Inability of CMIP5 Models to Simulate Recent Strengthening of the Walker Circulation: Implications for Projections

Precipitation variability has great economic, social, and environmental impacts across the globe, and in particular in China. This paper evaluates the historical precipitation variability based on 20 general circulation models from the Coupled Model Intercomparison Project Phase 5 (CMIP5) archive over the 20th century relative to two observational data sets and quantifies CMIP5 improvements over CMIP3. Multimodel ensemble means and individual models are assessed. Three future emission scenarios are used (representative concentration pathways (RCP) 8.5, RCP 4.5, and RCP 2.6), and 21st century CMIP5 estimates are put into context based on the 20th century biases. We find that CMIP5 models can reproduce the spatial pattern of precipitation over China during the 20th century, which represents an improvement over CMIP3. However, the models overestimate the magnitude of seasonal and annual precipitation in most regions of China, especially along the eastern edge of the Tibetan Plateau, and underestimate summer precipitation over southeastern China. For China as a whole, CMIP5's overestimation of annual precipitation is greater than CMIP3, which can be traced back to a greater underestimation of summer precipitation in CMIP3. There is a large spread among individual models, with the greatest uncertainties in simulating summer precipitation. Trends and correlations also suggest a better agreement of CMIP5 with observations than CMIP3. Throughout the 20th century, both the observations and models show an increasing trend in precipitation over parts of northwestern China and a decreasing trend over the Tibetan Plateau. There is poor agreement in precipitation trends over the southeast and northeast regions. In general, multimodel means cannot capture the amplitude of observed multidecadal precipitation variability. In the 21st century, precipitation is generally projected to increase across all of China under all three scenarios. RCP 8.5 exhibits the largest significant trend at a rate of +1.5 mm/yr, corresponding to 16% precipitation increase by the end of the century. The RCP 2.6 scenario shows the smallest increases, at +0.5 mm/yr (6%) by 2100. The greatest increases are projected to occur over the Tibetan Plateau and eastern China in summer, suggesting an altered monsoonal circulation in the future. However, due to the uncertainties in CMIP5, future precipitation projections should be interpreted with caution.

The double‐intertropical convergence zone (ITCZ) bias is one of the most outstanding errors in all previous generations of climate models. Here, the annual double‐ITCZ bias and the associated precipitation bias in the latest climate models for Coupled Model Intercomparison Project (CMIP) Phase 6 (CMIP6) are examined in comparison to their previous generations (CMIP Phase 3 [CMIP3] and CMIP Phase 5 [CMIP5]). All three generations of CMIP models share similar systematic annual multi‐model ensemble mean precipitation errors in the tropics. The notorious double‐ITCZ bias and its big inter‐model spread persist in CMIP3, CMIP5, and CMIP6 models. Based on several tropical precipitation bias indices, the double‐ITCZ bias is slightly reduced from CMIP3 or CMIP5 to CMIP6. In addition, the annual equatorial Pacific cold tongue persists in all three generations of CMIP models, but its inter‐model spread is reduced from CMIP3 to CMIP5 and from CMIP5 to CMIP6.

An assessment is made of the modes of interannual variability in the seasonal mean summer and winter Southern Hemisphere (SH) 500-hPa geopotential height in the twentieth century in models from the Coupled Model Intercomparison Project (CMIP) phase 5 (CMIP5) dataset. Modes of variability of both the slow (signal) and intraseasonal (noise) components in the CMIP5 models are evaluated against those estimated from reanalysis data. There is general improvement in the leading modes of the slow (signal) component in CMIP5 models compared with the CMIP phase 3 (CMIP3) dataset. The largest improvement is in the spatial structures of the modes related to El Niño–Southern Oscillation variability in SH summer. An overall score metric is significantly higher for CMIP5 over CMIP3 in both seasons. The leading modes in the intraseasonal noise component are generally well reproduced in CMIP5 models, and there are few differences from CMIP3. A new total overall score metric is used to rank the CMIP5 models over both seasons. Weighting the seasons by the relative spread of overall scores is shown to be suitable for generating multimodel ensembles for further analysis of interannual variability. In multimodel ensembles, it is found that an ensemble of size 5 or 6 is sufficient in SH summer to reproduce well the dominant modes. In contrast, about 13 models are typically are required in SH winter. It is shown that it is necessary that the selected models individually reproduce well the leading modes of the slow component.

