Weakening of the North American monsoon with global warming

Abstract

Future changes in the North American monsoon, a circulation system that brings abundant summer rains to vast areas of the North American Southwest, could have significant consequences for regional water resources. How this monsoon will change with increasing greenhouse gases, however, remains unclear, not least because coarse horizontal resolution and systematic sea-surface temperature biases limit the reliability of its numerical model simulations. Here we investigate the monsoon response to increased atmospheric carbon dioxide (CO2) concentrations using a 50-km-resolution global climate model which features a realistic representation of the monsoon climatology and its synoptic-scale variability. It is found that the monsoon response to CO2 doubling is sensitive to sea-surface temperature biases. When minimizing these biases, the model projects a robust reduction in monsoonal precipitation over the southwestern United States, contrasting with previous multi-model assessments. Most of this precipitation decline can be attributed to increased atmospheric stability, and hence weakened convection, caused by uniform sea-surface warming. These results suggest improved adaptation measures, particularly water resource planning, will be required to cope with projected reductions in monsoon rainfall in the American Southwest.

Full text

Accepted version1
This is the author’s version of the work. It is posted here for personal use, not for redistribution. The2
definitive version will be published in Nature Climate Change on Vol. 7, November issue.3
Weakening of the North American monsoon with global warming4
5
Salvatore Pascale1,2,∗William R. Boos3, Simona Bordoni4, Thomas L. Delworth2, Sarah B.6
Kapnick2, Hiroyuki Murakami1,2, Gabriel A. Vecchi5, Wei Zhang6
7
1Atmospheric and Oceanic Sciences Program, Princeton University, Princeton, NJ 08540, USA8
2Geophysical Fluid Dynamics Laboratory/NOAA, Princeton, NJ 08540, USA9
3Department of Earth and Planetary Science, University of California, Berkeley, and Climate and Ecosystem10
Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA11
4California Institute of Technology, Pasadena, CA 91125, USA12
5Department of Geosciences, Princeton University, Princeton, NJ 08544,USA13
6IIHR-Hydroscience and Engineering, The University of Iowa, Iowa City, IA 52242, USA14
∗Corresponding author’s address: Princeton University, and NOAA/Geophysical Fluid Dy-15
namics Laboratory, Princeton, New Jersey 08540, USA16
E-mail: Salvatore.Pascale@noaa.gov17
Previously at: California Institute of Technology, Pasadena, California 91125, USA18
1

Future changes in the North American monsoon, a circulation system that brings19
abundant summer rains to vast areas of the North American Southwest [1, 2], could20
have significant consequences for regional water resources [3]. How this monsoon21
will change with increasing greenhouse gases, however, remains unclear [4, 5, 6],22
not least because coarse horizontal resolution and systematic sea surface temper-23
ature biases limit the reliability of its numerical model simulations [5, 7]. Here we24
investigate the monsoon response to increased atmospheric carbon dioxide (CO2)25
concentrations using a 50 km-resolution global climate model which features a real-26
istic representation of the monsoon and its synoptic-scale variability [8]. It is found27
that the monsoon response to CO2doubling is sensitive to sea surface temperature28
biases. When minimising these biases, the model projects a robust reduction in mon-29
soonal precipitation over the southwestern United States, contrasting with previous30
multi-model assessments [4, 9]. Most of this precipitation decline can be attributed to31
increased atmospheric stability, and hence weakened convection, caused by uniform32
sea surface warming. These results suggest improved adaptation measures, partic-33
ularly water resource planning, will be required to cope with projected reductions in34
monsoon rainfall in the American Southwest.35
State-of-the-art general circulation models (GCMs) forced with greenhouse gas emission36
scenarios project a reduction of annual precipitation over a broad area of North America37
south of 35◦N [10]. While wintertime precipitation is robustly projected to decline in this38
region due to a poleward expansion of the subtropical dry zones [11], summertime precip-39
itation projections remain uncertain. This is due to a weak consensus across GCMs [10]40
and incomplete comprehension of the mechanisms through which global warming will im-41
pact the summertime North American monsoon (NAM). The NAM is shaped by both the42
complex regional geography (Supplementary Fig. 1) and remote larger-scale drivers [2, 12],43
which makes its simulation challenging [7, 13]. GCMs project a June-July reduction and44
2

a September-October increase in precipitation in the monsoon region [4, 9]. This early-to-45
late redistribution of rainfall has been conjectured to arise from two competing mechanisms46
[14]: a stronger tropospheric stability due to a remote sea surface temperature (SST) rise in47
spring that persists through early summer (a remote mechanism); and increased evapora-48
tion and near-surface moist static energy, driven by larger radiative fluxes at the surface (a49
local mechanism). The local mechanism is speculated to overcome the stabilizing effect of50
remote SST rise at the end of the summer [9]. However, the coarse horizontal resolution and51
existence of SST biases in coupled GCM simulations raise the question of how reliable such52
projections are for the NAM, which involves interactions across many spatial and temporal53
scales [12].54
Horizontal resolution is critical for adequately representing the NAM in models. It has55
been recently shown [8] that GCMs with horizontal grid spacing coarser than 100 km (as56
most models participating in the Coupled Model Intercomparison Project, Phase 3 and 5,57
CMIP3 and CMIP5) do not accurately resolve the summertime low-level flow along the Gulf58
of California (GoC), with detrimental impacts on simulated precipitation in parts of the south-59
western U.S. [1, 2]. For this reason, limited-area regional climate models have been used,60
suggesting drying of the monsoon region with warming [5]. Yet regional climate models lack61
two-way coupling with the larger-scale circulation and suffer from inherent boundary condi-62
tion biases [15], making them a questionable tool for studying the climate change response.63
GCM simulations of North American climate are affected by SST biases. In particu-64
lar, negative SST anomalies in the North Atlantic can substantially influence the North At-65
lantic subtropical high through the upstream influence of a Gill-type Rossby wave response66
[16, 17, 18]. This results in unrealistically strong easterly low-level moisture flux across the67
Caribbean region, causing the well-known monsoon retreat bias, i.e., excessive monsoonal68
precipitation in the fall [7, 13]. These biases are thus a substantial source of uncertainty for69
the projected NAM response to CO2forcing.70
3

