The Community Earth System Model version 2 (CESM2)74 The Community Earth System Model version 2 (CESM2) is the latest generation of the coupled 75 climate / Earth system models developed

Confidential manuscript submitted to Journal of Advances in Modeling Earth Systems

The Community Earth System Model version 2 (CESM2) 1

G. Danabasoglu1, J.-F. Lamarque1, J. Bacmeister1, D. A. Bailey1, A. K. DuVivier1, J. 2

Edwards1, L. K. Emmons2, J. Fasullo1, R. Garcia2, A. Gettelman1,2, C. Hannay1, M. M. 3

Holland1, W. G. Large1, P. H. Lauritzen1, D. M. Lawrence1, J. T. M. Lenaerts3, K. 4

Lindsay1, W. H. Lipscomb1, M. J. Mills2, R. Neale1, K. W. Oleson1, B. Otto-Bliesner1, A. S. 5

Phillips1, W. Sacks1, S. Tilmes2, L. van Kampenhout4, M. Vertenstein1, A. Bertini1, J. 6

Dennis5, C. Deser1, C. Fischer1, B. Fox-Kemper6, J. E. Kay7, D. Kinnison2, P. J. Kushner8, 7

V. E. Larson9, M. C. Long1, S. Mickelson5, J. K. Moore10, E. Nienhouse5, L. Polvani11, P. J. 8

Rasch12, W. G. Strand1 9

1Climate and Global Dynamics Laboratory, National Center for Atmospheric Research, Boulder, 10

CO, USA 11

2Atmospheric Chemistry Observations and Modeling Laboratory, National Center for 12

Atmospheric Research, Boulder, CO, USA 13

3Department of Atmospheric and Oceanic Sciences, University of Colorado Boulder, Boulder, 14

CO, USA 15

4Institute for Marine and Atmospheric Research Utrecht, Utrecht University, The Netherlands 16

5Computational and Information Systems Laboratory, National Center for Atmospheric 17

Research, Boulder, CO, USA 18

6Department of Earth, Environmental, and Planetary Sciences, Brown University, Providence, 19

RI, USA 20

7Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, 21

Boulder, CO, USA 22

8Department of Physics, University of Toronto, Toronto, ON, Canada 23

9Department of Mathematical Sciences, University of Wisconsin – Milwaukee, Milwaukee, WI, 24

USA 25

10Department of Earth System Science, University of California Irvine, Irvine, CA, USA 26

11Applied Mathematics / Earth and Environmental Sciences, Columbia University, New York, 27

NY, USA 28

12Atmospheric Science and Global Change Division, Pacific Northwest National Laboratory, 29

Richland, WA, USA 30

Corresponding author: Gokhan Danabasoglu ([email protected]) 31

mailto:[email protected])


2

Key Points: 32

• Community Earth System Model version 2 includes many substantial science and 33

infrastructure improvements since its previous version. 34

• Preindustrial control and historical simulations were performed with low-top and high-top 35

with comprehensive chemistry atmospheric models. 36

• Comparisons to observations are improved relative to previous versions, including major 37

reductions in radiation and precipitation biases. 38

39

Keywords: Community Earth System Model (CESM); Global coupled Earth system modeling; 40

Pre-industrial and historical simulations; Coupled model development and evaluation 41


3

Abstract 42

An overview of the Community Earth System Model version 2 (CESM2) is provided, including 43

a discussion of the challenges encountered during its development and how they were addressed. 44

In addition, an evaluation of a pair of CESM2 long pre-industrial control and historical ensemble 45

simulations is presented. These simulations were performed using the nominal 1° horizontal 46

resolution configuration of the coupled model with both the “low-top” (40 km, with limited 47

chemistry) and “high-top” (140 km, with comprehensive chemistry) versions of the atmospheric 48

component. CESM2 contains many substantial science and infrastructure improvements and new 49

capabilities since its previous major release, CESM1, resulting in improved historical 50

simulations in comparison to CESM1 and available observations. These include major reductions 51

in low latitude precipitation and short-wave cloud forcing biases; better representation of the 52

Madden-Julian Oscillation; better El Niño – Southern Oscillation-related teleconnections; and a 53

global land carbon accumulation trend that agrees well with observationally-based estimates. 54

Most tropospheric and surface features of the low- and high-top simulations are very similar to 55

each other, so these improvements are present in both configurations. CESM2 has an equilibrium 56

climate sensitivity of 5.1-5.3°C, larger than in CESM1, primarily due to a combination of 57

relatively small changes to cloud microphysics and boundary layer parameters. In contrast, 58

CESM2’s transient climate response of 1.9-2.0°C is comparable to that of CESM1. The model 59

outputs from these and many other simulations are available to the research community, and they 60

represent CESM2’s contributions to the Coupled Model Intercomparison Project phase 6 61

(CMIP6). 62


4

Plain Language Summary 63

The Community Earth System Model (CESM) is an open-source, comprehensive model used in 64

simulations of the Earth’s past, present, and future climates. The newest version, CESM2, has 65

many new technical and scientific capabilities ranging from a more realistic representation of 66

Greenland's evolving ice sheet, to the ability to model in detail how crops interact with the larger 67

Earth system, to improved representation of clouds and rain, and to the addition of wind-driven 68

waves on the model's ocean surface. The datasets from a large set of simulations that include 69

integrations for the pre-industrial conditions (1850s) and for the 1850-2014 historical period are 70

available to the community, representing CESM2’s contributions to the Coupled Model 71

Intercomparison Project phase 6 (CMIP6). 72

1 Introduction 73

The Community Earth System Model version 2 (CESM2) is the latest generation of the coupled 74

climate / Earth system models developed as a collaborative effort between scientists, software 75

engineers, and students from the National Center for Atmospheric Research (NCAR), 76

universities, and other research institutions. The CESM2 builds on many successes of its 77

predecessors. The first global coupled climate model that did not use any flux corrections, which 78

were previously required to stabilize coupled simulations even in the absence of changed 79

radiative forcing, was developed at NCAR. Washington and Meehl [1989] performed several 80

short (multi-decadal) simulations with this model to study climate sensitivity. The first coupled 81

climate model to achieve a stable present-day control simulation without any flux corrections 82

after long multi-century integrations was the Climate System Model [CSM; Boville and Gent, 83

1998]. This latter effort was followed by the Community Climate System Model version 2 84


5

[CCSM2; Kiehl and Gent, 2004], the CCSM3 [Collins et al., 2006], and the CCSM4 [Gent et al., 85

2011]. Additional capabilities such as interactive carbon – nitrogen cycling, global dynamic 86

vegetation and land use change due to anthropogenic activities, a marine biogeochemistry 87

module, a dynamic ice sheet model, and new chemical and physical processes for direct and 88

indirect aerosol effects were added to develop CCSM into an Earth system model. With these 89

new capabilities, the model was renamed the Community Earth System Model [CESM1; Hurrell 90

et al., 2013]. 91

These CCSM and CESM versions have been at the forefront of national and international efforts 92

to understand and predict the behavior of Earth's climate. Output from numerous simulations 93

using CCSM and CESM has been routinely used in studies to better understand the processes 94

and mechanisms responsible for climate variability and change. Most of these studies make use 95

of CCSM’s and CESM’s contributions to the various phases of the Coupled Model 96

Intercomparison Project (CMIP). Evaluation of those contributions has identified CCSM and 97

CESM as among the most realistic climate models in the world based on a few metrics that 98

compare the model outputs against present-day observationally-based datasets [Knutti et al., 99

2013]. In addition, CESM1 has also been used in key activities to advance our understanding of 100

the climate system and its variability and predictability, by supplementing CESM1’s 101

contributions to the CMIPs with other community-driven science efforts. These include the 102

CESM1 Large Ensemble [CESM1(LENS); Kay et al., 2015], the CESM Decadal Prediction 103

Large Ensemble [CESM-DPLE; Yeager et al., 2018], the CESM Last Millennium Ensemble 104

[CESM-LME; Otto-Bliesner et al., 2016], and the CESM Stratospheric Aerosol Geoengineering 105

Large Ensemble [GLENS; Tilmes et al., 2018]. 106


6

The CESM2 was released to the community in June 2018 (available at 107

www.cesm.ucar.edu:/models/cesm2/). Simulations with the model have been done as 108

contributions to the CMIP phase 6 [CMIP6; Eyring et al., 2016] using the nominal 1° horizontal 109

resolution configuration with both the so-called “low-top” (limited chemistry) and “high-top” 110

with comprehensive chemistry versions of the atmospheric model (see Section 2). Following the 111

notation introduced in Hurrell et al. [2013], these configurations will be referred to as 112

CESM2(CAM6) and CESM2(WACCM6), respectively. CMIP6-required core simulations are 113

the so-called Diagnostic, Evaluation, and Characterization of Klima (DECK) experiments that 114

consist of a long pre-industrial (PI) control simulation; an abrupt quadrupling of CO2 115

concentration simulation; a 1% per year CO2 concentration increase simulation; and an AMIP 116

(Atmospheric Model Intercomparison Project) simulation forced with prescribed observed sea 117

surface temperatures (SSTs) and sea-ice concentrations. These are complemented by multiple 118

simulations of the historical (1850-2014) period. In addition, CESM2 is participating in about 20 119

Model Intercomparison Projects (MIPs) within CMIP6, including the ScenarioMIP which 120

requests future-projection simulations driven by alternative plausible futures of emissions and 121

land use [O’Neill et al., 2016]. The output fields from these CESM2 simulations have been 122

posted on the Earth System Grid Federation (ESGF; https://esgf-node.llnl.gov/search/cmip6). To 123

expedite the use of CESM2 by the community primarily for CMIP6-related science and 124

simulations, two incremental releases of CESM2 with the same base code as in the June 2018 125

release were made available in December 2018 (CESM2.1.0) and in June 2019 (CESM2.1.1). 126

These two releases contain additional configurations from which many of the CMIP6 127

experiments can be run as out-of-the-box simulations. 128

http://www.cesm.ucar.edu/models/cesm2/


7

This manuscript provides an overview of the CESM2 as well as an evaluation of the solutions 129

from CESM2(CAM6) and CESM2(WACCM6) PI control simulations and from the subsequent 130

ensemble sets of historical simulations with 11 and 3 members, respectively. Section 2 131

summarizes the major new features and capabilities introduced in each component model since 132

CESM1. Section 3 briefly documents challenges encountered during the development process 133

and the strategies adopted to address them. Initial conditions and forcing datasets used in the 134

simulations are detailed in Sections 4 and 5, respectively. Information on model tuning, spin-up, 135

and drifts in the PI control simulations is presented in Section 6. Highlights from primarily the 136

historical simulations that include variables from many component models are presented in 137

Sections 7-11. CESM2 results are shown with respect to CESM1 PI control and CESM1(LENS) 138

historical simulations as well as available observations. Section 12 briefly discusses the new 139

model’s equilibrium climate sensitivity and transient climate response. Finally, Section 13 140

provides a summary and discussion. While much of our analysis makes use of the full set of 141

ensemble members available, in a few instances fewer (or even one) ensemble members are used 142

for clarity and simplicity as many of the simulation characteristics or improvements are robust 143

across the ensemble members. Solutions from the PI control and historical simulations as well as 144

many other CESM2 DECK and MIP simulations are analyzed and documented in more detail in 145

the manuscripts collected as part of the AGU CESM2 Virtual Special Issue. 146

2 CESM2 and its Components 147

The CESM2 is an open-source community coupled model consisting of ocean, atmosphere (both 148

low-top and high-top with comprehensive chemistry), land, sea-ice, land-ice, river, and wave 149

models that exchange states and fluxes via a coupler (Fig. 1). It contains many substantial 150

science and infrastructure improvements and capabilities since its previous version, CESM1. In 151


8

this section, each component model is introduced, and new developments are summarized. As 152

noted below, the individual version numbers across component models differ, reflecting their 153

independent development paths. 154

Figure 1. Schematic of the CESM2 component models. 155

2.1 Atmosphere 156

The Community Atmosphere Model version 6 (CAM6) uses the same Finite Volume (FV) 157

dynamical core [Lin and Rood, 1997] as in CCSM4 and CESM1, but it includes many changes to 158

the existing representation of physical processes. With the exception of radiation [Rapid 159

Radiative Transfer Model for General circulation models, RRTMG; Iacono et al., 2008], all 160

parameterizations have undergone some level of change. The primary change is the inclusion of 161

the unified turbulence scheme, Cloud Layers Unified By Binormals [CLUBB; Golaz et al., 2002; 162

Larson, 2017]. CLUBB unifies the description of cloudy turbulent layers by directly replacing 163

the separate shallow-convection, boundary layer, and grid-scale condensation schemes present in 164

CAM5 [Bogenschutz et al., 2013]. It is a high-order closure representation of moist turbulence 165

that closes many terms by novel use of a simple, multivariate binormal probability density 166

function (PDF) describing subgrid scale variations in temperature, humidity, and vertical 167


9

velocity. The Morrison-Gettelman cloud microphysics scheme [MG1; Morrison and Gettelman, 168

2008] has been replaced by an updated version [MG2; Gettelman and Morrison, 2015], which is 169

now able to forecast, instead of diagnose, the mass and number concentration of falling 170

condensed species (rain and snow). Notably, mixed phase ice nucleation depends on aerosols, 171

rather than just temperature, following Hoose et al. [2010], as implemented in CESM2 by Wang 172

et al. [2014] with modifications of Shi et al. [2015]. Aerosols are treated using the Modal 173

Aerosol Model version 4 [MAM4; Liu et al., 2016]. A new subgrid orographic form drag 174

parameterization [Beljaars et al., 2004] replaces the Turbulent Mountain Stress (TMS) scheme, 175

and now operates with a diagnosed profile that may extend above the surface level. An 176

anisotropic orographic gravity wave drag scheme following Scinocca and McFarlane [2000] 177

represents vertically propagating gravity waves and near-surface drag as a function of mountain 178

ridge orientation and height to more accurately represent the dependence on near-surface flow 179

direction and mountain orientation. The workhorse version of CAM6 uses a nominal 1° (1.25° in 180

longitude and 0.9° in latitude) horizontal resolution with 32 vertical levels and a model top at 181

2.26 hPa (about 40 km). This version of the model has relatively coarse stratospheric 182

representation (and no prognostic chemistry module for ozone and other stratospheric 183

constituents) and is thus referred to as a “low-top” model. 184

The Whole Atmosphere Community Climate Model version 6 (WACCM6) in CESM2 is 185

configured identically to CAM6, except that it uses 70 vertical levels and its model top is at 186

4.5x10-6 hPa (about 130 km). CAM6 and WACCM6 share the same vertical level structure 187

between the surface and 87 hPa. Compared to CAM6, this version of the model has superior 188

stratospheric representation and is thus referred to as a “high-top” model. WACCM6 features a 189

comprehensive chemistry mechanism with a description of tropospheric, stratospheric, and 190


10

upper-atmospheric chemistry. The default WACCM6 chemical mechanism contains 228 191

prognostic chemical species with reactions relevant for the whole atmosphere: Troposphere, 192

Stratosphere, Mesosphere, and Lower Thermosphere (TSMLT). In particular, it includes an 193

extensive representation of secondary organic aerosols [Tilmes et al., 2019]. MAM4 has been 194

modified to incorporate a new prognostic stratospheric aerosol capability [Mills et al., 2016]. The 195

modifications include mode width changes, growth of sulfate aerosol into the coarse mode, and 196

the evolution of stratospheric sulfate aerosol from natural and anthropogenic emissions of source 197

gases, including carbonyl sulfide (OCS) and volcanic sulfur dioxide (SO2). As described by 198

Gettelman et al. [2019a] and as also shown below, WACCM6 simulates nearly the same climate 199

as CAM6 in CESM2, with slight differences due to aerosol responses to chemistry, and upper 200

atmospheric processes above the CAM6 model top. WACCM6 also simulates internal variability 201

in the stratosphere, including Stratospheric Sudden Warming (SSW) events on the intraseasonal 202

timescales and the explicitly-resolved Quasi-Biennial Oscillation. 203

2.2 Ocean 204

The CESM2 ocean component is the same as in CCSM4 and CESM1, but features several 205

physics and numerical improvements; it is a level-coordinate model based on the Parallel Ocean 206

Program version 2 [POP2; Smith et al., 2010], as described in Danabasoglu et al. [2012]. The 207

physics advances include a new parameterization for mixing effects in estuaries, which improves 208

the representation of the exchange of freshwater between the terrestrial and marine branches of 209

the hydrologic cycle [Sun et al., 2019]; increased mesoscale eddy (isopycnal) diffusivities at 210

depth to improve the representation of passive tracers; use of prognostic chlorophyll for short-211

wave absorption; use of salinity-dependent freezing-point together with the sea-ice model [Assur, 212

1958]; and a new Langmuir mixing parameterization in conjunction with the new wave model 213


11

component [Li et al., 2016; also see below]. The numerical improvements include a new iterative 214

solver for the barotropic mode to reduce communication costs, particularly advantageous for 215

high-resolution simulations on large processor counts [Huang et al., 2016]; a tracer-conserving 216

time filtering scheme based on an adaption of the Robert – Asselin filter to enable sub-diurnal 217

coupling of the ocean model [Robert, 1966; Asselin, 1972; Williams, 2011]; and subsequently, 218

use of a one-hour coupling frequency to explicitly resolve (sub-)diurnal and inertial periods. In 219

addition, the K-Profile vertical mixing Parameterization [KPP; Large et al., 1994] is 220

incorporated via the Community ocean Vertical Mixing (CVMix) framework, and the Caspian 221

Sea is treated as a lake in the land model and thus it is no longer included in the ocean model as a 222

marginal sea. The model uses spherical coordinates in the Southern Hemisphere. In the Northern 223

Hemisphere, the grid North Pole is displaced into Greenland. The horizontal resolution is 224

nominal 1° with a uniform resolution of 1.125° in the zonal direction. The resolution varies 225

significantly in the meridional direction with the finest resolution of 0.27° at the equator. In the 226

Southern Hemisphere, the resolution monotonically increases to 0.53° at 32°S, remaining 227

constant further south. In the Northern Hemisphere high latitudes, the finest resolution is about 228

0.38°, occurring in the northwestern Atlantic Ocean, and the coarsest resolution is about 0.64°, 229

located in the northwestern Pacific Ocean. There are 60 levels in the vertical, with a maximum 230

depth of 5500 m. The vertical resolution is uniform at 10 m in the upper 160 m and increases 231

monotonically to a maximum of 250 m by a depth of 3500 m. 232

Ocean biogeochemistry in CESM2 is represented by the Marine Biogeochemistry Library 233

(MARBL), configured to implement an updated version of what has previously been known as 234

the Biogeochemistry Elemental Cycle [BEC; Moore et al., 2002; 2004; 2013]. MARBL 235

represents multiple nutrient co-limitation (N, P, Si, and Fe), includes three explicit phytoplankton 236


12

functional groups (diatoms, diazotrophs, and pico/nano phytoplankton), and one implicit group 237

(calcifiers); and there is one zooplankton group [Moore et al., 2004]. MARBL includes fully 238

prognostic carbonate chemistry and simulates sinking particulate organic matter implicitly, 239

subject to ballasting by mineral dust, biogenic CaCO3, and Si following Armstrong et al. [2002]. 240

Major updates relative to CESM1 include a representation of subgrid-scale variations in light 241

[Long et al., 2015], variable C:P stoichiometry [Galbraith and Martiny, 2015], burial of material 242

at the seafloor that matches riverine inputs in the pre-industrial climate, both semi-labile and 243

refractory dissolved organic material pools [Letscher et al., 2015], prognostic oceanic emission 244

of ammonia [Paulot et al., 2015], and an explicit ligand tracer that complexes iron. Atmospheric 245

deposition of iron is computed prognostically as a function of dust and black carbon deposition 246

simulated by CAM6. Riverine nutrient, carbon, and alkalinity fluxes are supplied to the ocean 247

from a dataset. These fluxes are held constant using data from GlobalNEWS [Mayorga et al., 248

2010] with the exception of inorganic nitrogen and phosphorus fluxes, which are taken from the 249

Integrated Model to Assess the Global Environment-Global Nutrient Model (IMAGE-GNM) 250