Abstract. By coordinating the design and distribution of global climate model simulations of the past, current, and future climate, the Coupled Model Intercomparison Project (CMIP) has become one of the foundational elements of climate science. However, the need to address an ever-expanding range of scientific questions arising from more and more research communities has made it necessary to revise the organization of CMIP. After a long and wide community consultation, a new and more federated structure has been put in place. It consists of three major elements: (1) a handful of common experiments, the DECK (Diagnostic, Evaluation and Characterization of Klima) and CMIP historical simulations (1850–near present) that will maintain continuity and help document basic characteristics of models across different phases of CMIP; (2) common standards, coordination, infrastructure, and documentation that will facilitate the distribution of model outputs and the characterization of the model ensemble; and (3) an ensemble of CMIP-Endorsed Model Intercomparison Projects (MIPs) that will be specific to a particular phase of CMIP (now CMIP6) and that will build on the DECK and CMIP historical simulations to address a large range of specific questions and fill the scientific gaps of the previous CMIP phases. The DECK and CMIP historical simulations, together with the use of CMIP data standards, will be the entry cards for models participating in CMIP. Participation in CMIP6-Endorsed MIPs by individual modelling groups will be at their own discretion and will depend on their scientific interests and priorities. With the Grand Science Challenges of the World Climate Research Programme (WCRP) as its scientific backdrop, CMIP6 will address three broad questions: – How does the Earth system respond to forcing? – What are the origins and consequences of systematic model biases? – How can we assess future climate changes given internal climate variability, predictability, and uncertainties in scenarios? This CMIP6 overview paper presents the background and rationale for the new structure of CMIP, provides a detailed description of the DECK and CMIP6 historical simulations, and includes a brief introduction to the 21 CMIP6-Endorsed MIPs.

Global and regional climate change assessments rely heavily on the general circulation model (GCM) outputs such as provided by the Coupled Model Intercomparison Project phase 5 (CMIP5). Here we evaluate the ability of 25 CMIP5 GCMs to simulate historical precipitation and temperature over central Africa and assess their future projections in the context of historical performance and intermodel and future emission scenario uncertainties. We then apply a statistical bias correction technique to the monthly climate fields and develop monthly downscaled fields for the period of 1948–2099. The bias‐corrected and downscaled data set is constructed by combining a suite of global observation and reanalysis‐based data sets, with the monthly GCM outputs for the 20th century, and 21st century projections for the medium mitigation (representative concentration pathway (RCP)45) and high emission (RCP85) scenarios. Overall, the CMIP5 models simulate temperature better than precipitation, but substantial spatial heterogeneity exists. Many models show limited skill in simulating the seasonality, spatial patterns, and magnitude of precipitation. Temperature projections by the end of the 21st century (2070–2099) show a robust warming between 2 and 4°C across models, whereas precipitation projections vary across models in the sign and magnitude of change (−9% to 27%). Projected increase in precipitation for a subset of models (single model ensemble (SME)) identified based on performance metrics and causal mechanisms are slightly higher compared to the full multimodel ensemble (MME) mean; however, temperature projections are similar between the two ensemble means. For the near‐term (2021–2050), neither the historical performance nor choice of models is related to the precipitation projections, indicating that natural variability dominated any signal. With fewer models, the “blind” MME approach will have larger uncertainties in future precipitation projections compared to projections by the SME models. We propose the latter a better approach in regions that lack quality climate observations. Our analyses also show that the choice of model and emission scenario dominate the uncertainty in precipitation projections, whereas the emission scenario dominates the temperature projections. Although our analyses are done for central Africa, the final Bias‐Corrected Spatially Downscaled data set is available for global land areas. The framework for climate change assessment and the data will be useful for a variety of climate assessment, impact, and adaptation studies.

