This study introduces CNRM‐ESM2‐1, the Earth system (ES) model of second generation developed by CNRM‐CERFACS for the sixth phase of the Coupled Model Intercomparison Project (CMIP6). CNRM‐ESM2‐1 ...offers a higher model complexity than the Atmosphere‐Ocean General Circulation Model CNRM‐CM6‐1 by adding interactive ES components such as carbon cycle, aerosols, and atmospheric chemistry. As both models share the same code, physical parameterizations, and grid resolution, they offer a fully traceable framework to investigate how far the represented ES processes impact the model performance over present‐day, response to external forcing and future climate projections. Using a large variety of CMIP6 experiments, we show that represented ES processes impact more prominently the model response to external forcing than the model performance over present‐day. Both models display comparable performance at replicating modern observations although the mean climate of CNRM‐ESM2‐1 is slightly warmer than that of CNRM‐CM6‐1. This difference arises from land cover‐aerosol interactions where the use of different soil vegetation distributions between both models impacts the rate of dust emissions. This interaction results in a smaller aerosol burden in CNRM‐ESM2‐1 than in CNRM‐CM6‐1, leading to a different surface radiative budget and climate. Greater differences are found when comparing the model response to external forcing and future climate projections. Represented ES processes damp future warming by up to 10% in CNRM‐ESM2‐1 with respect to CNRM‐CM6‐1. The representation of land vegetation and the CO2‐water‐stomatal feedback between both models explain about 60% of this difference. The remainder is driven by other ES feedbacks such as the natural aerosol feedback.
Key Points
This study introduces CNRM‐ESM2‐1 and describes its set‐up for CMIP6
Represented Earth system processes further impact the model response to external forcing than the model performance over present‐day
Represented Earth system processes damp future warming by up to 10%
Abstract
Some of the new generation CMIP6 models are characterised by a strong temperature increase in response to increasing greenhouse gases concentration
1
. At first glance, these models seem ...less consistent with the temperature warming observed over the last decades. Here, we investigate this issue through the prism of low-frequency internal variability by comparing with observations an ensemble of 32 historical simulations performed with the IPSL-CM6A-LR model, characterized by a rather large climate sensitivity. We show that members with the smallest rates of global warming over the past 6-7 decades are also those with a large internally-driven weakening of the Atlantic Meridional Overturning Circulation (AMOC). This subset of members also matches several AMOC observational fingerprints, which are in line with such a weakening. This suggests that internal variability from the Atlantic Ocean may have dampened the magnitude of global warming over the historical era. Taking into account this AMOC weakening over the past decades means that it will be harder to avoid crossing the 2 °C warming threshold.
The objective of this study is to evaluate the potential for History Matching (HM) to tune a climate system with multi‐scale dynamics. By considering a toy climate model, namely, the two‐scale ...Lorenz96 model and producing experiments in perfect‐model setting, we explore in detail how several built‐in choices need to be carefully tested. We also demonstrate the importance of introducing physical expertise in the range of parameters, a priori to running HM. Finally we revisit a classical procedure in climate model tuning, that consists of tuning the slow and fast components separately. By doing so in the Lorenz96 model, we illustrate the non‐uniqueness of plausible parameters and highlight the specificity of metrics emerging from the coupling. This paper contributes also to bridging the communities of uncertainty quantification, machine learning and climate modeling, by making connections between the terms used by each community for the same concept and presenting promising collaboration avenues that would benefit climate modeling research.
Plain Language Summary
Climate models are computer simulation codes that incorporate centuries of human knowledge of the physics of planet Earth. They are used to understand the past, the present and make projections about the future of our climate. To validate a climate model, scientists tune a number of its parameters so that it yields a simulated climate resembling real‐life observations as much as possible. The main challenge in this tuning task is the extreme cost of climate models which limits a lot the number of tuning experiments scientists can run. In this paper we are interested in a technique that uses artificial intelligence in order to replace the expensive climate model with a cheaper surrogate. We experiment on a simplified model to assess the strengths and weaknesses of this semi‐automatic technique, and show that it can be more efficient when combined with human expertise.
