Snow and poker

Snow and poker

Photo by Anastasia Yılmaz

Originally published 3 March 1986

Who could have pre­dict­ed this almost snow­less win­ter? Here it is March and I still haven’t tak­en my snow shov­el out of the base­ment. I checked the Old Farm­ers Almanac. I checked the news­pa­pers. As far as I can dis­cov­er, no fore­cast­er antic­i­pat­ed the remark­able deficit of snow in my part of the New England.

Suc­cess­ful­ly pre­dict­ing the weath­er is a lot like win­ning at pok­er: What it takes is part­ly sci­ence, part­ly art, and part­ly bluff. At the end of a sea­son, the TV mete­o­rol­o­gist cash­es in what­ev­er chips he has won and nev­er looks back. But the Nation­al Weath­er Ser­vice has a greater bur­den of pub­lic account­abil­i­ty. In 1985, the Ser­vice issued a report card on its own suc­cess. The report card came upon the 25th anniver­sary of the first weath­er satel­lite and coin­cid­ed with the instal­la­tion of a new CYBER 205 super­com­put­er at the Nation­al Mete­o­ro­log­i­cal Cen­ter near Wash­ing­ton, D.C. And what was the grade? I would say about a C+.

Even after a huge invest­ment in the best tech­nol­o­gy mon­ey can buy, the Nation­al Weath­er Ser­vice’s next day pre­dic­tions are still only 20 or 30 per­cent bet­ter than guess­es based on cli­ma­to­log­i­cal aver­ages. Fore­casts for 36 to 48 hours ahead are less suc­cess­ful. Twen­ty-five years of progress in mete­o­rol­o­gy have yield­ed only an 8 per­cent improve­ment in over­all accu­ra­cy for 48 hour forecasts.

The prob­lem is that the atmos­phere is an immense­ly com­pli­cat­ed sys­tem. Mete­o­rol­o­gists under­stand the prin­ci­ples of atmos­pher­ic dynam­ics well enough, but the busi­ness of apply­ing the prin­ci­ples to a con­ti­nent-wide (or world-wide) weath­er sys­tem remains impos­si­bly demanding.

In spite of past short­com­ings, satel­lites and super­com­put­ers con­tin­ue to offer the best hope for improv­ing long and mid­dle range weath­er forecasts.

A job for supercomputer

Since the mid-60s, satel­lites have pro­vid­ed almost con­tin­u­ous imag­ing of cloud cov­er over most of the earth­’s sur­face. Infrared sen­sors have extend­ed the cov­er­age into the night­time hours. The most recent satel­lites are equipped with instru­ments for mon­i­tor­ing atmos­pher­ic tem­per­a­ture, and water vapor, liq­uid water and ice in the atmos­phere. Satel­lite research con­cen­trates on atmos­pher­ic sound­ing: the remote sens­ing of phys­i­cal prop­er­ties of the atmos­phere as a func­tion of height above the surface.

And what becomes of this huge mass of data? Does it end up buried on mag­net­ic tapes? Can it be ana­lyzed fast enough to be of use in forecasting?

Only super­com­put­ers have the abil­i­ty to process weath­er data as fast as it is col­lect­ed. Some of the most pow­er­ful com­put­ers in the world are cur­rent­ly being used to mod­el weath­er here and in Europe, gob­bling up in the process tens of thou­sands of obser­va­tions a day.

New weather models

The new com­put­er mod­els of the Earth­’s atmos­phere are one of the most sophis­ti­cat­ed achieve­ments of con­tem­po­rary sci­ence. Work­ing mod­els of the Earth­’s atmos­phere have been con­struct­ed inside the pow­er­ful machines — with winds, and rain, and sun­light, and moun­tains, and con­ti­nents, and seas — and day by day the mod­els evolve accord­ing to the laws of physics. The weath­er­men strug­gle as best they can to make their lit­tle elec­tron­ic worlds match the real one.

All of this is part of a mam­moth multi­na­tion­al enter­prise aimed at pre­dict­ing the weath­er a few days down the line. Satel­lites soar, anten­nas beam data, tape dri­ves spin, microchips hum. And will it rain tomor­row? Well, it depends. At this time of the year in this part of the coun­try it has rained 10 per­cent of the time in the past. So if you fig­ure on a dry day tomor­row the chances are 9 out of 10 that you will be right. And the Nation­al Weath­er Ser­vice, with all of its satel­lites and super­com­put­ers, can do only mar­gin­al­ly better.

But the weath­er­men are try­ing. If they get a C+ for suc­cess, they deserve an A for effort. In some areas of fore­cast­ing they have made spec­tac­u­lar progress: Com­pare New Eng­land’s readi­ness for Hur­ri­cane Glo­ria with the almost com­plete sur­prise of the Hur­ri­cane of 1938.

I am will­ing to be patient and give the fore­cast­ers the tools they need to try to do bet­ter. In the mean­time, I’ll keep in mind what the Old Farmer’s Almanac up in Dublin, N.H., has to say: “We believe noth­ing in the uni­verse occurs hap­haz­ard­ly; there is a cause-and-effect pat­tern to all phe­nom­e­na, includ­ing weath­er. It is obvi­ous, how­ev­er, that nei­ther we nor any­one else has yet gained suf­fi­cient insight into the mys­ter­ies of the uni­verse to pre­dict weath­er with any­thing resem­bling total accuracy.”


Thir­ty years lat­er, the increased com­pu­ta­tion­al pow­er of today’s super­com­put­ers have result­ed in more accu­rate weath­er mod­el­ing and fore­casts. Still, bet­ter bring an umbrel­la, just in case. ‑Ed.

Share this Musing: