Cloud-based framework for inter-comparing submesoscale-permitting realistic ocean models
Takaya Uchida, Julien Le Sommer, Charles R. Stern, Ryan Abernathey, Chris Holdgraf, Aurélie Albert, Laurent Brodeau, Eric P. Chassignet, Xiaobiao Xu, Jonathan Gula, Guillaume Roullet, Nikolay Koldunov, Sergey Danilov, Qiang Wang, Dimitris Menemenlis, Clément Bricaud, Brian K. Arbic, Jay F. Shriver, Fangli Qiao, Bin Xiao, Arne Biastoch, René Schubert, Baylor Fox‐Kemper, William K. Dewar, Alan J. Wallcraft
Abstract
Abstract. With the increase in computational power, ocean models with kilometer-scale resolution have emerged over the last decade. These models have been used for quantifying the energetic exchanges between spatial scales, informing the design of eddy parametrizations, and preparing observing networks. The increase in resolution, however, has drastically increased the size of model outputs, making it difficult to transfer and analyze the data. It remains, nonetheless, of primary importance to assess more systematically the realism of these models. Here, we showcase a cloud-based analysis framework proposed by the Pangeo project that aims to tackle such distribution and analysis challenges. We analyze the output of eight submesoscale-permitting simulations, all on the cloud, for a crossover region of the upcoming Surface Water and Ocean Topography (SWOT) altimeter mission near the Gulf Stream separation. The cloud-based analysis framework (i) minimizes the cost of duplicating and storing ghost copies of data and (ii) allows for seamless sharing of analysis results amongst collaborators. We describe the framework and provide example analyses (e.g., sea-surface height variability, submesoscale vertical buoyancy fluxes, and comparison to predictions from the mixed-layer instability parametrization). Basin- to global-scale, submesoscale-permitting models are still at their early stage of development; their cost and carbon footprints are also rather large. It would, therefore, benefit the community to document the different model configurations for future best practices. We also argue that an emphasis on data analysis strategies would be crucial for improving the models themselves.