COSMO (weather model)
COSMO (short for COnsortium for Small-scale MOdelling) is a non-hydrostatic, limited-area numerical weather prediction model developed for operational forecasting and mesoscale research by a multinational consortium of national weather services.[1] The model originated at the Deutscher Wetterdienst (DWD) as the "Local Model" (LM) and became operational there on 1 December 1999.[2][3] The COSMO consortium was formed in October 1998 and adopted DWD's Local Model as the basis of a shared modeling system, coordinating its subsequent development under the COSMO name.[4] From the late 2010s, consortium members progressively transitioned from COSMO to the limited-area configuration of ICON, DWD's global model.[5] DWD replaced its final operational COSMO configurations with ICON-based systems on 10 February 2021.[6]
Operation
[edit]COSMO was deployed in separate regional configurations by participating weather services. The model domain, horizontal resolution, run frequency, forecast length, and source of boundary conditions varied among the operational centers.[2]
Because COSMO is a limited-area model, each forecast requires initial conditions and regularly updated lateral boundary conditions from a coarser global or regional model. The INT2LM preprocessing program interpolates fields from the driving model onto the COSMO grid.[7] The model uses one-way nesting at its lateral boundaries, meaning that the driving model supplies information to the COSMO domain but does not receive feedback from it.[8]
Initial conditions can be interpolated from a global analysis or produced by a local data-assimilation system. At DWD, the assimilation system included analyses of atmospheric conditions, sea-surface temperature, and snow depth, as well as a soil-moisture analysis for COSMO-EU.[2]
Principles
[edit]COSMO is based on the non-hydrostatic primitive thermo-hydrodynamical equations describing compressible flow in a moist atmosphere. The model equations are formulated in rotated geographical coordinates using a generalized terrain-following height coordinate, and the model is designed for both operational forecasting and scientific applications at the meso-β and meso-γ scale.[1] Unlike DWD's global GME and ICON models, which use icosahedral grids, COSMO uses a rotated latitude–longitude grid restricted to a limited area rather than the whole globe.[1]
Variants
[edit]DWD ran two principal COSMO-based configurations before retiring the model. The roughly 7 km European configuration originated with the LM application that became operational over Central Europe in December 1999. It was later operated as COSMO-EU until the ICON-EU nest replaced it in December 2016.[2]
COSMO-DE was DWD's convection-permitting configuration covering Germany and surrounding areas. It became operational in April 2007 at a horizontal grid spacing of 2.8 km and was followed by the ensemble system COSMO-DE-EPS in May 2012.[2][9] On 15 May 2018, DWD replaced COSMO-DE and COSMO-DE-EPS with COSMO-D2 and COSMO-D2-EPS. The new configuration expanded the model domain, reduced the horizontal grid spacing to 2.2 km, and increased the number of vertical levels from 50 to 65.[10] A broadly similar convection-permitting configuration using a 2.8 km grid and 50 vertical levels continued in operational use by other consortium members after DWD retired COSMO.[2]
Other consortium members developed their own operational COSMO configurations, generally with coarser-grid spacings of about 4–7 km or convection-permitting grid spacings of about 2.2–2.8 km, adapted to their own domains.[2] Notable examples include COSMO-LEPS, a limited-area ensemble prediction system operated by Italy's ARPAE-SIMC using ECMWF ensemble boundary conditions,[2] and COSMO-CLM, a regional climate-mode configuration of the model developed jointly by the consortium and a separate climate-modeling community (CLM-Community).[11]
Usage
[edit]The COSMO consortium was formed in October 1998 at the regular annual meeting of DWD and MeteoSwiss. A Memorandum of Understanding on scientific collaboration in non-hydrostatic modeling was signed in 1999 by the directors of the national weather services of Germany, Switzerland, Italy, and Greece; this was replaced by a formal COSMO Agreement, signed by the same four services on 3 October 2001.[4] Poland's IMGW joined the consortium in 2002, followed by Romania's NMA (2006), Russia's Roshydromet (2009), and the Israel Meteorological Service (2017).[4] The consortium's governing agreement was renewed and re-signed by member directors on 21 September 2009 and again on 19 June 2014.[4]