The process of estimating the effect of a changing climate on the severity of future ocean storms is plagued by large uncertainties; for safe design and operation of offshore structures, it is nevertheless important that best possible estimates of climate effects is made given the available data. We explore the variability in estimates of 100-year return value of significant wave height (HS), and changes in estimates over a period of time, for output of WAVEWATCH-III models from 7 representative Coupled Model Intercomparison Project (CMIP) Phase 5 General Circulation Models (GCMs), and the FIO-ESM v2.0 CMIP Phase 6 GCM. Non-stationary extreme value analysis of peaks-over-threshold and block maxima using Bayesian inference provide posterior estimates of return values as a function of time; MATLAB software for the extreme value analysis is provided. Best overall estimates for return values, and changes in return value over the period 1979-2100, are calculated by averaging estimates for individual GCMs. We focus attention on neighbourhoods of locations east of Madagascar and south of Australia where a previous study of CMIP5-derived output reported significant decrease and increase in HS respectively, under Representative Concentration Pathway (RCP) scenarios RCP4.5 and RCP8.5. There is large variation between return value estimates from different GCMs, and with longitude and latitude within each neighbourhood for estimates based on samples corresponding to ≤ 165 years of model output; these sources of uncertainty tend to be larger than those due to typical modelling choices (such as choice of threshold for peaks over threshold, or block length for block maxima). However, we also find that careful threshold choice and block length are critical east of Madagascar, because of the presence of a mixed population of storms there. Nevertheless, there is general evidence supporting the trends reported by others, but these findings are conditional on the choice of 8 GCMs being representative of climate evolution. We use simple randomisation testing to identify “significant” departures from steady climate. The long 700-year pre-industrial control (piControl) output of the CMIP6 GCM offers an excellent opportunity to quantify the apparent inherent variability in return value as a function of time, estimated using a subsample of output corresponding to a continuous time interval of between 20 and 160 years in length, where no climate forcing is present. We find large variation in return value estimates of approximately ±15% made from samples corresponding to periods of time of around 50 years drawn from piControl data.

Ensembles of climate change projections created by general circulation models (GCMs) with high resolution are increasingly needed to develop adaptation strategies for regional climate change. The Meteorological Research Institute atmospheric GCM version 3.2 (MRI-AGCM3.2), which is listed in the Coupled Model Intercomparison Project phase 5 (CMIP5), has been typically run with resolutions of 60 km and 20 km. Ensembles of MRI-AGCM3.2 consist of members with multiple cumulus convection schemes and different patterns of future sea surface temperature, and are utilized together with their downscaled data; however, the limited size of the high-resolution ensemble may lead to undesirable biases and uncertainty in future climate projections that will limit its appropriateness and effectiveness for studies on climate change and impact assessments. In this study, to develop a comprehensive understanding of the regional precipitation simulated with MRI-AGCM3.2, we investigate how well MRI-AGCM3.2 simulates the present-day regional precipitation around the globe and compare the uncertainty in future precipitation changes and the change projection itself between MRI-AGCM3.2 and the CMIP5 multiple atmosphere–ocean coupled GCM (AOGCM) ensemble. MRI-AGCM3.2 reduces the bias of the regional mean precipitation obtained with the high-performing CMIP5 models, with a reduction of approximately 20% in the bias over the Tibetan Plateau through East Asia and Australia. When 26 global land regions are considered, MRI-AGCM3.2 simulates the spatial pattern and the regional mean realistically in more regions than the individual CMIP5 models. As for the future projections, in 20 of the 26 regions, the sign of annual precipitation change is identical between the 50th percentiles of the MRI-AGCM3.2 ensemble and the CMIP5 multi-model ensemble. In the other six regions around the tropical South Pacific, the differences in modeling with and without atmosphere–ocean coupling may affect the projections. The uncertainty in future changes in annual precipitation from MRI-AGCM3.2 partially overlaps the maximum–minimum uncertainty range from the full ensemble of the CMIP5 models in all regions. Moreover, on average over individual regions, the projections from MRI-AGCM3.2 spread over roughly 0.8 of the uncertainty range from the high-performing CMIP5 models compared to 0.4 of the range of the full ensemble.