Key Points
The History Matching method is explained in detail then used for tuning a toy coupled model: the Lorenz 96 model
The importance of several design choices is demonstrated, especially when considering forced experiments such as Atmospheric Model Intercomparison Protocol and Ocean Model Intercomparison Project
We argue that this tuning method is semi‐automatic & highlight the importance of human expertise when considering it for real coupled models
The subpolar North Atlantic is a center of variability of ocean properties, wind stress curl, and air–sea exchanges. Observations and hindcast simulations suggest that from the early 1970s to the ...mid-1990s the subpolar gyre became fresher while the gyre and meridional circulations intensified. This is opposite to the relationship of freshening causing a weakened circulation, most often reproduced by climate models. The authors hypothesize that both these configurations exist but dominate on different time scales: a fresher subpolar gyre when the circulation is more intense, at interannual frequencies (configuration A), and a saltier subpolar gyre when the circulation is more intense, at longer periods (configuration B). Rather than going into the detail of the mechanisms sustaining each configuration, the authors’ objective is to identify which configuration dominates and to test whether this depends on frequency, in preindustrial control runs of five climate models from phase 5 of the Coupled Model Intercomparison Project (CMIP5). To this end, the authors have developed a novel intercomparison method that enables analysis of freshwater budget and circulation changes in a physical perspective that overcomes model specificities. Lag correlations and a cross-spectral analysis between freshwater content changes and circulation indices validate the authors’ hypothesis, as configuration A is only visible at interannual frequencies while configuration B is mostly visible at decadal and longer periods, suggesting that the driving role of salinity on the circulation depends on frequency. Overall, this analysis underscores the large differences among state-of-the-art climate models in their representations of the North Atlantic freshwater budget.
The variability of the circulation in the North Atlantic and its link with atmospheric variability are investigated in a realistic hindcast simulation from 1953 to 2003. The interannual-to-decadal ...variability of the subpolar gyre circulation and the Meridional Overturning Circulation (MOC) is mostly influenced by the North Atlantic Oscillation (NAO). Both circulations intensified from the early 1970s to the mid-1990s and then decreased. The monthly variability of both circulations reflects the fast barotropic adjustment to NAO-related Ekman pumping anomalies, while the interannual-to-decadal variability is due to the baroclinic adjustment to Ekman pumping, buoyancy forcing, and dense water formation, consistent with previous studies.
An original characteristic of the oceanic response to NAO is presented that relates to the spatial patterns of buoyancy and wind forcing over the North Atlantic. Anomalous Ekman pumping associated with a positive NAO phase first induces a decrease of the southern subpolar gyre strength and an intensification of the northern subpolar gyre. The latter is reinforced by buoyancy loss and dense water formation in the Irminger Sea, where the cyclonic circulation increases 1–2 yr after the positive NAO phase. Increased buoyancy loss also occurs in the Labrador Sea, but because of the early decrease of the southern subpolar gyre strength, the intensification of the cyclonic circulation is delayed. Hence the subpolar gyre and the MOC start increasing in the Irminger Sea, while in the Labrador Sea the circulation at depth leads its surface counterpart. In this simulation where the transport of dense water through the North Atlantic sills is underestimated, the MOC variability is well represented by a simple integrator of convection in the Irminger Sea, which fits better than a direct integration of NAO forcing.
The Institut Pierre‐Simon Laplace Climate Modeling Center has produced an ensemble of extended historical simulations using the IPSL‐CM6A‐LR climate model. This ensemble (referred to as IPSL‐EHS) is ...composed of 32 members over the 1850–2059 period that share the same external forcings but differ in their initial conditions. In this study, we assess the simulated decadal to multidecadal climate variability in the IPSL‐EHS. In particular, we examine the global temperature evolution and recent warming trends, and their consistency with ocean heat content and sea ice cover. The model exhibits a large low‐frequency internal climate variability. In particular, a quasi‐bicentennial mode of internal climate variability is present in the model and is associated with the Atlantic Meridional Overturning Circulation. Such variability modulates the global mean surface air temperature changes over the historical period by about ∼ 0.1K. This modulation is found to be linked to the phase present in the initial condition state of each member. This variability appears to decrease during the 1850–2018 period in response to external forcings. The analysis of the ocean heat content reveals furthermore an overestimation of the ocean stratification, which likely leads to an overestimation of the recent warming rate on average.
