π Science and technology
How revolutionary AI and satellites are changing weather predictions

New observation systems for better weather forecasting

Pierre Tabary, Deputy Director of Operations at the Direction des Opérations pour la Prévision (DirOP) of Météo France
On April 19th, 2023 |
4 min reading time
Pierre Tabary
Pierre Tabary
Deputy Director of Operations at the Direction des Opérations pour la Prévision (DirOP) of Météo France
Key takeaways
  • The first weather observation systems were based on ground-level observations.
  • Soon after, meteorologists introduced remote sensing, with radars emitting electromagnetic waves, for observations at altitude.
  • Observations are also made by satellites, which can be polar-orbiting or geostationary.
  • The number of meteorological satellites has increased significantly in recent years and ambitious European programmes have been launched.
  • There are also timely-based observations from infrastructures that were not designed for atmospheric measurements, such as telephone networks.

The first weath­er fore­cast­ing mod­els were devel­oped in the ear­ly 20th cen­tu­ry1 with the idea that if the ini­tial state of the atmos­phere – that is, the con­di­tions of wind, humid­i­ty, tem­per­a­ture and pres­sure at a giv­en time – are known pre­cise­ly, then its future state can be pre­dict­ed thanks to the equa­tions of physics gov­ern­ing the tem­po­ral evo­lu­tion of the atmos­pher­ic vari­ables. For a weath­er fore­cast to be use­ful, it needs to be of a cer­tain qual­i­ty and pro­vide pre­dic­tions for the short term: it can there­fore be thought of as a kind of race against time. More­over, as the weath­er fore­cast at a giv­en point depends on the weath­er con­di­tions observed else­where, weath­er fore­cast­ers set up col­lab­o­ra­tions ear­ly on in the his­to­ry of the field to exchange their obser­va­tion­al data in real time and com­pare the per­for­mance of their models.

Ground and upper air observations

The first obser­va­tion­al sys­tems con­sist­ed of net­works of instru­ments that mea­sured atmos­pher­ic con­di­tions in situ (at the exact loca­tion of the sen­sor) at ground lev­el, but mete­o­rol­o­gists soon realised that they also need­ed obser­va­tions at alti­tude. They thus devel­oped radioson­des – bal­loons car­ry­ing instru­ments to mea­sure tem­per­a­ture, pres­sure, humid­i­ty, and wind, launched sev­er­al times a day and reach­ing up to 20 km in alti­tude. Some bal­loons can also be launched from ships.

Obser­va­tion tech­nolo­gies have evolved con­sid­er­ably and we are now wit­ness­ing the emer­gence of obser­va­tion projects based on drones, which could be of oper­a­tional inter­est for sam­pling the atmos­phere at alti­tude, par­tic­u­lar­ly over the sea (where deploy­ment con­di­tions are less con­strained), fol­low­ing sam­pling strate­gies that are deter­mined by mete­o­ro­log­i­cal conditions.

The advent of remote sensing 

From the 1960–70s, remote sens­ing (the remote mea­sure­ment of atmos­pher­ic para­me­ters) began to be intro­duced into obser­va­tion net­works. Weath­er radar is an exam­ple of such remote sens­ing instru­ments. These devices emit elec­tro­mag­net­ic waves that prop­a­gate and inter­act with rain, snow and hail, and can thus be used to map pre­cip­i­ta­tion over areas of sev­er­al hun­dred kilo­me­tres or more when the instru­ments are net­worked. Today, about 200 weath­er radars are in oper­a­tion in Europe. The num­ber is sim­i­lar in the Unit­ed States and Japan. For many years, weath­er radar data have been assim­i­lat­ed into numer­i­cal weath­er pre­dic­tion mod­els, sig­nif­i­cant­ly improv­ing the qual­i­ty of pre­cip­i­ta­tion fore­casts2.

Anoth­er exam­ple of a remote sens­ing instru­ment is lidar, which is sim­i­lar to radar except that the waves emit­ted are light waves. These waves are sen­si­tive to aerosols – small dust par­ti­cles sus­pend­ed in the atmos­phere – or cloud droplets. Lidars can there­fore mea­sure the prop­er­ties of these par­ti­cles from a dis­tance and are now part of the oper­a­tional net­works used by mete­o­ro­log­i­cal ser­vices3.

Satel­lite observations 

Anoth­er major advance has of course been the emer­gence of satel­lite obser­va­tions, which make it pos­si­ble to observe large areas with the same instru­ment. Satel­lites can be divid­ed into two main fam­i­lies. The first are the geo­sta­tion­ary satel­lites, which remain per­ma­nent­ly above a sin­gle point on the equa­tor (at an alti­tude of about 36 000 km). As they can only observe half the globe, inter­na­tion­al col­lab­o­ra­tion is required to cov­er the whole planet.

The sec­ond type of satel­lite is the polar-orbit­ing satel­lite, which, as its name sug­gests, con­tin­u­al­ly orbits the Earth (at an alti­tude of between 300 and 800 km), typ­i­cal­ly cir­cling the globe in about 100 min­utes. These satel­lites can pro­vide obser­va­tions at all lat­i­tudes, includ­ing the poles, with the same spa­tial res­o­lu­tion, unlike geo­sta­tion­ary satel­lites. And because they are about ten times clos­er to the Earth, they offer bet­ter res­o­lu­tion of clouds and surfaces.

A wide vari­ety of instru­ments oper­at­ing in dif­fer­ent wave­lengths are car­ried on these satel­lites and allow the prop­er­ties of the atmos­phere to be mea­sured: clouds, pre­cip­i­ta­tion, aerosols, wind, tem­per­a­ture and humidity.

