Monday, December 23, 2013

The Truth Is, You Gave a Lousy Forecast

"Of course I couldn't possibly be talking about you."

I'm resurrecting a blog title from an old TCHE article in light of recent snowfall forecasts (for the 12-21/12-22 period), most of which originated on Facebook and Twitter.  I hesitate to call these "forecasts," as they were simply regurgitated images of model output posted at face value, without any meteorological interrogation.  As an example, here is a popular image that was viral on social media sites by late Monday afternoon:

156 hr. "Total Snowfall" forecast from the ECMWF model initialized 1200 UTC December 16, 2013    
This eye-candy was often accompanied with hyperbole like, "LOOK OUT central IL!", or "Watching a major winter storm coming this weekend!"  For the meteorological novices reading,  this image is a forecast of "total snowfall" 6 1/2 days in advance from the European (ECMWF) model.  The image paints a heavy stripe of snow (~15-25 in.) from north-central MO through central IL and into northern IN.  You might be saying to yourself, "Uh, that didn't happen!"'re right!  In fact, most of this region did not receive any snowfall from the most recent cyclone, as seen below:

Actual snowfall verification from the 12-21/22-13 cyclone

 As you can see, the heaviest axis of observed snow fell roughly 200 miles further northwest than the model forecast and also totaled nearly 10 in. less in terms of magnitude.  

So, what went wrong?

Meteorologically, nothing!

While this was obviously a poor numerical weather model forecast, error only arises when a human interprets this solution as reality.  This is not necessarily a ploy to get people to stop posting such images (although that would certainly be nice!), rather I would like to see interrogation of the meteorological fields and discussions of features that would support or refute the forecast in question.  This is obviously not feasible on Twitter, yet I keep seeing such images posted.  Every snowfall. Every severe weather outbreak.  Pictures of various model fields with hype abound.  Where are the discussions of vorticity, thermal advection, and frontogenesis; or analysis of variables such as pressure, temperature, and specific humidity that are actually simulated by the governing equations driving the "sexy" model fields?  

They're around, just not on social media sites.  In fact, I closely followed the Area Forecast Discussions leading up to this event from WFO KLOT (Chicago) and KDVN (Davenport), and I must say they were quite impressive (you can find such discussions here).  Both offices discussed uncertainty in the model fields and provided physical meteorological reasoning behind their forecasts.  While these forecasts were far more accurate (in the true sense of the word), they were not the first forecasts for the event.  Similar to what many are seeing in main-stream journalism, there seems to be a rush to get the word out first, regardless of fact-checking or care for accuracy/precision.  It seems that simply being the first to announce a snowstorm a week in advance is more important than getting the forecast correct!

As the event drew closer, images kept popping up of  "snowfall" plots slowly shifting the axis of heaviest snow further northwest with each run.  I put "snowfall" in quotation marks, as this is not actually a post-processed model output field for the ECMWF, rather a simple 10:1 ice:liquid ratio multiplication, which can itself be a source of significant error depending on the vertical thermal profile.  These slow shifts to the northwest were not surprising to me, especially after analyzing low-level potential vorticity fields.

The point here is to suggest better communication practices for meteorologists when dealing with folks who may have little/no meteorological training.  Posting an image of 240 hr. snowfall as your forecast may draw you more followers, but where's the verification?  You're lucky most people have a short memory!  These posts are doing disservice to meteorologists everywhere and helping to perpetuate "the weatherman is always wrong" stereotype.  Forecasting is an imperfect science that should be expressed with probabilistic forecasts and associated uncertainty.

Let's strive to be better weather journalists.


Unknown said...

So much for the ECMWF always being the "best" model with winter storms (or otherwise). When I saw these posted early last week, my first thought was that even the surface pressure pattern of the ECMWF did not support near so far south a snowfall axis. The thermal and pressure patterns seemed inconsistent, or at least highly unusual. Of course as I recall at that time the GFS hadn't a clue about cyclogenesis. the early days of NWP one could count on subsequent forecasts of approaching cyclogenesis to track further and further to the left. While not as predominant as was once the case, my impression is that it's still much more common than later forecasts of cyclogenesis that actually occurs tracking to the right.

Scott Sabol said...

Hello Professor, my name is Scott Sabol, morning meteorologist at WJW FOX8 in Cleveland. You made excellent points here. This is why I personally don't harp on trying to pin down specifics that far out. The general public doesn't differentiate between a 12 hour forecast and a 240 hour trend via the EURO or the GFS. Viewers/the public will hold their hat on that 240 hr map as Gospel even if after the event passes you attempt to rationalize your reasons for posting it. I write about the psychology of why the general public can't handle weather forecasts.

or here

Let me know what you think!

Joseph Cooper said...


I saw these as well. However, if you looked at the RAW model output from the EURO during these runs the areas that were "forecast" to get heavy snow had temps. both at 2m and 850mb above freezing(well above it in some cases. So it came down to, as you said, interrogating it. This was a great post and something the field of meteorology needs to hear!

P said...

That was a well-written post mortem of this event. I agree completely. I think so much of a forecast depends on subtle differences, such as are found in winter. There was a threat of a large storm (and there was.) The storm was associated with large QPF which there was. What trended away from earlier progs was the lack of cold air in place, thus keeping much of the precip liquid. And it seemed clear to me that the difference from snow to rain in this event was a matter of small forecast error.

I agree with Victor in the probabilistic understanding. But I also would say that there are some parts of the event that are more predictable than others. And it varies from case to case. Experience matters. So keep the posts coming, but put caveats on the forecast. What could go wrong? What is largely unknowable? *What* are the uncertainties?

Paul Sirvatka

Anonymous said...

Follow a reputable source like Tom Skilling or the NWS on Facebook or otherwise. Model data now is often quite accurate, even a few days in advance, and these graphics pop up to give an indication of what may happen. Tom Skilling used graphics like this (though I can't remember how far out the forecast was) to identify the massive snowstorms of 1999 and 2011 days before they hit Chicago. I also have to say, in my experience as a meteorologist observing models, that the models this time just happened to be way off, even very close to when the snow/rain/ice started falling. I think these graphics can add some interesting information and speculation if you have the right disclaimer.

The largest tornado outbreak ever recorded was predicted at least 5 days in advance by the SPC:

Just one example of at least being informed on the possibilities of what could happen, if you have the right context.

Rich Putnam said...

Worst of all, 95% of these people aren't even meteorologists! They are weekend warriors who couldn't make it through the program, got a subscription to weatherbell, and now make the rest of us look bad.