Spin‐up of UK Earth System Model 1 (UKESM1) for CMIP6. (4th August 2020)

• Record Type: Journal Article
• Authors: Yool, A.; Palmiéri, J.; Jones, C. G.; Sellar, A. A.; de Mora, L.; Kuhlbrodt, T.; Popova, E. E.; Mulcahy, J. P.; Wiltshire, A.; Rumbold, S. T.; Stringer, M.; Hill, R. S. R.; Tang, Y.; Walton, J.; Blaker, A.; Nurser, A. J. G.; Coward, A. C.; Hirschi, J.; Woodward, S.; Kelley, D. I.; Ellis, R.; Rumbold‐Jones, S.
• Is Part Of: Journal of advances in modeling earth systems. Volume 12:Number 8(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-08-04
• Subjects: Earth system model -- spin‐up -- CMIP6 -- carbon cycle -- equilibrium; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2019MS001933
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: For simulations intended to study the influence of anthropogenic forcing on climate, temporal stability of the Earth's natural heat, freshwater, and biogeochemical budgets is critical. Achieving such coupled model equilibration is scientifically and computationally challenging. We describe the protocol used to spin‐up the UK Earth system model (UKESM1) with respect to preindustrial forcing for use in the sixth Coupled Model Intercomparison Project (CMIP6). Due to the high computational cost of UKESM1's atmospheric model, especially when running with interactive full chemistry and aerosols, spin‐up primarily used parallel configurations using only ocean/land components. For the ocean, the resulting spin‐up permitted the carbon and heat contents of the ocean's full volume to approach equilibrium over 5, 000 years. On land, a spin‐up of 1, 000 years brought UKESM1's dynamic vegetation and soil carbon reservoirs toward near‐equilibrium. The end‐states of these parallel ocean‐ and land‐only phases then initialized a multicentennial period of spin‐up with the full Earth system model, prior to this simulation continuing as the UKESM1 CMIP6 preindustrial control (piControl). The realism of the fully coupled spin‐up was assessed for a range of ocean and land properties, as was the degree of equilibration for key variables. Lessons drawn include the importance of consistent interface physics across ocean‐ and land‐only models and the coupled (parent) model, the extreme simulation duration required to approach equilibration targets, and the occurrence of significant regional land carbon drifts despite global‐scale equilibration. Overall, the UKESM1 spin‐up underscores the expense involved and argues in favor of future development of more efficient spin‐up techniques. Plain Language Summary: Earth system models (ESMs) are an important tool for understanding the Earth and for projecting how climate change may affect natural and human systems. For simulations of ESMs to separate anthropogenic influences on climate from the background state, the stability of the unperturbed system is critical. However, achieving this equilibrium is both scientifically and computationally challenging. Here, we describe how this was achieved for one such model, UKESM1, for the sixth Coupled Model Intercomparison Project (CMIP6). Due to the cost of the full model, especially when running with atmospheric chemistry and aerosols, much of UKESM1's spin‐up to equilibrium made use of ocean‐ and land‐only configurations. Millennial‐scale spin‐up phases of these component‐only models were used to initialize a final centennial‐scale phase of the full model to reach preindustrial equilibrium targets. The stability and realism of UKESM1's spun‐up state was then evaluated across a broad range of properties. A number of lessons were drawn from this spin‐up including the extreme simulation duration required to reach equilibrium. A key conclusion is the importance of developing efficient techniques to spin‐up ESMs. Key Points: Earth system components and spin‐up protocol of UKESM1 for CMIP6 outlined Ocean‐only (5, 000 years) and land‐only (1, 000 years) phases used prior to fully coupled finalizing of spin‐up (500 year) Evaluation of spin‐up protocol presented, including cross‐component validation of piControl state and drift