[Beusen et al., 2015; 2016] and vary from 1900 through 2005. Outside this period, the fluxes are 251

held constant using the closest temporal value. 252

Version 3.14 of the NOAA WaveWatch-III ocean surface wave prediction model [Tolman, 2009] 253

is incorporated in CESM2 as a new component. The wave model is solved on a coarse 254

resolution, near-global grid (78.4°S – 78.4°N) with a resolution of 4° in longitude and 3.25° in 255

latitude with 25 frequency and 24 direction bins, optimized for climate applications of the Stokes 256

drift, called the ww3a grid. The waves are driven by winds, air-sea temperature differences, 257

mixed layer depth, and ocean currents provided through the coupler, and they deliver three new 258

benefits. First, the waves provide the forcing necessary for including Langmuir, or wave-driven, 259


13

turbulence in the KPP vertical mixing scheme in POP2 [Li et al., 2016]. Second, wave statistics 260

from CMIP6 experiments will be contributed to the Coordinated Ocean Wave Climate Project 261

(COWCLIP) experiment [Hemer et al., 2012], which studies changes in wave climate under 262

climate change. The CESM2 contribution is expected to be unique in COWCLIP comparison as 263

the waves will be run in a fully coupled system. Third, wave statistics are now available for 264

development of other wave-related process studies and parameterizations, such as drag and gas 265

transfer coefficients that depend on sea state [Reichl et al., 2014], waves in the marginal ice 266

zones [Roach et al., 2018], wave effects on the mean flow that become important at mesoscale-267

permitting ocean model resolutions [e.g., McWilliams and Fox-Kemper, 2013; Suzuki and Fox-268

Kemper, 2016], and other climate effects of waves [Cavaleri et al., 2012]. The CESM2 default 269

version of Langmuir mixing includes only enhanced mixing within the mixed layer [McWilliams 270

and Sullivan, 2000; Li et al., 2016], but a scheme with more accurate entrainment at the mixed 271

layer base is also available [Li et al., 2017]. When the prognostic wave model is not used, a 272

statistical theory prediction of the needed variables of the wave state based on empirical 273

relationships observed in surface waves can be used instead. This “theory wave” feature in 274

CVMix provides similar Langmuir mixing effects at a much lower cost [Li and Fox-Kemper, 275

2017; Reichl and Li, 2019], with a minimal loss of accuracy for the Langmuir mixing 276

application. 277

2.3 Sea-ice 278

CESM2 uses CICE version 5.1.2 [CICE5; Hunke et al., 2015] as its sea-ice component. 279

Compared to CESM1, the sea-ice model now incorporates mushy-layer thermodynamics [Turner 280

and Hunke, 2015], with a prognostic vertical profile of ice salinity. To accomplish this, the 281

model parameterizes the gravity drainage of brine, melt water flushing within the ice, and 282


14

salinity effects of snow-ice formation. The sea-ice vertical resolution has been enhanced from 283

four layers in CESM1 to eight layers in CESM2 to better resolve the salinity and temperature 284

profiles. Also, the snow model now resolves three layers to represent the vertical temperature 285

profile. The melt pond parameterization has been updated [Hunke et al., 2013] and now accounts 286

for the fact that ponds preferentially form on undeformed sea ice. In conjunction with the ocean 287

model, a salinity-dependent freezing temperature is incorporated [Assur, 1958]. CICE5 shares 288

the same horizontal grid with the ocean component. 289

2.4 Land-ice 290

The land-ice modeling capabilities introduced in CESM1 [Hurrell et al., 2013; Lipscomb et al., 291

2013] have been extended in CESM2. CESM1 used the Glimmer Community Ice Sheet Model (a 292

serial code with simplified shallow-ice dynamics) and was limited to one-way coupling; the ice 293

sheet could respond dynamically to surface mass balance (SMB) forcing from the land model, 294

but land surface topography and surface types could not evolve. The ice sheet component of 295

CESM2 is the Community Ice Sheet Model version 2.1 [CISM2.1; Lipscomb et al., 2019], a 296

state-of-the-art parallel model with higher-order velocity solvers (valid for fast-flowing ice 297

streams and ice shelves) and improved treatments of basal sliding, iceberg calving, and other 298

physical processes. The simulations in this manuscript assume fixed ice sheets, but CESM2 also 299

supports two-way coupling between the Greenland ice sheet and the land and atmosphere 300

models. For two-way coupling, land surface types and surface elevation evolve as the ice sheet 301

advances and retreats, with total water mass conserved. The ice sheet model typically is run on 302

the Greenland domain at 4 km grid resolution with a depth-integrated higher-order solver 303

[Goldberg, 2011], a pseudo-plastic basal sliding law [Aschwanden et al., 2016], and with floating 304

ice immediately calved. CESM2 simulations with dynamic Antarctic and paleo ice sheets are not 305


15

yet supported but are under development. With either fixed or dynamic ice sheets, the SMB and 306

surface temperature over ice sheets are computed in CLM5 in multiple elevation classes for each 307

glaciated grid cell [Lipscomb et al., 2013]. By default, these calculations are done for any 308

CESM2 simulation with an active land model, unlike CESM1 in which ice sheet SMB was 309

computed only in a few custom simulations [e.g., Vizcaíno et al., 2013]. For Greenland, the SMB 310

and surface temperature are passed to the coupler and downscaled to the finer CISM grid, with 311

linear interpolation between elevation classes in the vertical direction and bilinear interpolation 312

in the horizontal. For Antarctica, the SMB is computed in multiple elevation classes in CLM5 313

but is not downscaled by the coupler. 314

2.5 Land 315

The Community Land Model version 5 [CLM5; Lawrence et al., 2019] includes an extensive 316

suite of new and updated processes and parameterizations. Collectively, these developments 317

improve the model’s hydrological and ecological realism and enhance the representation of 318

anthropogenic land use activities on climate and the carbon cycle. Specifically, the main updates 319

are as follows: photosynthesis updates (canopy scaling, co-limitation [Bonan et al., 2011], and 320

temperature acclimation [Lombardozzi et al., 2015]); soil hydrology improvements (frozen soil 321

hydrology updates [Swenson et al., 2012], spatially-explicit soil depth [Brunke et al., 2016], dry 322

surface layer control on soil evaporation [Swenson and Lawrence, 2014], and updated 323

groundwater scheme [Swenson and Lawrence, 2015]); snow model updates (new snow cover 324

fraction parameterization [Swenson and Lawrence, 2012], new fresh snow density 325

parameterization based on temperature and wind resulting in more realistic denser snow packs, 326

new fresh grain size parameterization based on temperature, revised canopy snow interception, 327

canopy snow radiation, and snow unloading parameterizations, and 10 m snow water equivalent 328


16

maximum snow pack allowing for firn formation [van Kampenhout et al., 2017]); introduction of 329

a mechanistic plant hydraulics and hydraulic redistribution scheme which explicitly models 330

water transport through roots, stems, and leaves according to a simple hydraulic framework and 331

which replaces the empirical soil moisture stress function [Kennedy et al., 2019]; a completely 332

updated lake model [Subin et al., 2012]; vertically-resolved soil biogeochemistry scheme that 333

allows for simulation of permafrost carbon [Koven et al., 2013]; addition of a global crop model 334

that represents planting, harvest, grain fill and grain yields for six crop types and includes time-335

evolving, spatially-explicit fertilization rates and irrigated areas [Levis et al., 2018]; inclusion of 336

a new fire model that has representation of natural and anthropogenic fire triggers and 337

suppression as well as agricultural, deforestation, and peat fires [Li et al., 2013; Li and 338

Lawrence, 2017]; multiple urban classes and updated urban energy model [Oleson and Feddema, 339

2019]; and improved representation of plant N dynamics to address known deficiencies in 340

CLM4. This latter improvement is accomplished by: 1) introducing flexible plant carbon : 341

nitrogen (C:N) stoichiometry [Ghimire et al., 2016], which circumvents the problematic CLM4 342

separation of potential and actual productivity fluxes that effectively decoupled leaf-level C 343

exchange from associated water and energy fluxes; 2) explicitly simulating how photosynthetic 344

capacity responds to environmental conditions through the Leaf Utilization of Nitrogen for 345

Assimilation (LUNA) module [Ali et al., 2016]; and 3) accounting for how changes in N 346

availability have consequences for plant productivity through the Fixation and Uptake of 347

Nitrogen (FUN) module, which calculates the carbon costs of N acquisition and allocates C 348

expenditure on N uptake among biological fixation, active uptake, and retranslocation [Shi et al., 349

2016]. CLM5 receives instantaneous N fluxes directly through the coupler from WACCM6 350

when it is used. Otherwise, CLM5 uses monthly N-deposition derived from prior 351


17

CESM2(WACCM6) simulations. Also, irrigation is applied either with water taken from rivers, 352

if available, or diffusely from the ocean if there is not enough river water. Detailed description, 353

assessment, and characterization of CLM5 performance compared to prior CLM versions in 354

land-only configurations are provided in Lawrence et al. [2019], Wieder et al. [2019], Fisher et 355

al. [2019], and Bonan et al. [2019]. 356

2.6 River transport 357

The River Transport Model (RTM) used in CESM1 has been replaced with the Model for Scale 358

Adaptive River Transport [MOSART; Li et al., 2013]. In standard CESM2 configurations, the 359

MOSART grid resolution is 0.5° x 0.5°. In each MOSART grid cell, surface runoff received 360

from CLM5 is routed across hillslopes and then discharged along with subsurface runoff, also 361

received from CLM5, into a tributary subnetwork. This tributary subnetwork then feeds into the 362

main river channel of that grid cell that links river water from upstream grid cells to downstream 363

grid cells. A primary difference between RTM and MOSART is how river flow is calculated. 364

RTM uses a simple linear reservoir method wherein the flux of water transferred from an RTM 365

grid cell to its downstream neighbor is dependent only on the river water storage in that grid cell, 366

the mean distance between grid cells, and a globally constant effective water velocity. In 367

MOSART, river flow rates are calculated explicitly via the physically-based kinematic wave 368

method, a commonly used method in hydrology that is based on mass and momentum equations. 369

As a consequence, whereas RTM only simulates streamflow (m3 s-1), MOSART additionally 370

simulates time-varying main channel flow velocity (m s-1) and water depth, as well as subgrid 371

surface water flow in hillslopes and tributaries. MOSART requires detailed information on river 372

hydrography (i.e., parameters describing river and tributary width, depth, average slope, 373


18

roughness coefficient, and dominant river length) which is derived from the global 1 km 374

HydroSheds dataset [Lehner et al., 2008]. 375

2.7 Coupling 376

CESM2 is accompanied by a new collaborative software infrastructure for building and running 377

the model system as well as for controlling state and flux exchanges among the components. 378

This framework is known as the Common Infrastructure for Modeling the Earth (CIME; 379

http://github.com/ESMCI/cime). CIME contains a significantly re-written coupling infrastructure 380

that includes modularity, multi-instance capabilities for data assimilation, and new support for 381

land-ice coupling. It also includes re-written data components that have new Python capabilities 382

and are more easily extensible. As in CESM1, the data components can replace prognostic 383

components, thereby enabling component feedbacks to be easily controlled. Examples of such 384

configurations include ocean – sea-ice coupled integrations in which the atmospheric datasets are 385

provided by a data atmosphere model, and a land-only setup where the land model is the only 386

prognostic component. CIME also provides a new Case Control System (CCS) for configuring, 387

compiling, and executing complex Earth system model experiments. CCS is an object-oriented 388

set of Python utilities that provide many new capabilities for easier portability, case generation, 389

and user customization. It records details of the model version used as well as user customization 390

of the experimental setup. Additionally, CCS has system and unit testing functionality that leads 391

to a more robust and flexible code base. CIME can be ported and tested in a stand-alone mode, 392

without any of the CESM prognostic components, as a first step in porting CESM to other 393

machines. Finally, CIME contains additional tools, such as a new statistical consistency test, that 394

allow users to quickly verify if their port of CESM2 is successful. 395


19

The atmosphere, land, wave, and sea-ice components communicate state information and fluxes 396

through the coupler every atmospheric time step of 30 minutes. The only fluxes calculated in the 397

coupler are those between the atmosphere and ocean, and the coupler communicates them to the 398

ocean component every hour and to the atmosphere every 30 minutes. The land-ice component 399

communicates with the coupler once a day. The nominal 1° horizontal version of the 400

CESM2(CAM6) has a throughput of about 30 simulated years per day on 4320 processors on the 401

Cheyenne Supercomputer, corresponding to about 3450 cpu hours per simulation year. The 402

CESM2(WACCM6) version has a throughput of about 4 simulated years per day on 3780 403

processors on the same machine, corresponding to about 22700 cpu hours per simulation year, 404

roughly 7 times more expensive than CESM2(CAM6). In both CESM2(CAM6) and 405

CESM2(WACCM6) configurations, the atmosphere is scaled to near the maximum allowed by 406

the FV dycore: 1152 tasks by 3 threads. Although CAM6 and WACCM6 can run on more 407

processors, it is not cost effective to do so. Thus, the lower processor count in 408

CESM2(WACCM6) simulations is due to the ocean component which uses 256 tasks by 3 409

threads in CESM2(CAM6), but only 24 tasks by 3 threads in CESM2(WACCM6). This is 410

because the atmospheric model is much slower with WACCM6, so fewer processors are needed 411

for the ocean to keep pace. 412

3 Development Strategy and Challenges 413

Evaluations of the component model developments summarized above were done first by the 414

respective developers using both component-only and fully-coupled configurations with various 415

CCSM4 and CESM1 versions. This is in contrast to the process used during the CCSM4 416

development where initial assessments were done primarily in component-only simulations. 417

Preliminary versions of the component models were then brought together to start the first 418


20

CESM2 coupled simulations in October 2015. During this development and evaluation process 419

particular attention was paid to conservation of heat and freshwater across the component 420

models, climate drifts, reasonableness of mean climate and of variability in various fields such 421

as El Niño – Southern Oscillation (ENSO) and Atlantic meridional overturning circulation 422

(AMOC) as well as making sure that the model simulations were able to reproduce observed key 423

features of the 20th century climate such as evolution of global-mean surface temperatures, 424

present-day SSTs and sea-ice distributions and trends. Unfortunately, the path to a fully-coupled 425

CESM2 version that produced acceptable simulations for CMIP6 turned out to be much longer 426

than anticipated and very challenging. Along the way, errors and bugs were discovered and 427

addressed. Progress was particularly delayed due to two major unacceptable features that were 428

not obviously related to bugs or implementation errors. The first concerned the formation of 429

unrealistically extensive sea-ice cover over the Labrador Sea region in PI control simulations. 430

The second was that the global-mean surface temperature time series in historical simulations 431

showed a significant and quite unrealistic cooling particularly during the second half of the 20th 432

century, with end-of-the-century temperatures being actually colder than in the 1850s. 433

Substantial human and computational resources were dedicated to address these major 434

challenges, as well as less serious issues, as described below. After executing over 300 435

simulations, the vast majority being fully-coupled, ranging in duration from several decades to 436

centennial, a near-final CESM2 version was created in April 2018. 437

The formation of unrealistically extensive sea-ice cover over the Labrador Sea region, extending 438

into the northern North Atlantic, in PI control simulations proved to be very challenging to 439

address robustly. In fact, this feature – not supported by available observations – emerged twice 440

during the development process. Figure 2 shows 30-year average sea-ice concentration 441


21

distributions from two CESM2(CAM6) PI simulations with and without such an extensive 442

Labrador Sea ice cover. The more extensive Labrador Sea ice extent is also accompanied by 443

enhanced (and more unrealistic) sea-ice in the Sea of Okhotsk region. In simulations with 444

extensive ice cover, this feature emerges as early as around year 40 and as late as around year 445

105 relative to initialization of the coupled simulation. In no simulation did the Labrador Sea ice 446

cover become unrealistically extensive after year 105 if it had not already emerged earlier in the 447

simulation, indicating that the model likely has multiple equilibria and the trajectory through 448

phase-space following initialization passes near a bifurcation point. It was found that extensive 449

ice cover persisted when the simulations were extended by fifty or more years further. It was not 450

feasible, given time constraints, to undertake longer integrations to investigate long-term 451

behavior of this feature. Simulation characteristics when extensive ice cover is present include: 452

annual-mean sea-ice concentrations that are usually around 60% in the Labrador Sea; Labrador 453

Sea SSTs that remain around freezing point with salinities < 34 psu; curtailed Labrador Sea deep 454

water formation / convection; compensating enhanced deep-water formation to the east of 455

Greenland and in the Irminger Sea; and a slightly weaker AMOC maintained by the switch in the 456

location of deep-water formation. Detailed analyses of many simulations with extensive ice 457

cover in comparison to simulations without this feature, unfortunately, did not identify any 458

robust mechanisms involving a particular process or phenomenon such as surface fluxes, liquid 459

and solid runoff, lateral advection, vertical mixing, and ice drag as instigators or drivers. Indeed, 460

when ensemble simulations were performed using the same model configuration but with only 461

differences in their initial atmospheric potential temperatures introduced by round-off level 462

perturbations, some members showed extensive ice cover and others did not. Given this 463

experience, a practical – though not fully satisfactory – approach was adopted to use an ensemble 464


22

member without an extensive Labrador Sea ice cover after a sufficiently long simulation (see 465

Section 4) to provide ocean and sea-ice initial conditions going forward. 466

Figure 2. 30-year mean Arctic sea-ice concentration distributions (in %) from CESM2(CAM6) 467

PI simulations (left) without and (right) with excessive sea-ice formation in the Labrador Sea 468

region. The black lines are the 15% concentration values from SSM/I observations for the 1981-469

2005 period [Cavalieri et al., 1996]. 470

The second unacceptable feature was an unrealistic cooling of global-mean surface temperatures 471

in the historical simulations (see Fig. 7, green line). The cooling started around the 1920s, mildly 472

at first and then getting worse during roughly the 1950-1980 period, with surface temperature 473

anomalies exceeding –0.4°C with respect to the 1850s. Although the model warmed after the 474

1980s, the end-of-the-century temperatures remained unrealistically colder than in the pre-475

industrial 1850s. This behavior first appeared when the new CMIP6-based emissions datasets 476

were applied in our simulations. The same model versions with CMIP5-based emissions did not 477

cool in this way. The observational time series (see Fig. 7, black and gray lines) show decadal 478


23

fluctuations that have significant contributions from internal variability. The cooling described 479

above clearly exceeds this baseline internal variability range and this is what is addressed here. 480

The goal is not to reproduce exactly the 20th century global-mean surface temperature time series 481

in each realization, but to eliminate the unrealistic cooling behavior. Initial investigations 482

uncovered some inconsistencies and errors in the CMIP6 emissions datasets (which were 483

resolved in their final release) and their application to CAM6, particularly for anthropogenic 484

sulfur. Simulations with the corrected datasets did show some reductions of the unrealistic 485

cooling, but the cooling behavior remained, particularly during the mid-1940s-1980 period, with 486

end-of-the-century temperatures only slightly warmer than in the 1850s (not shown). Analysis of 487

CAM6 against recent volcanic sulfur emission events indicated a potentially excessive response 488

[Malavelle et al., 2017], which could be traced to the impact of aerosols on the liquid water path. 489

Several modifications were, therefore, made to aerosol – cloud interaction processes, focusing on 490

the warm rain formation process (autoconversion and accretion) to reduce the magnitude of these 491

effects. As a result of these modifications, the global historical (2000 minus 1850) aerosol 492

effective radiative forcing changed from –1.82 W m-2 in CAM5 with CMIP5-based emissions to 493

–1.67 W m-2 in CAM6 with CMIP6-based emissions (see Gettelman et al. [2019b] for further 494

details). These changes were instrumental in enabling the model to simulate a realistic 20th 495

century surface temperature time series as discussed below (see Section 7), but they did not 496

target a particular equilibrium climate sensitivity value (see Section 12). 497

4 Initial Conditions 498

The ocean and sea-ice initial conditions used in both CESM2(CAM6) and CESM2(WACCM6) 499

PI control simulations were obtained through the following series of integrations (Fig. 3). First, a 500

3-member ensemble set of simulations was started from the CESM1(LENS) PI control 501


24

integration [Kay et al., 2015] at year 402 (all simulations start from 01 January of a given year), 502

which itself was initialized from the January-mean potential temperature and salinity from the 503

Polar Science Center Hydrographic Climatology [PHC2; Steele et al., 2001; Levitus et al., 1998]. 504

Two ensemble members developed excessive Labrador Sea ice cover, starting at about year 80. 505

The third member (simulation 262c) did not show such a behavior. Therefore, it was decided to 506

use this simulation to provide ocean and sea-ice states for initializing subsequent simulations. 507

Indeed, the simulation was later extended to > 350 years to confirm that the Labrador Sea ice 508

cover remained realistic. A series of simulations with minor adjustments and bug fixes in the 509

coupled model were then performed, starting with a simulation (293) from year 161 of 510

simulation 262c. This simulation was followed by another (simulation 297) that started from year 511

