OptimESM has delivered developments in three interlinked areas, (i) Earth system models (ESMs), (ii) new emission scenarios and (iii) machine learning (ML)-based methods for regional downscaling, as well as performed sets of ESM simulations and analysis of these simulations and of CMIP6 data.
First, we have finalised our so-called “post-CMIP6” ESMs, starting from the versions of our four project ESMs that were available at the start of the project. We have used these new “post-CMIP6” ESMs to perform sets of CO2-emission-driven pre-industrial control, historical and idealised simulations, including ramp-up simulations to different global warming levels, zero emission simulations at these global warming levels and ramp-down simulations back to lower warming levels. The post-CMIP6 ESMs have been evaluated for the historical period using observations.
First results from our simulations indicate the importance of temperature overshoot on the regional climate and provide first new knowledge on occurrence of various abrupt changes at different global warming levels. We have also developed a catalogue of abrupt changes and state transitions using CMIP6 projections and tested early warning signals for tipping points.
The post-CMIP6 ESM versions have then been used as a starting point to further improve the representation of e.g. ice sheets, permafrost, fires, soil and vegetation. In addition, we have worked on hybrid resolution methods, where the physical components of a model run at higher spatial resolution than other ESM components, and tested methods to accelerate spin-up and calibration of the ESMs to reduce the computational needs.
The integrated assessment model REMIND-MAgPIE has been applied to develop new land and emission scenarios, including scenarios that reach the warming targets of the Paris agreement and some that overshoot them. They have been extended to 2300 and these extended scenarios have been run with a Simple Climate Model.
In order to provide more detailed information at the regional level, different ML-based tools have been developed to refine the regional representation of the ESM simulations. These ML-methods have been tested on first ESM simulations, and the results have been compared against data from dynamical high resolution regional climate models.
