DAMPEn · Data Assimilation and Machine learning for online Parameter Estimation

Линиите свързват координатора с партньорите.

Накратко на български

Цел на проекта

Over the past four decades the Arctic sea ice cover has undergone dramatic changes as a result of anthropogenic CO2 emissions, with precipitous declines in both sea ice thickness and area. Such changes have direct consequences for global climate and human populations, through impacts Ono mid-latitude weather, large-scale ocean circulation patterns and high-latitude climate feedbacks with regulate global-mean temperature. Our ability to quantify the impacts of continued sea ice loss on society and the environment depends on the accuracy with which climate models simulate the coupled interactions between the atmosphere, ocean, and sea ice. However, structural errors associated with the calibration of model physics parameters lead to systematic biases and uncertainty in future projections. For sea ice, one major source of parameter uncertainty comes from the representation of snow thermal conductivity. DAMPEn will demonstrate a unique and multi-disciplinary framework for deriving state-dependent parameter estimates of snow thermal conductivity in the large-scale sea ice model, SI3. For the first time, this framework will allow the estimation of spatially and temporally varying sea ice model parameters. This will be achieved by a three-step process: (1) Estimate snow conductivity parameter corrections via the assimilation of sate-of-the-art sea ice thickness and snow depth data into SI3. (2) Use convolutional neural networks to learn functional relationships between model state variables and snow conductivity corrections. (3) Replace the default (static) snow conductivity parameter in SI3 with the machine learning representation. This data-driven scheme is expected to reduce multi-decadal sea ice biases and uncertainty in future projections. Furthermore, it will provide the groundwork for future developments into the calibration of numerous sea ice model parameters, reducing the cost and carbon footprint associated with climate model calibration activities.

Оригинален текст от CORDIS (на английски).

Участници

• UNIVERSITY COLLEGE LONDON · LondonКоординаторОбединеното кралство

Връзки

• Виж в CORDIS
• DOI: 10.3030/101270625