130 of simulation 293. Finally, a third simulation (299) was initialized from year 249 of 512

simulation 297. The ocean and sea-ice states from year 134 of the final simulation (299) were 513

used as the initial conditions for the main CESM2(CAM6) and CESM2(WACCM6) PI control 514

simulations. Thus, in total, these ocean potential temperature and salinity initial conditions 515


25

represent states integrated for about 1070 years after initialization starting from the PHC2 516

datasets. 517

Figure 3. Schematic of the sequence of simulations used to obtain ocean and sea-ice initial 518

conditions for the CESM2 PI control simulations. The simulation names are given in blue. The 519

orange numbers represent the year of a simulation from which a subsequent simulation started. 520

The black horizontal lines represent the length of a given simulation in years. The red line shows 521

the cumulative integration time in years. The black and red thick marks indicate 100-year 522

intervals. CESM1(LENS) PI control simulation started from the PHC2 observations. 523

The runoff initial conditions also come from year 134 of simulation 299 for both 524

CESM2(CAM6) and CESM2(WACCM6) PI controls. Although their origins can be traced back 525

to year 161 of simulation 262c, their in-between paths differ from that of ocean and sea-ice 526

initial conditions. The atmospheric initial conditions come from year 134 of simulation 299 for 527


26

CESM2(CAM6) PI control and year 11 of a preliminary CESM2(WACCM6) PI simulation 528

(simulation 298) for the main CESM2(WACCM6) PI control. 529

Initial conditions for the land states and ocean biogeochemical tracers, including abiotic 530

radiocarbon, were obtained with long spin-up runs of the land and ocean models, respectively. In 531

these spin-up runs, the active atmospheric component was replaced with a data-atmosphere 532

component that simply read in and repeatedly cycled through multiple years of surface forcing 533

datasets that were needed to construct surface fluxes. Twenty-one years of forcing datasets 534

extracted from simulation 297 were used, in order to capture some aspects of interannual 535

variability. The land model was spunup from bare ground, utilizing the accelerated 536

decomposition mode to equilibrate the slow carbon pools, followed by several hundred years of 537

normal mode spin-up [Lawrence et al., 2019]. The ocean tracer spin-up was applied only to the 538

biogeochemical tracers of the ocean model. To avoid drift of the ocean physical state, it was reset 539

to its initial condition at the beginning of each 21-year cycle, keeping it synchronized with the 540

surface forcing. A subset of the ocean biogeochemical tracers was spun up using a Newton-541

Krylov based solver [Lindsay, 2017]. 542

The initial conditions for the 11 members of the CESM2(CAM6) historical simulations come 543

from years 501, 601, 631, 661, 691, 721, 751, 781, 811, 841, and 871 of the corresponding PI 544

control simulation (Fig. 4). The 3 members of the CESM2(WACCM6) historical simulations are 545

initialized from years 56, 61, and 71 of the corresponding PI control integration. The late and 546

early starts dates for the CESM2(CAM6) and CESM2(WACCM6) historical simulations simply 547

reflect their respective PI control integration lengths and are not intended to avoid or sample any 548

particular variability characteristics. 549


27

Figure 4. Schematic of the CESM2 PI control and historical simulations. CESM2(WACCM6) PI 550

simulation was started first (black) to obtain the forcing datasets used for the CESM2(CAM6) PI 551

integration (red). These forcings were based on the average of years 21-50 (green shading). 552

Three historical CESM2(WACCM6) simulations were started from years 56, 61, and 71 of the 553

corresponding PI simulation (gray lines). As described in Section 5, three-member ensemble 554

averages of these WACCM6 simulations (blue shading) were then used to obtain the forcings 555

needed for the CESM2(CAM6) historical simulations (orange lines). The CESM2(CAM6) 556

historical simulations started from years 501, 601, 631, 661, 691, 721, 751, 781, 811, 841, and 557

871 of the corresponding PI simulation. The black and red thick marks indicate 100-year 558

intervals. 559

5 Forcing Datasets for Pre-Industrial and Historical Simulations 560

CESM2(CAM6) PI control and historical simulations used several forcing datasets that were 561

obtained from the corresponding CESM2(WACCM6) simulations. This is a new approach in 562

CESM2 adopted to ensure that a physically consistent pair of low- and high-top atmospheric 563

models that share the same code base and contain the same physics in their common range of 564


28

altitude is used. It represents a degree of compatibility and consistency not available to most 565

modeling groups. CESM2(WACCM6) uses CMIP6-provided forcings wherever appropriate, but 566

calculates itself chemical and aerosol constituents from emissions that are based on the 567

anthropogenic and biomass burning inventories specified by CMIP6. Anthropogenic emissions 568

for 1850 to 2014 are provided from the Community Emissions Data System [CEDS; Hoesly et 569

al., 2018]. Biomass burning emissions are described by van Marle et al. [2017] and are all 570

specified at the surface (without any plume-rise or specified vertical distribution of the 571

emissions). Further details of the emissions and chemistry used in CESM2(WACCM6), 572

including the specification of ozone-depleting substances, are described in Gettelman et al. 573

[2019a]. CESM2(CAM6) simulations use the CMIP6-provided forcings except for i) 574

tropospheric oxidants for chemistry; ii) stratospheric ozone for radiative effects; iii) stratospheric 575

aerosols for radiative effects; iv) H2O production rates due to CH4 oxidation in the stratosphere; 576

and v) N deposition to the land and ocean components. These WACCM6-based forcings are 577

described below. 578

First, CESM2(CAM6) PI simulation 299 was run with prescribed forcings from a 579

CESM2(WACCM6) simulation (295). Then the CESM2(WACCM6) PI control was conducted, 580

using initial states for the ocean, sea-ice, and land components from simulation 299 (see Section 581

4). Distributions of tropospheric oxidants, stratospheric ozone, stratospheric aerosol, and 582

methane oxidation rates were derived from the average of years 21-50 of this 583

CESM2(WACCM6) PI simulation, and used in the subsequent CESM2(CAM6) PI control. For 584

historical simulations, an ensemble of three CESM2(WACCM6) simulations was run first for the 585

1850-2015 period. Corresponding forcings were derived from the CESM2(WACCM6) ensemble 586

average, for use in the CESM2(CAM6) historical simulations. The averaging of 30 years in the 587


29

PI control and of three ensemble members in the historical simulations is intended to reduce the 588

influence of natural variability on these forcings. A schematic summary of our integration 589

strategy to obtain the necessary forcing datasets to perform the CESM2(CAM6) simulations is 590

provided in Fig. 4. 591

The tropospheric oxidant species (O3, OH, NO3, and HO2) are prescribed in CESM2(CAM6), 592

which uses them in gas-phase and aerosol chemistry calculations to allow tropospheric aerosol 593

formation. Monthly-averaged values are used in both the PI control and historical simulations. 594

For the CESM2(CAM6) PI control, a repeating 12-month annual cycle is derived from averaged 595

values over years 21-50 of the CESM2(WACCM6) PI control simulation. CESM2(CAM6) 596

interpolates cyclically between mid-month date values. For CESM2(CAM6) historical 597

simulations, 12-month annual cycles are provided at quasi-decadal frequency (see Appendix A). 598

CESM2(CAM6) interpolates annual cycles for years between those provided. 599

Stratospheric ozone and stratospheric aerosol are radiatively active species important to the 600

radiative budget, temperature, general circulation, chemistry, and cloud physics of CESM2. In 601

CESM2(CAM6), stratospheric distributions of these species are prescribed from 5-day zonal 602

means of CESM2(WACCM6) output. Ozone volume mixing ratio is prescribed for the entire 603

vertical extent of the model for use in radiative transfer calculations. 604

A major advance in CESM2(WACCM6) is the development of prognostic stratospheric aerosols 605

derived from gas-phase sulfur emissions. As described in Mills et al. [2016], this new feature 606

provides CESM2 with stratospheric aerosol forcing that is much more consistent with 607

observations than previous climatologies developed for climate models. This self-consistent 608

treatment of volcanic aerosols and chemistry has been shown to produce realistic radiative 609


30

responses to eruptions over the 1979-2015 period [Mills et al., 2017; Schmidt et al., 2018]. For 610

CMIP6, volcanic SO2 emissions for 1850-2015 from VolcanEESM database [Neely and Schmidt, 611

2016] are used. In addition, observed carbonyl sulfide concentrations are specified as time-612

varying boundary conditions. As shown in Fig. 5, this leads to a realistic simulation of the 613

stratospheric aerosol optical depth (SAOD) compared to the CMIP6 data which are constrained 614

by satellite observations starting in 1979. Before this date, when the information on volcanic 615

emissions is sparser, CESM2(WACCM6) and CMIP6 SAOD can differ; for example, the 616

eruptions of Hekla (1947) and Bezymianny (1956) are missing from the CMIP6 database. Prior 617

to the satellite era, there is also a very small (~0.001) difference between CESM2(WACCM6) 618

and CMIP6 data in the background component of SAOD at 550 nm that persists during 619

volcanically quiescent periods. Historical variability in this background is largely due to 620

variability in the source gas OCS, which WACCM6 accounts for as a time-varying lower 621

boundary condition for OCS [Montzka et al., 2004]. CMIP6 prescribes a time-varying 622

background based on the 1999-2005 average, scaled following the calculations of Sheng et al. 623

[2015]. 624

Although CESM2(CAM6) does not include the interactive chemistry needed to realistically 625

simulate the development of stratospheric aerosol following a major volcanic eruption, model 626

consistency is maintained by prescribing aerosol properties above the tropopause in 627

CESM2(CAM6) using 5-day zonally averaged output from CESM2(WACCM6). Prescribed 628

stratospheric aerosol properties are the sulfate mass mixing ratio and diameter of each of the 629

three sulfate-bearing modes of MAM4, aerosol surface area density, and ice condensation nuclei 630

number concentration. Each of these properties is prescribed consistently with its use in 631

CESM2(WACCM6), which includes prognostic stratospheric aerosol in MAM4. As a result, the 632


31

radiative impacts of stratospheric aerosol from large volcanic eruptions are consistent between 633

CESM2(CAM6) and CESM2(WACCM6) simulations. For the CESM2(CAM6) PI control, a 5-634

day zonal-mean annual cycle was created from the average of years 21-50 of the 635

CESM2(WACCM6) PI control. For the CESM2(CAM6) historical simulations, 5-day zonal-636

mean values were calculated for each year over the 1850-2014 historical period from the average 637

of the three CESM2(WACCM6) historical ensemble members. The 5-day frequency allows 638

resolution of the impacts of major volcanic eruptions, as well as the annual development of polar 639

ozone loss [Neely et al., 2014]. The averaging of three CESM2(WACCM6) historical ensemble 640

members is especially important to the evolution of stratospheric aerosol resulting from volcanic 641

eruptions, which can be particularly sensitive to the circulation present at the time of the 642

eruption. 643

Figure 5. Time evolution of the globally-averaged stratospheric aerosol optical depth (SAOD) at 644

550 nm as simulated by CESM2(WACCM6) (3-member average) and prescribed in 645


32

CESM2(CAM6) (black) in comparison to that provided by CMIP6 (red). There is uncertainty in 646

the time, location, and intensity of volcanic emissions, especially further back in time. 647

The oxidation of methane (CH4) is an important source of water vapor to the middle atmosphere. 648

Because water vapor is a greenhouse gas, production rates of water vapor in CESM2(CAM6) are 649

prescribed using methane oxidation rates from CESM2(WACCM6). It is assumed that the 650

oxidation of each CH4 molecule results in the production of two molecules of H2O [Nicolet, 651

1970]. For the CESM2(CAM6) PI control, a repeating 12-month annual cycle is derived from 652

averaged values over years 21-50 of the CESM2(WACCM6) PI control simulation. For 653

CESM2(CAM6) historical simulations, 12-month annual cycles are provided at the same quasi-654

decadal frequency as the tropospheric oxidants (see above and Appendix A). 655

Regarding Chlorofluorocarbons (CFCs), in CESM2(CAM6), they are specified as a global 656

average mixing ratio derived from the lower boundary conditions provided by CMIP6. The 657

radiative transfer module in both CESM2(CAM6) and CESM2(WACCM6) considers only CFC-658

12 and CFC-11eq (equivalent). The CFC-11eq time series is supplied by CMIP6 [Meinshausen 659

et al., 2017]. Because CESM2(WACCM6) includes most of the chemical species contained in 660

CFC-11eq, the global distribution of CFC-11eq is first derived interactively in WACCM6. The 661

derived global average distribution at the surface is then scaled to match a given CMIP6 scenario 662

that contains additional species not included in the WACCM6 chemistry, i.e., CFC-11eq = CFC-663

11eq [CMIP6] / CFC-11eq [WACCM6]). 664

Updated land cover and land use data have been created for CLM5. The new dataset combines 665

updated versions of present-day satellite land cover descriptions with the past and future 666

transient land use time series from the Land Use Harmonization version 2 product (LUH2, 667


33

http://luh.umd.edu/). CLM5 (and ocean biogeochemistry) uses monthly N-deposition derived 668

from CESM2(WACCM6) simulations. Additional information on this dataset is provided in 669

Lawrence et al. [2019] and references therein. 670

6 Pre-Industrial Control Spin-up and Tuning 671

The CESM2(CAM6) and CESM2(WACCM6) PI control simulations were integrated for 1200 672

and 500 years, respectively. In such long control simulations, it is necessary to aim for a top-of-673

the-atmosphere (TOA) global- and time-mean energy balance of < |0.1| W m-2 to avoid 674

unrealistically warm or cold climate states towards the end of these simulations that will be 675

subsequently used to initialize historical integrations. Such a small TOA energy balance is 676

indeed desired in CESM2, because there are no additional external sources or sinks of energy, 677

such as geothermal heat fluxes in the ocean bottom. To achieve this level of TOA energy 678

balance, the coupled model was tuned using two sets of adjustments. The first involved very 679

minor changes in the so-called “gamma_coef” parameter in CLUBB, which controls how 680

bimodal the PDF of subgrid vertical velocity is. A larger value of gamma_coef implies a less 681

bimodal, longer-tailed PDF for vertical velocity. The longer tails of velocity, in turn, lead to 682

stronger entrainment at the top of the planetary boundary layer, which in turn reduces the amount 683

and thickness of low-clouds, particularly in marine stratocumulus layers. The second set of 684

adjustments were done in CICE5. Because the sea-ice was deemed to be too thick in the PI 685

simulations – in comparison to the present-day observations – to produce a realistic sea-ice 686

distribution at present-day, the bulk dry snow grain radius standard deviation was reduced to 687

1.25 from its CESM1 value of 1.75. In addition, the temperature for the onset of snow melt was 688

lowered from –1.0°C to –1.5°C and the maximum melting snow grain radius size was increased 689

from 1000 microns to 1500 microns, both adjustments again with respect to those used in 690

http://luh.umd.edu/


34

CESM1 simulations. All these adjustments were only applied to snow on sea-ice and they are 691

certainly within the observationally-based acceptable ranges of these tunable sea-ice parameters. 692

The changes collectively result in a lower snow and sea-ice albedo and hence thinner sea-ice 693

overall. These parameters were adjusted identically in the two PI control simulations, and no 694

additional changes were made throughout the control integrations. Furthermore, the tuning 695

parameters remained fixed at these values in all subsequent experiments. 696

The impacts of these tunings were assessed in decadal to multi-decadal length simulations prior 697

to the start of the main PI control simulations. Figure 6 (left panel) shows the time series of the 698

global-mean TOA energy imbalances (fluxes) for the two control simulations where the 699

imbalances are +0.05 and +0.06 W m-2 for CESM2(CAM6) and CESM2(WACCM6), 700

respectively, averaged over their integration lengths. These imbalances are +0.03 W m-2 during 701

the last 500 years of CESM2(CAM6) and +0.06 W m-2 during the last 200 years of 702

CESM2(WACCM6). While these TOA imbalances are much smaller than the imbalance of 703

about –0.15 W m-2 in CCSM4 [Gent et al., 2011], they are larger than the exceptionally small 704

TOA imbalance of near-zero in the CESM1(LENS) control simulation. In both CESM2(CAM6) 705

and CESM2(WACCM6) control simulations, this energy gain is solely reflected in the ocean 706

model, as the land and sea-ice components show minor energy losses. The global-mean energy 707

gain by the ocean model is about 0.07 W m-2 during the last 500 years of CESM2(CAM6) and 708

about 0.10 W m-2 during the last 200 years of CESM2(WACCM6). As a result, the global-mean 709

ocean potential temperature (Fig. 6, right panel) increases from its initial value of 3.92°C to 710

about 4.23°C by year 1200 in CESM2(CAM6). Similarly, it increases from 3.92°C to about 711

4.08°C in CESM2(WACCM6). Although the details of the depth ranges and magnitudes of this 712

warming differ across the ocean basins, depths below about 2 km show warming in all major 713


35

basins (not shown). The global-mean ocean salinity decreases slightly by about 5 x 10-5 psu 714

century-1 in both CESM2(CAM6) and CESM(WACCM6). This salinity trend is very similar to 715

those of CCSM4 and CESM1(LENS) PI control simulations. 716

Figure 6. Time series of (left) five-year average, global-mean TOA energy imbalance (flux) and 717

(right) annual- and volume-mean ocean potential temperature from the CESM2(CAM6) and 718

CESM2(WACCM6) PI control simulations. 719

7 Surface Temperatures 720

The model global-mean surface (2-m air) temperature anomaly time series for the historical 721

period are presented in Fig. 7 in comparison with observations. A reference period of 1850-1870 722

is chosen to easily identify warming since the pre-industrial conditions. For CESM2(CAM6), 723

both the ensemble-mean and its spread are shown. Only the ensemble-mean time series are 724

included from CESM2(WACCM6) and CESM1(LENS) simulations for clarity. The observations 725

are from the Hadley Centre – Climate Research Unit Temperature Anomalies [HADCRU4.5; 726


36

Jones et al., 2012]; the Goddard Institute for Space Studies Surface Temperature Analysis 727

[GISTEMP; Lenssen et al., 2019]; and the National Oceanic and Atmospheric Administration 728

merged land-ocean global surface temperature analysis [NOAAGlobalTemp; Vose et al., 2012]. 729

In general, the CESM2(WACCM6) time series tend to be slightly warmer than those of 730

CESM2(CAM6). Both sets of simulations capture the observed low-frequency variability, and 731

the observations are usually within the ensemble spread of CESM2(CAM6) simulations with a 732

few notable exceptions. One such example is the dip around 1910 evident in observations, but 733

absent in model simulations. This discrepancy may be related to incorrect and / or missing 734

representations of impacts of volcanic eruptions in both models and observations during the 735

1900-1910 period. Another major discrepancy is seen during the 2000-2014 period with model 736

ensemble-mean temperature anomalies being warmer than observed by about 0.2°C and 0.4°C at 737

the end of the integration period in CESM2(CAM6) and CESM2(WACCM6) simulations, 738

respectively. Although the warming trends between 1975-2000 are comparable between these 739

two model simulations and observations, no CESM2 ensemble member is able to capture the 740

observed slowdown in the warming trend during roughly 2000-2012. This is not surprising as 741

internal variability as well as anthropogenic and natural forcings are known to play a role in the 742

creation of this decadal-scale slowdown [e.g., Solomon et al., 2010; Kaufmann et al., 2011; 743

Kosaka and Xie, 2013; Meehl et al., 2014; Deser et al., 2017]. Indeed, as discussed in Meehl et 744

al. [2014], a much larger ensemble than presently used is needed as such a slowdown is present 745

only in a very small fraction of the available CMIP5 simulations (10 out of 262). In comparison 746

to CESM2 simulations and observations, CESM1(LENS) ensemble-mean time series show 747

generally colder temperature anomalies, usually remaining below or near the lower extent of the 748


37

CESM2(CAM6) spread. CESM1(LENS) does not capture the observed decadal-scale slowdown, 749

either. 750

Figure 7. Time series of the global-mean surface (2-m air) temperature anomaly relative to the 751

respective 1850-1870 averages during the historical period from observations, CESM2(CAM6), 752

CESM2(WACCM6), and CESM1(LENS) ensemble simulations. The solid red line and the pink 753

shading show the 11-member ensemble-mean and its spread for CESM2(CAM6) simulations. 754

The solid blue and dashed red lines show the 3-member and 40-member ensemble means for 755

CESM2(WACCM6) and CESM1(LENS) simulations, respectively. The green line, EXP(A), 756

shows a sample time series from an earlier simulation with unrealistic cooling particularly during 757

the second half of the 20th century. The CESM1 simulations use the Representative 758

