Many public forecasts in America (e.g., zone forecasts) now are produced not by humans directly typing them, but by software that “reads” human-manipulated, digital forecast grids. The programs automatically translate the (sometimes) modified forecast database to text segments. Unfortunately, what these bots spit forth often sees no human eyes before it hits the wires; and it shows!
For background, I didn’t just happen upon these troubles recently; it has been going on since the advent of NDFD (National Digital Forecast Database). Instead, I finally felt compelled by two outstanding (in a bad way) recent examples to offer my personal opinions on the topic.
Read the first 2/3 of this one slowly, as if you were the robot voice narrating this for weather radio:
.REST OF TODAY…MUCH COLDER. RAIN…DRIZZLE…SNOW LIKELY LATE
IN THE MORNING…THEN OCCASIONAL RAIN…DRIZZLE…LIGHT FREEZING
DRIZZLE…SNOW EARLY IN THE AFTERNOON. SNOW LIKELY…AREAS OF
LIGHT FREEZING DRIZZLE LATE IN THE AFTERNOON. LITTLE OR NO NEW SNOW
ACCUMULATION. HIGHS IN THE MID 30S. TEMPERATURES STEADY OR SLOWLY
FALLING IN THE AFTERNOON. NORTH WINDS AROUND 20 MPH. CHANCE OF
PRECIPITATION 80 PERCENT.
As a professional atmospheric scientist, it’s embarrassing, or at least should be, that such convoluted gibberish gets out at all, much less so often.
A self-contradictory zone forecast piece appeared the previous day from the same office. Read carefully the part after “TONIGHT.”
.REST OF TODAY…MOSTLY CLOUDY. HIGHS AROUND 70. TEMPERATURES FALLING INTO THE 50S IN THE AFTERNOON. SOUTHWEST WINDS 10 TO 15 MPH.
.TONIGHT…CLOUDY WITH A 30 PERCENT CHANCE OF RAIN. COLDER. STORM TOTAL SNOW ACCUMULATION UP TO 1 INCH. LOWS IN THE UPPER 30S. NORTH WINDS 15 TO 20 MPH.
I included the text before “TONIGHT” only to show that snow was not in the forecast in or before that mention of snow accumulations. My reaction upon reading that was, “Say what?”
These are not isolated examples, nor are they restricted to the issuing office. In fact, I’ve heard similarly disjointed, confusing and/or self-contradictory forecast text while listening to weather radio all across this land. It is not confined to any particular office or region. Nor is it necessarily the fault of the human forecasters who may never have the opportunity to catch such nonsense in all products before they go public. The cold, sad truth is that, in many cases, there is such a large volume of stuff being generated and sent out that the people on station simply don’t have time to proofread it all thoroughly, and fix any errors or nonsense that creeps in.
This is a matter of insufficient staffing of offices for the volume of output demanded, and a matter of expecting far greater volumes of forecast output then the forecaster can devote to on his own, or even proofread. The solution, of course, is either
1. Unpopular with users (cut back on the number of forecasts to a level that can be quality-controlled without sacrificing time for meteorological analysis), or
2. Expensive (add local staffing specifically for quality control and/or to loosen the data-processing workload enough to allow better QC).
NDFD can be a very good thing! But like any digital operation, it must be monitored and quality-controlled to ensure not only a loss of embarrassing foul-ups, not only for mere adequacy, but for excellence! Forecasts such as those above, which are quite common nationwide and in aggregate, do not exemplify excellence.
Welcome to the increasingly automated world of public weather forecasting. Expect more of this in the future as the role of humans becomes a smaller and smaller part of the forecast process.
Meanwhile, feel free to share examples or offer your thoughts on how to solve this embarrassing problem. The one rule is: Don’t name any particular offices or forecasters. It goes well beyond that.
Mike U says
Yikes, Roger. Unfortunately, there exists no equality from office to office with respect to GFE (graphical forecast editor) and Text Formatter expertise. Some offices have a great GFE/Text Formatter program, while others do not. As a GFE focal point at our office, those examples disturb me, and if you ever see something like that coming out of the DDC office, e-mail me immediately! A ZFP period should never be any longer than 4 lines, no matter how complicated the sensible weather may be.
The other part of the problem is the way these Wx grids are populated. One of the big problems the formatters struggle with seem to stem from how we populate Wx grids. We are required to attach a coverage/category with each Wx element we forecast, tied to the PoP, and if you want to show moving weather in the Grids, then any specific point or edit area (read Zone Group) could have as much as 3 different coverages/categories AND different Wx elements within one period! Then, this opens up a whole other can-o-worms regarding the definition of a “one hour PoP”, which some offices use (including ours). It’s a complicated mess, but the first thing I’d like to see us completely do away with is the ZFP product. There is no point is having both a ZFP worded output and another set of words being generated off-station which you see on every NWS website’s “Point and Click” forecast. Then the arguement becomes “well, one is a POINT forecast, and the other is an AREAL forecast”. Given the advancements of technology, the ZFP has outlived its usefulness.
We’re stuck in this age of Graphical forecast — Text forecast “tug of war” trying to satisfy every customer from those who like the new graphical forecasting and derived products from NDFD and those who are still reliant on text and are a little more unwilling to advance to better means of getting the information (a lot of this is also financially driven, of course). How long will we be in this “tug-of-war”? Who knows.
But you are right, it seems like quantity over quality when it comes to this kind of stuff. There is only so much we can quality control — and the more complicated and high-impact the weather is, the more products we issue… so you can see how easy it is for things to fall through the cracks. That said, the examples you show are inexcusable coming from the supposed leading meteorological agency in the world.
Anonymous says
Thanks for the excellent insights, Mike.
I’m somewhat surprised there’s that much variability from office to office with GFE.
I’ve heard the solution of getting rid of the ZFP (zone forecast product) before, and agree with it. In this era of digital and graphical forecasts, create great online graphics, and let the user (whether with TV, EM, private meteorology, newspapers, etc.) drink in the forecast grids to make what he wants out of them, tailored to his own purposes.
Speaking as a union person, of course, I see management as a whole has been
1. Willing to advance in fits and spurts and support some useful and improvements, like creating a digitized forecast database and implementing storm-based warnings, but on the flip side,
2. Unwilling to let go of outmoded, decreasingly used, legacy products that serve only the lowest common denominator — a minority of customers who (as Mike alluded) are not advancing beyond ancient technologies and text consumption.
The net result is more and more and more new stuff, but the old stuff isn’t being dropped at a commensurate pace. The outcome is predictable: greater and greater workload, less time for careful real-time analysis of weather, and less quality control over what goes out. Hence, the gaffes such as in the above BLOG entry.
[If you are a local forecaster, when was the last time you had time to analyze — by hand and with due detail — the full suite of upper air charts and at least one surface map before noon on a day shift, in order to most fully understand the weather of the day that initializes your forecast?]
Number Two also explains how we still see text forecasts in LOUD…RIDICULOUS…ALL CAPS FORMATS…WITHOUT PROPER PUNCTUATION SUCH AS COMMAS…SEMICOLONS…AND PARENTHESES. Why? Because someone, somewhere, perhaps, maybe, might still be using a teletype emulator, and heaven forbid we risk offending or alienating this thousandth of a percent or less of those who get forecasts.
Hello…this isn’t 1978! It’s long past time to cut the anchor on teletypes, folks, and as Mike advocates, probably time to get rid of the ZFP also.
Any other ideas or alternative views?