Bet­ter data and mi­cro­sta­tions are mak­ing fore­cast­ing bet­ter—but never per­fect

Newsweek International - - NEWS - BY KEVING MANEY @kmaney

The Weather Rev­o­lu­tion Will Be Tele­vised

IT’S EASY TO take weather fore­cast­ing for granted. Ev­ery goofy TV me­te­o­rol­o­gist told us more than a week ahead that Hur­ri­cane Irma was turn­ing into a gi­ant storm that would nail the U.S. East Coast. Given the in­com­pre­hen­si­ble com­plex­ity of weather, such a feat is like pre­dict­ing to­day who will win the 2020 pres­i­den­tial elec­tion. (Crowd­sourced site Paddy Power gives Oprah 33-to-1 odds.)

Over the next few years, tech­nol­ogy will make weather mod­el­ing even more pre­cise and use­ful, which is good news as the planet en­ters an era of worse storms driven by cli­mate change. Not only will mod­els get a bet­ter bead on storms that can wreck things, but su­per­speci c fore­casts will in­te­grate with ev­ery­day ac­tions. An app might read your cal­en­dar and au­to­mat­i­cally let you know that there is go­ing to be a rain cloud di­rectly over your pa­tio the mo­ment peo­ple ar­rive for that bar­be­cue a week from Satur­day.

Great weather mod­el­ing needs four key com­po­nents: data, com­put­ing power, math and sci­enti c understanding. They all feed o one an­other, and tech­nolo­gies such as arti cial in­tel­li­gence, robotics and the “in­ter­net of things” are hav­ing a big im­pact. While no one is ex­pect­ing an E=mc2–like break­through that sud­denly re­veals weather’s se­crets, fore­casts are steadily go­ing to get more ac­cu­rate fur­ther out. “We’re mak­ing im­prove­ments in­cre­men­tally, but if you look back over decades, it’s re­ally phe­nom­e­nal,”

Mary Glackin, head of science and fore­cast­ing for IBM’S Weather Com­pany, tells me.

Data is the most el­e­men­tal com­po­nent. The data rev­o­lu­tion in weather started al­most 100 years ago, when Pavel Molchanov in Rus­sia in­vented the ra­diosonde—a balloon that car­ried a few sen­sors and a ra­dio trans­mit­ter into the at­mos­phere, send­ing read­ings back to Earth. In 1960, NASA won the race to put up a weather satel­lite, TIROS 1. It sent back the rst pho­tos of the Earth’s cloud cover. Ever since, ma­jor coun­tries have been pack­ing the skies with weather satel­lites, lead­ing up to the GOES-16 satel­lite launched by the U.S. in late 2016. It trans­mits a con­stant stream of de­tailed im­ages of weather sys­tems. David No­vak of the U.S. Na­tional Weather Ser­vice calls GOES-16 “one of the big­gest ad­vance­ments we’ve seen.”

Satel­lites give us the macro view, and in­ter­net of things de­vices bring in mi­cro­data. These cheap sen­sors can re­lay read­ings from al­most any­where—street­lights, buoys, we­b­cams. The Weather Com­pany has 250,000 tiny weather sta­tions around the world. For per­spec­tive, Star­bucks has only about 25,000 stores world­wide. “We work with cit­i­zens that have these [sta­tions] in their back­yards,” Glackin says. “It helps ll in the blanks.” Drones and ro­bots will play more of a role too, car­ry­ing sen­sors to places where nei­ther hu­mans nor satel­lites can ven­ture.

All of this data has to get fed into math­e­mat­i­cal equa­tions that rep­re­sent the way sci­en­tists think the weather works. More data help the mod­els spit out bet­ter re­sults, and the re­sults in­form the sci­en­tists, who can then make bet­ter mod­els. But run­ning so much data through supremely com­pli­cated mod­els takes fan­tas­tic com­put­ing power. In 1922, Bri­tish math­e­ma­ti­cian Lewis Fry Richard­son set math­e­mat­i­cal weather mod­el­ing in mo­tion by pub­lish­ing his study, Weather Pre­dic­tion by Numer­i­cal Process. But, he noted, do­ing the cal­cu­la­tions fast enough would take 64,000 peo­ple work­ing in a room si­mul­ta­ne­ously. Com­put­ers have made that a lit­tle eas­ier. In early 2016, the Na­tional Oceanic and At­mo­spheric Ad­min­is­tra­tion turned on two su­per­com­put­ers that each tripled the speed of its pre­vi­ous ca­pa­bil­ity. These are some of the most pow­er­ful com­put­ers in the world.

The next step, hap­pen­ing now, in­volves AI learn­ing from mul­ti­ple mod­els. TV weath­ergeeks of­ten talk of com­pet­ing Euro­pean and Amer­i­can mod­els, but there are lots of mod­els with multi-let­tered ab­bre­vi­a­tions—cmc, NAVGEM, CFSV2, Can­sips—all work­ing di er­ently and churn­ing through di er­ent data sets. The new wrin­kle is that AI can suck up re­sults from all the mod­els and learn from them. Glackin’s group at IBM runs through re­sults from more than 160 mod­els mul­ti­ple times a day. “We ad­just fore­casts based on how di er­ent mod­els are per­form­ing at di er­ent points on Earth and on di er­ent time scales,” Glackin says. “So we take the best of the best and do ma­chine learn­ing on that.”

The pace of im­prove­ment in fore­casts is speed­ing up, which means we’ll get more ac­cu­racy fur­ther ahead of time, and fore­casts can get more gran­u­lar, pre­dict­ing the weather for a small patch of the planet. That, of course, is tremen­dously help­ful for pub­lic safety, but it also aids crit­i­cal de­ci­sions in ma­jor in­dus­tries like avi­a­tion, en­ergy, con­struc­tion and agri­cul­ture.

It will also help our ev­ery­day de­ci­sions. “We’re in the mid­dle of this big rev­o­lu­tion in how we use weather,” says Bill Gail, co-founder of Global Weather Cor­po­ra­tion in Boul­der, Colorado. “In a decade, those of us who al­ready use weather in­for­ma­tion will be us­ing it 100 times as of­ten and won’t even know it.” To­day, most peo­ple make plans for travel or a bike ride or build­ing a still in the back­yard, and then look at the weather, or vice versa. But go­ing for­ward, your


still-mak­ing app will un­der­stand what you want to ac­com­plish and how long it will take, look at your cal­en­dar and hy­per­local weather fore­casts, and tell you which day would be best to get set up to make a nice full-bod­ied rye.

Just don’t ex­pect weather fore­casts to get per­fect any­time soon. Like the global econ­omy and pol­i­tics, weather is too com­pli­cated and capri­cious to ever ex­actly model. “At some range in the fu­ture, it doesn’t mat­ter how good your com­puter mod­els are, the way the at­mos­phere is chaotic in nature, your pre­dictabil­ity will break down,” says Greg Carbin, fore­cast op­er­a­tions branch chief of the Na­tional Weather Ser­vice.

Tor­na­does, for in­stance, are the mod­el­ing equiv­a­lent of Don­ald Trump tweets—no amount of data or su­per­com­put­ing can fore­see what will cause them or how much dam­age they will do.

PIC­NIC PICK: Ad­vances in fore­cast­ing will let us use weather data 100 times as of­ten as we do now, and of­ten with­out even know­ing it. +

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.