Concentration Pathway 8.5 (RCP8.5) forcing for the 2006-2014 period. The observations are 759

from the HADCRU45, GISTEMP, and NOAAGlobalTemp datasets. The latter two time series 760

are referenced to the mean of HADCRU45 for the 1870-1890 period. 761


38

The 1985-2014 average SST (0-10 m average in the ocean) distributions from CESM2(CAM6), 762

CESM2(WACCM6), and CESM1(LENS) ensemble-means are presented in Fig. 8 in comparison 763

to the Extended Reconstructed Sea Surface Temperature version 5 dataset [ERSSTv5; Huang et 764

al., 2017]. There are two salient features in the figure, particularly evident in the model – 765

observations difference distributions. First, CESM2(CAM6) and CESM2(WACCM6) mean 766

SSTs are almost identical to each other. Second, both model SSTs are similarly warmer than 767

observations and those of CESM1(LENS). Indeed, CESM2(CAM6) and CESM2(WACCM6) 768

global-mean SSTs are warmer than observations by 0.39 and 0.33°C, respectively, while 769

CESM1(LENS) is colder than observations by 0.27°C. These global-mean differences essentially 770

reflect the SST differences in the corresponding PI control simulations from the ERSSTv5 771

dataset for the pre-industrial period, which are computed as 0.33°C for CESM2(CAM6), 0.18°C 772

for CESM2(WACCM6), and –0.18°C for CESM1(LENS) over 30-year periods close to the 773

mean initial start dates of the respective historical ensemble members. When root-mean-square 774

(rms) differences from observations are considered, both CESM2(CAM6) and 775

CESM2(WACCM6) have slight degradations of the mean SST simulations with rms differences 776

of 0.77 and 0.75°C, respectively, compared to 0.67°C for CESM1(LENS). The SST difference 777

distributions depict several persistent biases. The warm-cold SST difference dipole in the North 778

Atlantic associated with errors in the separation and path of the Gulf Stream – North Atlantic 779

Current system is present in all model simulations with comparable magnitudes. The warm 780

biases in the Eastern boundary upwelling regions, i.e., off the coasts of California, South 781

America, and South Africa, are more pronounced in CESM2(CAM6) and CESM2(WACCM6) 782

than in CESM1(LENS). These CESM2 biases are much larger in their magnitude (and spatial 783

extent) than the global- and time-mean SST differences between CESM2 and CESM1(LENS), so 784


39

the larger CESM2 biases cannot be solely attributed to the warmer mean state of these 785

simulations. Reducing these upwelling-related biases has been rather challenging in 1° horizontal 786

resolution models. In CESM1, the upwelling biases were reduced due to more realistic wind 787


40

stress curl in a model configuration with much higher horizontal resolutions in the ocean and 788

atmospheric models than used in the present simulations [Small et al., 2015]. 789

Figure 8. 1985-2014 average sea surface temperature (in °C) from (top) ERSST observations and 790

(left) model simulations. The right panels show the respective model minus observations 791

difference distributions. The model distributions represent full ensemble means. The global-792


41

means for each panel and the root-mean-square error for the difference distributions are also 793

provided. 794

Seasonal-average surface (2-m) air temperature model minus observations differences over land 795

for December-January-February (DJF) and June-July-August (JJA) for the 1985-2014 period 796

from CESM2(CAM6), CESM2(WACCM6), and CESM1(LENS) are shown in Fig. 9. The 797

observations are from the Berkeley Earth Surface Temperatures dataset [BEST; Rohde et al., 798

2013]. As in the SST distributions, CESM2(CAM6) and CESM2(WACCM6) distributions are 799

very similar to each other and, on average, CESM2 is warmer than CESM1. Both the global 800

mean bias and rms error are appreciably lower in CESM2 than in CESM1 in all seasons. For 801

example, the DJF mean bias (rms error) has been reduced from –2.24°C (1.90°C) in 802

CESM1(LENS) to 0.89°C (1.51°C) and 0.67°C (1.38°C) in CESM2(CAM6) and 803

CESM2(WACCM6), respectively. In CESM2, there is a notably damped annual cycle – warm 804

biases in the winter season and cold biases in the summer season – in the northern high latitudes. 805

There is also a large summertime warm bias over the central United States, which partially 806

coincides with a dry bias (see Fig. 10), and may be related to the absence of propagating 807

mesoscale convective systems at the 1° model resolution. Another interesting bias is the warm 808

bias over Northeast India which exists in all seasons except for March-April-May (MAM) (not 809

shown). This region is among the most heavily irrigated regions in the world and previous 810

studies have shown that the irrigation-induced cooling is likely mitigating against climate 811

warming in this area [Thiery et al., 2020]. Though CESM2 includes active irrigation by default, 812

irrigation in India is applied mainly in the MAM season due to a limitation in the crop planting 813

date scheme that incorrectly results in crops being planted in the spring over India. In reality, 814

over much of Northern India, the two main cropping and irrigation seasons are in summer and in 815


42

winter. Improved simulation of crop planting dates, therefore, may result in reduced surface air 816

temperature biases in this region. 817

Figure 9. Seasonal-average surface (2-m) air temperature model minus observations differences 818

(in °C) over land for DJF and JJA for the 1985-2014 period from CESM2(CAM6), 819

CESM2(WACCM6), and CESM1(LENS). The observations are from the BEST dataset. The 820

model distributions represent full ensemble means. The global-mean biases and the root-mean-821

square errors are also provided. 822

8 Atmospheric Fields 823

8.1 Precipitation 824


43

The numerous improvements incorporated into the atmospheric model components have a 825

positive impact on many aspects of the climate of the fully coupled CESM2 system. Figure 10 826

shows the 1985-2014 average precipitation from CESM2(CAM6), CESM2(WACCM6), and 827

CESM1(LENS) ensemble means in comparison to observational estimates from the Global 828

Precipitation Climatology Project [GPCP, Huffman et al., 2009]. As in the SST and surface 829

temperature over land distributions, CESM2(CAM6) and CESM2(WACCM6) simulations 830

produce nearly identical mean precipitation fields. Both sets of simulations clearly depict a 831

systematic reduction of most biases from CESM1(LENS) to CESM2, from a mean bias of 0.33 832

mm day-1 in CESM1(LENS) down to 0.25 and 0.23 mm day-1 in CESM2(CAM6) and 833

CESM2(WACCM6), respectively. The rms error also decreases substantially by about 20%, 834

from 0.87 mm day-1 in CESM1(LENS) to 0.71 mm day-1 in CESM2. These improvements arise 835

primarily from the reduction of the major tropical wet biases of the Indian Ocean, the South 836

Pacific Convergence Zone (SPCZ), and the tropical Atlantic Inter-Tropical Convergence Zone 837

(ITCZ). Over land, the dipole of Andean wet and Amazon dry biases is decreased, and excessive 838

precipitation over Australia and Western US is reduced. Nevertheless, there are a few degraded 839

regions, including a slightly larger negative bias off Brazil in CESM2 that extends into the South 840

Atlantic Ocean. 841


44

Figure 10. 1985-2014 average total precipitation (in mm day-1) from (top) GPCP observations 842

and (left) model simulations. The right panels show the respective model minus observations 843

difference distributions. The model distributions represent full ensemble means. The global-844

means for each panel and the root-mean-square error for the difference distributions are also 845

provided. 846


45

8.2 Short-wave and long-wave cloud forcing 847

A dramatic improvement is seen in the radiative properties and, in particular, the time-averaged 848

short-wave cloud forcing (Fig. 11). In comparison to the Clouds and the Earth’s Radiant Energy 849

System – Energy Balanced and Filled [CERES-EBAF; Loeb et al., 2009] dataset, the reflective 850

biases of the clouds in CESM1(LENS) throughout the tropics over land and ocean, often in 851

excess of –30 W m-2, are substantially alleviated in both CESM2(CAM6) and 852

CESM2(WACCM6). This improvement is due to several changes in different cloud regimes. The 853

interaction of CLUBB with the CAM6 deep convective parameterization enabled a re-adjustment 854

of tropical deep convection over land to reduce biases. Net rms for shallow clouds is reduced 855

with CLUBB. At higher latitudes, low clouds were insufficiently bright in CESM1(LENS). 856

Changes in mixed-phase cloud properties in the MG2 microphysics scheme produce more 857

supercooled liquid water in CESM2 and result in more low clouds and more reflective clouds. 858

Quantitatively, the global-mean rms error is substantially reduced from about 14 W m-2 in 859

CESM1(LENS) to about 9.1 W m-2 in CESM2(CAM6) and 9.7 W m-2 in CESM2(WACCM6). 860

The subtropical stratocumulus regions show a few large-scale improvements. For the most part 861

however, deficiencies in cloud forcing seen in CESM1(LENS) are still seen in the 862

CESM2(CAM6) and CESM2(WACCM6) stratocumulus regions. 863


46

Figure 11. 2000-2013 average short-wave cloud forcing (in W m-2) (a) from the CERES-EBAF 864

dataset and (b-d) for model minus observation differences. Only one ensemble member is used 865

for model simulations. 866

In contrast, the improvements in the time-averaged long-wave cloud forcing (Fig. 12) are much 867

more modest compared to those of the short-wave cloud forcing. In particular, weak cloud 868

forcing biases in CESM2(CAM6) and CESM2(WACCM6) remain similar to those of 869

CESM1(LENS) outside of the tropics, illustrating the continued challenge of representing ice 870

microphysics and their optical characteristics. Within the tropics, the Indo-Pacific improvements 871

in cloud forcing largely reflect the patterns of tropical precipitation changes and associated upper 872

tropospheric cloud and cloud water. These are most apparent in the reduction of the double ITCZ 873

bias in the Pacific Ocean and the east-west precipitation distribution in the Indian Ocean. 874

Quantitatively, the global-mean rms errors are about 8.3 W m-2 in CESM2(CAM6) and 7.6 W m-875


47

2 in CESM2(WACCM6), both representing modest reductions from the rms error of about 9.6 W 876

m-2 in CESM1(LENS). 877

Figure 12. Same as in Fig. 11 except for long-wave cloud forcing. 878

8.3 Madden-Julian Oscillation 879

An aspect of variability that was poorly represented in CESM1 is the Madden-Julian Oscillation 880

(MJO). In CESM2, minor changes to the Zhang-McFarlane deep convection scheme and the 881

addition of CLUBB have combined to generate a much improved MJO (Fig. 13). The 882

observation-based Tropical Rainfall Measuring Mission – European Centre for Medium-Range 883

Weather Forecast (ECMWF) Reanalysis-Interim [TRMM/ERAI; Kummerow et al., 2000; Dee et 884

al., 2011] product shows convectively-coupled coherent wind and precipitation features that 885

propagate in quadrature from the Indian Ocean into the West Pacific. In CESM1(LENS), these 886


48

features had incorrect westward propagation in the Indian Ocean and a lack of West Pacific 887

variability. In both CESM2(CAM6) and CESM2(WACCM6), these adverse characteristics are 888

largely absent, such that coherent eastward propagation now occurs from the Western Indian 889

Ocean through the Maritime Continent and into the Central Pacific. Correlations are slightly 890

weaker in CESM2(WACCM6) than in CESM2(CAM6). Remaining bias characteristics are 891

slightly weak coupling and phase speed. 892

Figure 13. Lag correlation relationship between total precipitation averaged in the Indian Ocean 893

and total precipitation (color) and 850-mb zonal wind (contour) at all Indo-Pacific longitudes 894

from (a) TRMM/ERAI datasets and (b-d) model simulations. Based on daily data filtered with a 895

20-100 day Lanczos filter. Only one ensemble member is used for model simulations. 896


49

8.4 El Niño – Southern Oscillation 897

ENSO is arguably the most important mode of tropospheric climate variability on interannual 898

timescales. Because it was one of the worst aspects in CCSM3 simulations with a dominant 899

period of 2 years, improving ENSO was the highest priority during CCSM4 development. As a 900

result of changes in the atmospheric deep convection parameterization, ENSO characteristics 901

were improved in CCSM4 and CESM1 [Richter and Rasch, 2008; Neale et al., 2008; Deser et 902

al., 2012]. These improvements are retained in CESM2. Figure 14 compares ENSO (El Niño 903

minus La Niña) spatial composites of DJF anomalies of precipitation, surface air temperature 904

over land and SST over ocean, and sea level pressure from observations and CESM2(CAM6), 905

CESM2(WACCM6), and CESM1(LENS) PI control simulations. Again, CESM2(CAM6) and 906

CESM2(WACCM6) are very similar to each other with the exception of a slightly stronger 907

ENSO amplitude in the former, and the spatial patterns are generally well simulated by all model 908

versions. (The ENSO composite value of DJF SST in the Niño 3.4 region is significantly 909

different between CESM2(CAM6) and CESM2(WACCM6) at the 95% confidence level based 910

on a 2-sided Student’s t-test (1.38°C vs. 1.26°C), for reasons which remain to be understood.) 911

There are two notable improvements in CESM2(CAM6) and CESM2(WACCM6) with respect 912

to CESM1(LENS). First, the orientation of the sea level pressure teleconnection over the North 913

Pacific, in particular its northwest-southeast tilt, is more realistic with implications for the 914

circulation-induced air temperature anomalies over western North America. Second, the 915

representation of the tropical precipitation anomaly pattern is improved, with more prominent 916

drying over the maritime continent and SPCZ, and a broader equatorial maximum over the 917

western half of the Pacific. These improvements are associated with a better representation of the 918

climatological precipitation distribution as evidenced by a more prominent northwest-southeast 919


50

oriented SPCZ that is confined to the western Pacific and less of a double ITCZ structure as 920

indicated by the contours in the top row. The precipitation anomaly pattern in CESM2(CAM6) 921

and CESM2(WACCM6) also shows a stronger east-west Walker Circulation component, more in 922

line with observations, compared to the dominant narrow Hadley Circulation response in 923

CESM1(LENS). 924

Figure 14. Spatial composites of DJF ENSO anomalies of precipitation (top row; in mm day-1), 925

surface air temperature over land and SST over ocean (bottom row; color shading; in °C), and 926

sea level pressure (bottom row; contours; –16 to 16 hPa by 2 hPa with negative contours dashed) 927

from observations, CESM2(CAM6) PI control for years 100-1199, CESM2(WACCM6) PI 928

control for years 50-499, and CESM1(LENS) PI control for years 500-1599. The 6 mm day-1 929

DJF climatological mean precipitation contour is shown in black for each dataset in the top row. 930

The normalized December Niño 3.4 timeseries are used to composite all years greater than 1 931

standard deviation (El Niño) and all years less than –1 standard deviation (La Niña), and the 932

ENSO composite is formed by subtracting the La Niña composite from the El Niño composite. 933

Observations are from GPCP for precipitation (1979-2017), ERSSTv5 for SST (1920-2017), 934


51

BEST for 2-m air temperature (1920-2017), and ECMWF Reanalysis for the 20th century 935

[ERA20C; Poli et al., 2016] combined with ERA-Interim [Berrisford et al., 2009] for sea level 936

pressure (1920-2017). 937

8.5 Stratospheric ozone 938

WACCM6 is able to accurately simulate the evolution of stratospheric ozone over the late 20th 939

century. In particular, WACCM6 can produce the halogen-induced ozone loss in the Southern 940

Hemisphere polar stratosphere in spring. This occurs due to the activation of chlorine and 941

bromine via heterogeneous chemistry on the surface of polar stratospheric cloud particles [WMO, 942

2010]. Figure 15 illustrates the annual evolution of the Southern Hemisphere polar cap Total 943

Column Ozone (TCO) in October. Coupled historical and AMIP-type simulations with 944

WACCM6 are able to broadly represent the ozone loss from the 1970s through 1990s. “Specified 945

Dynamics” (SD) simulations, wherein the model is constrained with MERRA-2 [Rienecker et 946

al., 2011] meteorology (SD-MERRA2), indicate that much of the variability in observations can 947

be explained by interannual variability of meteorology in the Southern Hemisphere polar vortex. 948


52

Figure 15. October 90°-60°S Total Column Ozone (TCO, in Dobson Units: DU) from 949

CESM2(WACCM6) simulations (blue), fixed SST (AMIP-type) simulations (green), Specified 950

Dynamics simulation (SD-MERRA2; purple), and observations (SBUV-MOD; black). For 951

CESM2(WACCM6) and AMIP, dots represent individual ensemble members while the 952

ensemble means are shown by the solid lines. 953

9 Atlantic Meridional Overturning Circulation 954

One of the climatically most important ocean circulations is the AMOC, representing a zonally-955

integrated view of the Atlantic Ocean circulation. Through its associated heat and salt transports, 956

AMOC significantly impacts spatial and temporal variability in the North Atlantic and 957

surrounding areas, even influencing the global climate. Many studies link the low-frequency 958

variability of the North Atlantic SSTs to fluctuations in AMOC [Zhang et al., 2019 and 959


53

references therein]. Figure 16 presents the 1995-2014 average AMOC distributions from 960

CESM2(CAM6) and CESM2(WACCM6) in comparison to CESM1(LENS). These are for the 961

Eulerian-mean, i.e., resolved flow, component only as the parameterized contributions are 962

relatively small. The figure shows that AMOC spatial structure and transports are 963

indistinguishable between CESM2(CAM6) and CESM2(WACCM6). The primary pattern is the 964

clockwise circulation cell (positive contours) associated with the southward flowing North 965

Atlantic Deep Water (NADW). The maximum NADW transport is about 24 Sv (1 Sv = 106 m3 s-966

1) in CESM2(CAM6) and CESM2(WACCM6), while it is about 26 Sv in CESM1(LENS). There 967

are no basin-scale observational estimates for AMOC, and only recently a few trans-basin 968

transport estimates have become available. Among these, the RAPID array at 26.5°N has the 969

longest record. For the April 2004 – February 2017 period, the mean and standard deviation of 970

the estimated NADW maximum transport at 26.5°N are 17.0 ± 3.3 Sv [Smeed et al., 2018; 971

Frajka-Williams et al., 2019]. Acknowledging the different averaging periods, the model 972

maximum transports at this latitude are larger than the RAPID estimate by about 2-3 Sv with 973

CESM1(LENS) showing the largest magnitude. 974

The penetration depth of the NADW cell as measured by the depth of the zero-contour line is 975

similarly shallower in CESM2(CAM6) and CESM2(WACCM6) than in CESM1(LENS). Given 976

that the ocean component in all model versions uses the same overflow parameterization 977

[Danabasoglu et al., 2010] to represent dense-water overflows through the Denmark Strait and 978

Faroe Bank Channel, the differences in penetration depth between the CESM1 and CESM2 979

simulations reflect differences in water mass properties, partially arising from surface flux 980

differences. 981


54

The counterclockwise (negative contours) circulation region in the abyssal ocean below the 982

NADW cell represents the northward flowing Antarctic Bottom Water (AABW). In comparison 983

to very limited observational estimates, all the simulations suffer from a chronic bias of very 984

anemic AABW transport due to too weak AABW formation, with CESM2(CAM6) and 985

CESM2(WACCM6) showing only a marginally broader and stronger AABW cell than in 986

CESM1(LENS). The observational range for the AABW transport into the Atlantic basin is 987

about 1-5 Sv at 24.5°N based on sparse datasets dating back to 1957 [Frajka-Williams et al., 988

2011]. Both CESM2 simulations have a maximum transport of about 0.5 Sv, compared to near 989

zero in CESM1(LENS). Such weak deep overturning circulations lead to weak ventilation at 990

depth evident in tracer distributions particularly in the North Pacific, compared to available 991

observations (not shown). 992

Figure 16. 1995-2014 average Eulerian-mean Atlantic meridional overturning circulation (in Sv) 993

from (left) CESM2(CAM6), (middle) CESM2(WACCM6), and (right) CESM1(LENS). The 994

model distributions represent full ensemble means. The positive and negative contours indicate 995

clockwise and counter-clockwise circulations, respectively. 996

10 Sea-Ice and Land-Ice Fields 997

10.1 Sea-ice thickness and extent 998


55

The annual-mean sea-ice thickness and extent are shown in Fig. 17, averaged for 1979-2014. For 999

the Arctic, sea-ice in both CESM2 simulations is thinner than in CESM1(LENS), which has a 1000

total annual-mean Arctic ice volume of 28.9x103 km3. The sea-ice is considerably thicker in the 1001

simulations with CESM2(WACCM6) than with CESM2(CAM6), with the respective total 1002

annual-mean Arctic ice volumes of 21.4x103 and 15.4x103 km3. The CESM2(WACCM6) 1003

simulations agree well with observationally-based estimates from the Pan-Arctic Ice Ocean 1004