Beyond consortium members, other national weather services could operate COSMO under license. Beginning in 2010, several services that had previously used DWD's hydrostatic HRM model migrated to COSMO. The annual fee for operational use was €20,000, although it was waived for countries classified as lower-middle-income economies by the World Bank.[2] Universities and research institutes could obtain a free research license.[12]
After consortium-wide model development shifted to ICON and ICON-LAM became the recommended successor to COSMO, the consortium shifted its external licensing program toward support for ICON users. As of 2025, new COSMO licenses were available only in exceptional cases.[5][13]
Accuracy
[edit]Peer-reviewed evaluations found that COSMO's forecast accuracy varied by model configuration, weather variable, region, and verification method. In a six-month evaluation over the Swiss radar domain, the convection-permitting COSMO-2 and COSMO-DE configurations equaled or outperformed their lower-resolution COSMO-7 and COSMO-EU counterparts in neighborhood-based precipitation verification. The largest improvements occurred for localized convective precipitation.[14]
A two-year evaluation of COSMO-DE and COSMO-EU over Germany found relatively small biases in integrated water vapor and satellite brightness temperature at 6.2 μm, but larger biases in cloud-base height, precipitation, and brightness temperature at 10.8 μm. COSMO-DE generally represented high clouds better than COSMO-EU, particularly during summer, although both configurations exhibited moisture and cloud biases that varied with weather regime.[15]
Development history
[edit]Origins and early operational use (1998–2007)
[edit]Development of the model that became COSMO began at DWD as the Lokalmodell (LM; English: "Local Model"), a non-hydrostatic alternative to DWD's existing hydrostatic regional models. Its formulation was described by Steppeler, Doms, Schättler, and colleagues.[3]
The COSMO consortium was formed in October 1998 to coordinate the development, improvement, and maintenance of a shared limited-area modeling system based on LM.[4] LM became operational at DWD on 1 December 1999, running alongside the global GME model over a Central European domain of 325×325 grid points. The initial configuration had a horizontal grid spacing of about 7 km and used 32 vertical layers.[2][1]
The consortium subsequently coordinated development of the shared software under the name COSMO-Model.[4][1]
Convection-permitting development (2003–2018)
[edit]Work on a convection-permitting version of the model, capable of explicitly representing deep convection at horizontal grid spacings of roughly 1–3 km, began in 2003 and centered on a new Runge–Kutta dynamical core.[2] The resulting convection-permitting configuration, COSMO-DE, became operational at DWD in April 2007 at a horizontal grid spacing of 2.8 km,[2] and its formulation and sensitivities were documented in a 2011 peer-reviewed paper.[9] An ensemble counterpart, COSMO-DE-EPS, followed in May 2012.[2]
DWD's data-assimilation system for the model was initially based on nudging. In March 2017, DWD introduced an ensemble-based approach developed through the COSMO Priority Project KENDA (kilometer-scale ensemble data assimilation).[7]
COSMO-D2 and COSMO-D2-EPS launched operationally at DWD on 15 May 2018, succeeding COSMO-DE and COSMO-DE-EPS as DWD's highest-resolution configuration until the model's replacement by ICON-D2 less than three years later.[10]
Transition to ICON (2016–2022)
[edit]DWD's coarse-grid COSMO-EU configuration was replaced by the ICON-EU nest of DWD's global ICON model in December 2016.[2] In 2018, the consortium launched Priority Project C2I ("COSMO Transition to ICON") to support a coordinated, four-year transition of consortium partners from the COSMO-Model to the limited-area configuration of ICON (ICON-LAM).[5]
On 10 February 2021, DWD replaced its remaining operational COSMO systems, COSMO-D2 and COSMO-D2-EPS, with ICON-D2 and ICON-D2-EPS, after which it no longer ran a COSMO-based model operationally.[6]
By 2022, ICON had become the consortium's recommended and supported modeling framework.[5][12] COSMO nevertheless remained in use after 2022 among licensees and at IMGW-PIB, where a COSMO-based rapid-update cycle was still in use in 2026.[12][16]
See also
[edit]- ICON – the global modeling framework whose limited-area configuration, ICON-LAM, succeeded COSMO across the consortium
- GME – DWD's former global model, run alongside COSMO/LM from 1999
- Numerical weather prediction
References
[edit]- 1 2 3 4 5 Schättler, Ulrich; Doms, Günther; Schraff, Christoph (September 2021). A Description of the Nonhydrostatic Regional COSMO-Model, Part VII: User's Guide (PDF) (Report). Consortium for Small-Scale Modelling. Retrieved 4 August 2026.