While most models project large increases in agricultural drought frequency and severity in the 21st century, significant uncertainties exist in these projections. Here, we compare the model-simulated changes with observation-based estimates since 1900 and examine model projections from both the Coupled Model Inter-comparison Project Phase 3 (CMIP3) and Phase 5 (CMIP5). We use the self-calibrated Palmer Drought Severity Index with the Penman-Monteith potential evapotranspiration (PET) (sc_PDSI_pm) as a measure of agricultural drought. Results show that estimated long-term changes in global and hemispheric drought areas from 1900 to 2014 are consistent with the CMIP3 and CMIP5 model-simulated response to historical greenhouse gases and other external forcing, with the short-term variations within the model spread of internal variability, despite that regional changes are still dominated by internal variability. Both the CMIP3 and CMIP5 models project continued increases (by 50–200 % in a relative sense) in the 21st century in global agricultural drought frequency and area even under low-moderate emissions scenarios, resulting from a decrease in the mean and flattening of the probability distribution functions (PDFs) of the sc_PDSI_pm. This flattening is especially pronounced over the Northern Hemisphere land, leading to increased drought frequency even over areas with increasing sc_PDSI_pm. Large differences exist in the CMIP3 and CMIP5 model-projected precipitation and drought changes over the Sahel and northern Australia due to uncertainties in simulating the African Inter-tropical convergence zone (ITCZ) and the subsidence zone over northern Australia, while the wetting trend over East Africa reflects a robust response of the Indian Ocean ITCZ seen in both the CMIP3 and CMIP5 models. While warming-induced PET increases over all latitudes and precipitation decreases over subtropical land are responsible for mean sc_PDSI_pm decreases, the exact cause of its PDF flattening needs further investigation.

Lake Chad lost more than 80% of its surface area over the past decades as a result of environmental change and climate variability. It is not yet known how climate change will affect water resources availability in the basin over the coming decades. In this study, the reliability ensemble averaging (REA) technique was used to evaluate the performance of Coupled Model Intercomparison Project phase 5 (CMIP5) models in simulating present‐day precipitation and temperature (1980–2005), and to quantify the uncertainties in future projections (2050–2075) under two representative concentration pathways (RCPs) in the Lake Chad basin (LCB). Analyses were carried out at both annual and seasonal timescales. Overall, the CMIP5 models simulated precipitation better than temperature in the study area. Although the models were able to simulate the annual precipitation cycle in the basin, most models overestimated precipitation during the dry season and underestimated it during the monsoon season. Future annual basin precipitation is projected to increase by 2.5 and 5%, respectively, under RCP4.5 and RCP8.5 scenario by the middle of the century by most of the models and most of the model projections are within the REA uncertainty range. Despite the increase in projected annual precipitation in the basin, most models project a decrease in monsoon precipitation under both RCPs. Although the uncertainty range for future precipitation projections for most models lie within the range of natural climate variability, additional analysis is needed for results to be useful for any future planning in the study area.

California is one of the major uncertainty hotspots for climate change, as climate models have historically been split between projecting wetter and drier future conditions over the region. We analysed the future (mid‐century and end‐century) projections of California's winter precipitation changes from the latest Coupled Model Intercomparison Project Phase 6 (CMIP6), and studied its respective model agreement in comparison to the previous CMIP5 projections. Over northern California more than two thirds of the models in each ensemble agree on wetter future conditions. However, over southern California both ensembles show highly uncertain precipitation changes, with model projections almost equally divided between wetter or drier conditions. Projected end‐century precipitation changes range from −30% to +70% in CMIP5 and −20% to +80% in CMIP6. The CMIP6 ensemble mean changes are generally wetter and show larger model disagreement compared to CMIP5. Distribution of year‐to‐year precipitation indicates more extremely wet or dry years over southern California in CMIP6 compared to CMIP5, with some models suggesting that the five wettest years account for as much as ~55% of the 20‐year rainfall, and the five driest for as little as ~5%. Dynamically, both ensembles project weakened subsidence over Baja California that is stronger in CMIP6 than in CMIP5, in line with the wetter mean conditions in CMIP6. In the western tropical Pacific we find strengthening of the Hadley circulation in CMIP6 that is not seen in CMIP5, and more El Niño than La Niña conditions in the equatorial Pacific. More CMIP6 models also project an increase in ENSO events compared to CMIP5, and a stronger impact of ENSO on California's precipitation is found in CMIP6 than in CMIP5. These factors also contribute to larger model disagreement and more extremely wet or dry years over southern California in CMIP6.