Modeling and Assimilation System [PIOMAS; Schweiger et al., 2011], which simulates a 1979-1005

2014 average Arctic ice volume of 20.2x103 km3. The thicker sea-ice in CESM2(WACCM6) is 1006

related to liquid Arctic clouds, which are thicker as a result of increased aerosol concentrations 1007

predicted from the interactive chemistry module [Gettelman et al., 2019a]. Specifically, the 1008

interactive oxidants with full chemistry in CESM2(WACCM6) mean that aerosols will evolve 1009

differently. Because the aerosols are not produced locally in remote regions such as the Arctic, 1010

the increase in aerosol concentrations over the Arctic is due to a lengthening of the aerosol 1011

lifetime. The anti-correlation between aerosols and oxidants is meteorologically dependent, and 1012

so may not be seen in monthly average oxidants fields used in CESM2(CAM6), leading to a 1013

different aerosol evolution, particularly in such remote areas. The differences in sea-ice thickness 1014

between the two model configurations have consequences for the annual cycle of ice extent. In 1015

particular, with the thin ice bias in the CESM2(CAM6) simulations, more melting occurs during 1016

the summer months, with a resulting low summer ice extent as compared to satellite passive-1017

microwave observations (Fig. 17g). In contrast, CESM2(WACCM6) and CESM1(LENS) have 1018

an excellent simulation of summer ice extent compared to observations. In both CESM2(CAM6) 1019

and CESM2(WACCM6) simulations, the ice extent is low in winter compared to observations. 1020


56

This is in contrast to CESM1(LENS) which simulates a winter ice extent that is in very good 1021

agreement with observations. 1022

In the Southern Hemisphere, the sea-ice is nearly identical in CESM2(CAM6) and 1023

CESM2(WACCM6), both somewhat thinner than the sea-ice simulated in CESM1(LENS). As in 1024

observations, a primarily seasonal ice pack is present and the ice remains thinner than 1 m in the 1025

annual mean over most of the sea-ice domain for all model versions. Sea-ice thicker than 1 m 1026

forms in the Amundsen/Bellingshausen and Weddell Seas. This is in reasonable agreement with 1027

the limited in-situ ice thickness observations [Worby et al., 2008]. The ice extent annual cycle is 1028

also reasonably well simulated in both CESM2 simulations. The maximum simulated ice cover is 1029

biased low by about 1.6x106 km2, and the ice melt-back is delayed by about one month relative 1030

to observations but reaches a similar minimum ice cover in February. This is in contrast to 1031

CESM1(LENS), which has a good simulation of the winter maximum ice extent but retains too 1032

much ice at the summer minimum. 1033


57

Figure 17. The annual-mean sea-ice thickness (in m) in CESM2(CAM6) (a and b), 1034

CESM2(WACCM6) (c and d), and CESM1(LENS) (e and f) simulations. The annual cycle of 1035

sea-ice extent (in km2) for the (g) Northern Hemisphere and (h) Southern Hemisphere. All values 1036


58

are averages for 1979-2014. The observations are estimates from the PIOMAS dataset. The 1037

analysis uses 6, 3, and 40 ensemble members for CESM2(CAM6), CESM2(WACCM6), and 1038

CESM1(LENS), respectively. In (h), CESM2(CAM6) and CESM2(WACCM6) lines overlap. 1039

10.2 Land-ice surface mass balance 1040

Figure 18 shows the simulated Greenland and Antarctic SMB for the 1961-1990 mean of 1041

CESM2(CAM6) and CESM2(WACCM6) historical simulations in comparison to the regional 1042

atmospheric climate model RACMO2 [Noël et al., 2018; van Wessem et al., 2018] and to 1043

CESM1(CAM4) [Vizcáino et al., 2013]. (We show CESM1(CAM4) results because ice-sheet 1044

SMB was not computed for CESM1(LENS) simulations.) CESM2(CAM6) and 1045

CESM2(WACCM6) solutions are very similar to each other, both reproducing the RACMO2 1046

SMB reasonably well for both ice sheets. Improvements relative to CESM1(CAM4) can be 1047

attributed in part to more realistic snow / firn physics in CLM5 [van Kampenhout et al., 2017], 1048

and to the Beljaars et al. [2004] orographic form drag scheme and new cloud microphysics in 1049

CAM6. Remaining biases for Greenland include excessive snowfall in the south, likely 1050

associated with unresolved topography, excessive rainfall, and a generally too positive SMB bias 1051

in the north. 1052


59

Figure 18. Surface mass balance (SMB in mm water equivalent yr-1) of the Greenland ice sheet 1053

(top row) and Antarctic ice sheet (bottom row) from statistically downscaled output of the 1054

regional atmospheric climate model RACMO2.3p2, and from CESM2(CAM6), 1055

CESM2(WACCM6), and CESM1(CAM4). The CESM results are from historical simulations 1056

averaged over 1961-1990, with eleven ensemble members for CESM2(CAM6), three members 1057

for CESM2(WACCM6), and one member for CESM1(CAM4). For Greenland, CESM2 SMB 1058

has been downscaled using multiple elevation classes to a 5-km grid for CESM1(CAM4) and to 1059

a 4-km CISM grid for both versions of CESM2. For Antarctica, both CESM1 and CESM2 SMB 1060

are shown on the land model grid without downscaling. RACMO2.3p2 data are averaged over 1061

1961-1990 for Greenland [Noël et al., 2018] and over 1979-2010 for Antarctica [van Wessem et 1062

al., 2018] at a resolution of 11 and 27 km, respectively. 1063


60

11 Land Fields 1064

The land model variables are assessed with the International Land Model Benchmarking 1065

[ILAMBv2.1; Collier et al., 2018] package. The ILAMBv2 used here integrates analysis for 30 1066

variables evaluated against more than 60 data products. ILAMB package results include 1067

statistics, maps, time series, and metrics for relative bias, rms error, seasonal cycle phase, spatial 1068

distribution, and interannual variability, as well as variable-to-variable assessments that evaluate 1069

functional relationships between variables. ILAMB summary diagrams, which integrate scores 1070

across several datasets and metrics for each variable and variable-to-variable comparisons, are 1071

shown in Fig. 19 for three ensemble members each of CESM1(LENS) and CESM2(CAM6) – 1072

CESM2(WACCM6) results are very similar to those of CESM2(CAM6). For most variables 1073

assessed, CESM2(CAM6) better matches the observations than CESM1(LENS). Except for 1074

precipitation, all the land climate variables (shown in grey in the left panel of Fig. 19) show 1075

modest improvement, indicating an improved surface climate simulation in CESM2(CAM6) due 1076

to a combination of improvements in CAM6 and CLM5. The land carbon, water, and energy 1077

variable improvements are also seen in land-only simulations forced with historical climate 1078

reconstructions (not shown, see Lawrence et al. [2019] for detailed assessment), which suggests 1079

that improvements in these variables are predominantly a result of CLM5 physics and 1080

biogeochemistry model development rather than an outcome of a better-simulated surface 1081

climate in CESM2(CAM6). The consistently improved variable-to-variable scores suggests 1082

improved process-representation in CLM5. These functional relationship scores are improved 1083

even for comparisons involving a mean quantity that is slightly degraded in CESM2(CAM6) – 1084

e.g., land precipitation is slightly degraded, but precipitation versus gross primary productivity 1085

and precipitation versus burned area are improved. This result supports the interpretation that the 1086


61

improvements derive mainly from CLM5 developments. Insight into the sources of improvement 1087

is provided in Lawrence et al. [2019] and Wieder et al. [2019]. A caveat here is that ILAMB 1088

results should be interpreted carefully and that an improvement or degradation in a particular 1089

ILAMB variable should not be overly interpreted as indicative of whether or not results that are 1090

dependent on that variable are reliable or not [see Lawrence et al., 2019 and Bonan et al., 2019]. 1091

For example, soil carbon is degraded according to the ILAMB metric, but other aspects of soil 1092

carbon such as the soil carbon turnover time appear to be improved [Lawrence et al., 2019], 1093

potentially reflecting a lack of consistency across observational datasets or a limitation in the 1094

applied ILAMB metric. A note of caution on the soil carbon field in CESM2 is that due to an 1095

error in the model spinup process, unrealistically large soil carbon stocks are seen for some high-1096

latitude grid cells where vegetation does not survive the cold climate (e.g., Canadian 1097

Archipelago). For these grid cells, the spinup process was supposed to set soil carbon stocks to 1098

near zero, but due to very small but non-zero vegetation stocks, the soil carbon stocks were not 1099

set to zero as intended. Model users can resolve this problem (which was fixed in CESM2.1.1) 1100

by screening out large soil carbon stocks when vegetation carbon is zero. 1101


62

Figure 19. ILAMB summary diagrams for three ensemble members each of CESM1(LENS) and 1102

CESM2(CAM6) historical simulations. The left panel shows the single-variable assessment 1103

summary for land variables. The upper part of the right panel shows assessment of the land 1104

forcing variables as simulated by the atmosphere model. The lower part of the right panel shows 1105

a variable-to-variable comparison summary. See Collier et al. [2018] for details on all metrics. 1106

Full ILAMB package results for both coupled and land-only simulations are available at 1107

http://www.cesm.ucar.edu/experiments/cesm2.0/land/diagnostics/clm_diag_ILAMB.html. 1108

One of the clearest improvements in CESM2 and CLM5 is the simulated global land carbon 1109

accumulation trends. CESM1/CLM4 produced an unrealistically strong nutrient limitation on 1110

photosynthesis, which limited the capacity in CESM1 for carbon uptake in response to increases 1111

in atmospheric CO2 concentrations. Consequently, in CESM1 land-use and land-cover change 1112


63

carbon loss fluxes dominate over the CO2 fertilization response, and CESM1 simulates an 1113

accumulated carbon loss over the period 1960 to 2004, whereas the observationally-based 1114

estimates indicate a carbon gain of ~40 PgC (Fig. 20). In CESM2(CAM6) and 1115

CESM2(WACCM6), the CO2 fertilization response appears to be more reasonable and, when 1116

this is combined with updates to the land-use and land-cover change carbon fluxes, results in a 1117

late 20th century accumulated carbon trend that agrees well with observationally-based estimates. 1118

Further discussion on the source of improvement for this feature of CESM2, which is also seen 1119

in land-only simulations, is provided in Lawrence et al. [2019], Wieder et al. [2019], and Bonan 1120

et al. [2019]. In addition to the improved accumulated carbon, CESM2 simulations also show 1121

evidence of considerable improvements in the amplitude of the annual cycle of net ecosystem 1122

exchange, especially at northern high latitudes (see Carbon Dioxide variable metrics in Fig. 19), 1123

which translate to improved annual cycle amplitudes of atmospheric CO2 concentrations in 1124

emissions-driven simulations. 1125


64

Figure 20. Land carbon accumulation for the period 1960-2014 for 3 historical ensemble 1126

members of CESM2(CAM6), CESM2(WACCM6), and CESM1(LENS) along with 1127

observationally-based estimates. The latter are from the Global Carbon Project [Le Quéré et al., 1128

2016]. Grey shading shows uncertainty range for the observational estimates, using an 1129

uncertainty of 0.8 Pg C yr-1 and calculated assuming that the errors are additive in time as in 1130

Koven et al. [2013]. 1131

12 Equilibrium Climate Sensitivity and Transient Climate Response 1132

An important emergent property of the coupled model is its Equilibrium Climate Sensitivity 1133

[ECS; e.g., Stocker et al., 2013], defined as the equilibrium change in global-mean surface 1134

temperature (SST over the ice-free oceans and a radiative temperature over land and sea-ice) 1135

after a doubling of CO2 concentration. ECS can be estimated using a slab ocean model (SOM) 1136

configuration, because 1) such a configuration equilibrates much faster compared to one with a 1137

full-depth ocean model (about 50 vs. > 1000 years), and 2) ECS values obtained with a SOM 1138


65

versus a full-depth ocean component are comparable to each other [Danabasoglu and Gent, 1139

2009]. Therefore, SOM configurations are used here to estimate the ECS of both model 1140

configurations. The control SOM experiments use a pre-industrial value of CO2 set at 280 ppmv. 1141

These respective control simulations are compared with a second set of SOM experiments in 1142

which the CO2 concentration is instantaneously doubled to 560 ppmv (2xCO2). All SOM 1143

experiments are integrated long enough to ensure that global-mean surface temperature has 1144

equilibrated and TOA radiative imbalances are near zero. Differencing the equilibrated surface 1145

temperatures between the respective control and 2xCO2 experiments, the ECS values are 1146

computed as 5.3°C for CESM2(CAM6) [Gettelman et al., 2019b] and 5.1°C for 1147

CESM2(WACCM6). These values represent significant increases from an ECS of around 4.0°C 1148

in the SOM version of CESM1(CAM5) [Gettelman et al., 2012]. 1149

Values of climate sensitivity derived from the first 150 years of fully-coupled 4xCO2 runs using 1150

the Gregory method [so called Effective Climate Sensitivity; Gregory et al., 2004] are 5.3°C for 1151

CESM2(CAM6) and 4.8°C for CESM2(WACCM6) compared to 4.3°C in CESM1(CAM5) 1152

obtained with the same method [Meehl et al., 2013]. However, this level of relatively good 1153

agreement with the above SOM-based estimates of ECS is deceptive, because the Gregory 1154

method estimate of climate sensitivity for CESM2 is highly sensitive to the number of simulation 1155

years used in the calculation, due to nonlinearities in the relationship between TOA energy 1156

imbalance and temperature, with larger estimates of climate sensitivity as more years are used. 1157

The evolution of CESM2's climate sensitivity through several development versions of the 1158

model is examined in detail in Gettelman et al. [2019b]. Both of these studies suggest that the 1159

increased climate sensitivity in CESM2 has arisen from a combination of relatively small 1160

changes to cloud microphysics and boundary layer parameters as well as from land parameter 1161


66

changes that were introduced late in the development process, along with the introduction of 1162

more realistic clouds with CLUBB and MG2. 1163

Another important emergent property of the coupled model is its Transient Climate Response 1164

(TCR). It is calculated from a pair of fully coupled simulations: a control simulation, again 1165

usually for PI conditions, and a 1% per year CO2 concentration increase simulation. Following 1166

the ESMValTool [Eyring et al., 2020] procedure, TCR in CESM2 is calculated as the difference 1167

of the 20-year average (years 60-79) global mean surface temperature, centered around the time 1168

of CO2 doubling at about year 70, from the 140-year linear trend (fit) of the corresponding PI 1169

control, overlapping the 1% per year CO2 increase run. The TCR values are obtained as 2.0°C for 1170

CESM2(CAM6) and 1.9°C for CESM2(WACCM6). These values are very comparable to a TCR 1171

value of 2.3°C for CESM1(CAM5) reported by Meehl et al. [2013]. Processes leading to 1172

CESM2’s TCR value are being investigated. Preliminary analysis suggests that, in CESM2, 1173

while tropics and the Southern Ocean warm faster, the Northern Hemisphere at mid- and high-1174

latitudes warms at a slower rate, in comparison to CESM1. 1175

It is important to stress that all estimates of ECS and TCR are subject to substantial uncertainties 1176

arising from the details of the calculation procedures. These uncertainties include the selection of 1177

reference periods for the PI simulations, the length of the averaging periods, and whether 1178

detrending is applied or not. For the above CESM values, such details may result in 0.1-0.2°C 1179

differences. 1180

13 Summary and Discussion 1181

The new version of the Community Earth System Model, CESM2, contains many substantial 1182

science and infrastructure improvements and capabilities. The CESM project is contributing 1183


67

simulations to the CMIP6 effort with two CESM2 model configurations differing only in their 1184

atmospheric components. While one configuration uses the low-top CAM6, CESM2(CAM6), the 1185

other employs WACCM6 with its high-top and comprehensive chemistry, CESM2(WACCM6). 1186

Both configurations use the same nominal 1° horizontal resolution. A noteworthy new approach 1187

in CESM2 is that CESM2(WACCM6) PI and historical simulations were performed first to 1188

obtain some of the necessary forcing datasets to run the corresponding CESM2(CAM6) 1189

simulations. This was critical to ensure that, for the first time for CESM, a physically consistent 1190

pair of low- and high-top atmospheric models are used. This manuscript documents the CESM2 1191

development process and evaluates solutions from the 1200-year CESM2(CAM6) and 500-year 1192

CESM2(WACCM6) PI control integrations and from the subsequent historical simulations. The 1193

focus has been primarily on the historical simulations, as they can be more easily and directly 1194

compared with modern observations. 1195

The new developments and capabilities in CESM2 result in many improvements in solutions in 1196

comparison to available observations and those of CESM1. These improvements include: major 1197

reductions in the tropical wet biases of the Indian Ocean, the SPCZ, and the Atlantic ITCZ, 1198

generally with less of a double ITCZ structure in both the Pacific and Atlantic Basins; reductions 1199

in the dipole of Andean wet and Amazon dry biases as well as in Australian and Western US 1200

excessive precipitation biases; substantial reductions in the reflective biases of the clouds 1201

throughout the tropics; a much better representation of MJO characteristics with coherent 1202

eastward propagation of wind and precipitation anomalies; more realistic representation of the 1203

northwest – southeast tilt of the sea level pressure anomaly teleconnection patterns associated 1204

with ENSO over the North Pacific; a global land carbon accumulation trend that matches the 1205

observationally-based trend remarkably well; and better representation of the SMB for both the 1206


68

Greenland and Antarctic ice sheets. An important finding is that CESM2(CAM6) and 1207

CESM2(WACCM6) historical simulations produce solutions that are very similar to each other 1208

in most variables. So, these improvements are present in both sets of simulations. An exception 1209

is the Arctic sea-ice thickness with CESM2(WACCM6) showing thicker sea-ice than in 1210

CESM2(CAM6) in better agreement with observations. The thicker sea-ice in 1211

CESM2(WACCM6) is related to liquid Arctic clouds, which are thicker as a result of increased 1212

aerosol concentrations predicted from the interactive chemistry module. 1213

Figure 21 presents a summary analysis comparing CESM2 solutions to those of CESM1(LENS). 1214

The figure shows mean pattern correlations across climatological mean, seasonal contrasts (JJA-1215

DJF), and ENSO teleconnections for the first three ensemble members of CESM2(CAM6), 1216

CESM2(WACCM6), and CESM1(LENS) historical simulations for 11 atmospheric fields with 1217

respect to many observational-based datasets. Indeed, CESM2 simulations have notable 1218

improvements in many of the metrics considered here compared to those of CESM1(LENS). 1219


69

Figure 21. Mean pattern correlations across climatological mean, seasonal contrasts (JJA-DJF), 1220

and ENSO teleconnections for three ensemble members of CESM2(CAM6), 1221

CESM2(WACCM6), and CESM1(LENS) for various fields: precipitable water (prw), sea level 1222

pressure (psl), 500 hPa relative humidity (hur500), outgoing longwave radiation (rlut), 500 hPa 1223

geopotential height (zg500), 2-m air temperature (tas), top of atmosphere net shortwave flux 1224

(rsnt), longwave and shortwave cloud forcing (lwcf, swcf), rainfall (pr), and 500 hPa vertical 1225

velocity (wap500). 1226

Undoubtedly, many biases still remain in the model solutions. These biases range from relatively 1227

minor ones, such as local precipitation errors without discernable global impacts to biases that 1228

can have significant climate impacts such as the thin Arctic sea-ice in CESM2(CAM6) 1229

simulations. The sea-ice bias can perhaps be tuned away at the expense of possible appearance of 1230

other biases. There are also much more persistent biases such as the incorrect separation and path 1231


70

of the Gulf Stream – North Atlantic Current system, with accompanying large SST and surface 1232

salinity biases. This latter class of biases can have significant climate impacts as well, but no 1233

robust remedy exists to alleviate them in non-eddy-permitting / -resolving ocean models usually 1234

used in climate simulations. 1235

The path to finalizing a CESM2 version that produced acceptable simulations took much longer 1236

than anticipated. The process was hindered particularly by two major unacceptable features in 1237

simulations. The first concerned formation of unrealistically extensive sea-ice cover over the 1238