To address these issues, here we investigate the response of the NAM to increased71
CO2and its sensitivity to both horizontal resolution and SST biases with the high resolu-72
tion (0.5◦×0.5◦in the land/atmosphere) Forecast-Oriented Low Ocean Resolution (FLOR)73
model [19, 20], developed at the National Oceanic and Atmospheric Administration (NOAA)74
Geophysical Fluid Dynamics Laboratory (GFDL). In addition to the standard configuration,75
the model can be run at coarser horizontal resolution (LOAR, 2◦×2◦in the land/atmosphere)76
or in a flux-adjusted version (FLOR-FA; see Methods).77
Compared to LOAR, increased horizontal resolution in FLOR allows for a better repre-78
sentation of the fall retreat at the end of the warm season (Fig. 1f) and a more realistic79
pattern of near-surface moist static energy (Supplementary Fig. 2). FLOR also better re-80
solves the seasonal cycle of low-level moisture flux along the GoC (Supplementary Fig. 3)81
and synoptic-scale variability within the monsoon [8]. These factors combine to create a82
more realistic simulation of the spatial pattern of mean rainfall (Fig. 1d) and the seasonal83
evolution of rainfall (Fig. 1f).84
To assess the impact of SST biases [7, 13], we contrast the free-running coupled FLOR85
with its flux-adjusted version, FLOR-FA. The flux adjustment adds a modification term to86
surface fluxes of enthalpy, momentum, and freshwater, reducing SST biases in the basic87
state (Supplementary Fig. 4b), and leading to a realistic GoC SST annual cycle (Supple-88
mentary Fig. 5). Globally, flux adjustment improves the simulations of tropical cyclones [20],89
trade winds, dry zones in the Pacific, and El Niño [21]. Specifically to the NAM, one impor-90
tant improvement is the more realistic representation of the monsoon retreat (Fig. 1f). Other91
regional improvements include better representation of the high near-surface moist static en-92
ergy along the GoC (Supplementary Fig. 2e), the GoC low-level jet (Supplementary Fig. 3),93
the Caribbean low-level jet, and the East Pacific Intertropical Convergence Zone. These94
results quantify that the separate impacts of both increased horizontal resolution and SST95
bias reduction enhance the simulation of the present-day NAM. The improvements seen96
4

in FLOR-FA suggest that this model is an excellent tool for investigations of the monsoon97
response to climate change.98
When atmospheric CO2concentration is doubled (2CO2_FLOR-FA vs. CTRL_FLOR-FA;99
Table 1), no statistically significant change is seen in mean June precipitation over the NAM100
region (Fig. 2a). A significant rainfall reduction is instead observed during July-August both101
in the core NAM region south of 28◦N and in its northern edge north of 28◦N (Supplemen-102
tary Fig. 6). Because of the large difference in mean summertime precipitation, this drying is103
substantial in percentage terms primarily in the northern edge of the monsoon (∼40%), be-104
coming increasingly smaller south of 28◦N (Fig. 2b). The drying persists – albeit weakened105
– over Arizona and northwestern Mexico during September-October, with no significant pre-106
cipitation changes seen along the monsoon coastal regions (Fig. 2c). Similar results are107
found in a second ensemble member, and in additional runs at 25 km atmospheric horizon-108
tal resolution (not shown). These trends are in line with observations, which suggest that109
precipitation has decreased in Arizona in recent decades [22].110
What determines the precipitation reduction over land during the mature monsoon sea-111
son? We answer this question by estimating changes in the vertical buoyancy [23]112
b=h10m−h∗(1)
induced by temperature and specific humidity changes. Here h10mis the near-surface moist113
static energy and h∗the saturation moist static energy (see Methods). Fig. 3 illustrates114
changes in buoyancy and cumulus convective mass flux under doubled CO2concentrations115
following a transect from the tropical eastern Pacific across the Sierra Madre Occidental116
into the southwestern U.S. (Fig. 1a). In June, convection is mostly unchanged over the117
western slopes of the Sierra Madre Occidental and south of 32◦N, consistent with modest,118
insignificant changes in vertical stability (Fig. 3a, d). In July-August, buoyancy decreases119
substantially between the lifted condensation level and the level of free convection over the120
most actively convecting regions on the Sierra Madre Occidental western slopes (Fig. 3b).121
5

Consistently, cumulus convective mass fluxes weaken substantially over the Sierra Madre122
Occidental western slopes (10-30%) and elevated terrain in Arizona (25-50%; Fig. 3e). In123
September-October, the region of negative buoyancy differences narrows and disappears124
almost everywhere except north of 30◦N. These patterns are consistent with those of con-125
vective mass flux changes (Fig. 3c,f).126
Importantly, when SST biases are not substantially reduced (i.e., 2CO2FLOR vs. CTRL_FLOR),127
the response to CO2doubling is different (Fig. 2d-f), with a drier (20-30% rainfall reduc-128
tion) June over both the southwestern U.S. and most of western Mexico (Supplementary129
Fig. 6), a substantially unaffected July-August (statistically insignificant differences), and a130
more pronounced tendency for larger rainfall rates along the coastal areas of western Mexico131
in September-October. This is consistent with the progressive increase from June to Octo-132
ber in evaporation anomalies (Supplementary Fig. 7a-f) and decrease in sensible heat flux133
anomalies (Supplementary Fig. 7g-l). The changes evident in FLOR without flux adjustment134
follow the consensus based on CMIP3 and CMIP5 model assessments [4, 14, 9], which in-135
vokes a late summer evaporation increase – and with it a near-surface moist static energy136
increase – that balances the larger radiative fluxes at the surface. This compensation results137
in the suppression or even reversal of the early summer rainfall reduction (local mechanism).138
This similarity between FLOR and most of the CMIP5 models may be due indeed to their139
similar SST biases [16].140
This picture is notably different in the southwestern U.S. and northwestern Mexico when141
SST biases are reduced (2CO2_FLOR-FA vs. CTRL_FLOR-FA): the strongest rainfall de-142
crease occurs in July-August (Fig. 2b) rather than in June. This more persistent drying in143
FLOR-FA reduces soil moisture availability and evaporation; hence, the local mechanism144
cannot reverse the drying, which persists until late summer. SST biases can thus substan-145
tially alter the intensity and effectiveness of the local mechanism [14, 9], leading to a change146
in the sign of the monsoon response to CO2forcing. One caveat is that the northernmost147
6