Abstract. Simulation of the carbon cycle in climate models is important due to its impact on climate change, but many weaknesses in its reproduction were found in previous models. Improvements in the representation of the land carbon cycle in Earth system models (ESMs) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) include the interactive treatment of both the carbon and nitrogen cycles, improved photosynthesis, and soil hydrology. To assess the impact of these model developments on aspects of the global carbon cycle, the Earth System Model Evaluation Tool (ESMValTool) is expanded to compare CO2-concentration- and CO2-emission-driven historical simulations from CMIP5 and CMIP6 to observational data sets. A particular focus is on the differences in models with and without an interactive terrestrial nitrogen cycle. Overestimations of photosynthesis (gross primary productivity (GPP)) in CMIP5 were largely resolved in CMIP6 for participating models with an interactive nitrogen cycle but remaining for models without one. This points to the importance of including nutrient limitation. Simulating the leaf area index (LAI) remains challenging, with a large model spread in both CMIP5 and CMIP6. In ESMs, the global mean land carbon uptake (net biome productivity (NBP)) is well reproduced in the CMIP5 and CMIP6 multi-model means. However, this is the result of an underestimation of NBP in the Northern Hemisphere, which is compensated by an overestimation in the Southern Hemisphere and the tropics. Carbon stocks remain a large uncertainty in the models. While vegetation carbon content is slightly better represented in CMIP6, the inter-model range of soil carbon content remains the same between CMIP5 and CMIP6. Overall, a slight improvement in the simulation of land carbon cycle parameters is found in CMIP6 compared to CMIP5, but with many biases remaining, further improvements of models in particular for LAI and NBP is required. Models from modeling groups participating in both CMIP phases generally perform similarly or better in their CMIP6 compared to their CMIP5 models. This improvement is not as significant in the multi-model means due to more new models in CMIP6, especially those using older versions of the Community Land Model (CLM). Emission-driven simulations perform just as well as the concentration-driven models, despite the added process realism. Due to this, we recommend that ESMs in future Coupled Model Intercomparison Project (CMIP) phases perform emission-driven simulations as the standard so that climate–carbon cycle feedbacks are fully active. The inclusion of the nitrogen limitation led to a large improvement in photosynthesis compared to models not including this process, suggesting the need to view the nitrogen cycle as a necessary part of all future carbon cycle models. Possible benefits when including further limiting nutrients such as phosphorus should also be considered.

This work presents updated hydrologic projections for the Upper Colorado River Basin (UCRB) using downscaled (approximately 12 km) General Circulation Model (GCM) output from Coupled Model Intercomparison Project – Phase 5 (CMIP5) with a comparison to CMIP3 GCMs. We use the Soil and Water Assessment Tool model to simulate the impacts of end-of-century climate change on the UCRB using 21 CMIP5 and 18 CMIP3 GCMs, collected into one CMIP5 ensemble and one CMIP3 ensemble, respectively. Previous CMIP3 studies have identified a drier climate for the UCRB because of projected increases in temperature and decreases/little change in precipitation. Hydrologic simulations from CMIP5 inputs suggest wetter conditions than simulations based on CMIP3 inputs, yet drier conditions than the historical climate. Both ensembles lead to timing shifts in peak streamflow during the snowmelt season from changes in snowmelt, but the higher CMIP5 projected precipitation leads to, on average, peak streamflows 200–300 m3 s−1 larger (25–40% difference) than the CMIP3 projections. This difference is largely generated in the northern UCRB region, where CMIP5 simulations project much more significant increases in streamflow than CMIP3. This increase is largely due to an overall larger rise in precipitation in the CMIP5 ensemble (57% of the total UCRB area) compared to the CMIP3 ensemble (5%). Even with projected increases in precipitation, snowmelt is projected to decrease dramatically throughout the UCRB for both ensembles. The increases in precipitation and decreases in snowmelt leads to significant differences in hydrologic flux components between the CMIP3 and CMIP5 ensembles, such as end-of-century rises in soil water content and evapotranspiration in the CMIP5 ensemble compared to the CMIP3 ensemble. The difference between the dry CMIP3 and the somewhat wetter CMIP5 projections may be critical for water management in the already over-allocated UCRB.