Labrador Sea region in PI control simulations. Unfortunately, extensive analysis of simulations 1239

did not reveal a particular process (or processes) as the culprit. To avoid further delays, a 1240

practical approach was adopted that involved performing several member ensemble simulations 1241

which differed at round-off levels in their atmospheric potential temperature initial conditions, 1242

and then using one of the members without an extensive Labrador Sea ice cover to provide the 1243

ocean and sea-ice initial conditions. The second issue was that the global-mean surface 1244

temperature time series in historical simulations showed a substantial and unrealistic cooling, 1245

particularly during the second half of the 20th century, with end-of-the-century temperatures 1246

being colder than in the 1850s. Adoption of updated CMIP6 emissions datasets, along with 1247

several modifications in the aerosol – cloud interaction processes based on observational data 1248

from recent volcanic sulfur emission events, resulted in surface temperature time series that 1249

evolve realistically during the historical period. Nevertheless, there are several discrepancies in 1250

the details of the modeled and observed time series. For example, the model surface temperatures 1251

are warmer in the early part of the 21st century than in observations by 0.2 and 0.4°C in 1252

CESM2(CAM6) and CESM2(WACCM6), respectively, due at least partly to the missing 1253

decadal-scale slowdown in warming during roughly 2000-2012 in the simulations. However, 1254


71

such a difference is not unexpected because the observed slowdown is associated with internal 1255

climate variability in addition to anthropogenic and natural forcings. This warm bias is also 1256

reflected in the SST and over-land surface temperature distributions. 1257

An emergent property of climate simulations is their ECS, which in CESM2 is calculated as 1258

5.3°C for CESM2(CAM6) and 5.1°C for CESM2(WACCM6) based on CO2 doubling 1259

experiments with CESM2 SOM configurations. These values represent a significant increase 1260

from an ECS of 4.0°C calculated for CESM1, also using a SOM configuration. Climate 1261

sensitivities based on the Gregory method from fully-coupled 4xCO2 runs are calculated as 1262

5.3°C for CESM2(CAM6) and 4.8°C for CESM2(WACCM6), again larger than the value of 1263

4.3°C in CESM1(CAM5) obtained with the same method [Meehl et al., 2013]. However, these 1264

latter estimates are highly sensitive to the number of simulation years used in the calculation due 1265

to nonlinearities in the relationship between TOA energy imbalance and temperature. Our 1266

analysis suggests that the increased climate sensitivity in CESM2 has arisen from a combination 1267

of seemingly small changes to cloud microphysics and boundary layer parameters as well as 1268

from land parameter changes that were introduced late in the development process, along with 1269

the introduction of more realistic clouds with CLUBB and MG2 [see Gettelman et al., 2019]. In 1270

contrast, the CESM2 TCR values, calculated as 2.0°C for CESM2(CAM6) and 1.9°C for 1271

CESM2(WACCM6), are comparable to a TCR value of 2.3°C for CESM1(CAM5) reported in 1272

Meehl et al. [2013]. 1273

The CESM community has been participating in about 20 CMIP6-endorsed MIPs. The datasets 1274

from these MIPs and all the DECK simulations are being made available on the ESGF site for 1275

use of the broader research community. These simulations, including the PI control and historical 1276

integrations presented here, are analyzed in more detail in over 70 manuscripts collected in the 1277


72

AGU CESM2 Virtual Special Issue. Our preliminary analyses suggest that the CESM2 1278

simulations exhibit agreement with satellite era observations of the climate mean state, seasonal 1279

cycle, and interannual variability that is among the closest of coupled climate models in the 1280

present CMIP6 archive. 1281

Model and Data Availability 1282

Previous and current CESM versions are freely available at www.cesm.ucar.edu:/models/cesm2/. 1283

The CESM datasets used in this study are also freely available from the Earth System Grid 1284

Federation (ESGF) at esgf-node.llnl.gov/search/cmip6 or from the NCAR Digital Asset Services 1285

Hub (DASH) at data.ucar.edu or from the links provided from the CESM web site at 1286

www.cesm.ucar.edu. 1287

Acknowledgments 1288

The CESM project is supported primarily by the National Science Foundation (NSF). This 1289

material is based upon work supported by the National Center for Atmospheric Research 1290

(NCAR), which is a major facility sponsored by the NSF under Cooperative Agreement No. 1291

1852977. Computing and data storage resources, including the Cheyenne supercomputer 1292

(doi:10.5065/D6RX99HX), were provided by the Computational and Information Systems 1293

Laboratory (CISL) at NCAR. We thank all the scientists, software engineers, and administrators 1294

who contributed to the development of CESM2. GPCP Precipitation data are provided by the 1295

NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their web site at 1296

https://www.esrl.noaa.gov/psd/. The implementation of the WaveWatch-III has benefitted from 1297

the input and collaborations with the wave model developers at the NOAA National Centers for 1298

Environmental Prediction (NCEP). 1299

http://www.cesm.ucar.edu/models/cesm2/

https://www.esrl.noaa.gov/psd/


73

Appendix A: Construction of forcing datasets from CESM2(WACCM6) historical 1300

simulations for use in historical simulations with CESM2(CAM6) 1301

For CESM2(CAM6) historical simulations, 12-month annual cycles are provided at quasi-1302

decadal frequency (Table A1). The first year in the forcing file is 1849, which is derived from 1303

years 21-50 of the CESM2(WACCM6) PI control simulation. Subsequent annual cycles are 1304

provided using the ensemble average of the three CESM2(WACCM6) historical simulations. 1305

One annual cycle per decade is provided from 1860 to 2010, each averaged over 9 years centered 1306

on the year. For example, oxidants provided for the mid-month date of 16 January 1860 are 1307

derived from the average of January averages for the years 1856-1864 of three 1308

CESM2(WACCM6) historical ensemble members. An annual cycle is also provided for 2015, 1309

averaged from years 2012-2014, as the end year needed for time interpolation. CESM2(CAM6) 1310

interpolates annual cycles for years between those provided. 1311

Table A1. Forcing datasets date information 1312

Seasonal cycle date WACCM6 experiment Years averaged

1849 PI control 21-50

1860 historical 1856-1864

















74



References 1313

Ali, A. A., C. Xu, A. Rogers, R. A. Fisher, S. D. Wullschleger, E. C. Massoud, J. A. Vrugt, J. D. 1314

Muss, N. G. McDowell, J. B. Fisher, P. B. Reich, and C. J. Wilson (2016), A global scale 1315

mechanistic model of photosynthetic capacity (LUNA V1.0), Geoscientific Model 1316

Development, 9, 587-606, doi: 10.5194/gmd-9-587-2016. 1317

Armstrong, R. A., C. Lee, J. I. Hedges, S. Honjo, and S. G. Wakeham (2002), A new, 1318

mechanistic model for organic carbon fluxes in the ocean based on the quantitative 1319

association of POC with ballast minerals, Deep-Sea Research, 49, 219-236, doi: 1320

10.1016/S0967-0645(01)00101-1. 1321

Aschwanden, A., M. A. Fahnestock, and M. Truffer (2016), Complex Greenland outlet glacier 1322

flow captured, Nat. Commun., 7, 10524, doi: 10.1038/ncomms10524. 1323

Asselin, R. (1972), Frequency filter for time integrations, Mon. Wea. Rev., 100, 487-490. 1324

Assur, A. (1958), Composition of sea ice and its tensile strength. In Arctic sea ice; Conference 1325

held at Easton, Maryland, February 24–27, 1958, v. 598 of Publs. Natl. Res. Coun. 1326

Wash., pages 106-138, Washington, DC, US. 1327

Beljaars, A. C. M., A. R. Brown, and N. Wood (2004), A new parameterization of turbulent 1328

orographic form drag, Q. J. Roy. Meteor. Soc., 130, 1327-1347, doi: 10.1256/qj.03.73. 1329

Berrisford, P., D. P. K. F. Dee, K. Fielding, M. Fuentes, P. Kallberg, S. Kobayashi, and S. 1330

Uppala (2009), The ERA-interim archive. ERA report series, (1), 1-16, Reading, U. K. 1331


75

Beusen, A. H. W., L. P. H. Van Beek, A. F. Bouwman, J. M. Mogollón, and J. J. Middelburg 1332

(2015), Coupling global models for hydrology and nutrient loading to simulate nitrogen 1333

and phosphorus retention in surface water – description of IMAGE-GNM and analysis of 1334

performance, Geosci. Model Dev., 8, 4045-4067, doi: 10.5194/gmd-8-4045-2015. 1335

Beusen, A. H. W., A. F. Bouwman, L. P. H. Van Beek, J. M. Mogollón, and J. J. Middelburg 1336

(2016), Global riverine N and P transport to ocean increased during the 20th century 1337

despite increased retention along the aquatic continuum, Biogeosciences, 13, 2441-2451, 1338

doi: 10.5194/bg-13-2441-2016. 1339

Bogenschutz, P. A., A. Gettelman, H. Morrison, V. E. Larson, C. Craig, and D. P. Schanen 1340

(2013), Higher-order turbulence closure and its impact on climate simulation in the 1341

Community Atmosphere Model, J. Climate, 26, 9655-9676, doi: 10.1175/JCLI-D-13-1342

00075.1. 1343

Bonan, G. B., P. J. Lawrence, K. W. Oleson, S. Levis, M. Jung, M. Reichstein, D. M. Lawrence, 1344

and S. C. Swenson (2011), Improving canopy processes in the Community Land Model 1345

version 4 (CLM4) using global flux fields empirically inferred from FLUXNET data, J. 1346

Geophys. Res. Biogeosciences, 116(G2), doi: 10.1029/2010jg001593. 1347

Bonan, G. B., D. L. Lombardozzi, W. R. Wieder, K. W. Oleson, D. M. Lawrence, F. M. 1348

Hoffman, and N. Collier (2019), Model structure and climate data uncertainty in 1349

historical simulations of the terrestrial carbon cycle (1850-2014), Global Biogeochemical 1350

Cycles, doi: 10.1029/2019GB006175. 1351

Boville, B. A., and P. R. Gent (1998), The NCAR Climate System Model, version one, J. 1352

Climate, 11, 1115-1130. 1353


76

Brunke, M. A., P. Broxton, J. Pelletier, D. Gochis, P. Hazenberg, D. M. Lawrence, L. R. Leung, 1354

G.-Y. Niu, P. A. Troch, and X. Zeng (2016), Implementing and evaluating variable soil 1355

thickness in the Community Land Model, version 4.5 (CLM4.5), J. Climate, 29, 3441-1356

3461, doi: 10.1175/JCLI-D-15-0307.1. 1357

Cavaleri, L., B. Fox-Kemper, and M. Hemer (2012), Wind waves in the coupled climate system, 1358

Bull. Amer. Meteor. Soc., 93, 1651-1661. 1359

Cavalieri, D. J., C. L. Parkinson, P. Gloersen, and H. J. Zwally (1996), updated yearly. Sea ice 1360

concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS passive microwave 1361

data, version 1. Boulder, Colorado USA. NASA National Snow and Ice Data Center 1362

Distributed Active Archive Center. doi: 10.5067/8GQ8LZQVL0VL. 1363

Collier, N., F. M. Hoffman, D. M. Lawrence, G. Keppel-Aleks, C. D. Koven, W. J. Riley, M. 1364

Mu, and J. T. Randerson (2018), The International Land 1 Model Benchmarking 1365

(ILAMB) system: Design, theory, and implementation. J. Advances in Modeling Earth 1366

Systems, doi: 10.1029/2018MS001354. 1367

Collins, W. D., C. M. Bitz, M. L. Blackmon, G. B. Bonan, C. S. Bretherton, J. A. Carton, P. 1368

Chang, S. C. Doney, J. J. Hack, T. B. Henderson, J. T. Kiehl, W. G. Large, D. D. 1369

McKenna, B. D. Santer, and R. D. Smith (2006), The Community Climate System Model 1370

version 3 (CCSM3), J. Climate, 19, 2122-2143. 1371

Danabasoglu, G., and P. R. Gent (2009), Equilibrium climate sensitivity: Is it accurate to use a 1372

slab ocean model? J. Climate, 22, 2494-2499, doi: 10.1175/2008JCLI2596.1. 1373

Danabasoglu, G., W. G. Large, and B. P. Briegleb (2010), Climate impacts of parameterized 1374

Nordic Sea overflows, J. Geophys. Res., 115, C11005, doi:10.1029/2010JC006243. 1375


77

Danabasoglu, G., S. C. Bates, B. P. Briegleb, S. R. Jayne, M. Jochum, W. G. Large, S. Peacock, 1376

and S. G. Yeager (2012), The CCSM4 ocean component, J. Climate, 25, 1361-1389, doi: 1377

10.1175/JCLI-D-11-00091.1. 1378

Dee, D. P., S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli, S. Kobayashi, U. Andrae, 1379

M. A. Balmaseda, G. Balsamo, P. Bauer, P. Bechtold, A. C. M. Beljaars, L. van de 1380

Berg, J. Bidlot, N. Bormann, C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. 1381

Haimberger, S. B. Healy, H. Hersbach, E. V. Holm, L. Isaksen, P. Kallberg, M. 1382

Kohler, M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J.-J. Morcrette, B.-K. 1383

Park, C. Peubey, P. de Rosnay, C. Tavolato, J.-N. Thepaut, and F. Vitart (2011), 1384

The ERA-Interim reanalysis: Configuration and performance of the data 1385

assimilation system, Q. J. R. Meteorol. Soc., 137, 553-597. 1386

Deser, C., A. S. Phillips, R. A. Tomas, Y. Okumura, M. A. Alexander, A. Capotondi, J. D. Scott, 1387

Y.-O. Kwon, and M. Ohba (2012), ENSO and Pacific decadal variability in Community 1388

Climate System Model version 4, J. Climate, 25, 2622-2651, doi: 10.1175/JCLI-D-11-1389

00301.1. 1390

Deser, C., R. Guo, and F. Lehner (2017), The relative contributions of tropical Pacific sea 1391

surface temperatures and atmospheric internal variability to the recent global warming 1392

hiatus, Geophys. Res. Lett., 44, doi: 10.1002/2017GL074273. 1393

Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor 1394

(2016), Overview of the Coupled Model Intercomparison Project phase 6 (CMIP6) 1395

experimental design and organization, Geosci. Model Dev., 9, 1937-1958, doi: 1396

10.5194/gmd-9-1937-2016. 1397


78

Eyring, V., L. Bock, A. Lauer, M. Righi, M. Schlund, B. Andela, E. Arnone, O. Bellprat, B. 1398

Brötz, L.-P. Caron, N. Carvalhais, I. Cionni, N. Cortesi, B. Crezee, E. Davin, P. Davini, 1399

K. Debeire, L. de Mora, C. Deser, D. Docquier, P. Earnshaw, C. Ehbrecht, B. K. Gier, N. 1400

Gonzalez-Reviriego, P. Goodman, S. Hagemann, S. Hardiman, B. Hassler, A. Hunter, C. 1401

Kadow, S. Kindermann, S. Koirala, N. V. Koldunov, Q. Lejeune, V. Lembo, T. Lovato, 1402

V. Lucarini, F. Massonnet, B. Müller, A. Pandde, N. Pérez-Zanón, A. Phillips, V. Predoi, 1403

J. Russell, A. Sellar, F. Serva, T. Stacke, R. Swaminathan, V. Torralba, J. Vegas-Regidor, 1404

J. von Hardenberg, K. Weigel, and K. Zimmermann (2020), ESMValTool v2.0 – 1405

Extended set of large-scale diagnostics for quasi-operational and comprehensive 1406

evaluation of Earth system models in CMIP, Geosci. Model Dev. Discuss., doi: 1407

10.5194/gmd-2019-291 (in review). 1408

Fisher, R. A., W. R. Wieder, B. M. Sanderson, C. D. Koven, K. W. Oleson, C. Xu, J. B. Fisher, 1409

M. Shi, A. P. Walker, and D. M. Lawrence (2019), Parametric controls on vegetation 1410

responses to biogeochemical forcing in the CLM5, J. Advances in Modeling Earth 1411

Systems, 11, doi: 10.1029/2019MS001609. 1412

Frajka-Williams, E., S. A. Cunningham, H. Bryden, and B. A. King (2011), Variability of 1413

Antarctic Bottom Water at 24.5°N in the Atlantic, J. Geophys. Res., 116, C11026, doi: 1414

10.1029/2011JC007168. 1415

Frajka-Williams, E., I. J. Ansorge, J. Baehr, H. L. Bryden, M. Paz Chidichimo, S. A. 1416

Cunningham, G. Danabasoglu, S. Dong, K. A. Donohue, S. Elipot, P. Heimbach, N. P. 1417

Holliday, R. Hummels, L. C. Jackson, J. Karstensen, M. Lankhorst, I. A. Le Bras, M. S. 1418

Lozier, E. L. McDonagh, C. S. Meinen, H. Mercier, B. I. Moat, R. C. Perez, C. G. 1419

Piecuch, M. Rhein, M. A. Srokosz, K. E. Trenberth, S. Bacon, G. Forget, G. Goni, D. 1420


79

Kieke, J. Koelling, T. Lamont, G. D. McCarthy, C. Mertens, U. Send, D. A. Smeed, S. 1421

Speich, M. van den Berg, D. Volkov, and C. Wilson (2019), Atlantic meridional 1422

overturning circulation: Observed transport and variability, Front. Mar. Sci., 6:260, doi: 1423

10.3389/fmars.2019.00260. 1424

Galbraith, E. D., and A. C. Martiny (2015), A simple nutrient-dependence mechanism for 1425

predicting the stoichiometry of marine ecosystems, Proceedings of the National Academy 1426

of Sciences of the United States of America, 112, 8199-8204, doi: 1427

10.1073/pnas.1423917112. 1428

Gent, P. R., G. Danabasoglu, L. J. Donner, M. M. Holland, E. C. Hunke, S. R. Jayne, D. M. 1429

Lawrence, R. B. Neale, P. J. Rasch, M. Vertenstein, P. H. Worley, Z.-L. Yang, and M. 1430

Zhang (2011), The Community Climate System Model version 4, J. Climate, 24, 4973-1431

4991, doi: 10.1175/2011JCLI4083.1. 1432

Gettelman, A., J. E. Kay, and K. M. Shell (2012), The evolution of climate sensitivity and 1433

climate feedbacks in the Community Atmosphere Model, J. Climate, 25, 1453-1469, doi: 1434

10.1175/JCLI-D-11-00197.1. 1435

Gettelman, A., and H. Morrison (2015), Advanced two-moment bulk microphysics for global 1436

models. Part I: Off-line tests and comparison with other schemes, J. Climate, 28, 1268-1437

1287, doi: 10.1175/JCLI-D-14-00102.1. 1438

Gettelman, A., M. J. Mills, D. E. Kinnison, R. R. Garcia, A. K. Smith, D. R. Marsh, S. Tilmes, F. 1439

Vitt, C. G. Bardeen, J. McInerny, H.-L. Liu, S. C. Solomon, L. M. Polvani, L. K. 1440

Emmons, J.-F. Lamarque, J. H. Richter, A. S. Glanville, J. T. Bacmeister, A. S. Philips, 1441

R. B. Neale, I. R. Simpson, A. K. DuVivier, A. Hodzic, and W. J. Randel (2019a), The 1442


80

Whole Atmosphere Community Climate Model version 6 (WACCM6), J. Geophys. Res. 1443

Atmos., 124, doi: 10.1029/2019JD030943. 1444

Gettelman, A., C. Hannay, and J. T. Bacmesiter, R. B. Neale, A. G. Pendergrass, G. 1445

Danabasoglu, J.-F. Lamarque, J. T. Fasullo, D. A. Bailey, and D. M. Lawrence (2019b), 1446

High climate sensitivity in the Community Earth System Model version 2 (CESM2), 1447

Geophys. Res. Lett., 46, 8329-8337, doi: 10.1029/2019GL083978. 1448

Ghimire, B., W. J. Riley, C. D. Koven, M. Q. Mu, and J. T. Randerson (2016), Representing leaf 1449

and root physiological traits in CLM improves global carbon and nitrogen cycling 1450

predictions, J. of Advances in Modeling Earth Systems, 8, 598-613, doi: 1451

10.1002/2015MS000538. 1452

Goldberg, D. N. (2011), A variationally derived, depth-integrated approximation to a higher-1453

order glaciological flow model, J. Glaciol., 57, 157-170, doi: 1454

10.3189/002214311795306763. 1455

Golaz, J.-C., V. E. Larson, and W. R. Cotton (2002), A PDF-based model for boundary layer 1456

clouds. Part I: Method and model description, J. Atmos. Sci., 59, 3540-3551. 1457

Gregory, J. M., W. J. Ingram, M. A. Palmer, G. S. Jones, P. A. Stott, R. B. Thorpe, J. A. Lowe, 1458