GoC is not resolved in FLOR [8]; this may artificially reduce precipitation in the Southwest148
U.S. [24] and weaken the impact of the local mechanism during the late summer season.149
The sensitivity of simulated rainfall changes to SST bias raises the question of how robust150
the projections shown in Fig. 2-3 are and what is the main driver of rainfall change. Although151
tropical precipitation changes produced by greenhouse gas warming are expected to be lo-152
cally correlated with SST changes [25], it has been argued that the precipitation response153
over land is insensitive to patterns of SST change [26]. To understand the cause of our sim-154
ulated precipitation changes, we use additional FLOR simulations in which SSTs are relaxed155
to a prescribed distribution (Table 1): (1) CLISST, where SSTs are relaxed to climatological156
1971-2012 observed values; (2) 2CO2, where CO2concentration is doubled and SSTs are157
relaxed to climatological values as in CLISST; (3) +2K, where SSTs are relaxed to climato-158
logical values augmented by a uniform 2 K anomaly; (4) 2CO2_+2K, which is a combination159
of +2K and 2CO2; and (5) 2CO2_pattern, where CO2concentration is doubled and SSTs160
are relaxed to climatological values augmented by a nonuniform anomaly pattern derived161
from the long-term 2CO2FLOR experiment, with global mean warming of +2.1 K. As shown162
in Fig. 4, the July-October NAM drying is in large part reproduced by 2CO2_pattern. Direct163
CO2forcing [27] causes a significant increase in June precipitation due to land and lower-164
troposphere warming [28], and compensates for the drying effect of SST rise. Although a165
uniform +2K warming generally increases convective inhibition over land and decreases pre-166
cipitation, the spatial structure of the SST rise (2CO2_pattern minus 2CO2_+2K) provides an167
important contribution to the total changes, as it leads to an additional and substantial reduc-168
tion of rainfall (Fig. 4b). This additional drying is explained by the impact of spatial variations169
in the SST rise, characterized by enhanced near-equatorial warming and off-equatorial rel-170
ative cooling in the eastern subtropical Pacific (Fig. 4c). As a consequence, subtropical171
subsidence intensifies as the sea surface warms more at the equator than in the subtropics.172
This response is in line with the “warmer-get-wetter” paradigm [25]; here we highlight the173
7

potential consequences of this response for the NAM region.174
The strong sensitivity of the NAM response to SST biases shows that these may be a175
large source of uncertainty for regional hydroclimate change [29]. Here we demonstrate176
that, when SST biases are reduced, a CO2increase causes a reduction of summertime177
precipitation in the NAM region, especially over northwestern Mexico and the southwestern178
U.S. (∼40%). These precipitation reductions are driven by the global mean SST rise, but,179
unlike what is seen in other tropical and subtropical land regions [26], they are substantially180
amplified by sea surface warming patterns. Interestingly, direct CO2radiative forcing [27, 28]181
has a negligible impact on the NAM, a circumstance that, along with the high interannual and182
interdecadal variability of NAM rainfall [2], may explain the difficulty to detect rainfall trends183
from historical observations [30].184
Although our results are based on a single climate model, this model is integrated in mul-185
tiple configurations and has a highly realistic representation of the monsoon compared to186
CMIP models. Our results highlight the possibility of a strong precipitation reduction in the187
northern edge of the monsoon in response to warming, with potential consequences for re-188
gional water resources, agriculture and ecosystems [3]. In addition to this mean precipitation189
response, changes in precipitation extremes [31] with warming will also have a significant190
impact in the monsoon region’s hydrology. We will explore them in future studies. Further191
study of the sensitivity to key parameterized processes such as cumulus convection and land192
surface physics will improve understanding of the monsoon response. Additional progress193
is within reach, as increasing horizontal resolution in state-of-the-art GCMs will soon allow194
new comparative and idealized studies in this critical region.195
References196
[1] Adams D. K. and A. C. Comrie. The North American monsoon. Bull. Amer. Meteor.197
Soc., 78:2197–2213, 1997.198
8

