The response of the climate system to increased greenhouse gases and other radiative perturbations is governed by a combination of fast and slow feedbacks. Slow feedbacks are typically activated in response to changes in ocean temperatures on decadal timescales and manifest as changes in climatic state with no recent historical analogue. However, fast feedbacks are activated in response to rapid atmospheric physical processes on weekly timescales, and they are already operative in the present-day climate. This distinction implies that the physics of fast radiative feedbacks is present in the historical meteorological reanalyses used to train many recent successful machine-learning-based (ML) emulators of weather and climate. In addition, these feedbacks are functional under the historical boundary conditions pertaining to the top-of-atmosphere radiative balance and sea-surface temperatures. Together, these factors imply that we can use historically trained ML weather emulators to study the response of radiative-convective equilibrium (RCE), and hence the global hydrological cycle, to perturbations in carbon dioxide and other well-mixed greenhouse gases. Without retraining on prospective Earth system conditions, we use ML weather emulators to quantify the fast precipitation response to reduced and elevated carbon dioxed concentrations with no recent historical precedent. We show that the responses from historically trained emulators agree with those produced by full-physics Earth System Models (ESMs). In conclusion, we discuss the prospects for and advantages from using ESMs and ML emulators to study fast processes in global climate.
In climate science, a central challenge is to predict the future evolution of the climate system given only observations available up to the present. The response to increased greenhouse gases (GHGs) is governed by a combination of fast and slow feedbacks. Slow feedbacks are activated in response to decadal changes in ocean temperatures and introduce changes in climatic state with no recent historical analogue. In contrast, fast feedbacks emerge from atmospheric radiative and convective processes operating on weekly timescales, and these are already fully operative in the present-day climate.
Radiative-convective equilibrium (RCE) is a key process in Earth’s troposphere that governs these fast feedbacks. RCE brings the troposphere into a global, time-mean thermal equilibrium by mostly balancing atmospheric radiative cooling with latent heating released when water vapor is lifted by convection and condenses to form clouds [1,2]. Based on the ratio of the global atmospheric water content to the global mean rainfall rate, RCE adjusts on a characteristic timescale of roughly one week, which is orders of magnitude faster than the thermal response of the upper ocean. Radiation itself responds almost instantly to radiatively active agents in the Earth’s atmosphere such as GHGs. Hence, the processes governing the response of RCE to an instantaneous pulse of GHGs are very fast compared to the slow thermally driven feedbacks in the climate system. These fast adjustments, first identified in simulations of instantaneous CO 2 forcing, have since been confirmed across multi-model ensembles and play a central role in shaping the hydrological response to climate forcing [3][4][5][6][7][8][9][10][11][12][13].
These processes can be illustrated using a representative subset of the physics-based Earth System Models (ESMs) submitted to the Coupled Model Intercomparison Project Phase 6 (CMIP6) [14]. In the abrupt4xCO2 CMIP6 experiment, an instantaneous quadrupling of atmospheric CO 2 from pre-industrial concentrations reduces atmospheric radiative cooling [2]. Figure 1a shows that global-mean sea surface temperatures (SSTs) in these ESMs change relative to the pre-industrial control simulations over the course of one month by only approximately 0.1 • C, to the abrupt4xCO2 experiment from the Coupled Model Intercomparison Project (CMIP) v6 [14].
Lead time is shown as days since CO 2 was quadrupled. Listed in Table B1, models were selected based on availability of data at daily resolution and sufficient metadata to determine the branch points of the instantaneous quadrupling simulations from the corresponding pre-industrial control simulations. Panel (a) shows the response of ocean surface temperature, and panel (b) shows the response of precipitation. For precipitation and temperature, the ∆ refers to the global-mean differences between the abrupt4xCO2 simulations and the pre-industrial control simulations from which these were branched. Shading denotes ±1 standard deviation across the multi-model ensemble.
a small fraction of the equilibrated response in SST. Therefore, during the first month, the upper and lower boundary conditions on the Earth’s atmosphere, i.e., the solar insolation and sea-surface temperatures, remain approximately the same as the boundary conditions in the preindustrial simulations. On ∼10-day timescales, the hydrological cycle responds to the reduced radiative cooling with a corresponding reduction in latent heating from condensation (and hence precipitation) [3]. Figure 1b shows this fast response in the CMIP6 multi-model ensemble.
Here, we test whether machine learning (ML) emulators of weather and climate can reproduce these fast adjustments of RCE without being trained on simulations with elevated CO 2 . Physics-based ESMs approach climate prediction by encoding fundamental laws of physics, chemistry, and biogeochemistry, approximating subgrid-scale physical processes, and extrapolating into future climate states. On the other hand, ML emulators are trained to emulate the time evolution of the atmosphere directly from historical reanalyses [15][16][17][18][19]. While they can be expensive to train, ML emulators are significantly faster and more energy-efficient during inference than their physics-based counterparts [20]. ML emulators also maintain similar accuracy on weather and climate timescales [16,17,[21][22][23][24]. Here we use the Allen Institute for Artificial Intelligence (Ai2) Climate Emulator (ACE) [25], an ML emulator that autoregressively predicts the three-dimensional atmospheric state with a 6-hour timestep and 1-degree horizontal resolution (Section 4.2). We use the version of ACE trained on 40 years of simulations from a physics-based atmosphere model called the Energy Exascale Earth System Model Atmosphere Model (EAMv2) [25,26]. In these simulations, EAMv2 uses an annually repeating cycle of climatological SSTs (2005-2014 average) as its boundary condition, and it has fixed gre
This content is AI-processed based on open access ArXiv data.