CMIP6 simulations with the compact Earth system model OSCAR v3.1

Yann Quilcaille, Thomas Gasser, Philippe Ciais, Oliviér Boucher

TL;DR

This paper evaluates OSCAR v3.1, a reduced-complexity Earth system model, against CMIP6 simulations and observational data. By applying observational constraints, the model's performance is improved, particularly in temperature projections, and its equilibrium climate sensitivity is brought into better alignment with other models. This work provides a crucial validation of OSCAR and sets the stage for its use in future climate modeling efforts.

What Problem Does This Paper Address?

Earth System Models (ESMs) are powerful tools for climate projections, but their high computational cost limits their use in large ensembles and detailed uncertainty analyses. Reduced-complexity models (RCMs), like OSCAR, offer a computationally efficient alternative. However, their simplicity can come at the expense of accuracy and spatial/temporal detail. Therefore, it's crucial to rigorously evaluate and validate these RCMs against observations and more complex ESMs to ensure their reliability for climate science. This paper addresses this by evaluating a new version of OSCAR (v3.1) using a comprehensive set of CMIP6 experiments and observational data.

Methods

The authors evaluate OSCAR v3.1 using a probabilistic framework, where each ensemble member represents a different configuration of the model's parameters. They employ a two-stage post-processing approach:

  1. Exclusion: Configurations leading to numerical divergence (e.g., in ocean carbon cycle simulations under high warming) are removed.
  2. Constraining: The remaining configurations are weighted based on their likelihood of reproducing observational data. This involves using observational constraints on global surface air temperature and compatible fossil-fuel emissions.

They also leverage a wide array of CMIP6 and RCMIP experiments, including idealized scenarios (e.g., abrupt CO2 changes) and emission/concentration-driven historical and future projections.

Datasets

The paper utilizes data from:

  • CMIP6 ESM simulations: A large ensemble of simulations from the Coupled Model Intercomparison Project Phase 6.
  • RCMIP Phase 2 data: Results from the Reduced Complexity Model Intercomparison Project.
  • Observational Datasets:
    • HadCRUT 4.6.0.0 for global surface air temperature.
  • Global Carbon Budget data for compatible fossil-fuel emissions.
  • Data on cumulative net ocean to atmosphere flux of CO2.

Key Insights and Ideas

  • Observational Constraints are Powerful: Applying observational constraints significantly improves OSCAR's performance, particularly by reducing the overestimation of historical warming and tightening uncertainty ranges. This highlights the value of integrating observational data into RCM calibration.
  • RCMs Can Emulate ESMs (with caveats): OSCAR, while not emulating any single ESM, successfully reproduces many of the qualitative responses seen in complex ESMs, demonstrating its utility as a complementary tool.
  • Model Structure Matters: Differences between OSCAR and ESMs are partly due to OSCAR's unique features (e.g., interactive atmospheric chemistry, endogenous permafrost emissions) and partly due to the limitations of its simplified modules (e.g., ocean carbon cycle instability).
  • ECS Calibration is Key: The Equilibrium Climate Sensitivity (ECS) of OSCAR, when constrained, aligns well with CMIP5 and CMIP6 estimates, but the unconstrained value is also within a reasonable range. This suggests that RCMs can capture key climate sensitivity metrics.

Key Results

  • Temperature Projections: Observational constraints reduced the overestimation of global surface air temperature over 2000-2019 (relative to 1961-1990) from 0.60±0.11 K to 0.55±0.04 K.
  • Equilibrium Climate Sensitivity (ECS): The unconstrained OSCAR has an ECS of 3.17±0.63 K, comparable to CMIP5 (3.2±0.7 K) and CMIP6 (3.7±1.1 K) models. After constraining, OSCAR's ECS is reduced to 2.78±0.47 K.
  • Carbon Cycle: While OSCAR qualitatively reproduces ESM responses, it shows a low sensitivity of the land carbon cycle and an instability in the ocean carbon cycle, identified as areas for improvement.
  • Attribution: OSCAR's simulations suggest that natural factors alone are unlikely to be causing the current observed climate change, consistent with IPCC conclusions.

Limitations and Future Work

  • Ocean Carbon Cycle Instability: The paper identifies an instability in the ocean carbon cycle module, particularly under high-warming scenarios, suggesting a need for its redesign.
  • Land Carbon Cycle Sensitivity: The land carbon cycle exhibits a low sensitivity to climate change and a wide transient response, necessitating improved calibration and potentially more parameters.
  • Simplified Climate Module: The climate module is described as "seemingly too simple," and the climate feedback involving short-lived species is too strong, indicating potential for further development.
  • Over-constraining: The authors acknowledge that the constraining approach might lead to over-constraining, particularly for temperature projections, and suggest exploring alternative constraining methods.
  • Future Work: The authors propose improving the climate module with extra parameters, enhancing the land carbon cycle calibration, and further investigating the impact of observational constraints. They also aim to use these simulations as a basis for more systematic assessments of OSCAR's performance and for comparisons with older versions.

Summary updated

Abstract

Reduced-complexity models, also called simple climate models or compact models, provide an alternative to Earth system models (ESMs) with lower computational costs, although at the expense of spatial and temporal information. It remains important to evaluate and validate these reduced-complexity models. Here, we evaluate a recent version (v3.1) of the OSCAR model using observations and results from ESMs from the current Coupled Model Intercomparison Project 6 (CMIP6). The results follow the same post-processing used for the contribution of OSCAR to the Reduced Complexity Model Intercomparison Project (RCMIP) Phase 2 regarding the identification of stable configurations and the use of observational constraints. These constraints succeed in decreasing the overestimation of global surface air temperature over 2000–2019 with reference to 1961–1900 from 0.60±0.11 to 0.55±0.04 K (the constraint being 0.54±0.05 K). The equilibrium climate sensitivity (ECS) of the unconstrained OSCAR is 3.17±0.63 K, while CMIP5 and CMIP6 models have ECSs of 3.2±0.7 and 3.7±1.1 K, respectively. Applying observational constraints to OSCAR reduces the ECS to 2.78±0.47 K. Overall, the model qualitatively reproduces the responses of complex ESMs, although some differences remain due to the impact of observational constraints on the weighting of parametrizations. Specific features of OSCAR also contribute to these differences, such as its fully interactive atmospheric chemistry and endogenous calculations of biomass burning, wetlands CH4 and permafrost CH4 and CO2 emissions. Identified main points of needed improvements of the OSCAR model include a low sensitivity of the land carbon cycle to climate change, an instability of the ocean carbon cycle, the climate module that is seemingly too simple, and the climate feedback involving short-lived species that is too strong. Beyond providing a key diagnosis of the OSCAR model in the context of the reduced-complexity models, this work is also meant to help with the upcoming calibration of OSCAR on CMIP6 results and to provide a large group of CMIP6 simulations run consistently within a probabilistic framework.
BibTeX
@inproceedings{quilcaille2023cmip,
  author = {Yann Quilcaille and Thomas Gasser and Philippe Ciais and Oliviér Boucher},
  title = {{CMIP6 simulations with the compact Earth system model OSCAR v3.1}},
  booktitle = {Geoscientific},
  year = {2023},
  doi = {10.5194/gmd-16-1129-2023},
}
Figure