How to reduce long-term drift in present-day and deep-time simulations?

TL;DR

This paper investigates methods to reduce long-term drift in climate models by analyzing the effects of model configuration and tuning procedures on the stability and accuracy of control runs, with implications for both present-day and deep-time simulations.

Contribution

It introduces a systematic approach to assess and improve climate model stability by examining the interplay between model configurations and tuning, applicable to geological past simulations.

Findings

Tuning procedures significantly influence climate model stability.

Different model configurations affect the magnitude of climate drift.

Robust methods can be applied across various climate models.

Abstract

Climate models are often affected by long-term drift that is revealed by the evolution of global variables such as the ocean temperature or the surface air temperature. This spurious trend reduces the fidelity to initial conditions and has a great influence on the equilibrium climate after long simulation times. Useful insight on the nature of the climate drift can be obtained using two global metrics, i.e. the energy imbalance at the top of the atmosphere and at the ocean surface. The former is an indicator of the limitations within a given climate model, at the level of both numerical implementation and physical parameterisations, while the latter is an indicator of the goodness of the tuning procedure. Using the MIT general circulation model, we construct different configurations with various degree of complexity (i.e. different parameterisations for the bulk cloud albedo, inclusion or not of friction heating, different bathymetry configurations) to which we apply the same tuning procedure in order to obtain control runs for fixed external forcing where the climate drift is minimised. We find that the interplay between tuning procedure and different configurations of the same climate model provides crucial information on the stability of the control runs and on the goodness of a given parameterisation. This approach is particularly relevant for constructing good-quality control runs of the geological past where huge uncertainties are found in both initial and boundary conditions. We will focus on robust results that can be generally applied to other climate models.