AI, a new asset for weather forecasters

Weath­er fore­cast­ing is now a com­mon prac­tice for which mete­o­rol­o­gists require atmos­pher­ic sim­u­la­tion mod­els. Météo-France reg­u­lar­ly rein­forces its inten­sive com­put­ing capac­i­ties, qual­i­ty of numer­i­cal weath­er pre­dic­tion mod­els, and the use of obser­va­tion data to cal­i­brate them.

The research car­ried out at CNRM (Cen­tre nation­al de recherch­es météorologiques), in col­lab­o­ra­tion with sev­er­al nation­al and inter­na­tion­al part­ners, focus­es on under­stand­ing the atmos­pher­ic envi­ron­ment and its inter­ac­tions with con­ti­nen­tal and ocean­ic sur­faces. The CNRM devel­ops inves­tiga­tive tools for observ­ing and improv­ing our under­stand­ing of process­es at obser­va­tion sites. It also works to devel­op mod­els for numer­i­cal sim­u­la­tions our ever-chang­ing atmos­phere and its inter­faces, doing so on a scale of hours, days or much longer. These mod­els oper­ate on mul­ti­ple spa­tial scales, from sev­er­al tens of metres to hun­dreds of kilo­me­tres – or even on a glob­al scale.

These weath­er mod­els are used to fore­cast the weath­er and study cli­mate change through cli­mate pro­jec­tions. There are sev­er­al mod­el­ling tools, such as Arpège and Arome. Arpège is used to sim­u­late the atmos­phere of the entire plan­et, cov­er­ing Europe with a mesh size of around 5 km and the rest of the globe with mesh sizes rang­ing from 5–24 km. Arome cov­ers a domain lim­it­ed to met­ro­pol­i­tan France and neigh­bour­ing coun­tries with a hor­i­zon­tal res­o­lu­tion of 1.3 km, and is also deployed over over­seas regions.

Arome has been designed to oper­ate over a part of the world only, which may be cen­tred on some over­seas ter­ri­to­ries and oth­er loca­tions of inter­est. The mod­el is used to improve the short-term fore­cast­ing of haz­ardous phe­nom­e­na such as heavy Mediter­ranean rain­fall (‘Mediter­ranean episodes’), severe thun­der­storms, fog or urban heat islands dur­ing heat waves. It allows sim­u­la­tions to be car­ried out with very fine res­o­lu­tion, by slic­ing the atmos­phere into small cubes (known as ‘mesh­es’) to solve the equa­tions gov­ern­ing the phys­i­cal process­es in the atmos­phere and at the interfaces. 

The raw results of these sim­u­la­tions must be processed to pro­duce fore­casts in an oper­a­tional frame­work. In this con­text, CNRM is increas­ing­ly using advanced sta­tis­ti­cal meth­ods, includ­ing arti­fi­cial intel­li­gence (AI), to com­bine past fore­casts with what has been observed, to some­how adjust the mod­els so that they are as rel­e­vant as pos­si­ble to the observations.

AI is a method­olog­i­cal tool that has been emerg­ing for sev­er­al years. It can speed up the exe­cu­tion of a weath­er mod­el and thus reduce its com­pu­ta­tion­al cost. For exam­ple, one of the most expen­sive aspects of mod­el exe­cu­tion is solv­ing the com­plex equa­tions that gov­ern radia­tive trans­fer through the atmos­phere, through clouds, and the inter­ac­tions of light rays with the land or ocean sur­face. Future research could speed up mod­el exe­cu­tion by sub­sti­tut­ing some explic­it res­o­lu­tions of the phys­i­cal equa­tions with results obtained through machine learn­ing. Some work goes so far as to sug­gest the out­right replace­ment of fore­cast­ing mod­els solv­ing phys­i­cal equa­tions with deep learn­ing emu­la­tion of fore­cast­ing systems.