T. C. Johns, and K. D. Williams (2004), A new method for diagnosing radiative forcing 1459

and climate sensitivity, Geophys. Res. Lett., 31, L03205, doi: 10.1029/2003GL018747. 1460

Hemer, M. A., X. L. Wang, R. Weisse, and V. R. Swail (2012), Advancing wind-waves climate 1461

science: The COWCLIP project, Bull. Amer. Meteor. Soc., 93, 791-796. 1462

Hoesly, R. M., S. J. Smith, L. Feng, Z. Klimont, G. Janssens-Maenhout, T. Pitkanen, J. J. 1463

Seibert, L. Vu, R. J. Andres, R. M. Bolt, T. C. Bond, L. Dawidowski, N. Kholod, J.-I. 1464


81

Kurokawa, M. Li, L. Liu, Z. Lu, M. C. P. Moura, P. R. O’Rourke, and Q. Zhang (2018), 1465

Historical (1750-2014) anthropogenic emissions of reactive gases and aerosols from the 1466

Community Emissions Data System (CEDS), Geosci. Model Dev., 11, 369-408, doi: 1467

10.5194/gmd-11-369-2018. 1468

Hoose, C., J. E. Kristjánsson, J.-P. Chen, and A. Hazra (2010), A classical-theory-based 1469

parameterization of heterogeneous ice nucleation by mineral dust, soot, and biological 1470

particles in a global climate model, J. Atmos. Sci., 67, 2483-2503, doi: 1471

10.1175/2010JAS3425.1. 1472

Huang, B., P. W. Thorne, V. F. Banzon, T. Boyer, G. Chepurin, J. H. Lawrimore, M. J. Menne, 1473

T. M. Smith, R. S. Vose, and H.-M. Zhang (2017), Extended Reconstructed Sea Surface 1474

Temperature, version 5 (ERSSTv5): Upgrades, validations, and intercomparisons, J. 1475

Climate, 30, 8179-8205, doi: 10.1175/JCLI-D-16-0836.1. 1476

Huang, X., Q. Tang, Y. Tseng, Y. Hu, A. H. Baker, F. O. Bryan, J. Dennis, H. Fu, and G. Yang 1477

(2016), P-CSI v1.0, An accelerated barotropic solver for the high-resolution ocean model 1478

component in the Community Earth System Model v2.0, Geosci. Model Dev., 9, 4209-1479

4225, doi: 10.5194/gmd-9-4209-2016. 1480

Huffman, G. J., R. F. Adler, D. T Bolvin, and G. Gu (2009), Improving the global precipitation 1481

record: GPCP Version 2.1, Geophys. Res. Lett., 36, L17808, doi: 1482

10.1029/2009GL040000. 1483

Hunke, E. C., D. A. Hebert, and O. Lecomte (2013), Level-ice melt ponds in the Los Alamos sea 1484

ice model, CICE, Ocean Modell., 71, 26-42, doi: 10.1016/j.ocemod.2012.11.008. 1485


82

Hunke, E. C, W. H. Lipscomb, A. K. Turner, N. Jeffery, and S. Elliott (2015), CICE: The Los 1486

Alamos Sea Ice Model. Documentation and Software User’s Manual. Version 5.1. T-3 1487

Fluid Dynamics Group, Los Alamos National Laboratory, Tech. Rep. LA-CC-06-012. 1488

Hurrell, J. W., M. M. Holland, P. R. Gent, S. Ghan, J. E. Kay, P. J. Kushner, J.-F. Lamarque, W. 1489

G. Large, D. Lawrence, K. Lindsay, W. H. Lipscomb, M. C. Long, N. Mahowald, D. R. 1490

Marsh, R. B. Neale, P. Rasch, S. Vavrus, M. Vertenstein, D. Bader, W. D. Collins, J. J. 1491

Hack, J. Kiehl, and S. Marshall (2013), The Community Earth System Model: A 1492

framework for collaborative research, Bull. Amer. Meteor. Soc., 94, 1339-1360, doi: 1493

10.1175/BAMS-D-12-00121.1. 1494

Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins 1495

(2008), Radiative forcing by long-lived greenhouse gases: Calculations with the AER 1496

radiative transfer models, J. Geophys. Res., 113, D13103, doi: 10.1029/2008JD009944. 1497

Jones, P. D., D. H. Lister, T. J. Osborn, C. Harpham, M. Salmon, and C. P. Morice (2012), 1498

Hemispheric and large-scale land surface air temperature variations: An extensive 1499

revision and an update to 2010, J. Geophys. Res., 117, D05127, doi: 1500

10.1029/2011JD017139. 1501

Kaufmann, R. K., H. Kauppi, M. L. Mann, and J. H. Stock (2011), Reconciling anthropogenic 1502

climate change with observed temperature 1998-2008, Proc. Nat. Academy Sci., 108, 1503

11790-11793, doi: 10.1073/pnas.1102467108. 1504

Kay, J. E., C. Deser, A. Phillips, A. Mai, C. Hannay, G. Strand, J. Arblaster, S. Bates, G. 1505

Danabasoglu, J. Edwards, M. Holland, P. Kushner, J.-F. Lamarque, D. Lawrence, K. 1506

Lindsay, A. Middleton, E. Munoz, R. Neale, K. Oleson, L. Polvani, and M. Vertenstein 1507


83

(2015), The Community Earth System Model (CESM) Large Ensemble project: A 1508

community resource for studying climate change in the presence of internal climate 1509

variability, Bull. Amer. Meteor. Soc., 96, 1333-1349, doi:10.1175/BAMS-D-13-00255.1. 1510

Kennedy, D., S. Swenson, K. W. Oleson, D. M. Lawrence, R. Fisher, A. C. L. da Costa, and P. 1511

Gentine (2019), Implementing plant hydraulics in the Community Land Model, version 5, 1512

J. Advances in Modeling Earth Systems, 11, 485-513, doi: 10.1029/2018ms001500. 1513

Kiehl, J. T., and P. R. Gent (2004), The Community Climate System Model, version 2. J. 1514

Climate, 17, 3666–3682. 1515

Knutti, R., D. Masson, and A. Gettelman (2013), Climate model genealogy: Generation CMIP5 1516

and how we got there, Geophys. Res. Lett., 40, 1194-1199, doi: 10.1002/grl.50256. 1517

Kosaka, Y., and S.-P. Xie (2013), Recent global-warming hiatus tied to equatorial Pacific surface 1518

cooling, Nature, 501, 403-407, doi: 10.1038/nature12534. 1519

Koven, C. D., W. J. Riley, Z. M. Subin, J. Y. Tang, M. S. Torn, W. D. Collins, G. B. Bonan, D. 1520

M. Lawrence, and S. C. Swenson (2013), The effect of vertically resolved soil 1521

biogeochemistry and alternate soil C and N models on C dynamics of CLM4, 1522

Biogeosciences, 10, 7109-7131, doi: 10.5194/bg-10-7109-2013. 1523

Kummerow, C., J. Simpson, O. Thiele, W. Barnes, A. T. Chang, E. Stocker, R. F. Adler, A. Hou, 1524

R. Kakar, F. Wentz, P. Ashcroft, T. Kozu, Y. Hong, K. Okamoto, T. Iguchi, H. Kuroiwa, 1525

E. Im, Z. Haddad, G. Huffman, B. Ferrier, W. S. Olson, E. Zipser, E.A. Smith, T. T. 1526

Wilheit, G. North, T. Krishnamurti, and K. Nakamura (2000), The status of the Tropical 1527

Rainfall Measuring Mission (TRMM) after two years in orbit, J. Appl. Meteor., 39, 1965-1528

1982, doi: 10.1175/1520-0450(2001)040<1965:TSOTTR>2.0.CO;2. 1529


84

Large, W. G., J. C. McWilliams, and S. C. Doney (1994), Oceanic vertical mixing: A review and 1530

a model with a nonlocal boundary layer parameterization, Rev. Geophys., 32, 363-403. 1531

Larson, V. E., (2017), CLUBB-SILHS: A parameterization of subgrid variability in the 1532

atmosphere. arXiv:1711.03675v2 [physics.ao-ph]. 1533

Lawrence, D. M., R. A. Fisher, C. D. Koven, K. W. Oleson, S. C. Swenson, G. Bonan, N. 1534

Collier, B. Ghimire, L. van Kampenhout, D. Kennedy, E. Kluzek, P. J. Lawrence, F. Li, 1535

H. Li, D. Lombardozzi, W. J. Riley, W. J. Sacks, M. Shi, M. Vertenstein, W. R. Wieder,, 1536

C. Xu, A. A. Ali, A. M. Badger, G. Bisht, M. A. Brunke, S. P. Burns, J. Buzan, M. Clark, 1537

A. Craig, K. Dahlin, B. Drewniak, J. B. Fisher, M. Flanner, A. M. Fox, P. Gentine, F. 1538

Hoffman, G. Keppel-Aleks, R., Knox, S. Kumar, J. Lenaerts, L. R. Leung, W. H. 1539

Lipscomb, Y. Lu, A., Pandey, J. D. Pelletier, J. Perket,, J. T. Randerson, D. M. Ricciuto, 1540

B. M., Sanderson, A. Slater, Z. M. Subin, J. Tang, R. Q. Thomas, M. Val Martin, and X. 1541

Zeng (2019), The Community Land Model version 5: Description of new features, 1542

benchmarking, and impact of forcing uncertainty, J. Advances in Modeling Earth 1543

Systems, 11, doi: 10.1029/2018MS001583. 1544

Lehner, B., K. Verdin, and A. Jarvis (2008), New global hydrography derived from spaceborne 1545

elevation data, AGU Eos Trans., 89, 93-94. 1546

Le Quéré, C., R. M. Andrew, J. G. Canadell, S. Sitch, J. I. Korsbakken, G. P. Peters, A. C. 1547

Manning, T. A. Boden, P. P. Tans, R. A. Houghton, R. F. Keeling, S. Alin, O. 1548

D. Andrews, P. Anthoni, L. Barbero, L. Bopp, F. Chevallier, L. P. Chini, P. Ciais, K. 1549

Currie, C. Delire, S. C. Doney, P. Friedlingstein, T. Gkritzalis, I. Harris, J. Hauck, V. 1550

Haverd, M. Hoppema, K. Goldewijk, A. K. Jain, E. Kato, A. Körtzinger, P. 1551

Landschützer, N. Lefèvre, A. Lenton, S. Lienert, D. Lombardozzi, J. R. Melton, N. 1552


85

Metzl, F. Millero, P. M. S. Monteiro, D. R. Munro, J. E. M. S. Nabel, S. Nakaoka, K. 1553

O’Brien, A. Olsen, A. M. Omar, T. Ono, D. Pierrot, B. Poulter, C. Rödenbeck, J. 1554

Salisbury, U. Schuster, J. Schwinger, R. Séférian, I. Skjelvan, B. D. Stocker, A. J. Sutton, 1555

T. Takahashi, H. Tian, B. Tilbrook, I. T. van der Laan-Luijkx, G. R. van der Werf, N. 1556

Viovy, A. P. Walker, A. J. Wiltshire, and S. Zaehle (2016), Global Carbon Budget 2016, 1557

Earth Syst. Sci. Data, 8, 605-649, doi:10.5194/essd-8-605-2016. 1558

Lenssen, N., G. Schmidt, J. Hansen, M. Menne, A. Persin, R. Ruedy, and D. Zyss (2019), 1559

Improvements in GISTEMP uncertainty model, J. Geophys. Res. Atmos., 124, 6307-1560

6326, doi: 10.1029/2018JD029522. 1561

Letscher, R. T., J. K. Moore, Y. C. Teng, and F. Primeau (2015), Variable C:N:P stoichiometry 1562

of dissolved organic matter cycling in the Community Earth System Model, 1563


Levis, S., A. Badger, B. Drewniak, C. Nevison, and X. L. Ren (2018), CLMcrop yields and 1565

water requirements: Avoided impacts by choosing RCP 4.5 over 8.5, Climatic Change, 1566

146, 501-515 doi: 10.1007/s10584-016-1654-9. 1567

Levitus, S., T. Boyer, M. Concright, D. Johnson, T. O’Brien, J. Antonov, C. Stephens, and R. 1568

Garfield (1998), Introduction, Vol. I, World Ocean Database 1998, NOAA Atlas 1569

NESDIS 18, 346 pp. 1570

Li, F., S. Levis, S., and D. S. Ward (2013), Quantifying the role of fire in the Earth system - Part 1571

1: Improved global fire modeling in the Community Earth System Model (CESM1), 1572



86

Li, F., and D. M. Lawrence (2017), Role of fire in the global land water budget during the 1574

twentieth century due to changing ecosystems, J. Climate, 30, 1893-1908, doi: 1575

10.1175/JCLI-D-16-0460.1. 1576

Li, H. Y., M. S. Wigmosta, H. Wu, M. Y. Huang, Y. H. Ke, A. M. Coleman, and L. R. Leung 1577

(2013), A physically based runoff routing model for land surface and Earth system 1578

models, J. Hydrometeorology, 14, 808-828, doi: 10.1175/Jhm-D-12-015.1. 1579

Li, Q., A. Webb, B. Fox-Kemper, A. Craig, G. Danabasoglu, W. G. Large, and M. Vertenstein 1580

(2016), Langmuir mixing effects on global climate: WAVEWATCH III in CESM, Ocean 1581

Modell., 103, 145-160, doi: 10.1016/j.ocemod.2015.07.020. 1582

Li, Q., B. Fox-Kemper, O. Breivik, and A. Webb (2017), Statistical models of global Langmuir 1583

mixing, Ocean Modell., 113, 95-114. 1584

Li, Q., and B. Fox-Kemper (2017), Assessing the effects of Langmuir turbulence on the 1585

entrainment buoyancy flux in the ocean surface boundary layer, J. Phys. Oceanogr., 47, 1586

2863-2886. 1587

Lin, S. J., and R. B. Rood (1997), An explicit Flux-Form Semi-Lagrangian shallow water model 1588

on the sphere, Q. J. Royal Meteor. Soc., 123, 2477-2498. 1589

Lindsay, K. (2017), A Newton-Krylov solver for fast spin-up of online ocean tracers, Ocean 1590

Modell., 109, 33-43, doi: 10.1016/j.ocemod.2016.12.001. 1591

Lipscomb, W. H., J. G. Fyke, M. Vizcaíno, W. J. Sacks, J. Wolfe, M. Vertenstein, A. Craig, E. 1592

Kluzek, and D. M. Lawrence (2013), Implementation and initial evaluation of the 1593

Glimmer Community Ice Sheet Model in the Community Earth System Model, J. 1594



87

Lipscomb, W. H., S. F. Price, M. J. Hoffman, G. R. Leguy, A. R. Bennett, S. L. Bradley, K. J. 1596

Evans, J. G. Fyke, J. H. Kennedy, M. Perego, D. M. Ranken, W. J. Sacks, A. G. Salinger, 1597

L. Vargo, and P. H. Worley (2019), Description and evaluation of the Community Ice 1598

Sheet Model (CISM) v. 2.1, Geosci. Model Dev., 12, 387-424, doi: 10.5194/gmd-12-387-1599

2019. 1600

Liu, X., P. L. Ma, H. Wang, S. Tilmes, B. Singh, R. C. Easter, S. J. Ghan, and P. J. Rasch (2016), 1601

Description and evaluation of a new four-mode version of the Modal Aerosol Module 1602

(MAM4) within version 5.3 of the Community Atmosphere Model, Geosci. Model Dev., 1603

9, 505-522, doi: 10.5194/gmd-9-505-2016. 1604

Loeb, N. A., B. A. Wielicki, D. R. Doelling, G. L. Smith, D. F. Keyes, S. Kato, N. Manalo-1605

Smith, and T. Wong (2009), Toward optimal closure of the Earth's top-of-atmosphere 1606

radiation budget, J. Climate, 22, 748-766. 1607

Lombardozzi, D. L., G. B. Bonan, N. G. Smith, J. S. Dukes, and R. A. Fisher (2015), 1608

Temperature acclimation of photosynthesis and respiration: A key uncertainty in the 1609

carbon cycle‐climate feedback. Geophys. Res. Lett., 42, 8624-8631. 1610

Long, M. C., K. Lindsay, and M. M. Holland (2015), Modeling photosynthesis in sea ice covered 1611

waters, J. Advances in Modeling Earth Systems, 7, 1189-1206, doi: 1612

10.1002/2015MS000436. 1613

Malavelle, F. F., J. M. Haywood, A. Jones, A. Gettelman, L. Clarisse, S. Bauduin, R. P. Allan, I. 1614

H. H. Karset, J. E. Kristjansson, L. Oreopoulos, N. Cho, D. Lee, N. Bellouin, O. Boucher, 1615

D. P. Grosvenor, K. S. Carslaw, S. Dhomse, G. W. Mann, A. Schmidt, H. Coe, M. E. 1616

Hartley, M. Dalvi, A. A. Hill, B. T. Johnson, C. E. Johnson, J. R. Knight, F. M. 1617


88

O’Connor, D. G. Partridge, P. Stier, G. Myhre, S. Platnick, G. L. Stephens, H. Takahashi, 1618

and T. Thordarson (2017), Strong constraints on aerosol – cloud interactions from 1619

volcanic eruptions, Nature, 546, 485-491, doi: 10.1038/nature22974. 1620

Mayorga, E., S. P. Seitzinger, J. A., Harrison, E. Dumont, A. H. W. Beusen, A. F. Bouwman, B. 1621

M. Fekete, C. Kroeze, and G. van Drecht (2010), Global Nutrient Export from 1622

WaterSheds 2 (NEWS 2): Model development and implementation, Environmental 1623

Modelling and Software[R], 25, 837-853, doi: 10.1016/j.envsoft.2010.01.007. 1624

McWilliams, J. C., and B. Fox-Kemper (2013), Oceanic wave-balanced surface fronts and 1625

filaments, J. Fluid Mech., 730, 464-490, doi: 10.1017/jfm.2013.348. 1626

McWilliams, J. C., and P. Sullivan (2000), Vertical mixing by Langmuir circulations, Spill 1627

Science and Technology Bull., 6, 225-237. 1628

Meehl, G. A., W. M. Washington, J. M. Arblaster, A. Hu, H. Teng, J. E. Kay, A. Gettelman, D. 1629

M. Lawrence, B. M. Sanderson, and W. G. Strand (2013), Climate change projections in 1630

CESM1(CAM5) compared to CESM4, J. Climate, 26, 6287-6308, doi: 10.1175/JCLI-D-1631

12-00572.1. 1632

Meehl, G. A., H. Teng, and J. M. Arblaster (2014), Climate model simulations of the observed 1633

early-2000s hiatus of global warming, Nature Climate Change, 4, 898-902, doi: 1634

10.1038/NCLIMATE2357. 1635

Meinshausen, M., E. Vogel, A. Nauels, K. Lorbacher, N. Meinshausen, D. M. Etheridge, P. J. 1636

Fraser, S. A. Montzka, P. J. Rayner, C. M. Trudinger, P. B. Krummel, U. Beyerle, J. G. 1637

Canadell, J. S. Daniel, I. G. Enting, R. M. Law, C. R. Lunder, S. O'Doherty, R. G. Prinn, 1638

S. Reimann, M. Rubino, G. J. M. Velders, M. K. Vollmer, R. H. J. Wang, and R. Weiss 1639


89

(2017), Historical greenhouse concentrations for climate modelling (CMIP6), Geosci. 1640

Model Dev., 10, 2057-2116, doi: 10.5194/gmd-10-2057-2017. 1641

Mills, M. J., A. Schmidt, R. Easter, S. Solomon, D. E. Kinnison, S. J. Ghan, R. R. Neely, D. R. 1642

Marsh, A. Conley, C. G. Bardeen, and A. Gettelman (2016), Global volcanic aerosol 1643

properties derived from emissions, 1990–2014, using CESM1(WACCM), J. Geophys. 1644

Res. Atmos., 121, 2332-2348, doi: 10.1002/2015JD024290. 1645

Mills, M. J., J. H. Richter, S. Tilmes, B. Kravitz, D. G. MacMartin, A. A. Glanville, J. J. Tribbia, 1646

J.-F. Lamarque, F. Vitt, A. Schmidt, A. Gettelman, C. Hannay, J. T. Bacmeister, and D. 1647

E. Kinnison (2017), Radiative and chemical response to interactive stratospheric sulfate 1648

aerosols in fully coupled CESM1(WACCM), J. Geophys. Res. Atmos., 122, doi: 1649

10.1002/2017JD027006. 1650

Montzka, S. A., M. Aydin, M. Battle, J. H. Butler, E. S. Saltzman, B. D. Hall, A. D. Clarke, D. 1651

