The role of atmospheric uncertainty in Arctic summer sea ice data assimilation and prediction

• Published in: Quarterly Journal of the Royal Meteorological Society
• Main Authors: Yang, Qinghua, Losa, Svetlana N., Losch, Martin, Jung, Thomas, Nerger, Lars
• Other Authors: German Federal Ministry of Education and Research, National Natural Science Foundation of China
• Format: Article in Journal/Newspaper
• Language: English
• Published: Wiley 2015
• Subjects: Arctic; National Snow and Ice Data Center; Sea ice
• Online Access: http://dx.doi.org/10.1002/qj.2523; https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fqj.2523; https://onlinelibrary.wiley.com/doi/pdf/10.1002/qj.2523; https://onlinelibrary.wiley.com/doi/full-xml/10.1002/qj.2523; https://rmets.onlinelibrary.wiley.com/doi/pdf/10.1002/qj.2523

The role of atmospheric uncertainty for the assimilation and prediction of Arctic sea ice is explored by running the Massachusetts Institute of Technology general circulation model (MITgcm) in data assimilation (DA) and prediction mode for summer 2010. The atmospheric ensemble forcing is taken from the UK Met Office (UKMO) system available through the TIGGE (THORPEX Interactive Grand Global Ensemble) database. The DA system is based on a Local Singular Evolutive Interpolated Kalman (LSEIK) filter, and Special Sensor Microwave Imager/Sounder (SSMIS) sea ice concentration operational products from the National Snow and Ice Data Center (NSIDC) are assimilated. Two kinds of experiments are carried out differing in the LSEIK configuration and forcing used: the first one uses a single deterministic control forcing and a forgetting factor necessary to inflate the ensemble spread in the DA phase; the second one uses 23 members from the UKMO atmospheric ensemble prediction system, thereby avoiding any additional ensemble inflation and making further tuning unnecessary. With both systems the model data misfit improves as expected, but the ensemble approach outperforms the deterministic filter. The ice concentration of 24 h forecasts is consistently closer to observations with the ensemble approach, because a larger and more realistic ensemble spread, representing model uncertainty, leads to a better adjustment. Fifteen‐day forecasts are also better with ensemble forcing than with deterministic forcing, because of both the larger spread and the better initial state in the ensemble forced system. The ensemble forcing can also improve poor initial states obtained with the deterministic control forcing, because the ensemble forcing introduces a larger spread that spans a larger range of model simulations. Ice thickness forecasts cannot be significantly improved with the ensemble forcing.