[2] Higgins R. W., Y. Yao, and X. L. Wang. Influence of the North American monsoon199
system on the U.S. summer precipitation regime. J. Climate, 10:298–306, 1997.200
[3] Ray A. J., G. M. Garfin, M. Wilder, M. Vasquez-León, M. Lenart, and A. C. Comrie.201
Applications of monsoon research: opportunities to inform decision making and reduce202
regional vulnerability. J. Climate, 20:1608–1627, 2007.203
[4] Cook B. I. and R. Seager. The response of the North American monsoon to increased204
greenhouse gas forcing. J. Geophys. Res., 118(4):1690–1699, 2013.205
[5] Bukovsky M. S., C. M. Carrillo, D. J. Gochis, D. M. Hammerling, R. R. McCrary, and L. O.206
Mearns. Toward assessing NARCCAP regional climate model credibility for the North207
American monsoon: Future climate simulations. J. Climate, 28:6707–6728, 2015.208
[6] Meyer J. D. D. and J. Jin. The response of future projections of the North American209
monsoon when combining dynamical downscaling and bias correction of CCSM4 out-210
put. Climate Dyn., 49:433–447, 2017.211
[7] Geil K. L., Y. L. Serra, and X. Zeng. Assessment of CMIP5 model simulations of the212
North American monsoon system. J. Climate, 26:8787–8801, 2013.213
[8] Pascale S., S. Bordoni, S. B. Kapnick, G. A. Vecchi, L. Jia, T. L. Delworth, S. Un-214
derwood, and W. Anderson. The Impact of horizontal resolution on North American215
monsoon Gulf of California moisture surges in a suite of coupled global climate models.216
J. Climate, 29:7911–7936, 2016.217
[9] Seth A., S. A. Rauscher, M. Rojas, A. Giannini, and S. J. Camargo. Enhanced spring218
convective barrier for monsoons in a warmer world? Clim. Chang., 104:403–414, 2011.219
[10] Maloney E. D., S. J. Camargo, E. Chang, B. Colle, R. Fu, K. L. Geil, Q. Hu, X. Jiang,220
N. Johnson, K. B. Karnauskas, J. Kinter, B. Kirtman, S. Kumar, B. Langenbrunner,221
K. Lombardo, L. N. Long, A. Mariotti, J. E. Meyerson, K. C. M, J. D. Neelin, Z. Pan,222
R. Seager, Y. Serra, A. Seth, J. Sheffield, J. Stroeve, J. Thibeault, S.-P. Xie, C. Wang,223
B. Wyman, and M. Zhao. North American climate in CMIP5 experiments: Part III:224
9

Assessment of twenty-first-century projections. J. Climate, 27:2230–2270, 2014.225
[11] Seager R. and G. A. Vecchi. Greenhouse warming and the 21st century hydroclimate226
of southwestern North America. Proc. Natl. Acad. Sci. (USA), 107:21277–21282, 2010.227
[12] Pascale S. and S. Bordoni. Tropical and extratropical controls of Gulf of California228
surges and summertime precipitation over the southwestern United States. Mon. Wea.229
Rev., 144:2695–2718, 2016.230
[13] Liang X.-Z., J. Zhu, K. E. Kunkel, M. Ting, and J. X. L. Wang. Do CGCMs simulate231
the North American monsoon precipitation seasonal-interannual variability? J. Climate,232
21:4424–4448, 2008.233
[14] Giannini A. Mechanisms of climate change in the semiarid African Sahel: The local234
view . J. Climate, 23:743–756, 2010.235
[15] Lorenz P. and D. Jacob. Influence of regional scale information on the global circulation:236
A two-way nesting climate simulation. Geophys. Res. Lett., 32:L18706, 2005.237
[16] Wang C., L. Zhang, S.-K. Lee, L. Wu, and C. R. Mechoso. A global perspective on238
CMIP5 climate model biases. Nature Climate Change, 4:201–205, 2014.239
[17] Sutton R. T. and D. L. R. Hodson. Climate response to basin-scale warming and cooling240
of the North Atlantic ocean. J. Climate, 20:891–907, 2006.241
[18] Kushnir Y., R. Seager, M. Ting, N. Naik, and J. Nakamura. Mechanisms of tropical242
Atlantic SST influence on North American precipitation variability. J. Climate, 23:5610–243
5628, 2010.244
[19] Delworth T. L., A. Rosati, W. Anderson, A. J. Adcroft, V. Balaji, R. Benson, K. Dixon,245
S. M. Griffies, H.-C. Lee, R. C. Pacanowski, G. A. Vecchi, A. T. Wittenberg, F. Zeng, and246
R. Zhang. Simulated climate and climate change in the GFDL CM2.5 high-resolution247
coupled climate model. J. Climate, 25:2755–2781, 2012.248
[20] Vecchi G. A., T. Delworth, R. Gudgel, S. Kapnick, A. Rosati, A. T. Wittenberg, F. Zeng,249
W. Andersona, V. Balaji, K. Dixon, L. Jia, H.-S. Kim, L. Krishnamurthy, R. Msadek,250
10

W. F. Stern, S. D. Underwood, G. Villarini, X. Yang, and S. Zhang. On the seasonal251
forecasting of regional tropical cyclone activity. J. Climate, 27:7994–8016, 2014.252
[21] Manganello J.V. and B. Huang. The influence of systematic errors in the Southeast253
Pacific on ENSO variability and prediction in a coupled GCM. Clim. Dyn., 32:1015–254
1034, 2009.255
[22] Luong T. M., C. L. Castro, H.-I Chang, T. Lahmers, D. K. Adams, and C. A. Ochoa-Moya.256
The more extreme nature of North American monsoon precipitation in the southwestern257
U.S. as revealed by a historical climatology of simulated severe weather events. J. Appl.258
Meteor. Climatol., pages doi.org/10.1175/JAMC–D–16–0358.1, in press, 2017.259
[23] Randall D. An introduction to the global circulation of the atmosphere. Princeton Uni-260
versity, Princeton, New Jersey, 456 pp, 2015.261
[24] Mitchell D. L., D. Ivanova, R. Rabin, T. J. Brown, and K. Redmond. Gulf of California262
sea surface temperatures and the North American monsoon: mechanistic implications263
from observations. J. Climate, 15:2261–2281, 2002.264
[25] Xie S.-P., C. Deser, G.A. Vecchi, J. Ma, H. Teng, and A.T. Wittenberg. Global warming265
pattern formation: Sea surface temperature and rainfall. J. Climate, 23:966–986, 2010.266
[26] Chadwick R. Which aspects of CO2forcing and SST warming cause most uncertainty267
in projections of tropical rainfall change over land and ocean? J. Climate, 29:2493–,268
2016.269
[27] Bony S., G. Bellon, D. Klocke, S. Sherwood, S. Fermepin, and S. Denvil. Robust di-270
rect effect of carbon dioxide on tropical circulation and regional precipitation. Nature271
Geoscience, 22:4213–4227, 2013.272
[28] Richardson T. B., P. M. Forster, T. Andrews, and D. J. Parker. Understanding the rapid273
precipitation response to CO2and aerosol forcing on a regional scale. J. Climate,274
29:583–594, 2016.275
[29] Zhou Z.-Q. and S.-P. Xie. Effects of climatological model biases on the projection of276
11