Plain Language Summary
The Institut Pierre‐Simon Laplace (IPSL) developed an ensemble of 32 simulations over the 1850–2059 period using the IPSL‐CM6A‐LR climate model. Such a large ensemble allows a better sampling of the internally generated variability. Moreover, the ensemble averaging provides an estimation of the forced variability induced by the greenhouse gases and the aerosol concentration used as boundary conditions. In this study, we assess the simulated decadal to multidecadal climate variability in the IPSL ensemble. Relative to the large variability of the model, the evolution of observed surface temperature and sea ice cover is within the range of possibilities of the ensemble. The oceanic circulation and sea surface temperature over the North Atlantic are key players in the low‐frequency internal variability of the model.
Key Points
A large part of the spread of temperature and sea ice trends in the IPSL ensemble is related to a large multicentennial internal variability
Some members of the IPSL ensemble are consistent with the observed surface temperature, sea ice variations, and ocean heat content evolution
The low‐frequency internal climate variability of IPSL‐CM6A‐LR decreases since the 2000s in response to external forcing
Large differences in the Atlantic meridional overturning circulation (AMOC) exhibited between the available ocean models pose problems as to how they can be interpreted for climate policy. A novel ...Lagrangian methodology has been developed for use with ocean models that enables a decomposition of the AMOC according to its source waters of subduction from the mixed layer of different geographical regions. The method is described here and used to decompose the AMOC of the Centre National de Recherches Météorologiques (CNRM) ocean model, which is approximately 4.5 Sv (1 Sv = 10⁶ m³ s−1) too weak at 26°N, compared to observations. Contributions from mixed layer subduction to the peak AMOC at 26°N in the model are dominated by the Labrador Sea, which contributes 7.51 Sv; but contributions from the Nordic seas, the Irminger Sea, and the Rockall basin are also important. These waters mostly originate where deep mixed layers border the topographic slopes of the Subpolar Gyre and Nordic seas. The too-weak model AMOC can be explained by weak model representations of the overflow and of Irminger Sea subduction. These are offset by the large Labrador Sea component, which is likely to be too strong as a result of unrealistically distributed and too-deep mixed layers near the shelf.
The trustworthiness of neural networks is often challenged because they lack the ability to express uncertainty and explain their skill. This can be problematic given the increasing use of neural ...networks in high stakes decision‐making such as in climate change applications. We address both issues by successfully implementing a Bayesian Neural Network (BNN), where parameters are distributions rather than deterministic, and applying novel implementations of explainable AI (XAI) techniques. The uncertainty analysis from the BNN provides a comprehensive overview of the prediction more suited to practitioners' needs than predictions from a classical neural network. Using a BNN means we can calculate the entropy (i.e., uncertainty) of the predictions and determine if the probability of an outcome is statistically significant. To enhance trustworthiness, we also spatially apply the two XAI techniques of Layer‐wise Relevance Propagation (LRP) and SHapley Additive exPlanation (SHAP) values. These XAI methods reveal the extent to which the BNN is suitable and/or trustworthy. Using two techniques gives a more holistic view of BNN skill and its uncertainty, as LRP considers neural network parameters, whereas SHAP considers changes to outputs. We verify these techniques using comparison with intuition from physical theory. The differences in explanation identify potential areas where new physical theory guided studies are needed.
Plain Language Summary
Understanding ocean dynamics and how they are affected by global heating is crucial for understanding climate change impacts. Neural networks are ideally suited to this problem, but do not explain how they make predictions nor express how certain they are of the predictions' accuracy, which considerably limits their trustworthiness for ocean science problems. Here, we address both issues by using a “Bayesian Neural Network” (BNN), which directly expresses prediction uncertainty, and applying explainable AI (XAI) techniques to explain how the BNN arrives at its prediction. The BNN provides a comprehensive overview more suited to addressing the core problem than that provided by classical neural networks. We also apply two XAI techniques (SHAP and LRP) to the BNN and evaluate their trustworthiness by comparing the similarities and differences between their explanations with intuition from physical theory. Any differences offer an opportunity to develop physical theory guided by what the BNN considers important.