Simulation of the carbon cycle in climate models is important due to its impact on climate change, but many weaknesses in its reproduction were found in previous models. Improvements in the representation of the land carbon cycle in Earth system models (ESMs) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) include the interactive treatment of both the carbon and nitrogen cycles, improved photosynthesis, and soil hydrology. To assess the impact of these model developments on aspects of the global carbon cycle, the Earth System Model Evaluation Tool (ESMValTool) is expanded to compare CO2-concentration- and CO2-emission-driven historical simulations from CMIP5 and CMIP6 to observational data sets. A particular focus is on the differences in models with and without an interactive terrestrial nitrogen cycle. Overestimations of photosynthesis (gross primary productivity (GPP)) in CMIP5 were largely resolved in CMIP6 for participating models with an interactive nitrogen cycle but remain for models without one. This points to the importance of including nutrient limitation in models. Simulating the leaf area index (LAI) remains challenging, with a large model spread in both CMIP5 and CMIP6. The global mean land carbon uptake (net biome productivity (NBP)) is well reproduced in the CMIP5 and CMIP6 multi-model means. This is the result of an underestimation of NBP in the Northern Hemisphere, compensated by an overestimation in the Southern Hemisphere and the tropics. Models from modeling groups participating in both CMIP phases generally perform similarly or better in their CMIP6 version compared to their CMIP5 version. Emission-driven simulations perform just as well as the concentration-driven models, despite the added process realism. Due to this, we recommend that ESMs in future Coupled Model Intercomparison Project (CMIP) phases perform emission-driven simulations as the standard so that climate–carbon cycle feedbacks are fully active. The inclusion of the nitrogen limitation led to a large improvement in photosynthesis compared to models not including this process, suggesting the need to view the nitrogen cycle as a necessary part of all future carbon cycle models. Overall, a slight improvement in the simulation of land carbon cycle parameters is found in CMIP6 compared to CMIP5, but with many biases remaining, further improvements of models in particular for LAI and NBP is required. Due to the inclusion of the study in ESMValTool, the analysis can easily be repeated on the upcoming CMIP7 models to evaluate the progress from CMIP6.

The changes in the winter climatology and variability of the East Asian winter monsoon (EAWM) for the late 21st century (2070–2099) under the Representative Concentration Pathway (RCP) 4.5 and 8.5 scenarios are projected in terms of EAWM indices (EAWMIs). Firstly, the capability of the climate models participating in the Coupled Model Intercomparison Project phase 5 (CMIP5) in simulating the boreal winter climatology and the interannual variability of the EAWM for the late 20th century (1971–2000) is examined. Nine of twenty-three climate models are selected based on the pattern correlations with observation and a multi-model ensemble is applied to the nine model data. Three of twelve EAWMIs that show the most significant temporal correlations between the observation and CMIP5 surface air temperatures are utilized. The ensemble CMIP5 is capable of reproducing the overall features of the EAWM in spite of some biases in the region. The negative correlations between the EAWMIs and boreal winter temperature are well reproduced and 3–5 years of the major interannual variation observed in this region are also well simulated according to power spectral analyses of the simulated indices. The fields regressed onto the indices that resemble the composite strong winter monsoon pattern are simulated more or less weakly in CMIP5 compared to the observation. However, the regressed fields of sea level pressure, surface air temperature, 500-hPa geopotential height, and 300-hPa zonal wind are well established with pattern correlations above 0.83 between CMIP5 and observation data. The differences between RCPs and Historical indicate strong warming, which increases with latitude, ranging from 1 to 5 °C under RCP4.5 and from 3 to 7 °C under RCP8.5 in the East Asian region. The anomalous southerly winds generally become stronger, implying weaker EAWMs in both scenarios. These features are also identified with fields regressed onto the indices in RCPs. The future projections reveal that the interannual variability of the indices will be maintained with an intensity similar to that of the present. The correlation between monsoon indices and Arctic Oscillation increases over time. On the other hand, the correlation between monsoon indices and North Atlantic Oscillation decreases.