tropical climate change. J. Climate, 28:9909–9917, 2015.277
[30] Anderson B. T., J. Wang, G. Salvucci, S. Gopal, and S. Islam. Observed trends in278
summertime precipitation over the southwestern United States. J. Climate, 23:1937–279
1944, 2010.280
[31] Pfahl S., P. A. O’Gorman, and E. M. Fischer. Understanding the regional pattern of281
projected future changes in extreme precipitation. Nature Climate Change, 7:423–427,282
2017.283
Methods284
Experiments. We use the NOAA GFDL coupled Forecast-Oriented Low Ocean Resolution285
(FLOR) model [20], derived from the GFDL Coupled Model version 2.5 (CM2.5) [19]. CM2.5286
features a 0.5◦×0.5◦atmospheric horizontal resolution with 32 vertical levels and has been287
successfully used for studies of regional hydroclimate change [1, 2]. FLOR is identical to288
CM2.5 but features a coarser ocean horizontal resolution (1◦×1◦versus 0.25◦×0.25◦). The289
land model component is the Land Model, version 3 [3], with a horizontal resolution equal290
to that of the atmospheric model. The sea ice model is the Sea Ice Simulator, version 1,291
as in [19]. A second model called LOAR (Low Ocean Atmosphere Resolution) is also used292
to test the impact of atmospheric horizontal resolution. The LOAR model has a horizontal293
atmospheric resolution of 2◦×2◦and is otherwise identical to FLOR [4].294
As in most of CMIP5 models [16], FLOR features positive (negative) SST bias in the295
eastern (western) North Pacific and a negative SST bias in the North Atlantic (Supplemen-296
tary Fig. 4). SST biases have a negative impact on simulations of the NAM in present-day297
climate [13] and are a source of uncertainty for projected changes in the tropics [29]. To re-298
duce them, we use a flux-adjusted version of FLOR. In this configuration, which is otherwise299
identical to the standard FLOR configuration, fluxes of momentum, enthalpy and freshwater300
12

are “adjusted” to bring the model’s climatology of SST, as well as surface wind stress and301
salinity, closer to observational estimates. We refer to this configuration as FLOR-FA. De-302
tails about the flux adjustment procedure can be found in [20]. FLOR-FA features reduced303
SST biases as compared to FLOR, especially in the Pacific and Atlantic oceans (Fig. S4).304
In both FLOR and FLOR-FA, long-term control simulations are performed with atmospheric305
CO2concentration held fixed at 1990 values. In the 2CO2experiments, we increase CO2
306
concentration at 1% per year starting from 1990 levels. After it has doubled (after approxi-307
mately seventy years), we hold it constant and let the model run for additional two hundred308
years. In this experiment, the flux adjustment correction term remains the same as in the309
control run. As for freely-coupled models (i.e., developing systematic SST biases), the un-310
derlying assumption for applying the same adjustment correction under CO2forcing is that311
the emergent error in the SST climatology is the same in present and future climates.312
Nudged-SST simulations. Mechanisms of NAM changes in response to CO2doubling are313
investigated with additional nudged-SST numerical simulations. In these simulations, sim-314
ulated SSTs are restored toward a given field SST0while allowing high-frequency (i.e., on315
timescales smaller than the restoration timescale) SST fluctuations and ocean-atmosphere316
interactions. This is obtained by adding a restoration term (SST0−SST)/τ to the SST317
tendency equation:318
d SS T/dt = (d SST /dt)C+ (SST0−SST)/τ (2)
where τ= 10 days is the restoration timescale and (d SST /dt)Cthe SST tendency as com-319
puted in the coupled model. Specifically, we perform five nudged-SST simulations in which:320
(1) SST0is the observed 1971-2012 climatological monthly-varying mean and CO2concen-321
trations are held constant at 1990 values (CLISST); (2) SST0is the observed climatolog-322
ical monthly-varying SST mean and CO2concentration is doubled relative to 1990 values323
(2CO2); (3) SST0is the observed climatological monthly-varying SST increased globally by324
2K and CO2concentration is kept at 1990 values (+2K); (4) SST0is the observed climatolog-325
13

ical monthly-varying SST increased globally by 2K and CO2concentration is doubled relative326
to 1990 values (2CO2_+2K); (5) SST0is the observed climatological monthly-varying SST327
plus a nonuniform SST anomaly taken from the long-term 2CO2FLOR climatology and CO2
328
is doubled relative to 1990 values (2CO2_pattern). Further details about these nudged-SST329
simulations and their purpose can be found in Table 1.330
Observations. To validate the FLOR and FLOR-FA simulations, we use several obser-331
vational datasets. For precipitation, we use the Global Precipitation Climatology Centre332
(GPCC) dataset [5]. GPCC is based on statistically interpolated in situ rain measurements333
and cover all land areas at monthly temporal resolution for the period 1901−2010. GPCC334
monthly precipitation data were obtained at 0.5◦×0.5◦horizontal resolution from the NOAA335
Physical Science Division Climate and Weather data website (www.esrl.noaa.gov/psd/data/).336
We use the Modern Era Retrospective-analysis for Research and Applications (MERRA) [6]337
for monthly and daily precipitation, near-surface moisture and winds. MERRA is a reanalysis338
with improved representation of the atmospheric branch of the hydrological cycle developed339
by NASA’s Global Modeling and Assimilation Office (NASA Earth Observing System Data340
and Information System website: https://earthdata.nasa.gov/). Finally, the observed SST0
341
field from the Met Office Hadley Centre Sea Ice and SST dataset [7] is used for the nudged-342
SST runs (Eq. 2) and to evaluate FLOR SST biases (Supplementary Fig. 4).343
Buoyancy and convection diagnostics. The buoyancy of a saturated ascending air par-344
cel, as measured by the difference between its temperature Tcand the temperature of the345
environment T, is proportional to the difference between the saturation moist static energy346
of the environment and the moist static energy of the ascending cloudy air [23]:347
cp(Tc−T) = hc−h∗
1 + γ,(3)
where h=cpT+g z +L q is the moist static energy, h∗the saturation moist static energy, hc
348
the moist static energy of the ascending parcel, qis the specific humidity, gis the gravitational349
acceleration, cp= 1004 J K−1kg−1is the isobaric specific heat of dry air, L= 2.5×106J kg−1
350
14