- 1 2 3 4 5 6 7 8 9 10 11 12 13 14 "Operational Applications within COSMO". Consortium for Small-scale Modelling. 23 February 2021. Retrieved 4 August 2026.
- 1 2 Steppeler, J.; Doms, G.; Schättler, U.; et al. (2003). "Meso-gamma scale forecasts using the nonhydrostatic model LM". Meteorology and Atmospheric Physics. 82 (1–4): 75–96. Bibcode:2003MAP....82...75S. doi:10.1007/s00703-001-0592-9.
- 1 2 3 4 5 6 "Consortium". Consortium for Small-scale Modelling. 5 June 2025. Retrieved 4 August 2026.
- 1 2 3 4 "ICON Transition". Consortium for Small-scale Modelling. 26 February 2024. Retrieved 4 August 2026.
- 1 2 "Operational NWP System: ICON: Replacement of COSMO-D2 / COSMO-D2 EPS with ICON-D2 / ICON-D2-EPS". Deutscher Wetterdienst. 10 February 2021. Retrieved 4 August 2026.
- 1 2 Schättler, Ulrich; Blahak, Ulrich (February 2018). A Description of the Nonhydrostatic Regional COSMO-Model, Part V: Initial and Boundary Data for the COSMO-Model (PDF) (Report). Consortium for Small-Scale Modelling. Retrieved 4 August 2026.
- ↑ Schraff, Christoph; Hess, Reinhold (September 2021). A Description of the Nonhydrostatic Regional COSMO-Model, Part III: Data Assimilation (PDF) (Report). Consortium for Small-Scale Modelling. Retrieved 4 August 2026.
- 1 2 Baldauf, Michael; Seifert, Axel; Förstner, Jochen; et al. (2011). "Operational Convective-Scale Numerical Weather Prediction with the COSMO Model: Description and Sensitivities". Monthly Weather Review. 139 (12): 3887–3905. Bibcode:2011MWRv..139.3887B. doi:10.1175/MWR-D-10-05013.1.
- 1 2 "Ersetzung von COSMO-DE /-EPS durch COSMO-D2 /-EPS" (PDF) (in German). Deutscher Wetterdienst. 15 May 2018. Retrieved 4 August 2026.
- ↑ Rockel, Burkhardt; et al. (2020). "Regional Climate Modelling with COSMO-CLM: History and Perspectives". Atmosphere. 11 (11): 1250. Bibcode:2020Atmos..11.1250S. doi:10.3390/atmos11111250.
- 1 2 3 "Operational Applications within COSMO". Consortium for Small-scale Modelling. 15 December 2022. Retrieved 4 August 2026.
- ↑ "COSMO Licencing". Consortium for Small-scale Modelling. 4 September 2025. Retrieved 4 August 2026.
- ↑ Weusthoff, Tanja; Ament, Felix; Arpagaus, Marco; Rotach, Mathias W. (2010). "Assessing the Benefits of Convection-Permitting Models by Neighborhood Verification: Examples from MAP D-PHASE". Monthly Weather Review. 138 (9): 3418–3433. Bibcode:2010MWRv..138.3418W. doi:10.1175/2010MWR3380.1.
- ↑ Böhme, Tim; Stapelberg, Stefan; Akkermans, Tom; et al. (2011). "Long-term evaluation of COSMO forecasting using combined observational data of the GOP period". Meteorologische Zeitschrift. 20 (2): 119–132. Bibcode:2011MetZe..20..119B. doi:10.1127/0941-2948/2011/0225.
- ↑ "Priority Project APOCS". Consortium for Small-scale Modelling. 23 January 2026. Retrieved 4 August 2026.