miles. If the National Weather Service puts out a warning for a single county, then even if warning is spot-on, 97% of that county is not directly affected (there is, of course, the aftermath, which we’re learning daily can be horrific). Put another way, that 1000-square-mile area is also about the size of one grid cell in a global numerical model, so the embedded tornado is really a needle in a haystack.

This has significant societal implications. Andy cites a recent paper in the Weather, Climate and Society [a publication of the American Meteorological Society], by Kevin Simmons and Daniel Sutter[1], and their findings to the effect that tornadoes occurring in areas of higher false-alarm rate tend to kill and injure more people.

Thus tornadoes pose a significant challenge for the National Weather Service, for emergency managers and other state and local officials, and for all those responsible for running facilities of special vulnerability such as hospitals and schools. Each piece of the approach is problematic, yet represents a bit of a success story – a success to date, because tornado warnings have improved with respect to accuracy and lead time over decades, and a success-in-the-making, because warnings should [see below for a caveat] continue to improve in coming years.

The basic idea is to build up community alertness and readiness as the actual tornado risk builds. To see what’s going on here, contrast what happens in weather with what happens with earthquakes. You and I both know that when we’re in Oakland, California, we’re at risk from earthquake, and when we’re in Norman, Oklahoma, we’re at risk from tornado. The difference is that when we’re in Oakland, we have no way of telling which days might be dicier than others. By contrast, in Norman, we can know, as far as several days out, that the period, say, 4-5 days from now is a period of heightened risk.

[This wasn’t always the case. A decade or two ago, the five-day weather forecast might suggest that, but it might well be contradicted by the next day’s four-day forecast for the same period. A day later, the three-day forecast might suggest the danger was back on. Whenever the forecasts vacillated in this way as the period of threat approached, they lost any utility they might have had. Years of attention to the observations of the atmosphere and to the computerized weather prediction models using those observations have reduced the incidence of this kind of problem. Note that the same challenge applies by the way, to forecasts of winter storms, or hurricanes, or spring flooding.]

This is where the “threat level” concept comes in. Picture the threat level going from green to amber during this 4-5-day period. That’s possible for the tornado outlook. It’s not possible for the earthquake problem, at least not today, in the year 2011. Note that it also doesn’t seem to be possible for the TSA color-coded security alerts at airports (can you remember the last time the security threat was anything but amber?).

All this leads up to “the day of.” The National Weather Service morning forecasts point out the possibility of severe thunderstorms. [After all, storm severity can manifest itself in other ways. Heavy rainfalls and flash flooding. Hail. Straight-line winds. Lightning.]. Later in the day, that may give way to “watches” of various kinds (think amber). Then, once the tornado has been spotted by radar through its vortex signature, or spotted visually, the watch gives way to a warning (think red!).

A pretty good plan! But things can go wrong; a safer future is not a slam dunk!

Some of the bumps in the road are inherent to this logic as just outlined; others are external. Let’s start with the internal issues, and the caveat hinted-at above. Turns out that the big ticket item for the Earth observations needed for the numerical weather prediction models is the Joint Polar Satellite System or JPSS; NOAA’s polar-orbiting satellites. The $800M needed to maintain continuity of these satellites was axed in the 2011 budget discussions – collateral damage in the partisan wrangling, continuing resolutions, and threatened government shutdown. Therefore (think amber) we can expect the eye-in-the-sky to go blind around 2017 if not before. Funding for these NOAA programs can’t be herky-jerky year-to-year without making the warning system on which we all depend herky-jerky again – back the way it was decades ago. Given everything going on with the tornado season, the Mississippi flooding, the Icelandic volcanoes, etc., such intermittent policy attention – and intermittent protection – is not what most Americans would want.

A similar issue is emerging; the network of weather radars on which the last stages of the warnings depend is aging. Those radars need to be replaced! The march of science and engineering has been such that when new, phased-array radars are deployed, tornado warning accuracies will continue to improve. But we need to appropriate funding, perhaps $1.5B spread over 5-10 years.

How about those externalities? Here are two. Let’s start with building codes. Stronger construction will not save the homes and buildings directly in the path of those 200-mph winds. But it will decrease the width of any damage swath.  And it is possible to equip homes with tornado shelters or safe rooms that can withstand not just the winds but the wind-born debris.

This distinction is vital. When thinking of this threat, don’t visualize your house  in a 200 mph wind. It won’t survive, but that’s actually the rosy scenario. Instead, picture your house exposed to an open-air rock tumbler, where the “rocks,” that is the projectiles (lumber, hunks of cinder block, roofs, cars) are flying around at 200 mph.

The second externality? Social realities. Even on the day of the tornado, families, small-business owners, and local governments have many additional urgencies on their minds. Getting the kids to school, medical care, keeping the shop from going bankrupt, holding that urban planning meeting. Ensuring that weather warnings take their rightful, useful place in this larger context requires that we be as disciplined about our social needs as we are about those physical threats. We need more social science, including communication of risk, and communication of uncertainty.

Overwarn? I hope I didn’t overstate.