latent heat of condensation, q∗(T, p)the saturation specific humidity that we calculate using351
the August-Roche-Magnus formula [8] and γ= (L/cp)(∂q∗/∂ T )p. Since the ascending parcel352
is lifted adiabatically from near surface, and thus lifted conserves its moist static energy, hc
353
is well approximated by the near-surface moist static energy, i.e. hcp
≈h10m=cpT10m+354
g z10m+L q10m, here computed at the model’s reference height z10m=10 m. The parameter355
γis positive and of order 1 [23], thus h10m−h∗is approximately twice the buoyancy value.356
To detect changes in the atmospheric convective instability, we estimate the buoyancy index357
b=h10m−h∗at each horizontal grid point xand vertical level pabove the lifted condensation358
level, and then the buoyancy index anomaly ∆bas:359
∆b= ∆(h10m−h∗),(4)
where the difference ∆is taken between the perturbed and the control simulation and posi-360
tive (negative) values of bindicating upward (downward) acceleration.361
Changes in the intensity of convection are assessed through changes in the diagnosed362
cumulus convective mass flux from the relaxed-Arakawa-Schubert scheme [9] employed in363
the GFDL models.364
Statistical significance. We estimate statistical significance for differences shown in Fig. 2-365
3 and in Supplementary Fig. 7 using a two-sided Student’s t-test at the 95% significance366
level. Confidence intervals for the mean differences shown in Fig. 4 are determined through367
applying 104bootstrap resampling, as we randomly reshuffle the two time series (forced and368
control run) 10,000 times and the construct a probability distribution for the mean difference.369
Data availability The data that support the findings of this study are available from the370
corresponding author upon request.371
15

References372
[1] Kapnick S. B., T. L. Delworth, M. Ashfaq, S. Malyshev, and P. C. D. Milly. Snowfall less373
sensitive to warming in Karakoram than in Himalayas due to a unique seasonal cycle.374
Nature Geoscience, 7:834 – 840, 2014.375
[2] Delworth T. L. and F. Zeng. Regional rainfall decline in Australia attributed to anthro-376
pogenic greenhouse gases and ozone levels. Nature Geoscience, 7:583–587, 2014.377
[3] Milly P. C. D., S. L. Malyshev, E. Shevliakova, K. A. Dunne, K. L. Findell, T. Gleeson,378
Z. Liang, P. Phillipps, R. J. Stouffer, and S. Swenson. An enhanced model of land water379
and energy for global hydrologic and earth-system studies. J. Hydrometeor., 15:1739–380
1761, 2014.381
[4] van der Wiel K., S. B. Kapnick, G. A. Vecchi, W. F. Cooke, T. L. Delworth, L. Jia, H. Mu-382
rakami, S. Underwood, and F. Zeng. The resolution dependence of contiguous U.S.383
precipitation extremes in response to CO2forcing. J. Climate, 29:7991–8012, 2016.384
[5] Schneider U., E. Becker, P. Finger, A. Meyer-Christoffer, M. Ziese, and B. Rudolf.385
GPCC’s new land surface precipitation climatology based on quality-controlled in situ386
data and its role in quantifying the global water cycle. Theoretical and Applied Clima-387
tology, 115(1-2):15–40, 2013.388
[6] Rienecker M. M., M. J. Suarez, R. Gelaro, R. Todling, J. Bacmeister, E. Liu, M. G.389
Bosilovich, S. D. Schubert, L. Takacs, G.-K. Kim, S. Bloom, J. Chen, D. Collins,390
A. Conaty, A. da Silva, W. Gu, J. Joiner, R. D. Koster, R. Lucchesi, A. Molod, T. Owens,391
S. Pawson, P. Pegion, C. R. Redder, R. Reichle, F. R. Robertson, A. G. Ruddick,392
M. Sienkiewicz, and J. Woollen. MERRA: NASA’s Modern-Era Retrospective analy-393
sis for Research and Applications. J. Climate, 24:3624–3648, 2011.394
[7] Rayner N., D. E. Parker, E. Horton, C. Folland, L. Alexander, D. Rowell, E. Kent, and395
A. Kaplan. Global analyses of sea surface temperature, sea ice, and night marine air396
temperature since the late nineteenth century. J. Geophys. Res., 108:4407, 2003.397
16

[8] World Meteorological Organization. Technical Regulations. Volume 1, WMO-No. 49,398
Geneva, 1988.399
[9] Moorthi S. and M J Suarez. Relaxed Arakawa-Schubert: A parameterization of moist400
convection for general circulation models . Mon. Wea. Rev., 120:978–1002, 1992.401
17

