Leo EdelNansen Environmental and Remote Sensing Center
Leo Edel’s current work focuses on improving sea-ice thickness estimations in an ocean model (called TOPAZ), using a combined approach of artificial intelligence and data assimilation. Improving the estimation of sea-ice thickness can provide crucial information to make better predictions on future sea-ice trends. It can also help mitigate the consequences of sea-ice loss on relevant sectors such as navigation, oil exploration, as well as fishing industries, in addition to the livelihoods of Arctic communities that rely on sea ice. |
 |
TalksReconstruction of Arctic sea ice thickness (1992-2010) based on a hybrid machine learning and data assimilation approach24-Apr-24Arctic sea ice thickness (SIT) remains one of the most crucial yet challenging parameters to estimate. Satellite data generally presents temporal and spatial discontinuities, which constrain studies focusing on long-term evolution. Since 2011, the combined satellite product CS2SMOS enables more accurate SIT retrievals that significantly decrease the modelled SIT errors during assimilation. Can we extrapolate the benefits of data assimilation to past periods without SIT observations? In this study, we train a machine learning (ML) algorithm to learn the systematic SIT errors between two versions of the model TOPAZ4 over 2011-2022, with and without CS2SMOS assimilation, to predict the SIT error and extrapolate the SIT prior to 2011. The ML algorithm relies on SIT coming from the two versions of TOPAZ4, various oceanographic variables, and atmospheric forcings from ERA5. Over the test period 2011-2013, the ML method outperforms TOPAZ4 without CS2SMOS assimilation when compared to TOPAZ4 assimilating CS2SMOS. The root mean square error of Arctic averaged SIT decreases from 0.42 to 0.28 meters and the bias from -0.18 to 0.01 meters. Relative to independent mooring data in the Beaufort Gyre between 2001 and 2010, mean SIT bias reduces from 0.21 meters to 0.02 meters when using the ML algorithm. Ultimately, the ML-adjusted SIT reconstruction reveals an Arctic mean SIT of 1.61 meters in 1992 compared to 1.08 meters in 2022. This corresponds to a decline of total sea ice volume from 19,690 to 12,700 km^3 with an associated trend of -3,153 km^3/decade. These changes are accompanied by a distinct shift in SIT distribution. Our innovative approach proves its ability to correct a significant part of the primary biases of the model by combining data assimilation with machine learning. Once this new reconstructed SIT dataset is assimilated in TOPAZ4, the correction can be further propagated to the other sea ice and ocean variables. |