Anoth­er appli­ca­tion of AI is the post-pro­cess­ing of mod­el­ling results for weath­er fore­casts. This process allows the cor­rec­tion of cer­tain sys­tem­at­ic errors in the out­put of the mod­els to make the fore­cast more con­sis­tent with obser­va­tions in sim­i­lar cir­cum­stances. It also makes it pos­si­ble to detect struc­tures cor­re­spond­ing to phe­nom­e­na at stake, which is all the more dif­fi­cult as fore­casts are increas­ing­ly made with­in the frame­work of ensem­ble fore­casts, mak­ing it pos­si­ble to bet­ter account for the uncer­tain­ties of the fore­casts. Among the work in progress at the CNRM, we should men­tion, for exam­ple, work aimed at improv­ing the detec­tion of par­tic­u­lar­ly vio­lent thun­der­storms in the results of numer­i­cal sim­u­la­tions, which are extreme­ly impor­tant phe­nom­e­na to detect but dif­fi­cult to fore­cast¹.

One approach to pro­duc­ing increas­ing­ly rich weath­er fore­casts is to reduce the spa­tial scale on which they oper­ate, there­by rep­re­sent­ing phys­i­cal process­es that can be solved on a small­er scale. As men­tioned ear­li­er, the Arome fore­cast sys­tem, for exam­ple, cur­rent­ly oper­ates at a hor­i­zon­tal res­o­lu­tion of 1.3 km. A high­er res­o­lu­tion would allow even small­er scale process­es to be rep­re­sent­ed, for exam­ple those lead­ing to very localised high-impact thun­der­storms. How­ev­er, a high-res­o­lu­tion mod­el does not guar­an­tee that the loca­tion of the most intense thun­der­storms can be accu­rate­ly iden­ti­fied. Thun­der­storms are scat­tered phe­nom­e­na, not wide­spread events over large parts of the ter­ri­to­ry, as are heat waves, and are there­fore more dif­fi­cult to pre­dict in terms of their occur­rence, loca­tion, and inten­si­ty. Obser­va­tions play a fun­da­men­tal role in ini­tial­iz­ing weath­er fore­cast­ing sys­tems, with meth­ods known as ‘data assimilation’.

Work is under­way to exploit more inno­v­a­tive obser­va­tions such as so-called ‘oppor­tu­ni­ty’ data, col­lect­ed by par­tic­i­pa­to­ry tools such as indi­vid­ual con­nect­ed weath­er sta­tions. These data are col­lect­ed by the com­pa­ny that mar­kets these instru­ments and allow us to enrich our knowl­edge even if the intrin­sic qual­i­ty is not at the same lev­el as that of the con­ven­tion­al network.

There are also devel­op­ments in the obser­va­tions obtained from air­craft and ships, as well as from bal­loons, equipped with probes, which are sent into the atmos­phere sev­er­al times a day. And then, of course, we have many obser­va­tions made by satel­lites, either geo­sta­tion­ary or defil­ing, which pro­vide a view on very large spa­tial scales. The for­mer pro­vides us with a per­ma­nent view of the face of the Earth that con­cerns us with infrared sen­sors, for exam­ple. The lat­ter cir­cle the Earth sev­er­al times a day and pro­vide obser­va­tions along their tra­jec­to­ries with more detailed obser­va­tions. Sci­en­tif­ic and tech­no­log­i­cal advances make it pos­si­ble to dis­tin­guish small­er and small­er mete­o­ro­log­i­cal objects from space, which is cru­cial for char­ac­ter­is­ing the state of the atmos­phere and improv­ing fore­casts of its evolution.

Interview by Isabelle Dumé

Fur­ther reading:

• Mounier, A., et al., Detec­tion of Bow Echoes in Kilo­me­ter-Scale Fore­casts Using a Con­vo­lu­tion­al Neur­al Net­work, Arti­fi­cial Intel­li­gence for the Earth Sys­tems, 1(2), e210010. https://journals.ametsoc.org/view/journals/aies/1/2/AIES-D-21–0010.1.xml, 2022.
• Rap­port Recherche 2021 de Météo-France qui fait référence à plusieurs travaux récents au CNRM sur les mod­èles de prévi­sion météorologique : http://​www​.umr​-cnrm​.fr/​I​M​G​/​p​d​f​/​r​_​r​_​2​0​2​1​-​f​r​_​w​e​b.pdf