A new gen­er­a­tion of Euro­pean geo­sta­tion­ary satel­lites has been launched.

The num­ber of mete­o­ro­log­i­cal satel­lites has increased dra­mat­i­cal­ly in recent years and ambi­tious pro­grammes, par­tic­u­lar­ly in Europe, have been launched. Last Decem­ber, for exam­ple, a new gen­er­a­tion of Euro­pean geo­sta­tion­ary satel­lites was launched4, as part of the Euro­pean Agen­cy’s pro­gramme for oper­a­tional weath­er satellites.

Time­ly observations 

Final­ly, there are the so-called time­ly obser­va­tions – that is, those that come from an infra­struc­ture that was not orig­i­nal­ly designed to car­ry out atmos­pher­ic mea­sure­ments. An impor­tant exam­ple: mobile phone net­works. These net­works use anten­nas (there are sev­er­al thou­sand of them in France) that com­mu­ni­cate with each oth­er at microwave wave­lengths. Oper­a­tors quick­ly realised that the lev­el of recep­tion between anten­nas was reduced when there were areas (or “cells”) of rain­fall between them. This is because microwave wave­lengths are affect­ed by precipitation. 

« Oper­a­tors react­ed to this prob­lem by equip­ping the sys­tems with a capac­i­ty to increase recep­tion lev­els in the event of observed atten­u­a­tion, but for mete­o­rol­o­gists, the mea­sure­ment of this observed atten­u­a­tion was very inter­est­ing because it pro­vid­ed infor­ma­tion on the inten­si­ty of pre­cip­i­ta­tion occur­ring in a giv­en area, » explains Pierre Tabary, Deputy Direc­tor of Oper­a­tions at Météo France. This indi­rect infor­ma­tion, prop­er­ly processed, can help to improve pre­cip­i­ta­tion maps5. « Nobody had imag­ined at the out­set that mobile phone net­works could be used in this way.

Posi­tion­ing satel­lites – the GPS sys­tem in the US and Galileo in Europe – are anoth­er exam­ple of a time­ly-based mea­sure­ment. « These ded­i­cat­ed posi­tion­ing satel­lites are con­stant­ly emit­ting sig­nals. So, the inge­nious idea was to put oth­er, much small­er satel­lites, that oppor­tunis­ti­cal­ly pick up these sig­nals, into orbit. The sig­nals that are detect­ed are slight­ly refract­ed on their way from the trans­mit­ting satel­lite to the receiv­ing satel­lite as they pass through the atmos­phere, so that they are some­what ‘bent’. The lev­el of this bend­ing can be quan­ti­fied by the receiv­ing satel­lite and indi­rect­ly pro­vides valu­able infor­ma­tion about ther­mo­dy­nam­ic con­di­tions, such as humid­i­ty, in the stratos­phere and upper troposphere.

Researchers have ver­i­fied the rel­e­vance of this mea­sure­ment prin­ci­ple (called radio occul­ta­tion) and there are now sev­er­al dozen recep­tion satel­lites of this type in ser­vice, the data from which are analysed by oper­a­tional weath­er fore­cast­ing mod­els6. « Here again, we are able to car­ry out mea­sure­ments at low­er cost: we do not emit waves into the atmos­phere our­selves, but we exploit the waves already emit­ted by others.”

Today, about 90% of the data input into glob­al weath­er fore­cast­ing mod­els comes from satel­lites and this trend will con­tin­ue. « That said, ground-based mea­sure­ments will always be impor­tant – espe­cial­ly for cal­i­brat­ing satel­lite data, » con­cludes Pierre Tabary.

 Isabelle Dumé
1Lynch, Peter ; Les orig­ines de la prévi­sion numérique du temps et de la mod­éli­sa­tion cli­ma­tique, La Météorolo­gie, 2008, N° 63 ; p. 14–24 10.4267/2042/21887
2Wat­trelot, Eric, Olivi­er Cau­mont, Jean-Fran­cois Mah­fouf. Oper­a­tional Imple­men­ta­tion of the 1D13D-Var Assim­i­la­tion Method of Radar Reflec­tiv­i­ty Data in the AROME Mod­el. Month­ly Weath­er Review, 2013, 142, pp.1852–1871. ⟨10.1175/MWR-D-13–00230.1⟩⟨meteo-01001390⟩
3Rey, Gérard ; Traullé, Olivi­er ; Bour­cy, Thomas ; Dubouchet, Elisa. Un nou­veau réseau de lidars aérosols à Météo-France. La Météorolo­gie, 2016, 95, p. 11–14 10.4267/2042/61610
4Stuhlmann, Rolf, Ken­neth Holm­lund, Johannes Schmetz, Hervé Roquet et al.Obser­va­tions depuis l’orbite géo­sta­tion­naire avec Meteosat troisième généra­tion et EUMETSAT – https://​www​.eumet​sat​.int/
5Alpert, P., Mess­er, H. & David, N. Mobile net­works aid weath­er mon­i­tor­ing. Nature 537, 617 (2016). https://​doi​.org/​1​0​.​1​0​3​8​/​5​3​7617e 
6Kursin­s­ki et al. 1997. Observ­ing the Earth­’s atmos­phere with radio occul­ta­tion mea­sure­ments using the Glob­al Posi­tion­ing Sys­tem. J. Geo­phys. Res. 102:23.429–23.465. 

Our world explained with science. Every week, in your inbox.

Get the newsletter