Automation, weather warnings, and social media

As Twitter, Facebook, and other social media began to take hold in weather centers and news rooms across the country a few years ago, one of the earliest conversations was about whether and to what extent automation should play a role in sharing weather information, especially during severe weather situations.  This is a bigger deal than it appears at first glance: The interactions between the immediate, reliable, and consistent provision of warning information and the trust relationships between communicators and those in the danger zone are complex and not as well-understood as we would like.  

Before I go further, I want to be very clear about my purpose here: I am not writing to criticize folks who are trying to get critical information out to people in harm’s way, often doing so as early adopters with few resources and little backing.  However, in watching Sunday’s tornado outbreak unfold in the midwest primarily via radar and Twitter, I was struck by how poorly some automated social media updates come across.  More importantly, there’s a flaw in the system that could leave some social media users to believe they’re not in harm’s way when they actually are.  I show numerous examples, and where I’ve been critical, I’ve taken steps to anonymize the source.  My goal here isn’t to call anyone to the carpet but to raise an important issue in communicating severe weather information.

To Be or Not To Be (Automated)

When it comes to automating weather warning information on social media, there appear to be two camps.  Some folks, like social media juggernaut James Spann, prefer typing out warnings on Twitter, saying that it allows him to communicate the warning in his own voice.  Others argue automation has a role to play for at least two reasons: Getting the word out as soon as possible and backstopping a meteorologist in providing critical information across multiple platforms when they might be too busy to tweet in a timely fashion (never mind when someone may be away from social media entirely).  Even Spann uses some automation – warnings are automated to his Facebook fan page, but he’s on record as saying Facebook is terrible for severe weather updates, and he’s got a point – but that’s another post.

Warning tweet from @wralweather

Warning tweet from @wralweather

Personally, I am squarely in the automation camp, but if and only if it is done properly. My definition of “properly” here is a blend of good emergency radio communication (in an emergency, “where” and “what” come first, everything else secondary – if that’s all you’re able to get out, crews can still get to you) and something akin to the Turing Test.  To be most effective, severe weather updates need to be immediate, relevant, and readable – and they need to stand on their own. We can’t count on people being both willing (Twitter’s click-through rate is south of 6%) and able (device limits, bandwidth, etc.) to click the link.

When I worked with our tech team at WRAL to automate warnings to the station’s @wralweather account on Twitter almost five years ago, we took the warning feed, created a bare-bones format that allows us the most characters for the important information (“where” and “what”), and had the program write tweets that read like a human typed them.¹  I and other meteorologists and reporters supplement the automated tweets with additional information across platforms, including photos of the storm or damage, radar images, storm reports, ETAs, and the like.  But the automated tweets provide an important backstop to what we do manually, they’re faster than any of us could type manually, and if that’s the only thing someone gets, they would at least know of the danger and the need to shelter or at least seek additional information.

Automatic for the People

Many broadcasters and emergency management sources felt that automation was an important tool and looked to add it to their suite of offerings.  A handful of services entered this space early on, but none took hold quite like iNWS – Interactive National Weather Service.  These alerts, offered only to NWS core partners like media and EMAs, are somewhat customized emails and text messages.  They are fast and, most importantly for cash-strapped departments, free.  They’re a great way to get severe weather information direct from the source.

Some enterprising folks then figured out that these iNWS alerts could, through a hack involving third-party services like TwitterMail or Tumblr, be automated to social media outlets.  This arrangement met a couple of primary goals: Getting the word out automatically and reliably and doing so quickly.  However, when it comes to being easy to read, they leave something to be desired:

Sample automated alert.

Sample automated alert.

This is especially clear when these heavily formatted alerts come from different sources – or when they’re seen next to a more clearly formatted message communicating the same thing:

Comparison of alerts

Comparison of alerts

One of these is not like the others – and in a very good way.  Neither of the automated alerts would pass a Turing Test; they are clearly computer-generated and they are difficult to read at a glance. Beside that, they exceed the 140-character maximum for Twitter in large part because they contain extraneous information.  “New event” is superfluous, and since tornado warnings are issued on detection, they’re valid from issuance until expiration.  The issue time is irrelevant, and if your system is working well, redundant, since it should be within moments of the time the tweet was published.

Contrast that with the alert from the Weather Channel’s breaking account, @TWCBreaking.  There’s still some formatting there, but the key information – the “what” and “where” – are front and center.  They’ve even included the expiration time.  Like the automated tweets, TWC’s includes a link to more information, which is fine, but the tweet stands on its own.

A Fatal Flaw?

My beef with these machine-generated, user-unfriendly alerts is nothing new, but on Sunday, I noticed something far more important for the first time.  At one point, I was scanning my timeline and I saw these two tweets back-to-back:

Sample severe weather tweets.

Sample severe weather tweets.

If you look closely, you’ll see the one on the bottom has three counties, while the one on the top only listed two. I took a closer look, and sure enough, they both referred to the same warning.  In other words, one tweet left a county out of the warning!

Because of the extraneous information – “New event”, “from 11/17/2013 5:43PM to” and repetitions of the word “County” – the tweet was too long to fit in Twitter’s 140-character limit and was abbreviated by TwitterMail, removing vital information:

Automated tweet with a missing county

Automated tweet with a missing county

This alert was generated by iNWS and delivered to Twitter via a service called TwitterMail, which detected the overage and included a link to the full message:

What comes up when you click the link to the above tweet

What comes up when you click the link to the above tweet

Notice the tweet left off Dearborn County entirely.  Were I a resident there scanning my timeline, I would have missed that my county was under a tornado warning!