Acknowledgements402
S. P. was supported by the NOAA Climate and Global Change Postdoctoral Fellowship Pro-403
gram, administered by the University Corporation for Atmospheric Research, Boulder, Col-404
orado and by the NOAA CICS grant - NA14OAR4320106. S.B. acknowledges support from405
the Caltech Davidow Discovery Fund. The authors thank Nathaniel Johnson and Honghai406
Zhang for comments on the manuscript.407
Author contributions408
S.P. designed the research and performed the analysis of the data. S. P. lead the writing409
with the assistance of S.B., S.B.K. and W.R.B.. S.P., W.R.B., S.B. and T.L.D. contributed to410
define the methods and to interpret the results. All authors took part in the discussion of the411
results and refined and improved the manuscript. H. M. and G. A. V. designed the model412
experiments. H. M. and W. Z. performed the simulations.413
Competing financial interests414
The authors declare no competing financial interests.415
18

1 Figures416
19

110W 100W 90W 80W 80W
10N
20N
30N
40N
Observed precipitation--GPCC
120W120W 120W
120W120W
110W 100W 90W
10N
20N
30N
40N
MERRA
80g/kg*ms-1
80
80g/kg*ms-1
110W 100W 90W 80W
10N
20N
30N
40N
LOAR
110W 100W 90W 80W
10N
20N
30N
40N
FLOR
110W 100W 90W 80W
10N
20N
30N
40N
FLOR-FA
0.5 1 1.5 2 2.5 3 4 6 8 10 12 15
Precipitation (mm/day)
NAM
domain
a b c
d e f
2 4 6 8 10 12
0
50
100
150
Rainfall seasonality
monthly precipitation (mm)
GPCC
LOAR
FLOR-FA
FLOR
Months
Figure 1: High-resolution flux-adjusted models better capture regional features of the North
American monsoon. a, Time-mean (July-August) observed precipitation from GPCC (1971-2010).
The blue contour delimits the area used for averaging over the North American monsoon in fand the
magenta line the transect used for vertical cross-sections in Fig. 3. Precipitation (shading) and 10m-
moisture flux (vectors) in b, MERRA reanalysis (1979-2010); c, LOAR, d; FLOR and e, FLOR-FA
control runs (see Table 1 for description of experiments). f, Seasonal cycle of monthly precipitation
averaged over the North American monsoon domain in observations and models. Shading denotes
the interannual variability spread in observations.
20

110W 100W 90W 80W
10N
20N
30N
40N
June
1
1
4
4
4
4
12
12
110W 100W 90W 80W
10N
20N
30N
40N
July-August
1
1
4
4
4
12
12
110W 100W 90W 80W
10N
20N
30N
40N
September-October
1
1
4
4
12
12
110W 100W 90W 80W
10N
20N
30N
40N
June
1
4
4
4
4
12
110W 100W 90W 80W
10N
20N
30N
40N
July-August
1
1
1
4
4
4
4
12
12
12
110W 100W 90W 80W
10N
20N
30N
40N
September-October
1
4
4
12
12
-50 -40 -30 -25 -20 -15 -10 -2 2 10 15 20 25 30 40 50
Precipitation Change (%)
a b c
d e f
1
4
1
Figure 2: Impact of increased CO2concentration and SST biases on the North American mon-
soon precipitation. Percent precipitation change induced by CO2doubling in FLOR-FA simula-
tions (%, color shading; 2CO2_FLOR-FA minus CTRL_FLOR-FA) in aJune, b, July-August, and c,
September-October. d-f, As in a-cbut for FLOR simulations (2CO2_FLOR minus CTRL_FLOR).
Grey contours denote climatological values of precipitation (mm/day) in the respective control runs.
Stippling indicates regions where precipitation differences are statistically significant at the 5 % level
on the basis of a t-test.
21

0
5 10 15 20 25 30 35 40
1000
800
600
400
200
Pressure (hPa)
0
10
5 10 15 20 25 30 35 40
1000
800
600
400
200
Pressure (hPa)
5 10 15 20 25 30 35 40
Latitude
1000
800
600
400
200
Pressure (hPa)
1
1
3
5
7
9
5 10 15 20 25 30 35 40
3
5
7
9
11
5 10 15 20 25 30 35 40
1
1
3
7
5 10 15 20 25 30 35 40
Latitude
-5 -4 -3 -2 -1 -0.5 -0.1 0 0.1 0.5 1 2 3 4 5
kJ kg 10 kg m s
SMO
SMO
SMO
SMO
SMO
SMO
LFC
LFC
LFC
LCL
LCL
LCL
-30
-20
-15
-10
-5
0
5
-15
-20
-5
5
-5
5
-10
-5
0
-20
-15
-5
5
0
-10
-5
-40
-30
-20
-15
-5
10
-10
1
3
1
3
3
3
5
9
7
1
3
5
9
11
13
3 -2 -1
-1
a d
b e
c f
Figure 3: CO2-induced warming strengthens convective inhibition and weakens convection
over land. Difference in a, June, b, July-August and c, September-October mean buoyancy between
doubled CO2and control FLOR-FA simulations (color shading; see Methods for details on buoyancy
calculations). Stippling denotes statistical significance, black lines denote climatological values of
buoyancy, LFC the level of free convection (zero buoyancy), and LCL the lifted condensation level.
Buoyancy values below the LCL are not shown because the relationship between buoyancy and moist
static energy does not hold for an unsaturated parcel. d-f, As in a-cbut for the cumulus convective
mass flux. The vertical transect is at 108◦W (pink line in Fig. 1a) and intersects the Sierra Madre
Occidental (SMO) at approximately 28◦N. The blue line encircles areas over land where there is a
significant buoyancy negative anomaly. 22