Key Points
Novel use of a Bayesian Neural Network (BNN) to quantify uncertainty in ocean dynamical regime classifications, giving a holistic prediction
Explaining the skill of a BNN using two techniques originating from two different classes of explainable AI: SHapley Additive exPlanation (SHAP) and Layer‐wise Relevance Propagation (LRP)
Trustworthiness is evaluated by comparing similarities and differences between SHAP and LRP explanations with intuition from physical theory
The Atlantic Meridional Overturning Circulation (AMOC) is a crucial element of the Earth climate. It is a complex circulation system difficult to monitor and to model. There is considerable debate ...regarding its evolution over the last century as well as large uncertainty about its fate at the end of this century. We depict here the progress since the IPCC SROCC report, offering an update of its chapter 6.7. We also show new results from a high-resolution ocean model and a CMIP6 model to investigate the impact of Greenland Ice Sheet (GrIS) melting, a key uncertainty for past and future AMOC changes. The ocean-only simulation at 1/24° resolution in the Arctic-North Atlantic Ocean performed over the period 2004–2016 indicates that the spread of the Greenland freshwater runoff toward the center of the Labrador Sea, where oceanic convection occurs, seems larger in this model than in a CMIP6 model. Potential explanations are related to the model spatial resolution and the representation of mesoscale processes, which more realistically transport the freshwater released around the shelves and, through eddies, provides strong lateral exchanges between the fine-scale boundary current and the convective basin in the Labrador Sea. The larger freshening of the Labrador Sea in the high-resolution model then strongly affects deep convection activity. In the simulation including GrIS melting, the AMOC weakens by about 2 Sv after only 13 years, far more strongly than what is found in the CMIP6 model. This difference raises serious concerns on the ability of CMIP6 models to correctly assess the potential impact of GrIS melting on the AMOC changes over the last few decades as well as on its future fate. To gain confidence in the GrIS freshwater impacts on climate simulations and therefore in AMOC projections, urgent progress should be made on the parameterization of mesoscale processes in ocean models.
The Tuning Strategy of IPSL‐CM6A‐LR Mignot, Juliette; Hourdin, Frédéric; Deshayes, Julie ...
Journal of advances in modeling earth systems,
20/May , Volume:
13, Issue:
5
Journal Article
Peer reviewed
Open access
The assessment of current and future risks for natural and human systems associated with climate change largely relies on numerical simulations performed with state‐of‐the‐art climate models. Various ...steps are involved in the development of such models, from development of individual components of the climate system up to free parameter calibration of the fully coupled model. Here, we describe the final tuning phase for the IPSL‐CM6A‐LR climate model. This phase alone lasted more than 3 years and relied on several pillars: (i) the tuning against present‐day conditions given a small adjustment of the ocean surface albedo to compensate for the current oceanic heat uptake, (ii) the release of successive versions after adjustments of the individual components, implying a systematic and recurrent adjustment of the atmospheric energetics, and (iii) the use of a few metrics based on large scale variables such as near‐global mean temperature, summer Arctic sea‐ice extent, as targets for the tuning. Successes, lessons and prospects of this tuning strategy are discussed.
Plain Language Summary
Evaluating current and future risks for natural and human systems associated with climate change is largely based on numerical simulations performed with models of the climate system, which includes the atmosphere, the land, the ocean, the cryosphere, and the oceanic and terrestrial biosphere. Various steps are involved in the development of such models. First, models for individual components are developed and tested. Second, many aspects are represented with parameterizations that summarize the effect of a missing process, such as those happening on scales that are smaller than the model grid sizes. The parameterizations in turn involve many parameters, sometimes poorly estimated from observations, that have to be calibrated. Here, we describe the final tuning phase of the IPSL‐CM6A‐LR climate model, which includes several novel aspects: first, the choice to calibrate the model against present‐day observations, which implies taking into account the transient nature of the observed climate; second, the systematic and recurrent adjustment of the atmospheric radiative budget; third, the use of a few large scale observable variables as targets. Successes, lessons and prospects of this tuning strategy are discussed.
Key Points
The tuning process of IPSL‐CM6A‐LR under present‐day control conditions is described
The associated continuous atmospheric energetics adjustment is presented
Successes, lessons and prospects of the IPSL‐CM6A‐LR tuning strategy are discussed