This was far from the only instance. Kenton county was left off this tweet:

Tornado warning tweet with missing county

Tornado warning tweet with missing county

Full information about the warning, showing all counties affected

Full information about the warning, showing all counties affected

This one left two counties out of the tweet:

Tornado warning tweet with a missing county

Tornado warning tweet with two missing counties

Full information about the warning, showing all counties affected

Full information about the warning, showing all counties affected

This wasn’t exclusive to warnings delivered by TwitterMail.  Alerts via Tumblr suffered from much the same issue, leaving Trigg county out:

Tornado warning tweet with missing county

Tornado warning tweet with missing county

Full information about the warning, showing all counties affected

Full information about the warning, showing all counties affected

Alerting Everyone, Automatically

Again, my goal here isn’t to beat up on people trying to spread the word of dangerous weather.  It is, however, to point out that we could be unintentionally making it harder for people in harm’s way to determine if they’re in a warning, and potentially, we could be leaving people with the impression they aren’t under a warning when they actually are.

This is not what any of us want, least of all the folks who figured out these connections to automate the warnings in the first place.  Weather folks are often tech early adopters, and in the early days of Twitter, often it was the weather folks who were first in their station to jump on the platform.  The duct-tape-and-baling-wire solution for automating tweets was fine in the early days, but it has got to go if we’re to consistently communicate severe weather threats to the public.

There are some that do it well:

Example of a complete warning tweet

Example of a complete warning tweet

Initially, I thought this was automated, but it was a copy-and-paste from NWSChat. A little more formatting than I’d like, but the extra information is useful:

Example of a complete warning tweet

Example of a complete warning tweet

Here’s another batch of good – as in “complete” and “easy to read” – tweets of tornado warnings:

Series of complete warning tweets

Series of complete warning tweets

(To be honest, I’m still surprised there hasn’t been a solution offered by any of the TV weather graphics vendors or the various notification service/app providers for this.  In the meantime, if it’s worth doing at all, it’s worth doing well, and we need to make sure we’re not unintentionally leaving folks out or giving the wrong impression.)

___

¹  They’re not perfect – I’d like to have the text be more informed by polygons, so we could tune them geographically: “northern Durham and Orange counties”, and perhaps even “including the city of Durham”, instead of just “Durham and Orange counties”.  I would like to include the expiration time, as well, when space allows, but toss it when I need the characters for places.  Regardless, the point here is the technology existed then and definitely exists to automate warnings in such a way that they are readable and useful.

Advertisements

About Nate Johnson

Meteorologist, instructor, blogger, and podcaster.
This entry was posted in Uncategorized and tagged , , . Bookmark the permalink.

7 Responses to Automation, weather warnings, and social media

  1. I’m not in TV but am squarely in social media with our site. You have some points. One BIG (and I mean that) problem with warnings is the era of polygons. A tornado warning often includes only a portion of a county or counties, not the whole. For instance, up here in Philly a couple of counties are 30-40 miles from east to west. If we tweet out “Tornado Warning until 6 for Berks County” when the warning is for the western part, not the eastern…the assumption is people in the east may think they’re under the warning when the tornado may be tracking away north of them and not towards them.

    I really think the warning over twitter message is muddled when we list whole counties. I know in automation there isn’t much of a choice but I do think for some geographically larger counties posting a whole county as being under a warning is crying wolf and may drive some folks into thinking “it’ll never happen.” Hopefully, in the future we drill past the “county” and more into town with warnings over twitter.

    I think it’s better to be as specific as possible with warnings and I do appreciate James Spann’s approach in using his own voice as much as possible. In the last couple of years I’ve gone more into naming towns that are in the path of a tornado warning or areas covered by a severe thunderstorm warning. I hope the technology is such that we can get see more of the community/part of the county under warning getting automated in the future. My followers and readers appreciate that…and yes, it sends a clearer message to the end users who are increasingly using twitter as a medium to get their news.

    If any lesson can be gathered on the TV station side — your producer(s) should be able to quickly transfer a warning from the NWS and get it geographically specific whenever possible if the producer is on duty and can tweet out. Sometimes, automation is the only course of action but I really believe that the less automation and the more specific you can get the warning, the better.

    In the end, it’s about saving lives. 5-10 seconds does matter but I also think pinning down the specific towns that are in the path of a tornado is just as important, if not moreso, than tweeting out automatically. I’m probably in the minority on this but I think the more social takes over in this realm, the better, more accurate, and specific the information has to be.

    Tom
    Phillyweather.net

  2. RDale says:

    My question – is there any evidence that people use this information for life safety decisions? I know they get RT’d a ton, but is there anyone who is just sitting around the computer and decides to head to Twitter to do a search for “tornado warning toledo”? I’m not saying that social media is useless when doing this, and Nate is spot-on that those automated ones are useless (and I stopped following people that might have stuff I want to read because they mix the automated junk in there) but I am looking for any studies.

  3. Also, if you had followed just ONE of these accounts Sunday, you would have received at least 130 tweets that were tornado warnings (400 if all warnings were enabled) or hundreds more if updates were included. That’s got to lose you some subscribers and most people don’t care unless they live in that county.

  4. Pingback: Simple, Sneaky and Clevage | Digital Meteorologist

  5. Pingback: Warnings, Warnings and More Warnings | Digital Meteorologist

  6. Pingback: Automation, weather warnings, and social media | Kelsey L. Reed

  7. Pingback: Tornado Quest Gee-O-Science Links For Nov. 17 – 24, 2013 | Welcome To Tornado Quest

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s