ClimaOcean: A GPU-powered Ocean-Only Model which enables Large Ensemble Climate Simulations
Taimoor Sohail
We showcase ClimaOcean, a GPU-native global ocean model, which achieves orders of magnitude speed-up over traditional CPU-based ocean models. The significant reduction in computational cost afforded by ClimaOcean enables us to conduct high-resolution (1/6th degree and higher) large ensemble experiments over a range of climate scenarios. In this presentation, we highlight the ability of a repeat-year forced ClimaOcean simulation to capture fine-scale ocean processes accurately with minimal computational cost. We explore ocean internal variability, eddy dynamics, and heat transport in a suite of simulations of a ‘neutral’ global ocean model. This emerging ensemble of simulations allows us to explore internal variability across a vast range of spatiotemporal scales. Focussing our analysis to the tropics and sub-tropics, we show the impact of fine-scale ocean eddies and jets on heat and salt content variability across daily, seasonal, interannual and decadal timescales. Our results highlight ClimaOcean’s ability to make ensemble simulations more accessible by reducing their computational cost while maintaining scientific rigour.
Please use this thread for discussion about this talk.