Multimodel Superensemble for Weather and Climate

点击翻译按钮获取中文摘要

This work will build on recent work by the PI on the multimodel superensemble approach for weather and climate forecasts. The principle of this technique contains a partitioning of the timeline with a control (training) and a forecast phase. Past results based on this technique have indicated that the superensemble outperforms in skill, multimodel forecasts, ensemble mean and climatology. The research will focus on the following areas: (1) to understand the basis for improved results from the multimodel superensemble for global numerical weather prediction, including hurricanes and forecasts of seasonal climate; (2) to further improve the methodology for the construction of superensemble forecasts; (3) to improve the initial state, i.e. superensemble of day 0 input for all variables for training and model validation; (4) to design and carry-out coupled atmosphere-ocean model based superensemble forecasts on seasonal to multiseasonal time scales; (5) to address the issue of sensitivity of superensemble forecast skills to perturbation of the training data base; and (6) to continue the preparation of real-time multimodel superensemble forecast data sets for global numerical weather prediction including hurricane track and intensity forecasts. The work is important because it has potential implications for prediction, both on weather and seasonal timescales, leading to great societal benefits.