-40
-30
-20
-10
0
10
June July/August September/October
precipitation change (mm)
+2K
2CO2
2CO2_pattern
2CO2_+2K
2CO2 FLOR-FA coupled (ens1)
2CO2 FLOR-FA coupled (ens2)
a
120W 100W 80W 60W 40W
0
10N
20N
30N
-50
-40
-30
-25
-20
-15
-10
-2
2
10
15
20
25
30
50
40
1
1
1
1
1
4
4
4
4
4
4
1
2
1
6
60W140W
b
%
120W 100W 80W 60W 40W
0
10N
20N
30N
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
294
296
298
298
300
300
300
302
302
302
302
140W
K
c
Figure 4: Attribution of projected North American monsoon precipitation changes. a, North
American monsoon area-averaged (defined in Fig. 1) precipitation change attributed to each experi-
ment (Table 1): 2CO2(red), +2K (green), 2CO2_+2K (blue), 2CO2_pattern (brown) and the coupled
2CO2_FLOR-FA simulations (yellow for the ensemble member 1, orange for the ensemble member
2). Error bars denote the 95% confidence interval. b, Percent July precipitation change induced by
patterns of SST anomalies (2CO2_pattern minus 2CO2_+2K). Yellow contours denote the 2CO2_+2K
climatology (mm/day). c, Areas of SST cooling and warming in the 2CO2_pattern run relative to the
2CO2_+2K run (uniform +2 K rise). Pink contours denote the 2CO2_+2K climatology (K). In both b
and c, stippling indicates regions where precipitation differences are statistically significant at the 5 %
level on the basis of a t-test. 23

List of Figures417
1High-resolution flux-adjusted models better capture regional features of418
the North American monsoon. a, Time-mean (July-August) observed pre-419
cipitation from GPCC (1971-2010). The blue contour delimits the area used420
for averaging over the North American monsoon in fand the magenta line the421
transect used for vertical cross-sections in Fig. 3. Precipitation (shading) and422
10m-moisture flux (vectors) in b, MERRA reanalysis (1979-2010); c, LOAR,423
d; FLOR and e, FLOR-FA control runs (see Table 1 for description of exper-424
iments). f, Seasonal cycle of monthly precipitation averaged over the North425
American monsoon domain in observations and models. Shading denotes426
the interannual variability spread in observations. . . . . . . . . . . . . . . . 20427
2Impact of increased CO2concentration and SST biases on the North428
American monsoon precipitation. Percent precipitation change induced by429
CO2doubling in FLOR-FA simulations (%, color shading; 2CO2_FLOR-FA mi-430
nus CTRL_FLOR-FA) in aJune, b, July-August, and c, September-October.431
d-f, As in a-cbut for FLOR simulations (2CO2_FLOR minus CTRL_FLOR).432
Grey contours denote climatological values of precipitation (mm/day) in the433
respective control runs. Stippling indicates regions where precipitation differ-434
ences are statistically significant at the 5 % level on the basis of a t-test. . . . 21435
24

3CO2-induced warming strengthens convective inhibition and weakens436
convection over land. Difference in a, June, b, July-August and c, September-437
October mean buoyancy between doubled CO2and control FLOR-FA simula-438
tions (color shading; see Methods for details on buoyancy calculations). Stip-439
pling denotes statistical significance, black lines denote climatological values440
of buoyancy, LFC the level of free convection (zero buoyancy), and LCL the441
lifted condensation level. Buoyancy values below the LCL are not shown be-442
cause the relationship between buoyancy and moist static energy does not443
hold for an unsaturated parcel. d-f, As in a-cbut for the cumulus convective444
mass flux. The vertical transect is at 108◦W (pink line in Fig. 1a) and inter-445
sects the Sierra Madre Occidental (SMO) at approximately 28◦N. The blue446
line encircles areas over land where there is a significant buoyancy negative447
anomaly. ...................................... 22448
4Attribution of projected North American monsoon precipitation changes.449
a, North American monsoon area-averaged (defined in Fig. 1) precipitation450
change attributed to each experiment (Table 1): 2CO2(red), +2K (green),451
2CO2_+2K (blue), 2CO2_pattern (brown) and the coupled 2CO2_FLOR-FA452
simulations (yellow for the ensemble member 1, orange for the ensemble453
member 2). Error bars denote the 95% confidence interval. b, Percent July454
precipitation change induced by patterns of SST anomalies (2CO2_pattern455
minus 2CO2_+2K). Yellow contours denote the 2CO2_+2K climatology (mm/day).456
c, Areas of SST cooling and warming in the 2CO2_pattern run relative to the457
2CO2_+2K run (uniform +2 K rise). Pink contours denote the 2CO2_+2K cli-458
matology (K). In both band c, stippling indicates regions where precipitation459
differences are statistically significant at the 5 % level on the basis of a t-test. 23460
25

Experiment yrs Radiative forcing/boundary conditions Purpose
a) CTRL_FLOR 200 CO2constant at 1990 levels Control run
b) CTRL_FLOR-FA 200 CO2constant at 1990 levels Control run; Reduce SST biases
c) 2CO2_FLOR 200 CO2doubles in 70 yrs, then constant CO2forcing
d) 2CO2_FLOR-FA 200 CO2doubles in 70 yrs, then constant CO2forcing; Reduce SST biases
1) CLISST 50 Model SST restored to observed climatological (1971-2012) values Remove SST biases
2) 2CO250 Model SST restored as in CLISST; atmospheric CO2concentration is Impact of 2CO2only
doubled relative to 1990 levels
3) +2K 50 Model SST restored to observed climatological SST plus 2K (no warming Impact of mean SST increase only
pattern); CO2concentration is held at 1990 values
4) 2CO2_+2K 50 Model SST restored to observed climatological SST plus 2K (no warming Combined impact of mean
pattern); CO2is doubled relative to 1990 levels SST increase and 2CO2
5) 2CO2_pattern 50 Model SST restored to observed climatological SST plus warming pattern Combined impact of nonuniform
from a long coupled 2CO2run; CO2is doubled relative to 1990 levels SST anomaly and 2CO2
Table 1: Description of the coupled (a-d) and nudged-SST (1-5) experiments used in this study (see Methods for further details). Two ensemble members
are available for experiments CTRL_FLOR, CTRL_FLOR-FA, 2CO2_FLOR and 2CO2_FLOR-FA.
26