Mondeel, and J. W. Elkins (2004), A 350-year atmospheric history for carbonyl sulfide 1652

inferred from Antarctic firn air and air trapped in ice, J. Geophys. Res. Atmos., 109, 1653

D22302, doi: 10.1029/2004JD004686. 1654

Moore, J. K., S. C. Doney, J. A. Kleypas, D. M. Glover, and I. Y. Fung (2002), An intermediate 1655

complexity marine ecosystem model for the global domain, Deep Sea Res., 49, 403-462, 1656

doi: 10.1016/S0967-0645(01)00108-4. 1657

Moore, J. K., S. C. Doney, and K. Lindsay (2004), Upper ocean ecosystem dynamics and iron 1658

cycling in a global three-dimensional model, Global Biogeochemical Cycles, 18, 1659

GB4028, doi: 10.1029/2004GB002220. 1660


90

Moore, J. K., K. Lindsay, S. C. Doney, M. C. Long, and K. Misumi (2013), Marine Ecosystem 1661

Dynamics and Biogeochemical Cycling in the Community Earth System Model 1662

[CESM1(BGC)]: Comparison of the 1990s with the 2090s under the RCP4.5 and RCP8.5 1663

scenarios, J. Climate, 26, 9291-9312, doi: 10.1175/JCLI-D-12-00566.1. 1664

Morrison, H., and A. Gettelman (2008), A new two-moment bulk stratiform cloud microphysics 1665

scheme in the NCAR community atmosphere model (CAM3), Part I: Description and 1666

numerical tests, J. Climate, 21, 3642-3659. 1667

Neale, R. B., J. H. Richter, and M. Jochum (2008), The impact of convection on ENSO: From a 1668

delayed oscillator to a series of events, J. Climate, 21, 5904-5924. 1669

Neely III, R. R., and A. Schmidt (2016), VolcanEESM: Global volcanic sulphur dioxide (SO2) 1670

emissions database from 1850 to present - Version 1.0. Centre for Environmental Data 1671

Analysis, 04 February 2016, doi: 10.5285/76ebdc0b-0eed-4f70-b89e-55e606bcd568. 1672

Neely III, R. R., D. R. Marsh, K. L. Smith, S. M. Davis, and L. M. Polvani (2014), Biases in 1673

Southern Hemisphere climate trends induced by coarsely specifying the temporal 1674

resolution of stratospheric ozone, Geophys. Res. Lett., 41, 8602-8610, doi: 1675

10.1002/2014GL061627. 1676

Nicolet, M. (1970), Ozone and hydrogen reactions, Annales de Géophysique, 26, 531-546. 1677

Noël, B., W. J. van de Berg, J. M. van Wessem, E. van Meijgaard, D. van As, J. T. M. Lenaerts, 1678

S. Lhermitte, P. Kuipers Munneke, C. J. P. P. Smeets, L. H. van Ulft, R. S. W. van de 1679

Wal, and M. R. van den Broeke (2018), Modelling the climate and surface mass balance 1680

of polar ice sheets using RACMO2 – Part 1: Greenland (1958–2016), The Cryosphere, 1681

12, 811-831, doi: 10.5194/tc-12-811-2018. 1682


91

Oleson, K. W., and J. Feddema (2019), Parameterization and surface data improvements and 1683

new capabilities for the Community Land Model Urban (CLMU). J. Advances in 1684

Modeling Earth Systems, doi: 10.1029/2018MS001586. 1685

O'Neill, B. C., C. Tebaldi, D. P. van Vuuren, V. Eyring, P. Friedlingstein, G. Hurtt, R. Knutti, E. 1686

Kriegler, J.-F. Lamarque, J. Lowe, G. A. Meehl, R. Moss, K. Riahi, and B. M. Sanderson 1687

(2016), The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6, Geosci. 1688

Model Dev., 9, 3461-3482. 1689

Otto-Bliesner, B. L., E. C. Brady, and J. Fasullo (2016), Climate variability and change since 850 1690

CE: An ensemble approach with the Community Earth System Model, Bull. Amer. 1691

Meteor. Soc., 97, 735-754, doi: 10.1175/BAMS-D-14-00233.1. 1692

Paulot, F., D. J. Jacob, M. T. Johnson, T. G. Bell, A. R. Baker, W. C. Keene, I. D. Lima, S. C. 1693

Doney, and C. A. Stock (2015), Global oceanic emission of ammonia: Constraints from 1694

seawater and atmospheric observations: Ocean Ammonia Emissions, Global 1695

Biogeochemical Cycles, 29, 1165-1178, doi: 10.1002/2015GB005106. 1696

Poli, P., H. Hersbach, D. P. Dee, P. Berrisford, A. J. Simmons, F. Vitart, P. Laloyaux, D. G. H. 1697

Tan, C. Peubey, J.-N. Thépaut, Y. Trémolet, E. V. Hólm, M. Bonavita, L. Isaksen, and 1698

M. Fisher (2016), ERA-20C: An atmospheric reanalysis of the twentieth century, J. 1699


Reichl, B. G., T. Hara, and I. Ginis (2014), Sea state dependence of the wind stress over the 1701

ocean under hurricane winds, J. Geophys. Res. Oceans, 119, 30-51. 1702


92

Reichl, B. G., and Q. Li (2019), A parameterization with a constrained potential energy 1703

conversion rate of vertical mixing due to Langmuir turbulence, J. Phys. Oceanogr., doi: 1704

10.1175/JPO-D-18-0258.1. 1705

Richter, J. H., and P. J. Rasch (2008), Effects of convective momentum transport on the 1706

atmospheric circulation in the Community Atmosphere Model, version 3, J. Climate, 21, 1707

1487-1499. 1708

Rienecker, M. M., M. J. Suarez, R. Gelaro, R. Todling, J. Bacmeister, E. Liu, M. G. Bosilovich, 1709

S. D. Schubert, L. Takacs, G.-K. Kim, S. Bloom, J. Chen, D. Collins, A. Conalty, A. da 1710

Silva and Coauthors (2011), MERRA: NASA’s Modern-Era Retrospective Analysis for 1711

Research and Applications, J. Climate, 24, 3624-3648, doi: 10.1175/JCLI-D-11-00015.1. 1712

Roach, L. A., C. Horvat, S. M. Dean, and C. M. Bitz, (2018), An emergent sea ice floe size 1713

distribution in a global coupled ocean‐sea ice model, J. Geophys. Res. Oceans, 123, 1714

4322-4337. 1715

Robert, A. J. (1966), The integration of a low order spectral form of the primitive meteorological 1716

equations, J. Meteor. Soc. Japan, 44, 237-245. 1717

Rohde, R., R. Muller, R. Jacobsen, S. Perlmutter, A. Rosenfeld, J. Wurtele, J. Curry, C. 1718

Wickham, and S. Mosher (2013), Berkeley earth temperature averaging process, 1719

Geoinformatics & Geostatistics: An Overview, 1, doi: 10.4172/gigs.1000103. 1720

Schmidt, A., M. J. Mills, S. Ghan, J. M. Gregory, R. P. Allan, T. Andrews, C. G. Bardeen, A. 1721

Conley, P. M. Forster, A. Gettelman, R. W. Portmann, S. Solomon, and O. B. Toon 1722

(2018), Volcanic radiative forcing from 1979 to 2015, J Geophys Res-Atmos., 123, 1723

12491-12508, doi: 10.1029/2018JD028776. 1724


93

Schweiger, A., R. Lindsay, J. Zhang, M. Steele, and H. Stern (2011), Uncertainty in modeled 1725

Arctic sea ice volume, J. Geophys. Res., doi: 10.1029/2011JC007084. 1726

Scinocca, J., and N. Mcfarlane (2000), The parametrization of drag induced by stratified flow 1727

over anisotropic orography, Q. J. Roy. Meteor. Soc., 126, 2353-2394, doi: 1728

10.1256/smsqj.56801. 1729

Sheng, J.-X., D. K. Weisenstein, B.-P. Luo, E. Rozanov, A. Stenke, J. Anet, H. Bingemer, and T. 1730

Peter (2015), Global atmospheric sulfur budget under volcanically quiescent conditions: 1731

Aerosol-chemistry-climate model predictions and validation, J. Geophys. Res. Atmos., 1732

120, 256-276, doi: 10.1002/2014JD021985. 1733

Shi, M., J. B. Fisher, E. R. Brzostek, and R. P. Phillips (2016), Carbon cost of plant nitrogen 1734

acquisition: Global carbon cycle impact from an improved plant nitrogen cycle in the 1735

Community Land Model, Glob. Chang. Biol., 22, 1299-1314, doi: 10.1111/gcb.13131. 1736

Shi, X., X. Liu, and K. Zhang (2015), Effects of preexisting ice crystals on cirrus clouds and 1737

comparison between different ice nucleation parameterizations with the Community 1738

Atmosphere Model (CAM5), Atmospheric Chemistry and Physics, 15, 1503-1520, doi: 1739

10.5194/acp-15-1503-2015. 1740

Small, R. J., E. Curchitser, K. Hedstrom, B. Kauffman, and W. G. Large (2015), The Benguela 1741

upwelling system: Quantifying the sensitivity to resolution and coastal wind 1742

representation in a global climate model, J. Climate, 28, 9409-9432, doi: 10.1175/JCLI-1743

D-15-0192.1. 1744

Smeed, D. A., S. A. Josey, C. Beaulieu, W. E. Johns, B. I. Moat, E. Frajka-Williams, D. Rayner, 1745

C. S. Meinen, M. O. Baringer, H. L. Bryden, and G. D. McCarthy (2018), The North 1746


94

Atlantic Ocean is in a state of reduced overturning, Goephys. Res. Lett., 45, 1527-1533, 1747

doi: 10.1002/2017GL076350. 1748

Smith, R., P. Jones, B. Briegleb, F. Bryan, G. Danabasoglu, J. Dennis, J. Dukowicz, C. Eden, B. 1749

Fox-Kemper, P. Gent, M. Hecht, S. Jayne, M. Jochum, W. Large, K. Lindsay, M. 1750

Maltrud, N. Norton, S. Peacock, M. Vertenstein, and S. Yeager (2010), The Parallel 1751

Ocean Program (POP) reference manual, Ocean component of the Community Climate 1752

System Model (CCSM), LANL Tech. Report, LAUR-10-01853, 141 pp. 1753

Solomon, S., K. H. Rosenlof, R. W. Portmann, J. S. Daniel, S. M. Davis, T. J. Sanford, and G.-K. 1754

Plattner (2010), Contributions of stratospheric water vapor to decadal changes in the rate 1755

of global warming, Science, 327, 1219-1223. 1756

Steele, M., R. Morley, and W. Ermold (2001), PHC: A global ocean hydrography with a high 1757

quality Arctic Ocean, J. Climate, 14, 2079-2087. 1758

Stocker, T. F., D. Qin, G. K. Plattner, M. Tignor, S. K. Allen, J. Boschung, A. Nauels, Y. Xia, B. 1759

Bex, and B. M. Midgley (2013), IPCC, 2013: Climate change 2013: The physical science 1760

basis. Contribution of working group I to the fifth assessment report of the 1761

intergovernmental panel on climate change. 1762

Subin, Z. M., W. J. Riley, and D. Mironov (2012), An improved lake model for climate 1763

simulations: Model structure, evaluation, and sensitivity analyses in CESM1. J. of 1764

Advances in Modeling Earth Systems, 4, M02001, doi: 10.1029/2011ms000072. 1765

Sun, Q., M. M. Whitney, F. O. Bryan, and Y.-H. Tseng (2019), Assessing the skill of the 1766

improved treatment of riverine freshwater in the Community Earth System Model 1767


95

(CESM) relative to a new salinity climatology, J. Advances in Modeling Earth Systems, 1768

11, doi: 10.1029/2018MS001349. 1769

Suzuki, N., and B. Fox-Kemper (2016), Understanding Stokes forces in the wave-averaged 1770

equations, J. Geophys. Res. Oceans, 121, 1-18, doi: 10.1002/2015JC011566. 1771

Swenson, S. C., and D. Lawrence (2012), A new fractional snow‐covered area parameterization 1772

for the Community Land Model and its effect on the surface energy balance. J. Geophys. 1773

Res. Atmos., 117, D21107, doi:10.1029/2012JD018178. 1774

Swenson, S. C., D. M. Lawrence, and H. Lee (2012), Improved simulation of the terrestrial 1775

hydrological cycle in permafrost regions by the Community Land Model. J. of Advances 1776

in Modeling Earth Systems, 4, doi: 10.1029/2012ms000165. 1777

Swenson, S., and D. Lawrence (2014), Assessing a dry surface layer‐based soil resistance 1778

parameterization for the Community Land Model using GRACE and FLUXNET‐MTE 1779

data, J. Geophys. Res. Atmos., 119, 10299-10312, doi: 10.1002/2014JD022314. 1780

Swenson, S., and D. Lawrence (2015), A GRACE‐based assessment of interannual groundwater 1781

dynamics in the Community Land Model, Water Resources Research, 51, 8817-8833. 1782

Thiery, W., A. J. Visser, E. M. Fischer, M. Hauser, A. L. Hirsch, D. M. Lawrence, Q. Lejeune, 1783

E. L. Davin, and S. I. Seneviratne (2020), Warming of hot extremes alleviated by 1784

expanding irrigation, Nature Communications (in press). 1785

Tilmes, S., J. H. Richter, B. Kravitz, D. G. MacMartin, M. J. Mills, I. R. Simpson, A. S. 1786

Glanville, J. T. Fasullo, A. S. Phillips, J.-F. Lamarque, J. Tribbia, J. Edwards, S. 1787

Mickelson, and S. Gosh (2018), CESM1(WACCM) Stratospheric Aerosol 1788


96

Geoengineering Large Ensemble (GLENS) Project, Bull. Amer. Meteor. Soc., doi: 1789

10.1175/BAMS-D-17-0267.1. 1790

Tilmes, S., A. Hodzic, L. K. Emmons, M. J. Mills, A. Gettelman, D. E. Kinnison, M. Park, J.-F. 1791

Lamarque, F. Vitt, M. Shrivastava, P. Campuzano-Jost, J. Jiminez, and X. Liu (2019), 1792

Climate forcing and trends of organic aerosols in the Community Earth System Model 1793

(CESM2), J. Advances in Modeling Earth Systems, 11, doi: 10.1029/MS001827. 1794

Tolman, H. L. (2009), User manual and system documentation of WAVEWATCH III TM 1795

version 3.14. Technical note, MMAB Contribution, 276, p.220. 1796

Turner, A. K., and E. C. Hunke (2015), Impacts of a mushy-layer thermodynamic approach in 1797

global sea-ice simulations using the CICE sea ice model, J. Geophys. Res. Oceans, 120, 1798

1253-1275, doi: 10.1002/2014JC010358. 1799

van Kampenhout, L., J. T. M. Lenaerts, W. H. Lipscomb, W. J. Sacks, D. M. Lawrence, A. G. 1800

Slater, and M. R. van den Broeke (2017), Improving the representation of polar snow and 1801

firn in the Community Earth System Model, J. Advances in Modeling Earth Systems, 9, 1802

2583-2600, doi: 10.1002/2017/MS000988. 1803

van Kampenhout, L., A. M. Rhoades, A. R. Herrington, C. M. Zarzycki, J. T. M. Lenaerts, W. J. 1804

Sacks, and M. R. van den Broeke (2019), Regional grid refinement in an Earth system 1805

model: Impacts on the simulated Greenland surface mass balance, The Cryosphere, doi: 1806

10.5194/tc-13-1547-2019. 1807

van Marle, M. J. E., S. Kloster, B. I. Magi, J. R. Marlon, A.-L. Daniau, R. D. Field, A. Arneth, 1808

M. Forrest, S. Hantson, N. M. Kehrwald, W. Knorr, G. Lasslop, F. Li, S. Mangeon, C. 1809

Yue, J. W. Kaiser, and G. R. van der Werf (2017), Historic global biomass burning 1810


97

emissions for CMIP6 (BB4CMIP) based on merging satellite observations with proxies 1811

and fire models (1750–2015), Geosci. Model Dev., 10, 3329-3357, doi:10.5194/gmd-10-1812

3329-2017. 1813

van Wessem, J. M., W. J. van de Berg, B. P. Y. Noël, E. van Meijgaard, C. Amory, G. 1814

Birnbaum, C. L. Jakobs, K. Krüger, J. T. M. Lenaerts, S. Lhermitte, S. R. M. Ligtenberg, 1815

B. Medley, C. H. Reijmer, K. van Tricht, L. D. Trusel, L. H. van Ulft, B. Wouters, J. 1816

Wuite, and M. R. van den Broeke (2018), Modelling the climate and surface mass 1817

balance of polar ice sheets using RACMO2 – Part 2: Antarctica (1979–2016), The 1818

Cryosphere, 12, 1479-1498, doi: 10.5194/tc-12-1479-2018. 1819

Vizcaíno, M., W. H. Lipscomb, W. J. Sacks, J. H. van Angelen, B. Wouters, and M. R. van den 1820

Broeke (2013), Greenland surface mass balance as simulated by the Community Earth 1821

System Model. Part I: Model evaluation and 1850-2005 results, J. Climate, 26, 7793-1822

7812, doi: 10.1175/JCLI-D-12-00615.1. 1823

Vose, R. S., D. Arndt, V. F. Banzon, D. R. Easterling, B. Gleason, B. Huang, E. Kearns, J. H. 1824

Lawrimore. M. J. Menne, T. C. Peterson, R. W. Reynolds, T. M. Smith, C. N. Williams 1825

Jr., and D. B. Wuertz (2012), NOAA’s merged land – ocean surface temperature analysis, 1826

Bull. Amer. Meteor. Soc., 1677-1685, doi: 10.1175/BAMS-D-11-00241.1. 1827

Wang, Y., X. Liu, C. Hoose, and B. Wang (2014), Different contact angle distributions for 1828

heterogeneous ice nucleation in the Community Atmospheric Model version 1829

5, Atmospheric Chemistry and Physics, 14, 10411-10430, doi: 10.5194/acpd-14-10411-1830

2014. 1831


98

Washington, W. M., and G. A. Meehl (1989), Climate sensitivity due to increased CO2: 1832

Experiments with a coupled atmosphere and ocean general circulation model. Climate 1833

Dynamics, 4, 1-38. 1834

Wieder, W. R., D. M. Lawrence, R. A. Fisher, G. B. Bonan, S. J. Cheng, C. L. Goodale, A. S. 1835

Grandy, C. D. Koven, D. L. Lombardozzi, K.W. Oleson, and R. Q. Thomas (2019), 1836

Beyond static benchmarking: Using experimental manipulations to evaluate land model 1837

assumptions, Global Biogeochemical Cycles, doi: 10.1029/2018GB006141. 1838

Williams, P. D. (2011), The RAW filter: An improvement to the Robert – Asselin filter in semi-1839

implicit integrations, Mon. Wea. Rev., 139, 1996-2007. 1840

Worby, A. P., C. Geiger, M. J. Paget, M. van Woert, S. F Ackley, and T. DeLiberty (2008), 1841

Thickness distribution of Antarctic sea ice, J. Geophys. Res., 113, C05S92, doi: 1842

10.1029/2007JC004254. 1843

World Meteorological Organization (2010), Scientific Assessment of Ozone Depletion, WMO 1844

Report. Geneva: World Meteorological Organization. 1845

Yeager, S. G., G. Danabasoglu, N. A. Rosenbloom, W. Strand, S. C. Bates, G. A. Meehl, A. R. 1846

Karspeck, K. Lindsay, M. C. Long, H. Teng, and N. S. Lovenduski (2018), Predicting 1847

near-term changes in the Earth System: A large ensemble of initialized decadal prediction 1848

simulations using the Community Earth System Model, Bull. Amer. Meteor. Soc., 99, 1849

1867-1886, doi: 10.1175/BAMS-D-17-0098.1. 1850

Zhang, R., R. Sutton, G. Danabasoglu, Y.-O. Kwon, R. Marsh, S. G. Yeager, D. E. Amrhein, and 1851

C. M. Little (2019), A review of the role of the Atlantic meridional overturning 1852


99

circulation in Atlantic multidecadal variability and associated climate impacts, Rev. 1853

Geophys., 57, 316-375, doi: 10.1029/2019RG000644. 1854

Documents

The Community Earth System Model version 2 (CESM2)74 The Community Earth System Model version 2 (CESM2) is the latest generation of the coupled 75 climate / Earth system models developed