Update, 3:05 p.m.: See update at bottom of post for clarification on editing features in The Weather Channel’s app that allow it to override automation
From 10:09 a.m.
Just because your favorite weather app says it’s so, doesn’t make it right.
In a driving rain storm Monday morning, some Washingtonians saw a surprising forecast on their smartphones: 5-8 or even 6-10 inches of the white stuff was in the works according to the The Weather Channel app. Yet hardly any snow ultimately fell.
— Cam Ayoub (@CamAyoub21) February 3, 2014
The Weather Channel app forecast diverged from official predictions from the National Weather Service, television outlets and the Capital Weather Gang, which called for no snow to a few sloppy inches at the most. Understandably, the discrepancy led to confusion.
Why were forecasts so different, numerous Capital Weather Gang readers wondered?
@capitalweather the weather channel app predicts reston getting 5-8 slushy inches today. Thoughts?
— Jackie Laurenzi (@jlaurenzi1114) February 3, 2014
And the answer? The forecast on many of your favorite smartphone weather apps is not created by a human being. Instead, it is generated by a computer. It does not harness the expertise of a local forecaster who understands a region’s weather peculiarities, has a wealth of historical knowledge and can correct for obvious computer errors.
Yes, Monday’s forecast was a tough one with the D.C. area near the rain-snow line. Computer model simulations were all over the map – some forecasting heavy rain, others heavy snow, and most a mix. But given the recent warmth and tendency for some models to be “too cold” in simulations, human forecasters were able to largely discount the snowy solution The Weather Channel app was touting.
I don’t mean to pick on The Weather Channel (it just provides a current example) whose app is well-designed and contains many useful features. Most app forecasts – irrespective of the supplier – are computer-generated and subject to spitting out head-scratching information.
In another baffling case Monday, CWG Twitter follower Erin Coyle said her Dark Sky app – which forecasts precipitation time of arrival on a visually stunning display – said rain was going to fall in Frederick, Md. when it was snowing all morning.
@capitalweather dark sky app thought that it was raining all morning in Frederick but I don’t have a screen cap
— Erin Coyle (@eccoyle) February 3, 2014
James Spann, the celebrated Birhimgham, Al. broadcast meteorologist, endearingly calls these tools “(blank) apps” (where the blank is a word rhyming with app).
“Unfortunately most weather apps use automated computer model data with no quality control, and you wind up with very poor quality forecasts that are not only useless at times, but also very deceiving,” Spann wrote in a rant on his blog in August, 2012.
In my view, these apps are fine for very general forecast information on your average weather days. But when the forecast is complex or a lot is at stake in the forecast, I don’t think an automated app can replace a local forecaster as a trusted source. An app cannot provide the depth of a local forecaster or a sense of the forecast confidence. (Disclaimer: I do realize this argument is self-serving, given the information we provide, but is also speaks to the value provided by TV weather teams, the National Weather Service, consulting meteorologists, etc.).
For its part, The Weather Channel is heavily investing in improving its app and related digital products. On Sunday, I toured The Weather Channel’s headquarters in Atlanta and was shown its impressive eighth floor, home of multiple digital development teams – working on products for iPhone, iPad, and Android.
The Weather Channel – leveraging crowd-sourced data from the personal weather station network of Weather Underground (one if its sister companies) and other sources – is working to refine local predictions across digital and television platforms.
“Forecast on Demand is a new [Weather Channel] technology that incorporates elements of nowcasting but is able to create a detailed forecast–at the request of a user–for more than 2 billion points around the globe,” describes a recent Weather Channel technology profile in the publication Fast Company. “…Forecast on Demand instantly generates real-time forecasts for that specific geographical point, using the freshest information available from its more than 75,000 data sources. This upends traditional forecasting, which relies on pregenerated predictions [from models].”
The integration of crowd-sourced data into weather models offers considerable promise for improving localized forecasts. But when these apps can overcome some of the thornier forecast problems like D.C. experienced Monday remains to be seen.
Even when the day comes when such apps can make a perfect forecast, human expertise will still be needed to help interpret the forecast and guide the best weather-related decisions. This is a challenging problem in its own right as the gridlocked highways outside The Weather Channel’s headquarters one week ago made abundantly clear – a solid forecast notwithstanding.
Update, 3:05 p.m.: Bruce Rose, a scientist at the Weather Channel, emailed me to clarify that forecasts provided on its apps originate from automation but go through an editing process.
“We have a robust human intervention capability called DiGit (Digital Intervention Tool) in Atlanta…it is manned 24×7 by 3 shifts of very skilled forecasters,” Rose said in an email.
He added: “In retrospect, I might call this a poor forecast [on Monday] … or the GFC [Global Forecast Center] might have too much on their plate (so many points to worry about in active weather situations); but I wouldn’t say the problem was automation.”
Although The Weather Channel’s forecast was off early that morning and the day before, Rose said forecasters remedied the situation. “The forecast center changed …. forecasts quite dramatically in the period between 4 a.m. and 9 a.m. to jibe with current trends,” he said.