All Climate Events

Friday, April 6, 2018. 1:00PM. Webinar: Toward seasonal to multi-annual marine biogeochemical prediction using GFDLs Earth System ModelCapes. Jong-Yeon Park, Princeton University / GFDL. Sponsored by Center for Satellite Applications and Research. More information here.

   While physical ocean prediction systems routinely assimilate observations and produce seasonal to decadal forecasts, ocean biogeochemical (BGC) prediction systems are less mature due to additional challenges. These include insufficient global-scale BGC observations to inform model initialization, uncertainties from both physical and BGC processes in earth system models, and properties of BGC variables that challenge data assimilation approaches (e.g., non-Gaussian, complex patterns of cross-correlation). A first impediment, however, is the high BGC sensitivity to transient momentum imbalances that arise during physical data assimilation. In this study, we develop a strategy to robustly integrate the GFDL’s ocean BGC model (i.e. COBALT) with the ensemble coupled-climate data assimilation (ECDA) system used for GFDL’s seasonal to decadal global climate predictions. The ocean and atmosphere data constraints in the assimilation system are optimally modified to reduce BGC biases caused by momentum imbalances while retaining the information of observed physical states. We then performed retrospective prediction runs by initializing the model with the output from our ECDA run coupled with BGC model and investigated seasonal to multi-annual prediction skills of nutrient anomalies, oxygen, phytoplankton and zooplankton over 1991 to 2016. We found that BGC variables generally show a higher and longer-term predictability than sea surface temperature although predictability varies regionally and with initialization month. While the predictability results here suggest a promising strategy of using earth system models for future operational BGC prediction systems, assessment of BGC predictions against satellite datasets shows a considerable gap between potential predictability and achieved prediction skills. Opportunities to close this gap will be also discussed.