Long-range weather forecasts don't do better than 'predicting' the same as
historical records
Q: How accurate are the long-term (10 to 15 day) weather forecasts on the
Internet (such as AccuWeather and The Weather Channel)? Seems to me they are
little better than wild guesses after the first couple of days. Lanney,
Sandia Park, New Mexico
 South
of Dimmitt, Texas. 1995. The lead time for tornado warnings has more than
doubled in the last decade with improved radar and satellite images. [National
Oceanic and Atmospheric Administration (NOAA)/National Severe Storms Laboratory]
A: This is one hard question. And I’m not the only one who has had trouble.
Who could better determine the facts than the weather folk themselves? Yet, in
1997, National Severe Storms of the NOAA stated, "Information on the quality of
public weather forecasts is difficult if not impossible to obtain." Amen.
In 1998, the American Meteorological Society, (AMS) analyzed how well
forecasters predict weather. Weathermen use various "skill" systems to score
forecasts, where "skill" has a technical meaning: The skill score is the
percentage improvement of the forecasts compared with historical climate
records. The AMS found:
- Extended forecasts (beyond Day 14) show no skill. "No verifiable skill
exists or is likely to exist for forecasting day-to-day weather changes beyond
two weeks. Claims to the contrary should be viewed with skepticism."
- Long-term forecasts (Day 7 to 14) demonstrate little skill. John Q Public
could do as well simply checking what the long-term average weather is for
that date.
- Short-term forecasts (Days 1 to 7) are better. Temperature predictions
show "considerable skill on day 3, with skill decreasing with time until
generally only marginal skill remains by day 7."
The 1998 AMS report, however, needs updating to reflect accelerating
improvement in forecasts, says Bob McDavitt, a meteorologist and weather
ambassador at MetService in New Zealand. McDavitt finds "useful skill" out to
about Day 7 now.
A forecast is simply an extrapolation from present conditions, based on a
weather model. The mathematical model uses atmospheric-physics theory to predict
the weather. Past weather patterns contribute to the picture, as do current
observations. "We gather data, extract a pattern, experiment with the results
(up to100 times), and publish the output of these experiments — the weather
forecast," says McDavitt.
Most people find the results useful, even for extended forecasts. The
National Weather Service (NWS) asked folks in Buffalo, New York what they
thought of the NWS extended forecast. Of the 500 that responded, most said
"good" — a remarkable achievement, considering what weather entails.
Real weather mixes chaos with pattern. Moreover, no model can include the
myriad possibilities inherent in chaos — it’s too vast. An iceberg calves off
Greenland’s coast. That chaotic event affects Burma’s weather and throws the
Burma forecast off, a bit.
The National Severe Storms of the NOAA, however, has useful hints for we who
struggle with forecast uncertainties. Which — of the various TV, newspaper, and
Internet forecasts — to believe?
- If all the forecasts agree, they are much more likely to be correct than
if they disagree.
Average the predictions. The AMS found, for example, that the average of
all forecasted maximum temperatures (for day 2 and beyond) has a lower error
than that of any individual forecast.
Finally, an encouraging word: forecasting skill on Day 5 has more than
doubled since the late 1970s. Forecasts improve as weathermen use bigger smarter
computers to calculate model predictions in greater detail. The models better
reflect reality, as we better understand the physics of weather. Observations
are more timely and comprehensive, as we see more with ever-growing networks of
radar, satellites, and automated equipment.
Further Reading:
American Meteorological Society (AMS): Weather Analysis and Forecasting, 1998
National Severe Storms of the NOAA: Verification of public weather forecasts
available via the media by Harold E. Brooks, Arthur Witt, and Michael D. Eilts
Wind & Weather, weather instruments: Barometer for $60
National
Weather Service Forecast Office in Buffalo, New York: Perceived accuracy
(Answered Aug. 27, 2004)
Click for printer version.
| 