Abstract. By regulating the global transport of heat, freshwater, and carbon, the Atlantic meridional overturning circulation (AMOC) serves as an important component of the climate system. During the late 20th and early 21st centuries, indirect observations and models suggest a weakening of the AMOC. Direct AMOC observations also suggest a weakening during the early 21st century but with substantial interannual variability. Long-term weakening of the AMOC has been associated with increasing greenhouse gases (GHGs), but some modeling studies suggest the build up of anthropogenic aerosols (AAs) may have offset part of the GHG-induced weakening. Here, we quantify 1900–2020 AMOC variations and assess the driving mechanisms in state-of-the-art climate models from the Coupled Model Intercomparison Project phase 6 (CMIP6). The CMIP6 forcing (GHGs, anthropogenic and volcanic aerosols, solar variability, and land use and land change) multi-model mean shows negligible AMOC changes up to ∼ 1950, followed by robust AMOC strengthening during the second half of the 20th century (∼ 1950–1990) and weakening afterwards (1990–2020). These multi-decadal AMOC variations are related to changes in North Atlantic atmospheric circulation, including an altered sea level pressure gradient, storm track activity, surface winds, and heat fluxes, which drive changes in the subpolar North Atlantic surface density flux. To further investigate these AMOC relationships, we perform a regression analysis and decompose these North Atlantic climate responses into an anthropogenic aerosol-forced component and a subsequent AMOC-related feedback. Similar to previous studies, CMIP6 GHG simulations yield robust AMOC weakening, particularly during the second half of the 20th century. Changes in natural forcings, including solar variability and volcanic aerosols, yield negligible AMOC changes. In contrast, CMIP6 AA simulations yield robust AMOC strengthening (weakening) in response to increasing (decreasing) anthropogenic aerosols. Moreover, the CMIP6 all-forcing AMOC variations and atmospheric circulation responses also occur in the CMIP6 AA simulations, which suggests these are largely driven by changes in anthropogenic aerosol emissions. More specifically, our results suggest that AMOC multi-decadal variability is initiated by North Atlantic aerosol optical thickness perturbations to net surface shortwave radiation and sea surface temperature (and hence sea surface density), which in turn affect sea level pressure gradient and surface wind and – via latent and sensible heat fluxes – sea surface density flux through its thermal component. AMOC-related feedbacks act to reinforce this aerosol-forced AMOC response, largely due to changes in sea surface salinity (and hence sea surface density), with temperature-related (and cloud-related) feedbacks acting to mute the initial response. Although aspects of the CMIP6 all-forcing multi-model mean response resembles observations, notable differences exist. This includes CMIP6 AMOC strengthening from ∼ 1950 to 1990, when the indirect estimates suggest AMOC weakening. The CMIP6 multi-model mean also underestimates the observed increase in North Atlantic ocean heat content, and although the CMIP6 North Atlantic atmospheric circulation responses – particularly the overall patterns – are similar to observations, the simulated responses are weaker than those observed, implying they are only partially externally forced. The possible causes of these differences include internal climate variability, observational uncertainties, and model shortcomings, including excessive aerosol forcing. A handful of CMIP6 realizations yield AMOC evolution since 1900 similar to the indirect observations, implying the inferred AMOC weakening from 1950 to 1990 (and even from 1930 to 1990) may have a significant contribution from internal (i.e., unforced) climate variability. Nonetheless, CMIP6 models yield robust, externally forced AMOC changes, the bulk of which are due to anthropogenic aerosols.